Journal of Applied Ecology volume 48 tahun 2011

Journal of Applied Ecology, 48, 1–2 doi: 10.1111/j.1365-2664.2010.01938.x EDITORIAL Practitioner’s perspectives: intr...

Author: Edited by: Jane Memmott | EJ Milner-Gulland | Phil Hulme | Marc Cadotte and Mark Whittingham with Gill Kerby

9 downloads 531 Views 16MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Journal of Applied Ecology, 48, 1–2

doi: 10.1111/j.1365-2664.2010.01938.x

EDITORIAL

Practitioner’s perspectives: introducing a different voice in applied ecology Philip E. Hulme* The Bio-Protection Research Centre, Lincoln University, PO Box 84, Canterbury, New Zealand

Most researchers working in applied ecology are aware that much of what is published in leading ecological and environmental science journals makes little difference to the day-today management of species and ecosystems (Nature 2007). In recent years, several editorials have made this point and attempted to identify a way forward to bridge ‘The Great Divide’ (Born, Boreux & Lawes 2009; Milner-Gulland et al. 2010; Memmott et al. 2010). The onus has largely been on the scientific community to communicate the value of its science more clearly, become increasingly involved in extension activities and, heaven forbid, step down from their ivory towers and get their hands dirty. Yet the reality is that many scientists are doing this already, several very successfully (Possingham 2009). A more serious concern is that academic journals are simply not the best medium to communicate practical messages to a wide audience who need specific solutions to particular problems that have to be delivered on a tight budget. We should not be surprised; much academic research aims to be innovative, internationally competitive and globally relevant – aims which are not always congruent with finding practical solutions. It is against these criteria that many editorial decisions regarding whether or not to accept a paper for publication are made. Irrespective of how much hand-wringing might take place among editors this situation is unlikely to change as publishers judge the viability of journals using bibliometric indices and numbers of institutional subscriptions. Yet, potentially there is another way. Communication is a two-way street, even if much of the academic traffic is heading in one direction with no clear destination. So how do scientific researchers hear about the concerns and needs of those tackling problems in the field? Surveys of stakeholders are certainly one way to confirm that they feel the scientific community is not listening to them (e.g. Andreu, Vila` & Hulme 2009) but questionnaires do not address the problem. An alternative is to provide an opportunity within the pages of academic journals for non-standard pieces to be written by individuals who have a different perspective on what is needed in applied ecology research and whether the papers published in academic journals get anywhere near it. With this aim, the Journal of Applied Ecology launches its first ‘Practitioner’s Perspective’. These ‘prick our conscience’ pieces can be contributed by anyone who has a strong opinion on the current state of applied ecology research, whether academic or not, as long as they can provide an original perspective and a constructive

way forward. Although practitioners have been identified as a distinct group of actors in applied ecology that ‘buy land, put up fences, set fires, put out fires, lobby politicians, negotiate with farmers, spray invasive weeds, poison rats and guard against poachers’ (Nature 2007), we are not placing restrictions on who is or is not a ‘practitioner’. Thus, we welcome pieces from academics (at least those with a bit of dirt under their fingernails) as well as civil servants, environmental consultants, park managers and environmental lobbyists. The truth is we are unsure what to expect in terms of submissions under this new feature, hopefully provocative pieces from writers whose voices are rarely heard in our journal. To kick-start this initiative we have commissioned a few articles that might give a flavour of the pieces we would like to see published in the future. Our greatest challenge to date has been to prevent these pieces from becoming advertorials for the activities of NGOs, conservation groups or consultancies. This is certainly not what we want, but we do welcome examples of best practice that may not have made it into the wider academic literature. The first Practitioner’s Perspective appears in this issue (Goulson et al. 2011) and illustrates the viewpoint of the Bumblebee Conservation Trust, although the lead author is a senior academic at Stirling University, UK. Hopefully, in addition to highlighting how science informs the conservation of bumblebees it will challenge readers to consider what more needs to be done. We encourage future submissions under Practitioner’s Perspectives but please be sure to contact the Editors to discuss your piece beforehand. There is no prescribed structure to Practitioner’s Perspectives apart from our hope that they will be thought-provoking and challenge the science community to consider the perspectives of those individuals addressing applied ecological issues. However, authors may wish to consider covering the activities of the individual or organization with regard to ecological management, the key issues they are addressing (see Sutherland et al. 2006, 2009 for a range of key questions), the extent to which applied ecological research has supported their activities (if at all), how future research might assist them to address ecological problems more effectively and how this might best be achieved (e.g. through greater dialogue, joint projects, new research techniques etc.).

References Andreu, J., Vila`, M. & Hulme, P.E. (2009) An assessment of stakeholder perceptions and management of alien plants in Spain. Environmental Management, 43, 1244–1255.

*Correspondence author. E-mail: [email protected] Ó 2011 The Author. Journal of Applied Ecology Ó 2011 British Ecological Society

2 Editorial Born, J., Boreux, V. & Lawes, M.J. (2009) Synthesis: sharing ecological knowledge – the way forward. Biotropica, 41, 586–588. Goulson, D., Rayner, P., Dawson, R. & Darvill, B. (2011) Translating research into action; bumblebee conservation as a case study. Journal of Applied Ecology, 48, 3–8. Memmott, J., Cadotte, M., Hulme, P.E., Kerby, G., Milner-Gulland, E.J. & Whittingham, M.J. (2010) Editorial: putting applied ecology into practice. Journal of Applied Ecology, 47, 1–4. Milner-Gulland, E.J., Fisher, M., Browne, S., Redford, K.H., Spencer, M. & Sutherland, W.J. (2010) Do we need to develop a more relevant conservation literature? Oryx, 44, 1–2. Nature (2007) The great divide. Nature, 450, 135–136. Possingham, H. (2009) Dealing with ‘The great divide’. Decision Point, 28, 2. Sutherland, W.J., Armstrong-Brown, S., Armsworth, P. R., Brereton, T., Brickland, J., Campbell, C.D., Chamberlain, D. E., Cooke, A.I., Dulvy, N.K., Dusic, N.R., Fitton, M., Freckleton, R.P., Godfray, H.C., Grout, N., Harvey, H.J., Hedley, C., Hopkins, J.J., Kift, N.B., Kirby, J., Kunin, W.E.,

MacDonald, D.W., Markee, B., Naura, M., Neale, A.R., Oliver, T., Osborn, D., Pullin, A.S., Shardlow, M.E.A., Showler, D.A., Smith, P.L., Smithers, R.J., Solandt, J.-L., Spencer, J., Spray, C.J., Thomas, C.D., Thompson, J., Webb, S.E., Yalden, D.W. & Watkinson, A.R. (2006) The identification of 100 ecological questions of high policy relevance in the UK. Journal of Applied Ecology, 43, 617–627. Sutherland, W.J., Adams, W.M., Aronson, R.B., Aveling, R., Blackburn, T.M., Broad, S., Ceballos, G., Coˆte´, I.M., Cowling, R.M., da Fonseca, G.A.B., Dinerstein, E., Ferraro, P.J., Fleishman, E., Gascon, C., Hunter Jr, M., Hutton, J., Kareiva, P., Kuria, A., Macdonald, D.W., MacKinnon, K., Madgwick, F.J., Mascia, M.B., McNeely, J., Milner-Gulland, E.J., Moon, S., Morley, C.G., Nelson, S., Osborn, D., Pai, M., Parsons, E.C.M., Peck, L.S., Possingham, H., Prior, S.V., Pullin, A.S., Rands, M.R.W., Ranganathan, J., Redford, K.H., Rodriguez, J.P., Seymour, F., Sobel, F., Sodhi, N.S., Stott, A., Vance-Borland, K. & Watkinson, A.R. (2009) One hundred questions of importance to the conservation of global biological diversity. Conservation Biology, 23, 557–567.

Ó 2011 The Author. Journal of Applied Ecology Ó 2011 British Ecological Society, Journal of Applied Ecology, 48, 1–2

Journal of Applied Ecology 2011, 48, 163–173

doi: 10.1111/j.1365-2664.2010.01890.x

Assessing spatial patterns of disease risk to biodiversity: implications for the management of the amphibian pathogen, Batrachochytrium dendrobatidis Kris A. Murray1*, Richard W. R. Retallick2, Robert Puschendorf3, Lee F. Skerratt4, Dan Rosauer5,6, Hamish I. McCallum7, Lee Berger4, Rick Speare4 and Jeremy VanDerWal3 1

The Ecology Centre, School of Biological Sciences, University of Queensland, Brisbane, Queensland 4072, Australia; GHD Pty Ltd, 8 ⁄ 180 Lonsdale Street, Melbourne, Victoria 3000, Australia; 3Centre for Tropical Biodiversity and Climate Change Research, School of Marine and Tropical Biology, James Cook University, Townsville, Queensland 4811, Australia; 4Amphibian Disease Ecology Group, School of Public Health, Tropical Medicine and Rehabilitation Sciences, James Cook University, Townsville, Queensland 4811, Australia; 5School of Biological, Earth and Environmental Sciences, University of New South Wales, Sydney, New South Wales 2052, Australia; 6Centre for Plant Biodiversity Research, GPO Box 1600, Canberra, Australian Capital Territory 2601, Australia; and 7School of Environment, Griffith University, Nathan Campus, Queensland 4111, Australia 2

Summary 1. Emerging infectious diseases can have serious consequences for wildlife populations, ecosystem structure and biodiversity. Predicting the spatial patterns and potential impacts of diseases in freeranging wildlife are therefore important for planning, prioritizing and implementing research and management actions. 2. We developed spatial models of environmental suitability (ES) for infection with the pathogen Batrachochytrium dendrobatidis, which causes the most significant disease affecting vertebrate biodiversity on record, amphibian chytridiomycosis. We applied relatively newly developed methods for modelling ES (Maxent) to the first comprehensive, continent-wide data base (comprising >10000 observations) on the occurrence of infection with this pathogen and employed novel methodologies to deal with common but rarely addressed sources of model uncertainty. 3. We used ES to (i) predict the minimum potential geographic distribution of infection with B. dendrobatidis in Australia and (ii) test the hypothesis that ES for B. dendrobatidis should help explain patterns of amphibian decline given its theoretical and empirical link with organism abundance (intensity of infection), a known determinant of disease severity. 4. We show that (i) infection with B. dendrobatidis has probably reached its broad geographic limits in Australia under current climatic conditions but that smaller areas of invasion potential remain, (ii) areas of high predicted ES for B. dendrobatidis accurately reflect areas where population declines due to severe chytridiomycosis have occurred and (iii) that a host-specific metric of ES for B. dendrobatidis (ES for Bdspecies) is the strongest predictor of decline in Australian amphibians at a continental scale yet discovered. 5. Synthesis and applications. Our results provide quantitative information that helps to explain both the spatial distribution and potential effects (risk) of amphibian infection with B. dendrobatidis at the population level. Given scarce conservation resources, our results can be used immediately in Australia and our methods applied elsewhere to prioritize species, regions and actions in the struggle to limit further biodiversity loss. Key-words: amphibian declines, bioclimatic modelling, chytrid fungus, chytridiomycosis, infectious disease, Maxent, species distribution model

*Correspondence author. E-mail: [email protected] 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

164 K. A. Murray et al.

Introduction Emerging infectious diseases can have serious consequences for wildlife populations, ecosystem structure and biodiversity (Crowl et al. 2008). Alarmingly, their incidence appears to be rising as a result of anthropogenic influences that favour the growth, dispersal and transmission of pathogens (Daszak, Cunningham & Hyatt 2000; Jones et al. 2008). Assessing the extent, effects and dynamics of diseases in host populations are therefore important for predicting disease emergence and its consequences, and to plan, prioritize and implement research and management actions. Arguably the most serious wildlife disease impacting vertebrate biodiversity at this time is chytridiomycosis. This disease, caused by infection with the fungal pathogen Batrachochytrium dendrobatidis Longcore, Pessier & Nichols (1999), has been implicated in many rapid and recent amphibian declines and extinctions (Stuart et al. 2004; Skerratt et al. 2007; Bielby et al. 2008; Wake & Vredenburg 2008). Batrachochytrium dendrobatidis (hereafter Bd) appears to have undergone recent global expansion after the outbreak of a single clonal lineage, the origin of which remains uncertain (Morehouse et al. 2003; Rachowicz et al. 2005; but see Goka et al. 2009; James et al. 2009). As an international notifiable disease, reporting of Bd detection to the World Organisation for Animal Health (OIE) is now obligatory for member countries (World Organisation for Animal Health 2008). Bd is currently known from hundreds of amphibian species and from all continents where amphibians occur (Speare & Berger 2000; Kusrini et al. 2008; Olson & Ronnenberg 2008). Given its broad host range, many more species are likely to be suitable hosts and this number will rise as search effort and reporting increases. The potential distribution of Bd is, however, still relatively poorly understood; the native range has not been delineated and Bd may still be expanding its range worldwide (Lips et al. 2008; Rohr et al. 2008; James et al. 2009). In Australia, it is now widely accepted that the invasion and spread of Bd is the probable cause of many frog declines (Skerratt et al. 2007). Despite this, little quantitative data on risk of disease have been available to researchers and managers at broad spatial scales, hampering efforts to pinpoint areas and species warranting immediate management attention. Tools for predicting the spread or establishment of Bd and for identifying areas of high disease risk are therefore critical for policy makers, researchers and managers charged with detecting this pathogen, developing management actions and prioritizing resource expenditure (Gascon et al. 2007; Skerratt et al. 2008). Predicting the dispersal and potential range of organisms is commonly approached by characterizing environmental suitability (ES) with correlative species distribution models (SDMs) (Guisan & Thuiller 2005; Kearney & Porter 2009). ‘Presence-only’ SDMs are being used increasingly for their application to species occurrence data sets for which no reliable absence records may be available (e.g. museum ⁄ herbarium collections, atlases, non-targeted surveys etc.) (Pearce & Boyce 2006). Rarely used in studies of infectious disease, presence-only SDMs appear well suited to investigating the

distribution of infection with some pathogens because, analogous to verifying true absence of rare or endangered species (Gibson, Barrett & Burbidge 2007), it is a statistical and sampling challenge to assert ‘freedom from disease’ (Digiacomo & Koepsell 1986; Ziller et al. 2002; Skerratt et al. 2008). This challenge is rarely met for wildlife pathogens because the cost of sufficient sampling (including diagnostics, personnel, logistics, etc.) at broad spatial scales is typically prohibitive. Furthermore, pathogen prevalence may be low in a host population or may fluctuate temporally, and the host itself may be difficult to detect, particularly if the pathogen has resulted in host declines as has been the case with Bd (e.g. Lips et al. 2006). Correlative SDMs will only be appropriate where the distribution of infection with a pathogen is expected to be regulated by spatially quantifiable predictors that capture ES, such as climate or habitat type. For many pathogens, this may be inappropriate if hosts provide a highly regulated ‘habitat’ in which to grow and no stage of the life cycle is exposed to external environmental conditions (e.g. for internal, directly transmitted pathogens of endotherms). In the case of Bd, however, infections occur on ectothermic amphibian hosts and there is a direct effect of the environment (particularly temperature and moisture) on growth and survival of both free-living and parasitic life stages (Johnson & Speare 2003; Berger et al. 2004; Piotrowski, Annis & Longcore 2004; Woodhams et al. 2008). SDMs should thus be highly suited to characterizing ES for infection with Bd to provide important insights into its potential distribution, shed light on the probability of pathogen establishment following invasion into previously naı¨ ve areas [as has been demonstrated for other invasive species (Ficetola, Thuiller & Miaud 2007), and to help improve detection probability while reducing cost and effort of surveying for the pathogen in the future (Guisan et al. 2006)]. In an adaptive management context, such models are ideally suited to tailoring future data collection, which can in turn be used to iteratively improve the model (Wintle, Elith & Potts 2005). We hypothesized that modelling ES for infection with Bd may also provide useful information about the risk to amphibian populations posed by chytridiomycosis. Recently, VanDerWal et al. (2009b) demonstrated that modelling ES broadly predicts an organism’s abundance. For chytridiomycosis, Bd abundance (infection intensity) on the host is a direct determinant of disease development, severity and population effects (Carey et al. 2006; Voyles et al. 2007; Briggs, Knapp & Vredenburg 2010). Indeed, seasonal and elevational variation in the prevalence, intensity and virulence of Bd infections has long implicated climatic suitability as a major factor governing its effects in the wild (Berger 2001; Berger et al. 2004; Woodhams & Alford 2005; Kriger & Hero 2007), and this has been consistently supported by laboratory infection experiments (Woodhams, Alford & Marantelli 2003; Berger et al. 2004; Carey et al. 2006). We would thus expect that our ES results not only reflect proliferation of Bd on the host at the time scale used in model training (average annual) but also the risk of severe chytridiomycosis to populations as a whole, a link we

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 163–173

Spatial patterns of disease risk 165 test herein by examining patterns of disease-induced amphibian population declines. A published SDM already exists for Bd (Ron 2005), in which a correlative, presence-only SDM (GARP) with relatively few (n = 44) presence records in the New World was used to predict the global potential range of this pathogen. While this was of great use at the time of publication, the model appears to exaggerate suitable area in Australia (Fig. S1, Supporting Information), is at a spatial scale too coarse to be useful for regional management or for predicting population declines, and is likely to suffer from several sources of uncertainty inherent to correlative SDMs (e.g. extrapolation beyond the training region, limited sample size, algorithm nuances, inappropriate pseudo-absence selection; see e.g. Araujo & Guisan 2006; Pearson et al. 2007; Peterson, Papes & Eaton 2007; VanDerWal et al. 2009a). These issues together necessitate the development of independent, regionally specific predictions for planning future research and management actions on Bd at finer spatial scales. To this end, we applied a relatively novel SDM method (Maxent) (Phillips, Anderson & Schapire 2006) to the most comprehensive, continent-wide data base available to date on the occurrence of infection with Bd (Murray et al. 2010) to model ES for this pathogen. We employed novel methodologies to deal with common but rarely addressed sources of SDM uncertainty to provide maximum robustness in our predictions of ES in Australia given the available data. The predictions were used to estimate the minimum potential geographic distribution of infection with Bd in Australia and to test the hypothesis linking ES to disease risk as indicated by patterns of disease-induced population declines. We used our results to identify where chytridiomycosis may pose the greatest risk to endangered species, allowing prioritization of species, regions and actions when considering research and management options given scarce conservation funds (Wilson et al. 2007).

Materials and methods MODEL DESCRIPTION

The SDM software used was Maxent (ver. 3.3.0), for which the underlying theory and assumptions have been described in detail elsewhere

(Phillips, Anderson & Schapire 2006; Dudik, Phillips & Schapire 2007). Briefly, Maxent has been shown to generally outperform other correlative (both presence-only and presence-absence) SDM algorithms (Elith et al. 2006; Peterson, Papes & Eaton 2007; Graham et al. 2008; Wisz et al. 2008). It requires presence records only (but uses random background points to sample available environmental space), accounts for interactions among variables and identifies areas that fall beyond the range of environmental conditions used during training when making projections (identified as ‘clamped’ areas). The output of Maxent corresponds with an index of ES for the organism, where higher values correspond to a prediction of better conditions (Phillips, Anderson & Schapire 2006). We used Bd occurrence records from Murray et al. (2010). Full details of these data and their collection methods are described in the Metadata provided therein. Briefly, this newly compiled data set represents the first comprehensive, continent-wide data base describing occurrence patterns of Bd on wild amphibian hosts. The data base comprises 821 sites in Australia at which frogs or tadpoles have been tested for Bd and includes 10 183 records from >80 contributors spanning collection dates from 1956 to 2007. Bd was detected on 63 (55%) of the 115 species in the data set (c. 28% of Australia’s 223 species) (Table S1, Supporting Information). Two hundred and eighty-four Bd-positive sites had sufficient geographic accuracy for inclusion in the model (Table 1, Fig. S2, Supporting Information). Few localities in the data base comprise statistically defensible absence records given the difficulty of asserting freedom from chytridiomycosis. The data base represents records of clinical and aclinical infection with Bd, which by definition is considered synonymous with chytridiomycosis (ranging from severe and clinical to benign and aclinical) by disease authorities (sensu Berger et al. 1998 and as per the ‘Definitions’ of the OIE’s Aquatic Animal Health Code; see http://www.oie.int/eng/normes/fcode/en_chapitre_1.1.1.htm) but distinct from including records of the free-living stage which may also be detected off-host (Kirshtein et al. 2007; Walker et al. 2007). Bd’s current occurrence pattern in Australia is highly consistent with the hypothesis that environmental characteristics, such as climate or habitat type, place direct limits on its distribution. Its extensive distribution nation-wide (Fig. S2; Murray et al. 2010) demonstrates that it has had sufficient opportunity to spread great distances and into new geographic areas from its hypothesized point(s) of introduction (major ports) (Murray et al. 2010). The large number of known hosts and the spectrum of potentially susceptible amphibian hosts nationally (e.g. Litoria spp.) in currently uninfected regions strongly suggest that Bd is not limited in Australia by the unavailability of susceptible amphibian host species. Similarly, its presence in some remote and sparsely populated regions of the

Table 1. Summary of Batrachochytrium dendrobatidis (Bd) data base records. Geo-referenced Bd+ sites are those where the pathogen was detected and an accurate geographic coordinate was obtained for input to the distribution model. Individuals tested is a minimum estimate; many site records in the database did not include total number of individuals tested (see Fig. S2 for map and key to State names)

State

Database records

Individuals tested

Sites with records

Bd+ sites

Georeferenced Bd+ sites

ACT NSW NT QLD SA TAS VIC WA Australia

77 494 14 6660 42 146 26 2647 10 106

77 887 14 8789 42 574 32 2446 12 861

7 79 2 359 16 122 11 225 821

1 39 0 165 8 45 6 76 340

1 39 0 165 8 43 6 22 284


166 K. A. Murray et al. country and its absence in some populated regions suggest that it is not simply dependent on humans for its establishment and persistence, although in some cases human aided spread seems likely (Morgan et al. 2007; Skerratt et al. 2007). In contrast, Adams et al. (2010) report that Bd occurrence in Oregon and California, USA, does not correlate with any hypothesized environmental factors, but that Bd detectability increases with human influence on the landscape. We thus also evaluated the predictive power of a human-influence hypothesis for predicting Bd’s current occurrence pattern in Australia and compared it with our ES model (Fig. S8, Supporting Information). We used 19 bioclimatic variables (all continuous), one geo-physical variable (distance to water; continuous) and one vegetation type variable (categorical) at a resolution of c. 250 m (9 arc-seconds) for our models (Table S2, Supporting Information). We knew a priori that many of the variables were correlated and potentially meaningful contributors to the model; to avoid over-sized models (including variables with no predictive value) or over-fitted models (too many parameters for the data set) (Parolo, Rossi & Ferrarini 2008), we first selected the top ranking variables that together contributed c. 90% of the information to a full model run. We then re-ran a ‘pruned’ model with the most important variables (Table S2). Model accuracy was assessed with the area under the curve (AUC) of the receiver operator characteristic (ROC), which is a single measure of discrimination ability (presence from random background, where a value of 1 = perfect prediction, 0Æ5 = prediction no better than random) of the models (Fielding & Bell 1997). To incorporate uncertainty into our predictions, we used bootstrapping (N = 100) with unique sets of training and testing data (70 : 30% respectively). Many presence-only SDMs require background points (or pseudo-absences), the selection of which can influence the outcome of the models (Phillips et al. 2009; VanDerWal et al. 2009a). We provide an extended discussion of our background point selection in Fig. S2 which we used in order to limit as far as possible the effects of unquantifiable sampling bias and modelling an organism with considerable invasion potential.

DISEASE RISK

To investigate the hypothesized relationship between ES for Bd and the risk of chytridiomycosis to susceptible amphibian populations, we assessed whether our results were consistent with descriptions of population decline attributed to severe chytridiomycosis (Berger et al. 1998, 2004). We anticipated that decline sites would be strongly skewed towards higher values of ES for Bd. Declines attributed to chytridiomycosis have been best described from uplands in the Australian Alps and from montane rainforest areas in Queensland, where ill and dead frogs have been rigorously diagnosed as dying from chytridiomycosis at the time of declines (Berger et al. 1998, 2004; Hines, Mahony & McDonald 1999; McDonald & Alford 1999; Osborne, Hunter & Hollis 1999) (Fig. S3, Supporting Information). We next averaged our ES predictions across amphibian occurrence records for each species in the data set described by Slatyer, Rosauer & Lemckert (2007 updated 2009, D. Rosauer unpubl. data) to derive a species-specific metric of ES for Bd that we termed ‘ES for Bdspecies’. Slatyer et al.’s extensive data set comprises 291 942 occurrence records for all of Australia’s amphibian species. We removed duplicate records from the same locality (leaving 140 897 records; mean per species = 640) for calculations. Further details of the metric are provided in Fig. S4, Supporting Information. Species range size has previously been identified as the major risk factor for decline and extinction in Australian amphibians after controlling for other lifehistory and ecological factors (Murray & Hose 2005). We thus tested

for an effect of ES for Bdspecies, controlling for the range size effect, in contributing to whether amphibians have experienced declines or not. Amphibian trend classifications were sourced from the IUCN (2008). Range sizes were calculated from extent of occurrence polygons developed for the Global Amphibian Assessment (GAA) (Stuart et al. 2004). Finally, we calculated mean ES values across Australia’s biogeographic regions to identify those most suitable for infection with the pathogen (Fig. S5, Supporting Information). We related these results to amphibian species richness and endemism statistics from the study of Slatyer, Rosauer & Lemckert (2007) to indicate where infection with Bd most threatens anuran biodiversity in Australia (Fig. S6, Supporting Information).

Results MODEL SELECTION, VALIDATION AND VARIABLE CONTRIBUTIONS

After the pruning step, mean test AUC was 0Æ900 (range 0Æ874–0Æ925) and the model contained eight variables. The jack-knife procedure, which examines the effect of individual variables, indicated that mean diurnal temperature range and annual precipitation had the most useful information as single variables on training data (highest gain scores in isolation) as well as the highest predictive power (highest AUC in isolation) (Fig. 1). Response curves characterizing the relationships between ES and each of the two most influential predictor variables are shown in Fig. S7, Supporting Information. In the comparative analysis incorporating human population density (HPD), predictive performance of the full model was unchanged (0Æ903, range = 0Æ852–0Æ936) and HPD had inferior predictive power in isolation (AUC = 0Æ763, range = 0Æ716–0Æ811) relative to many of the environmental predictors. For subsequent analyses, we thus used the model incorporating environmental variables only (see Fig. S8, Supporting Information for results and further discussion).

PREDICTED DISTRIBUTION

The model suggested that infection with Bd should be largely restricted to the eastern and southern seaboards of Australia, with nearly all of inland and northern Australia unsuitable. Figure 2 represents the average Maxent predictions of ES (available for download in Appendix S2, Supporting Information). Clamping indicated that all of Australia fell within the environmental limits used to train the model (data not shown).

DISEASE RISK

Decline sites (mean ESdecline = 0Æ758; 95% CI = 0Æ714–0Æ802, n = 39) were highly skewed towards higher ES values compared to all sites used for model training and testing (mean ESall = 0Æ577, 95% CI = 0Æ550–0Æ604, n = 284) (Fig. 4a,b,d). Mean ES for Bdspecies varied between population trend categories (Fig. 3a); three extinct species had the highest value, 42 declining species had an intermediate value and 151 stable


Spatial patterns of disease risk 167 Only variable

(a)

Without variable

Full Vegetation type Precip cold quart Precip warm quart Precip dry quart Annual precip Mean temp dry quart Temp ann range Mean diurnal range 0

0·2

0·6

Only variable

(b)

Fig. 1. Variable contributions to (a) training gain and (b) AUC of the final ‘pruned’ model for Batrachochytrium dendrobatidis in Australia. ‘Only variable’ indicates the results of the model when a single variable is run in isolation; ‘without variable’ indicates the effect of removing a single variable from the full model (jack-knife). Values are means from 100 replicates. See Table S2 for full variable names and descriptions.

0·4

0·8 1 Training gain

1·2

1·4

1·6

Without variable

Full Vegetation type Precip cold quart Precip warm quart Precip dry quart Annual precip Mean temp dry quart Temp ann range Mean diurnal range

species exhibited a comparatively low value. In a logistic model in which species were grouped by whether they had declined or not (unknown trend species omitted), ES for Bdspecies was a highly significant predictor of decline (Ddev = 20Æ932, d.f. = 1, P < 0Æ001), even after controlling for a significant influence of narrow species range size (Ddev = 22Æ831, d.f. = 1, P < 0Æ001). The best model in terms of AIC contained ES for Bdspecies as a highly significant term (P < 0Æ001), range size as a marginally significant term (P = 0Æ098) and no interaction term. Table S3, Supporting Information presents a list of priority species for research and management indicating where investigation of Bd as a potential threatening process is critical. Table S4, Supporting Information presents the full list of Australian species. Mean ES varied considerably across biogeographic regions (Fig. 3b); the Wet Tropics (see also Fig. 4a) was predicted to have the highest mean suitability for Bd, followed by the Central Mackay Coast (Fig. 4b), Tasmania’s southern ranges, northern slopes, north-east (Ben Lomond) and King Island (Fig. 4e) and the NSW north coast (Fig. 4c). South-east Queensland (Fig. 4c), the Australian Alps (Fig. 4d), the Swan Coastal Plain (around Perth) (Fig. 4f) and the Tasmanian south-east also showed high mean ES values. Many regions with low mean ES nevertheless showed limited areas of very high ES as indicated by their maximum values (e.g. Brigalow Belts, Einasleigh Uplands, NSW south-western slopes) (Fig. 3b).

Discussion Infection with B. dendrobatidis occurs across a broad range of climates in Australia, in areas that are at times very hot, cold,

0·65

0·7

0·75

0·8 AUC

0·85

0·9

0·95

dry or wet. Those locations range from the hot, humid coastal lowlands of north-eastern Australia to the highest peaks of the Australian Alps, where winter snow occurs. Despite its broad tolerance of conditions, the model suggested that specific environmental conditions will restrict infection with Bd to the generally cooler and wetter areas of Australia (Figs 2 and 4). In this respect, our model was highly consistent with that of Ron (2005) (Fig. S1; Fig. 2); however, our results suggested that Bd should be more restricted, with the majority of central (arid) Australia being broadly unsuitable for Bd persistence (see also Fig. S8). The model indicated that ES increased with annual precipitation (with a minimum extreme of c. 500 mm) (Fig. S7). This is not surprising since desiccation is known to rapidly kill Bd in vitro (Berger 2001; Johnson et al. 2003) and the presence of permanent water is known to be an important feature for sustaining Bd, probably because the transmission stage for Bd is an aquatic zoospore (Berger et al. 1998). The model also suggested that mean diurnal temperature range was an important variable; the response curve indicated that ES declined rapidly in highly variable temperature regimes, where the difference in daily maxima and minima is greater than c. 11 C. Variation in temperature of itself has not previously been shown to affect chytridiomycosis (Woodhams, Alford & Marantelli 2003). However, high temperatures are known to be lethal to Bd and the effect of temperature variability may be explained by the observation that areas with higher temperature variability (e.g. the arid ⁄ semi-arid interior of the country) also typically exhibit very high maximum temperatures. This suggestion is supported by the response of Bd to maximum temperature of the warmest month, which showed maximum ES in the range of



Fig. 2. Model predictions of environmental suitability (after bootstrapping N = 100) for Batrachochytrium dendrobatidis in Australia.

maximum temperatures 18–30 C beyond which there is a precipitous decrease (data not shown). Our results are thus highly consistent with those of previous studies indicating that high temperatures are detrimental to Bd (Kriger & Hero 2007; Muths, Pilliod & Livo 2008; Puschendorf et al. 2009).

(termed ES for Bdspecies) was a very strong predictor of amphibian decline at a national level. These findings support the hypothesis that ES for infection with the pathogen as modelled here is broadly predictive of suitability for, and severity of, the disease chytridiomycosis via a theoretical and empirical link with intensity of infection (VanDerWal et al. 2009b). We thus interpret our ES for Bd results as being a highly useful source of quantitative information relevant to explaining the potential effects of infection with B. dendrobatidis (disease risk). This association should be interpreted cautiously, however, as in order for ES for Bdspecies to translate to risk of decline a number of other conditions relevant to the epidemiology of chytridiomycosis must be fulfilled, most importantly transmission. Species with more aquatic life-histories and an association with permanent water are most susceptible and at greatest risk of severe disease (Berger et al. 1998, 2004). Further, species inhabiting different micro-habitats can vary in their relative risk of infection within a single location (Woodhams & Alford 2005; Skerratt et al. 2008). As such, actual disease risk will be a product of the ES for the pathogen, the susceptibility of the species and the factors that make it susceptible to decline (e.g. see Bielby et al. 2008) given the former. This is an important consideration as it will be necessary to stratify host life-history traits for prioritization purposes (Table S3). Bielby et al. (2008) found that small range size, altitude and an aquatic life stage are risk factors for rapid decline in Bd-positive species. However, applying these risk factors to all species as they do is a considerable extrapolation because not all species are equally susceptible to infection. For example, very high-risk values in that study were assigned to many species with largely terrestrial life-histories, including many of Australia’s microhylid frogs (e.g. Cophixalus sp). While some of these also exhibit high ES for Bdspecies values as described herein, neither index identifies actual risk from Bd as species in this group appear far less susceptible than stream-dwelling and permanent water-associated species from the same region (N = 557 negative results in areas that are Bd-positive; K. Hauselberger & D. Mendez et al. unpubl. data; Skerratt et al. 2008). A more sophisticated risk analysis can be performed when more information is available about the innate susceptibilities of different amphibian species to chytridiomycosis and to decline. Integrating host-life history and ecological traits with the pathogen’s environmental requirements (as modelled here) to predict infection and decline is the focus of our current research efforts (Murray et al. in press).

DISEASE RISK

Two key results from this study are that (i) our predictions of ES are strikingly consistent with known associations between Bd and amphibian population declines in Queensland ⁄ New South Wales (Fig. S3 and Fig. 4a–c) (Hines, Mahony & McDonald 1999; McDonald & Alford 1999) and in the Australian Alps (Osborne, Hunter & Hollis 1999; Berger et al. 2004) (Fig. 4d) and (ii) the species-specific metric of ES for Bd

MANAGEMENT IMPLICATIONS

We have shown that ES for Bdspecies is a strong predictor of decline at a continental scale. This result was independent of a previously reported, dominant effect of narrow geographic range size. Our study provides a species-specific metric, representing the environmental requirements of the pathogen, with which to begin to assess this risk and calls for targeted vigilance


(a)

Environmental suitability for Bd (95%CI)

Spatial patterns of disease risk 169 1 0·9 0·8 0·7 0·6 0·5 0·4 0·3 0·2 0·1 0 Extinct (3)

Declining (42)

Stable (151)

Unknown (17)

IUCN trend category Environmental suitability for Bd (±1SD)

(b) 1·0 MEAN Max

0·9 0·8 0·7 0·6 0·5 0·4 0·3 0·2 0·1 0·0

IBRA Ecoregion

Fig. 3. Mean environmental suitability for Batrachochytrium dendrobatidis across: (a) species with different IUCN trend classifications, (b) IBRA ecoregions (averaged across entire region; only regions where the maximum or mean value >0Æ2 are shown). See Fig. S5 for map.

in sampling for this disease and monitoring for its potentially insidious effects (Murray et al. 2009; Pilliod et al. 2010). Bd records exist from most regions that were deemed suitable by the model, indicating that it has probably reached its broad geographic limits on this continent. There are, however, at least two areas that show marginal suitability for Bd beyond the known range of the pathogen where testing has failed to detect it: Cape York (Skerratt et al. 2008) (Fig. 4a) and Tasmania’s World Heritage south-west (Pauza & Driessen 2008) (Fig. 4e). It is possible that Bd has simply not yet dispersed to these regions, as they are at the extreme limits of the distribution in northern and southern Australia. However, the results of this study may also suggest that establishment or disease risk could be relatively low in these regions. Our results provide a testable hypothesis and surveys should continue in these areas where suitability is predicted to be highest (Fig. 4a,e). Prevention of spread nevertheless remains the best management strategy and these areas should not be regarded as areas in which Bd could not establish and cause mortalities (Fig. 2). Hygiene protocols should therefore be enforced for people entering these areas (Phillott et al. 2010). Several other marginal to highly suitable regions exist where no sampling has occurred. These regions represent important

areas for future sampling to establish the actual geographic limit of Bd in Australia and to establish amphibian population health. Identification of naive populations at high disease risk is a particular priority. Examples include uplands in the far north of Queensland (Cape York; Fig. 4a), upland areas in the Brigalow Belt (Fig. 4c), a large expanse of the western slopes of the Great Dividing Range in NSW (Fig. 4c), the south-west central tablelands of NSW and the regions surrounding Mount Gambier and the Mt Lofty Ranges in South Australia (Fig. 2). Conversely, our results suggest that several declining species for which chytridiomycosis is a suspected threatening process may have relatively low ES for Bdspecies (e.g. Litoria piperata and Litoria castanea), which provides a way forward when considering management and research activities. Similarly, some declining species may be at high risk of disease only in a subset of populations (e.g. Litoria spenceri, as indicated by maximum vs. mean ES values – Table S3). We are not asserting that Bd will be absent from these species or populations, but that other factors may also be involved in their decline, an example of where our results provide some useful and testable hypotheses that should be pursued. Regions of high disease risk in association with high host endemism should be the highest priority for population moni-



(a)

(b)

(c)

(d)

(e)

(f)

toring and ⁄ or management activity. Of the 11 centres of exceptional anuran endemism identified by Slatyer, Rosauer & Lemckert (2007), six occur in regions predicted to be highly suitable for Bd, including the Wet Tropics (Fig. 4a), Central Mackay Coast (Mackay ⁄ Eungella; Fig. 4b), Gladstone (Kroombit Tops), South-east Queensland (Gympie-Coffs Harbour; Fig. 4c) and south-west Western Australia (Walpole and Bunbury-Augusta; Fig. 4f) (see Fig. S5 for ecoregion names and Fig. S6 for endemism ⁄ richness). Records of Bd exist from all of these areas. An additional two areas (Townsville and Cape York) are predicted to have more restricted regions that are marginally or highly suitable for Bd (Fig. 4a). Three endemism hotspots are predicted to be at negligible risk from Bd (Kakadu and the Arnold River region in the NT and the Mitchell Plateau in WA). Establishing and maintaining a disease-free status should be their regional priority. The methods and results from this study can be used as a tool for establishing cost-sharing arrangements, prioritizing future efforts to detect and manage this pathogen (e.g. disease

Fig. 4. Selected regions in Australia predicted to have high average environmental suitability for Batrachochytrium dendrobatidis (see Fig. 2 for key to colours). Stars = ill and dead frogs positive for Bd in association with population declines (Qld ⁄ Aust. Alps). Grey lines = IBRA ecoregion boundaries (see Fig. S5 for map and key).

surveys, preventing further spread to naı¨ ve areas), for prioritizing monitoring programmes for Bd and Australia’s anuran fauna (e.g. Skerratt et al. 2008) and for identifying priority species for potential emergency captive-breeding programmes (Gascon et al. 2007) (Table S3). We envisage this to be an iterative process, with models such as ours regularly updated and scrutinized as new systematically collected data accrue (Wintle, Elith & Potts 2005). Critically, our methods can be directly and rapidly applied to other regions of the world experiencing amphibian declines; such results will aid in the task of developing informed management and surveillance decisions for Bd (Skerratt et al. 2008) and will help to make the most of limited conservation funds for prioritizing species, regions and actions for biodiversity conservation outcomes (Wilson et al. 2007).

LIMITATIONS AND FUTURE DIRECTIONS

While our model had high predictive performance and clamping indicated a well sampled environmental space, relatively lit-


Spatial patterns of disease risk 171 tle sampling has occurred on the western margins of the Great Dividing Range and in inland Australia, and Queensland and Western Australia were over represented compared with other regions. In addition, frogs may take refuge in environments that are not captured by interpolated bioclimatic or vegetation mapping data (see also Fig. S8) and we have limited ability to incorporate microclimatic features into our models given the enormous diversity of amphibian hosts and their habitats in this country. Similarly, beyond considerations of HPD (Fig. S8) we were unable to incorporate models of pathogen dispersal given very limited knowledge regarding how this pathogen is spread. We model the realized niche of this invasive species in an invaded range; there is thus the possibility that Bd’s distribution in Australia has not approached an equilibrium state, potentially resulting in an underprediction of its potential range. We consider this an unlikely source of major bias in our results given Bd’s extensive distribution nation-wide and the spectrum of potentially susceptible amphibian hosts (e.g. Litoria spp.) and hypothesized vectors (e.g. humans) in currently uninfected regions. Nevertheless, our model represents a baseline, minimum potential distribution rather than a finite prediction of this organism’s fundamental niche; we encourage scrutiny and ongoing iteration (e.g. integrated use of new systematically collected data), particularly to increase representation of apparently disease-free areas into future models. Dispersal models should also be a future priority, particularly in areas that are newly invaded. Finally, genetic differentiation has been noted geographically (Morgan et al. 2007; James et al. 2009), and strains may undergo local adaptation (Fisher et al. 2009) and ⁄ or show strain specific differences in adaptive plasticity so distribution in Australia with respect to available environmental space may not necessarily correspond exactly to other regions or to the results of other predictive models. Comparison of these and future studies will thus identify important areas and avenues for further research and it is imperative that the predictions of any SDM be independently compared with other SDM methods and data sources (see Elith et al. 2006), other methods (e.g. mechanistic models; K.A.M. unpubl. data) (Morin & Thuiller 2009) and by comprehensive field surveys during sampling periods that maximize detection probability (Skerratt et al. 2008).

Acknowledgements We are indebted to the many authors and contributors named in Murray et al. (2010) for the production of the Bd occurrence data base. In particular, we thank K. McDonald, K. Aplin, H. Hines, D. Mendez, A. Felton, P. Kirkpatrick, D. Hunter, R. Campbell, M. Pauza, M. Driessen, S. Richards, M. Mahony, A. Freeman, A. Phillott, J-M. Hero, K. Kriger and D. Driscoll. KAM thanks D. Segan, M. Watts and C. Klein for GIS wisdom and spatial data, R. Wilson and H. Possingham for lab space, B. Sutherst, M. Zalucki and D. Kriticos for fruitful discussions and M. Araujo and R. Pearson for running a timely SDM workshop at the University of Queensland. We also thank Dr Marc Cadotte, Professor Christl Donnelly and three anonymous reviewers for excellent comments and discussion on earlier versions of the manuscript. KAM was supported by an Australian Postgraduate Award, an Australian Biosecurity CRC professional development award and a Wildlife Preservation Society of Australia student research award. Part of this work was conducted when RWRR was supported by the Australian Research Council, the School of Public Health and Tropical Medicine, James Cook University and a National Science Foundation Integrated Research Challenges in Environmen-

tal Biology grant awarded to J. Collins at Arizona State University, USA. RWRR thanks J. Collins and C. Carey.

References Adams, M.J., Chelgren, N.D., Reinitz, D., Cole, R.A., Rachowicz, L.J., Galvan, S., McCreary, B., Pearl, C.A., Bailey, L.L., Bettaso, J., Bull, E.L. & Leu, M. (2010) Using occupancy models to understand the distribution of an amphibian pathogen, Batrachochytrium dendrobatidis. Ecological Applications, 20, 289–302. Araujo, M.B. & Guisan, A. (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 1677–1688. Berger, L. (2001) Diseases in Australian frogs. PhD, James Cook University, Townsville. Berger, L., Speare, R., Daszak, P., Green, D.E., Cunningham, A.A., Goggin, C.L., Slocombe, R., Ragan, M.A., Hyatt, A.D., McDonald, K.R., Hines, H.B., Lips, K.R., Marantelli, G. & Parkes, H. (1998) Chytridiomycosis causes amphibian mortality associated with population declines in the rain forests of Australia and Central America. Proceedings of the National Academy of Sciences of the United States of America, 95, 9031–9036. Berger, L., Speare, R., Hines, H.B., Marantelli, G., Hyatt, A.D., McDonald, K.R., Skerratt, L.F., Olsen, V., Clarke, J.M., Gillespie, G., Mahony, M., Sheppard, N., Williams, C. & Tyler, M.J. (2004) Effect of season and temperature on mortality in amphibians due to chytridiomycosis. Australian Veterinary Journal, 82, 434–439. Bielby, J., Cooper, N., Cunningham, A., Garner, T. & Purvis, A. (2008) Predicting susceptibility to future declines in the world’s frogs. Conservation Letters, 1, 82–90. Briggs, C., Knapp, R.A. & Vredenburg, V.T. (2010) Enzootic and epizootic dynamics of the chytrid fungal pathogen of amphibians. Proceedings of the National Academy of Sciences of the United States of America, 107, 9695– 9700. Carey, C., Bruzgul, J.E., Livo, L.J., Walling, M.L., Kuehl, K.A., Dixon, B.F., Pessier, A.P., Alford, R.A. & Rogers, K.B. (2006) Experimental exposures of boreal toads (Bufo boreas) to a pathogenic chytrid fungus (Batrachochytrium dendrobatidis). EcoHealth, 3, 5–21. Crowl, T.A., Crist, T.O., Parmenter, R.R., Belovsky, G. & Lugo, A.E. (2008) The spread of invasive species and infectious disease as drivers of ecosystem change. Frontiers in Ecology and the Environment, 6, 238–246. Daszak, P., Cunningham, A. & Hyatt, A. (2000) Emerging infectious diseases of wildlife – threats to biodiversity and human health. Science, 287, 443–449. Digiacomo, R.F. & Koepsell, T.D. (1986) Sampling for detection of infection or disease in animal populations. Journal of the American Veterinary Medical Association, 189, 22–23. Dudik, M., Phillips, S.J. & Schapire, R.E. (2007) Maximum entropy density estimation with generalized regularization and an application to species distribution modeling. Journal of Machine Learning Research, 8, 1217–1260. Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S. & Zimmermann, N.E. (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography, 29, 129–151. Ficetola, G.F., Thuiller, W. & Miaud, C. (2007) Prediction and validation of the potential global distribution of a problematic alien invasive species – the American bullfrog. Diversity and Distributions, 13, 476–485. Fielding, A.H. & Bell, J.F. (1997) A review of methods for the assessment of prediction errors in conservation presence ⁄ absence models. Environmental Conservation, 24, 38–49. Fisher, M.C., Bosch, J., Yin, Z., Stead, D.A., Walker, J., Selway, L., Brown, A.J.P., Walker, L.A., Gow, N.A.R., Stajich, J.E. & Garner, T.W.J. (2009) Proteomic and phenotypic profiling of the amphibian pathogen Batrachochytrium dendrobatidis shows that genotype is linked to virulence. Molecular Ecology, 18, 415–429. Gascon, C., Collins, J.P., Moore, R.D., Church, D.R., McKay, J.E. & Mendelson, J.R. III (eds) (2007) Amphibian Conservation Action Plan: Proceedings: IUCN ⁄ SSC Amphibian Conservation Summit 2005. The World Conservation Union (IUCN), Gland, Switzerland. Gibson, L., Barrett, B. & Burbidge, A. (2007) Dealing with uncertain absences in habitat modelling: a case study of a rare ground-dwelling parrot. Diversity and Distributions, 13, 704–713. Goka, K., Yokoyama, J., Une, Y., Kuroki, T., Suzuki, K., Nakahara, M., Kobayashi, A., Inaba, S., Mizutani, T. & Hyatt, A.D. (2009) Amphibian chy-


172 K. A. Murray et al. tridiomycosis in Japan: distribution, haplotypes and possible route of entry into Japan. Molecular Ecology, 18, 4757–4774. Graham, C.H., Elith, J., Hijmans, R.J., Guisan, A., Peterson, A.T., Loiselle, B.A. & Group, N.P.S.W. (2008) The influence of spatial errors in species occurrence data used in distribution models. Journal of Applied Ecology, 45, 239–247. Guisan, A. & Thuiller, W. (2005) Predicting species distribution: offering more than simple habitat models. Ecology Letters, 8, 993–1009. Guisan, A., Broennimann, O., Engler, R., Vust, M., Yoccoz, N.G., Lehmann, A. & Zimmermann, N.E. (2006) Using niche-based models to improve the sampling of rare species. Conservation Biology, 20, 501–511. Hines, H., Mahony, M. & McDonald, K. (1999) An assessment of frog declines in wet subtropical Australia. Declines and Disappearances of Australian Frogs (eds A. Campbell), pp. 44–63. Environment Australia, Canberra. IUCN (2008) The IUCN red list of threatened species. IUCN. http:// www.iucnredlist.org, accessed July 2008. James, T.Y., Litvintseva, A.P., Vilgalys, R., Morgan, J.A.T., Taylor, J.W., Fisher, M.C., Berger, L., Weldon, C., du Preez, L. & Longcore, J.E. (2009) Rapid global expansion of the fungal disease chytridiomycosis into declining and healthy amphibian populations. PLoS Pathogens, 5, e1000458. Johnson, M.L. & Speare, R. (2003) Survival of Batrachochytrium dendrobatidis in water: quarantine and disease control implications. Emerging Infectious Diseases, 9, 922–925. Johnson, M.L., Berger, L., Philips, L. & Speare, R. (2003) Fungicidal effects of chemical disinfectants, UV light, desiccation and heat on the amphibian chytrid Batrachochytrium dendrobatidis. Diseases of Aquatic Organisms, 57, 255–260. Jones, K.E., Patel, N.G., Levy, M.A., Storeygard, A., Balk, D., Gittleman, J.L. & Daszak, P. (2008) Global trends in emerging infectious diseases. Nature, 451, 990–994. Kearney, M. & Porter, W. (2009) Mechanistic niche modelling: combining physiological and spatial data to predict species’ ranges. Ecology Letters, 12, 334–350. Kirshtein, J.D., Anderson, C.W., Wood, J.S., Longcore, J.E. & Voytek, M.A. (2007) Quantitative PCR detection of Batrachochytrium dendrobatidis DNA from sediments and water. Diseases of Aquatic Organisms, 77, 11–15. Kriger, K.M. & Hero, J.M. (2007) Large-scale seasonal variation in the prevalence and severity of chytridiomycosis. Journal of Zoology, 271, 352–359. Kusrini, M.D., Skerratt, L.F., Garland, S., Berger, L. & Endarwin, W. (2008) Chytridiomycosis in frogs of Mount Gede Pangrango, Indonesia. Diseases of Aquatic Organisms, 82, 187–194. Lips, K.R., Brem, F., Brenes, R., Reeve, J.D., Alford, R.A., Voyles, J., Carey, C., Livo, L., Pessier, A.P. & Collins, J.P. (2006) Emerging infectious disease and the loss of biodiversity in a neotropical amphibian community. Proceedings of the National Academy of Sciences of the United States of America, 103, 3165–3170. Lips, K.R., Diffendorfer, J., Mendelson, J.R. & Sears, M.W. (2008) Riding the wave: reconciling the roles of disease and climate change in amphibian declines. PLoS Biology, 6, e72. Longcore, J.E., Pessier, A.P. & Nichols, D.K. (1999) Batrachochytrium dendrobatidis gen et sp nov, a chytrid pathogenic to amphibians. Mycologia, 91, 219–227. McDonald, K. & Alford, R. (1999) A review of declining frogs in Northern Queensland. Declines and Disappearances of Australian Frogs (eds A. Campbell), pp. 9–22. Environment Australia, Canberra. Morehouse, E.A., James, T.Y., Ganley, A.R.D., Vilgalys, R., Berger, L., Murphy, P.J. & Longcore, J.E. (2003) Multilocus sequence typing suggests the chytrid pathogen of amphibians is a recently emerged clone. Molecular Ecology, 12, 395–403. Morgan, J.A.T., Vredenburg, V.T., Rachowicz, L.J., Knapp, R.A., Stice, M.J., Tunstall, T., Bingham, R.E., Parker, J.M., Longcore, J.E., Moritz, C., Briggs, C.J. & Taylor, J.W. (2007) Population genetics of the frog-killing fungus Batrachochytrium dendrobatidis. Proceedings of the National Academy of Sciences of the United States of America, 104, 13845–13850. Morin, X. & Thuiller, W. (2009) Comparing niche- and process-based models to reduce prediction uncertainty in species range shifts under climate change. Ecology, 90, 1301–1313. Murray, B.R. & Hose, G.C. (2005) Life-history and ecological correlates of decline and extinction in the endemic Australian frog fauna. Austral Ecology, 30, 564–571. Murray, K.A., Skerratt, L.F., Speare, R. & McCallum, H. (2009) Impact and dynamics of disease in species threatened by the amphibian chytrid fungus, Batrachochytrium dendrobatidis. Conservation Biology, 23, 1242–1252. Murray, K.A., Retallick, R., McDonald, K., Mendez, D., Aplin, K., Kirkpatrick, P. et al. (2010) The distribution and host range of the pandemic disease

chytridiomycosis in Australia spanning surveys from 1956 to 2007. Ecology, 91, 1557. Murray, K.A., Rosauer, D., McCallum, H. & Skerratt, L.F. (in press) Integrating species traits with extrinsic threats: closing the gap between predicting and preventing species declines. Proceedings of the Royal Society B-Biological Sciences. Muths, E., Pilliod, D.S. & Livo, L.J. (2008) Distribution and environmental limitations of an amphibian pathogen in the Rocky Mountains, USA. Biological Conservation, 141, 1484–1492. Olson, D. & Ronnenberg, K. (2008) Global Bd Mapping Project: Update – 17 January 2008. Available from http://www.parcplace.org/bdmap2008update. html (accessed January 2008). USDA Forest Service, Pacific Northwest Research Station, Corvallis, Oregon. Osborne, W., Hunter, D. & Hollis, G. (1999) Population declines and range contraction in Australian alpine frogs. Declines and Disappearances of Australian Frogs (eds A. Campbell), pp. 145–157. Environment Australia, Canberra. Parolo, G., Rossi, G. & Ferrarini, A. (2008) Toward improved species niche modelling: Arnica montana in the Alps as a case study. Journal of Applied Ecology, 45, 1410–1418. Pauza, M., Driessen, M. & Skerratt, L.F. (2010) Distribution and risk factors for spread of amphibian chytrid fungus Batrachochytrium dendrobatidis in the Tasmanian Wilderness World Heritage Area, Australia. Diseases of Aquatic Organisms, published online September 23 2010, 1–7. Pearce, J.L. & Boyce, M.S. (2006) Modelling distribution and abundance with presence-only data. Journal of Applied Ecology, 43, 405–412. Pearson, R.G., Raxworthy, C.J., Nakamura, M. & Peterson, A.T. (2007) Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. Journal of Biogeography, 34, 102– 117. Peterson, A.T., Papes, M. & Eaton, M. (2007) Transferability and model evaluation in ecological niche modeling: a comparison of GARP and Maxent. Ecography, 30, 550–560. Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231–259. Phillips, S.J., Dudik, M., Elith, J., Graham, C.H., Lehmann, A., Leathwick, J. & Ferrier, S. (2009) Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecological Applications, 19, 181–197. Phillott, A.D., Speare, R., Hines, H.B., Meyer, E., Skerratt, L.F., McDonald, K.R., Cashins, S.D., Mendez, D. & Berger, L. (2010) Minimising exposure of amphibians to pathogens during field studies. Diseases of Aquatic Organisms, in press. doi: 10.3354/dao02162. Pilliod, D.S., Muths, E., Scherer, R.D., Bartelt, P.E., Corn, P.S., Hossack, B.R., Lambert, B.A., McCaffery, R. & Gaughan, C. (2010) Effects of amphibian chytrid fungus on individual survival probability in wild boreal toads. Conservation Biology, 24, 1259–1267. Piotrowski, J.S., Annis, S.L. & Longcore, J.E. (2004) Physiology of Batrachochytrium dendrobatidis, a chytrid pathogen of amphibians. Mycologia, 96, 9– 15. Puschendorf, R., Carnaval, A.C., VanDerWal, J., Zumbado-Ulate, H., Chaves, G., Bolanos, F. & Alford, R.A. (2009) Distribution models for the amphibian chytrid Batrachochytrium dendrobatidis in Costa Rica: proposing climatic refuges as a conservation tool. Diversity and Distributions, 15, 401–408. Rachowicz, L.J., Hero, J.M., Alford, R.A., Taylor, J.W., Morgan, J.A.T., Vredenburg, V.T., Collins, J.P. & Briggs, C.J. (2005) The novel and endemic pathogen hypotheses: competing explanations for the origin of emerging infectious diseases of wildlife. Conservation Biology, 19, 1441–1448. Rohr, J.R., Raffel, T.R., Romansic, J.M., McCallum, H. & Hudson, P.J. (2008) Evaluating the links between climate, disease spread, and amphibian declines. Proceedings of the National Academy of Sciences of the United States of America, 105, 17436–17441. Ron, S.R. (2005) Predicting the distribution of the amphibian pathogen Batrachochytrium dendrobatidis in the New World. Biotropica, 37, 209–221. Skerratt, L.F., Berger, L., Speare, R., Cashins, S., McDonald, K.R., Phillott, A.D., Hines, H.B. & Kenyon, N. (2007) Spread of chytridiomycosis has caused the rapid global decline and extinction of frogs. EcoHealth, 4, 125– 134. Skerratt, L.F., Berger, L., Hines, H.B., McDonald, K.R., Mendez, D. & Speare, R. (2008) Survey protocol for detecting chytridiomycosis in all Australian frog populations. Diseases of Aquatic Organisms, 80, 85–94. Slatyer, C., Rosauer, D. & Lemckert, F. (2007) An assessment of endemism and species richness patterns in the Australian Anura. Journal of Biogeography, 34, 583–596.


Spatial patterns of disease risk 173 Speare, R. & Berger, L. (2000) Global Distribution of Chytridiomycosis in Amphibians. Amphibian Diseases Research Group, Townsville. http:// www.jcu.edu.au/shool/phtm/PHTM/frogs/chyglob.htm, accessed July 2007. Stuart, S.N., Chanson, J.S., Cox, N.A., Young, B.E., Rodrigues, A.S.L., Fischman, D.L. & Waller, R.W. (2004) Status and trends of amphibian declines and extinctions worldwide. Science, 306, 1783–1786. VanDerWal, J., Shoo, L.P., Graham, C. & William, S.E. (2009a) Selecting pseudo-absence data for presence-only distribution modeling: how far should you stray from what you know? Ecological Modelling, 220, 589– 594. VanDerWal, J., Shoo, L.P., Johnson, C.N. & Williams, S.E. (2009b) Abundance and the environmental niche: environmental suitability estimated from niche models predicts the upper limit of local abundance. American Naturalist, 174, 282–291. Voyles, J., Berger, L., Young, S., Speare, R., Webb, R., Warner, J., Rudd, D., Campbell, R. & Skerratt, L.F. (2007) Electrolyte depletion and osmotic imbalance in amphibians with chytridiomycosis. Diseases of Aquatic Organisms, 77, 113–118. Wake, D.B. & Vredenburg, V.T. (2008) Are we in the midst of the sixth mass extinction? A view from the world of amphibians. Proceedings of the National Academy of Sciences of the United States of America, 105, 11466– 11473. Walker, S.F., Salas, M.B., Jenkins, D., Garner, T.W.J., Cunningham, A.A., Hyatt, A.D., Bosch, J. & Fisher, M.C. (2007) Environmental detection of Batrachochytrium dendrobatidis in a temperate climate. Diseases of Aquatic Organisms, 77, 105–112. Wilson, K.A., Underwood, E.C., Morrison, S.A., Klausmeyer, K.R., Murdoch, W.W., Reyers, B., Wardell-Johnson, G., Marquet, P.A., Rundel, P.W., McBride, M.F., Pressey, R.L., Bode, M., Hoekstra, J.M., Andelman, S., Looker, M., Rondinini, C., Kareiva, P., Shaw, M.R. & Possingham, H.P. (2007) Conserving biodiversity efficiently: what to do, where, and when. PLoS Biology, 5, 1850–1861. Wintle, B.A., Elith, J. & Potts, J.M. (2005) Fauna habitat modelling and mapping: a review and case study in the Lower Hunter Central Coast region of NSW. Austral Ecology, 30, 719–738. Wisz, M.S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A. & NCEAS (2008) Effects of sample size on the performance of species distribution models. Diversity and Distributions, 14, 763–773.

Woodhams, D.C. & Alford, R.A. (2005) Ecology of chytridiomycosis in rainforest stream frog assemblages of tropical Queensland. Conservation Biology, 19, 1449–1459. Woodhams, D.C., Alford, R.A. & Marantelli, G. (2003) Emerging disease of amphibians cured by elevated body temperature. Diseases of Aquatic Organisms, 55, 65–67. Woodhams, D.C., Alford, R.A., Briggs, C.J., Johnson, M. & Rollins-Smith, L.A. (2008) Life-history trade-offs influence disease in changing climates: strategies of an amphibian pathogen. Ecology, 89, 1627–1639. World Organisation for Animal Health (2008) Aquatic Animal Code 2008. World Organisation for Animal Health, Paris. Available from http:// www.oie.int/eng/normes/fcode/en_sommaire.htm (accessed July 2008). Ziller, M., Selhorst, T., Teuffert, J., Kramer, M. & Schluter, H. (2002) Analysis of sampling strategies to substantiate freedom from disease in large areas. Preventive Veterinary Medicine, 52, 333–343. Received 31 May 2010; accepted 28 September 2010 Handling Editor: Marc Cadotte

Supporting Information Additional Supporting Information may be found in the online version of this article: Appendix S1. Supporting Information (Tables S1–S4, Figs S1–S8). Appendix S2. Results of the model (Bd_in_Australia.asc). As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.



doi: 10.1111/j.1365-2664.2010.01891.x

FORUM

Modelling the future distribution of the amphibian chytrid fungus: the influence of climate and human-associated factors Jason R. Rohr*, Neal T. Halstead and Thomas R. Raffel University of South Florida, Department of Integrative Biology, Tampa, FL 33620, USA

Summary 1. Many of the global losses of amphibians are believed to be caused by the chytrid fungus, Batrachochytrium dendrobatidis (Bd). Hence, determining its present and future environmental suitability should help to inform management and surveillance of this pathogen and curtail the amphibian biodiversity crisis. 2. In this issue of Journal of Applied Ecology, Murray et al. (2011) offer an important step in this direction by providing a species distribution model that projects the environmental suitability of Bd across Australia and predicts locations of chytridiomycosis and amphibian declines. Batrachochytrium dendrobatidis presence was predicted by diurnal temperature range (a measure of temperature variability) and mean precipitation. Human population density, a positive predictor of Bd, accounted for the most variation when removed from the statistical model. 3. This work represents an invaluable case study and has great potential for managing chytridiomycosis and associated amphibian declines, but its value in practice will depend on how well managers understand the limitations of species distribution models. 4. Synthesis and applications. To improve the management of chytridiomycosis, amphibian-chytrid research should attempt to understand how humans may affect the distribution of Bd, how climatic means and variances affect Bd transmission, how much variation in the distribution of Bd is unique to and shared among climate, human, and other factors, whether human-related factors and climate statistically interact, and how these potentially correlated factors and any interactions affect the predictability of species distribution models. In response to the swift spread of Bd and our rapidly changing planet, we encourage the application of Bd distribution models to other regions of the globe and predictions of Bd’s distribution under future climate change scenarios. Key-words: Batrachochytrium dendrobatidis, bioclimatic envelope models, biotic interactions, chytridiomycosis, climate change, disease, dispersal, diurnal temperature range, management, species distribution models

Introduction Alarmingly, almost a third of amphibian species are considered threatened and more than 43% are experiencing some form of population decline (Stuart et al. 2004). Many of these amphibian losses are believed to be driven by the disease chytridiomycosis, caused by the pathogenic chytrid fungus Batrachochytrium dendrobatidis (Bd; Wake & Vredenburg 2008). Given that Bd may be spreading and ⁄ or emerging (Rohr et al. 2008; James et al. 2009) and that climate is changing, one of *Correspondence author. E-mail: [email protected] Conflicts of interest: No conflicts declared

the priorities for managing this devastating pathogen is to determine its present and future environmental limitations at global and local scales. If this can be done accurately, it might help to predict the future distribution of Bd, identify where Bd poses the greatest future threats, and facilitate prioritization of species and locations for monitoring and management given scarce conservation funds. In this issue of Journal of Applied Ecology, Murray et al. (2011) offer an important step in this direction by providing a species distribution model that projects the environmental suitability for Bd across the entire continent of Australia. Murray et al. (2011) base their models on an impressive spatiotemporal database consisting of 821 sites in Australia where

2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society

Future distribution of the amphibian chytrid 175 115 amphibian species and 10 183 individuals were tested for Bd from 1956–2007. Previous large-scale databases associated with Bd-related declines have proven to be useful in understanding the biology of this pathogen (Lips et al. 2008; Rohr et al. 2008; Rohr & Raffel 2010), and there is little question that this new database will also be valuable for addressing the amphibian biodiversity crisis and questions of disease ecology in general. From 19 climatic variables, as well as information on distance to water, vegetation type, and human population density, Murray et al. (2011) identified annual precipitation and diurnal temperature range (a measure of temperature variation) as important predictors of Bd presence. Moisture has been well established as a crucial factor for Bd persistence (Johnson et al. 2003; Berger et al. 2004). Evidence for the importance of diurnal temperature range is interesting in light of a recent study that revealed that temperature variability, in general, and diurnal temperature range specifically, might drive amphibian declines putatively caused by chytridiomycosis (Rohr & Raffel 2010). Using recent advances in bioclimatic envelope modelling (Phillips, Anderson & Schapire 2006), Murray et al. (2011) estimated the range of environments suitable for Bd (its bioclimatic envelope) and then predicted its minimum potential geographic distribution across Australia. Importantly, sites with documented amphibian declines associated with severe chytridiomycosis had high Bd environmental suitability values, and environmental suitability values specific to each of Australia’s 196 amphibian species were significant positive predictors of whether species experienced declines (even after controlling for species’ range sizes). Hence, the developed species distribution model appears to be capable of predicting locations of high risk for both chytridiomycosis and amphibian losses. Some caution should be used interpreting these correlations, however, because sampling for Bd was likely biased towards locations where amphibians were in decline and ⁄ or showing signs of chytridiomycosis, so environmental suitability estimates might also be biased toward these locations. Finally, in an effort to facilitate and target management and monitoring, Murray et al. (2011) importantly identified amphibian species that have high environmental suitability scores for Bd and regions of Australia that have both high suitability scores and high amphibian species richness and endemism. Although this work clearly has great potential for managing chytridiomycosis and associated amphibian declines, its value in practice will depend on how well managers understand the limitations of bioclimatic envelope models (BEMs) and how well the assumptions of these models were met by this study. BEMs are strictly correlational and do not directly model abiotic and biotic interactions, dispersal, or evolution, all of which can be important for predicting the effects of climate change on biodiversity (Davis et al. 1998; Rohr & Madison 2003; Araujo & Luoto 2007; Harmon, Moran & Ives 2009). Further, BEMs often have considerable uncertainty despite systematically overestimating model fits during model validation (Hampe 2004). The over-fitting occurs because BEMs generally do not account for spatial autocorrelation

among their data (a pseudoreplication issue; Segurado, Araujo & Kunin 2006). Finally, the results of BEMs can be quite sensitive to the model parameterization and model selection procedures that were implemented (Araujo & Guisan 2006). Ideally, several parameterization and selection procedures should be used to evaluate the robustness of the BEM results. Nevertheless, BEMs provide a useful first approximation and working hypothesis when identifying a species’ future distribution (Pearson & Dawson 2003). These approximations should be improved upon with additional data and adaptive management approaches, as advocated by Murray et al. (2011). Murray et al. (2011) also offer some important insights into how their model might be improved upon. Intriguingly, Murray et al. (2011) discovered that human population density (HPD) was a positive predictor of the presence of Bd, with Bd almost exclusively being found near port cities and the highways connecting them (their Fig. S1). Furthermore, the removal of HPD from their statistical model resulted in the greatest change in variation relative to all other predictors (their Fig. S8), indicating that HPD accounted for the greatest unique variation in the distribution of Bd. While it is possible that the correlation is driven by humans and Bd simply preferring the same climate, this seems unlikely given that the effect of humans was still evident after accounting for variation due to the climatic factors. Hence, the relationship between HPD and Bd presence suggests that, in addition to climate, human-associated factors might affect the distribution of this pathogen. This is not surprising given that humans are believed to be a major dispersal agent for Bd (Skerratt et al. 2007), a hypothesis that received support from two molecular studies concluding that the distribution of Bd was consistent with human-assisted migration (Morgan et al. 2007; James et al. 2009), and from a survey in Oregon and Northern California revealing that detectability of Bd increased markedly with human influence on the landscape (Adams et al. 2010). However, if humans are regularly introducing Bd into areas of low environmental suitability, this could violate the underlying assumption of the BEM, that the climatic conditions where Bd is presently found will either match the conditions of its future range or at least be adequate surrogates for the factors that dictate its future distribution (Pearson & Dawson 2003). Research on source-sink dynamics and metapopulations, and more recently neutral theory, has shown that organisms can often appear in suboptimal habitats when immigration is high (Davis et al. 1998; Hanski 1998; Hubbell 2005). The fact that Bd is predominantly found in and around coastal cities might further exacerbate this concern because the coastline essentially functions as a giant drift fence, forcing dispersal along and away from the coast. If there are multiple introductions at nearby port cities and crossing waves of dispersal along the coast, then BEMs will provide greater weight to these coastal regions, even if they do not represent optimal Bd conditions. It remains to be seen how much of a concern this is to the reliability of Murray et al.’s (2011) predictions, but the potential influence of dispersal and human-associated factors point to some major challenges for future BEMs and amphibianchytrid work.


176 J. R. Rohr, N. T. Halstead & T. R. Raffel Future research on BEMs should evaluate whether accounting for spatial autocorrelation using traditional approaches adequately accounts for dispersal barriers and limitations and spread from known or presumed introduction sites. This is particularly important given that the goal of BEMs, in many cases, is to model invasive species, which are often introduced at locations of high human population density, such as at cities with ports and airports. Amphibian-chytrid research should attempt to understand how humans affect the distribution of Bd; how much variation in the distribution of Bd is unique to and shared among climate, human, and other factors; whether human-related factors and climate interact statistically; and how these correlated factors and any interactions affect the predictability of BEMs. Finally, amphibian-chytrid research should better validate BEMs by determining how accurately they predict the spread of Bd. Despite their limitations, BEMs represent an important tool for predicting the future distributions of species and Murray et al. (2011) provide an invaluable case study that should guide others in applying these tools worldwide. Undoubtedly, the reliability of these models for predicting the distribution of Bd will improve with a better understanding of the factors that dictate the persistence and transmission of this pathogen. In response to the swift spread of Bd (Skerratt et al. 2007; Rohr et al. 2008) and our rapidly changing planet, we encourage the application of BEMs to other regions of the globe and the use of BEMs and ensemble climate models to predict the distribution of Bd under future climate change scenarios (e.g. Lawler et al. 2009). Global climate change is creating a climatically more variable world (Raisanen 2002) and thus research must consider how changes to both the mean and variance of climatic variables affect Bd-amphibian interactions and species interactions, in general (Raffel et al. 2006; Rohr & Raffel 2010). These proposed efforts should help inform management and surveillance, and will hopefully curtail our amphibian biodiversity crisis.

Acknowledgements We thank Marc Cadotte for providing us with the opportunity to write this commentary and for discussing with us the reviewers’ praise and comments on this paper. Funds were provided by a National Science Foundation (DEB 0516227) grant to J.R.R., US Department of Agriculture (NRI 2006-01370, 2009-35102-05043) grants to J.R.R., and a US Environmental Protection Agency STAR grant to J.R.R. and T.R.R. (R833835).

References Adams, M.J., Chelgren, N.D., Reinitz, D., Cole, R.A., Rachowicz, L.J., Galvan, S., McCreary, B., Pearl, C.A., Bailey, L.L., Bettaso, J., Bull, E.L. & Leu, M. (2010) Using occupancy models to understand the distribution of an amphibian pathogen, Batrachochytrium dendrobatidis. Ecological Applications, 20, 289–302. Araujo, M.B. & Guisan, A. (2006) Five (or so) challenges for species distribution modelling. Journal of Biogeography, 33, 1677–1688. Araujo, M.B. & Luoto, M. (2007) The importance of biotic interactions for modelling species distributions under climate change. Global Ecology and Biogeography, 16, 743–753. Berger, L., Speare, R., Hines, H.B., Marantelli, G., Hyatt, A.D., McDonald, K.R., Skerratt, L.F., Olsen, V., Clarke, J.M., Gillespie, G., Mahony, M., Sheppard, N., Williams, C. & Tyler, M.J. (2004) Effect of season and

temperature on mortality in amphibians due to chytridiomycosis. Australian Veterinary Journal, 82, 434–439. Davis, A.J., Jenkinson, L.S., Lawton, J.H., Shorrocks, B. & Wood, S. (1998) Making mistakes when predicting shifts in species range in response to global warming. Nature, 391, 783–786. Hampe, A. (2004) Bioclimate envelope models: what they detect and what they hide. Global Ecology and Biogeography, 13, 469–471. Hanski, I. (1998) Metapopulation dynamics. Nature, 396, 41–49. Harmon, J.P., Moran, N.A. & Ives, A.R. (2009) Species response to environmental change: impacts of food web interactions and evolution. Science, 323, 1347–1350. Hubbell, S.P. (2005) Neutral theory in community ecology and the hypothesis of functional equivalence. Functional Ecology, 19, 166–172. James, T.Y., Litvintseva, A.P., Vilgalys, R., Morgan, J.A.T., Taylor, J.W., Fisher, M.C., Berger, L., Weldon, C., du Preez, L. & Longcore, J.E. (2009) Rapid global expansion of the fungal disease chytridiomycosis into declining and healthy amphibian populations. PLoS Pathogens, 5, e1000458. Johnson, M.L., Berger, L., Philips, L. & Speare, R. (2003) Fungicidal effects of chemical disinfectants, UV light, desiccation and heat on the amphibian chytrid Batrachochytrium dendrobatidis. Diseases of Aquatic Organisms, 57, 255–260. Lawler, J.J., Shafer, S.L., White, D., Kareiva, P., Maurer, E.P., Blaustein, A.R. & Bartlein, P.J. (2009) Projected climate-induced faunal change in the Western Hemisphere. Ecology, 90, 588–597. Lips, K.R., Diffendorfer, J.E., Mendelson, J.R. & Sears, M.W. (2008) Riding the wave: reconciling the roles of disease and climate change in amphibian declines. PLoS Biology, 6, 441–454. Morgan, J.A.T., Vredenburg, V.T., Rachowicz, L.J., Knapp, R.A., Stice, M.J., Tunstall, T., Bingham, R.E., Parker, J.M., Longcore, J.E., Moritz, C., Briggs, C.J. & Taylor, J.W. (2007) Population genetics of the frog-killing fungus Batrachochytrium dendrobatidis. Proceedings of the National Academy of Sciences of the United States of America, 104, 13845–13850. Murray, K.A., Retallick, R.W.R., Puschendorf, R., Skerratt, L.F., Rosauer, D., McCallum, H.I., Berger, L., Speare, R. & VanDerWal, J. (2011) Assessing spatial patterns of disease risk to biodiversity: implications for the management of the amphibian pathogen, Batrachochytrium dendrobatidis. Journal of Applied Ecology, 48, 163–173. Pearson, R.G. & Dawson, T.P. (2003) Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology and Biogeography, 12, 361–371. Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006) Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190, 231–259. Raffel, T.R., Rohr, J.R., Kiesecker, J.M. & Hudson, P.J. (2006) Negative effects of changing temperature on amphibian immunity under field conditions. Functional Ecology, 20, 819–828. Raisanen, J. (2002) CO2-induced changes in interannual temperature and precipitation variability in 19 CMIP2 experiments. Journal of Climate, 15, 2395–2411. Rohr, J.R. & Madison, D.M. (2003) Dryness increases predation risk in efts: support for an amphibian decline hypothesis. Oecologia, 135, 657–664. Rohr, J.R. & Raffel, T.R. (2010) Linking global climate and temperature variability to widespread amphibian declines putatively caused by disease. Proceedings of the National Academy of Sciences of the United States of America, 107, 8269–8274. Rohr, J.R., Raffel, T.R., Romansic, J.M., McCallum, H. & Hudson, P.J. (2008) Evaluating the links between climate, disease spread, and amphibian declines. Proceedings of the National Academy of Sciences of the United States of America, 105, 17436–17441. Segurado, P., Araujo, M.B. & Kunin, W.E. (2006) Consequences of spatial autocorrelation for niche-based models. Journal of Applied Ecology, 43, 433– 444. Skerratt, L.F., Berger, L., Speare, R., Cashins, S., McDonald, K.R., Phillott, A.D., Hines, H.B. & Kenyon, N. (2007) Spread of chytridiomycosis has caused the rapid global decline and extinction of frogs. EcoHealth, 4, 125–134. Stuart, S.N., Chanson, J.S., Cox, N.A., Young, B.E., Rodrigues, A.S.L., Fischman, D.L. & Waller, R.W. (2004) Status and trends of amphibian declines and extinctions worldwide. Science, 306, 1783–1786. Wake, D.B. & Vredenburg, V.T. (2008) Are we in the midst of the sixth mass extinction? A view from the world of amphibians. Proceedings of the National Academy of Sciences of the United States of America, 105, 11466– 11473. Received 10 September 2010; accepted 29 September 2010 Handling Editor: Marc Cadotte



doi: 10.1111/j.1365-2664.2010.01893.x

Optimizing search strategies for invasive pests: learn before you leap Peter W.J. Baxter* and Hugh P. Possingham The University of Queensland, The Ecology Centre and Centre for Applied Environmental Decision Analysis, School of Biological Sciences, St. Lucia, Qld 4072, Australia

Summary 1. Strategic searching for invasive pests presents a formidable challenge for conservation managers. Limited funding can necessitate choosing between surveying many sites cursorily, or focussing intensively on fewer sites. While existing knowledge may help to target more likely sites, e.g. with species distribution models (maps), this knowledge is not flawless and improving it also requires management investment. 2. In a rare example of trading-off action against knowledge gain, we combine search coverage and accuracy, and its future improvement, within a single optimization framework. More specifically we examine under which circumstances managers should adopt one of two search-and-control strategies (cursory or focussed), and when they should divert funding to improving knowledge, making better predictive maps that benefit future searches. 3. We use a family of Receiver Operating Characteristic curves to reflect the quality of maps that direct search efforts. We demonstrate our framework by linking these to a logistic model of invasive spread such as that for the red imported fire ant Solenopsis invicta in south-east Queensland, Australia. 4. Cursory widespread searching is only optimal if the pest is already widespread or knowledge is poor, otherwise focussed searching exploiting the map is preferable. For longer management timeframes, eradication is more likely if funds are initially devoted to improving knowledge, even if this results in a short-term explosion of the pest population. 5. Synthesis and applications. By combining trade-offs between knowledge acquisition and utilization, managers can better focus – and justify – their spending to achieve optimal results in invasive control efforts. This framework can improve the efficiency of any ecological management that relies on predicting occurrence. Key-words: adaptive management, containment, eradication, invasive species, optimal management, receiver operating characteristic (ROC) curve, species distribution models, statedependent management, stochastic dynamic programming (SDP), value of information

Introduction Invasive species comprise one of the main threats to global biodiversity (Sala et al. 2000) and their annual economic impact is substantial (Pimentel et al. 2001). While considerable economic resources can be allocated to invasive species management, it is important to strategise spending in a coherent decision-making framework, to maintain cost-efficiency as well as increase the likelihood of programme success (Regan et al. 2006; Bogich & Shea 2008). Such a framework should ideally take account of all economic factors in the programme, includ-

*Correspondence author. E-mail: [email protected]

ing investing in knowledge acquisition to improve future management. This notion of learning now, to make better decisions later, underpins adaptive management (Walters 1986) and theories of learning in animal behaviour (Stephens & Krebs 1986). In this paper we investigate how best to allocate a restricted budget among options for research and control of an invasive pest when we have some information about its distribution, as well as the ability to improve that information. For any detection-and-control programme constrained by time, budget and human resources, trade-offs exist between the different search strategies and the acquisition of information to inform future searches. Therefore managers are confronted with the following questions: How many sites should we search, which ones, and at what intensity? Should we invest


Optimizing search strategies for invasives 87 resources in refining our methods for choosing sites, to improve future search success? The first questions have been addressed to some extent in the literature. For example, controlling new satellite populations may be preferable to reducing the density of the core pest population (Moody & Mack 1988), although this may not be optimal when costs are taken into account (Whittle, Lenhart & Gross 2007). For pest metapopulations, it may be optimal to attempt eradication of mediumdensity sub-populations, while still balancing colonization from, and containment of, high-density patches (Bogich & Shea 2008). With enough money, however, prioritizing highdensity patches can become optimal (Taylor & Hastings 2004). Within a patch, detection probability depends on both the search method and search intensity or coverage, which in turn depend on budget (Cacho, Hester & Spring 2007). The tradeoff between cursory and focussed searching has not yet been examined in the context of (mis-) information about the distribution of the species. Predicting a pest’s distribution and spread often involves a species distribution model of its likely occurrence (e.g. Baret et al. 2006; Hauser & McCarthy 2009), allowing some degree of strategic searching within a region. The predictive accuracy of species distribution models is commonly assessed using Receiver Operating Characteristic (ROC) curves (Pearce & Ferrier 2000; Wintle, Elith & Potts 2005; Latimer et al. 2009), which graph the rates of occurrence of true vs. false positives. In species distribution models (henceforth, ‘maps’) this translates as rates of classifying occupied sites as being occupied, vs. rates of misclassifying unoccupied sites as occupied, implying when searches would be worthwhile or futile, respectively. As maps can be improved at some cost, namely redirecting funds from active control (Murray et al. 2009), optimal investment in research can be determined if we know the future benefit of having better predictive maps that will result from the short term reduction in on-ground effort. The trade-off of knowledge gain vs. immediate action is implicit in every area of applied ecology. Despite the practical benefits and broader implications of exploring such a trade-off, however, this has not yet been done in a theoretical or practical framework. Although some cost-benefit analysis has been applied to the use of ROC curves in clinical settings (Swets & Pickett 1982; Metz 1986), this has focussed on the direct costs of treating false-positive vs. true-positive diagnoses rather than deferring treatments while diagnostic tests are improved. As no map can predict a pest’s distribution with 100% precision, some unoccupied sites will inevitably be searched (Vaćlavı´ k & Meentemeyer 2009). As the overall search area increases we expect to reach more of the occupied sites, but we also experience a concomitant increase in the proportion of empty sites that are searched. In the extreme case, exhaustive searching entails looking for the pest throughout the entire area of unsuitable habitat as well as in all the more likely sites. Therefore it may be better to select fewer sites and increase the search effort at each site to increase the probability of detecting the pest, similar to intensive vs. extensive search modes of animal foraging behaviour (Fortin 2002). The search strategy for an invasive species will ideally incorporate some idea of how

accurate the distribution predictions are, and therefore what proportion of searches are likely to be futile due to an incorrect choice of site. Consequently it may even be beneficial to redirect resources to refining our knowledge of the organism’s expected distribution, to better identify candidate sites for future searches. We address these issues here by optimizing the trade-offs among widespread and more focussed search areas (allowing low and high search intensity per site, respectively), against knowledge acquisition to improve future searches. We use stochastic dynamic programming (SDP), a procedure that identifies optimal strategies by considering the possible changes in the states of a system over time (Bellman 1957). We also compare, by simulation under different assumptions, the relative performances of the SDP recommendations and alternative management strategies. SDP is commonly applied in behavioural ecology (Mangel & Clark 1988; McNamara & Houston 1996), including examining when foraging organisms should learn about resource distribution by moving between patches (Eliassen et al. 2009). It is being used increasingly to solve state-dependent management problems in ecology: choosing between fire management options (Richards, Possingham & Tizard 1999; McCarthy, Possingham & Gill 2001), how to allocate management effort within or among sites (Baxter et al. 2007; McDonald-Madden, Baxter & Possingham 2008), and when to cease management or monitoring altogether (Regan et al. 2006; Chade`s et al. 2008). We frame our analysis around the invasion of red imported fire ants Solenopsis invicta Buren in the Brisbane, Queensland, region, dating from February 2001 (Jennings & McCubbin 2004). Native to South America, their establishment as an invasive alien species is greatly facilitated by anthropogenic disturbance (King & Tschinkel 2008), and they are capable of considerable environmental, social and economic damage (Williams 1994; Callcott & Collins 1996). The fire ant incursion into Australia is therefore potentially very serious, given extensive suitable habitat (Moloney & Vanderwoude 2002; Sutherst & Maywald 2005), and the invasion has been listed as a Key Threatening Process under the 1999 Commonwealth Environment Protection and Biodiversity Conservation Act. The Queensland invasion dynamics have been modelled previously (Scanlan & Vanderwoude 2006), allowing reasonable biological parameterisation for our purposes. In the absence of detailed management cost data, however, we keep our approach general and demonstrate the method in a form broadly applicable to invasive species management (and indeed to applied ecology in general), rather than presenting a specific case analysis of the fire-ant invasion. This novel approach shows how current and future prediction capability should affect current and future search strategies to optimize invasive species control and planning.

Materials and methods In order to model the maps’ predictive quality, we use a family of ROC curves. In practice, ROC curves can take any shape between (0,0) and (1,1), with one measure of map quality being the area under


88 P. W. J. Baxter & H. P. Possingham the curve (AUC); the closer AUC gets to one, the better the map (Swets 1988; but see Lobo, Jimeńez-Valverde & Real 2008). We assume a family of ROC curves given by y ¼ x1=a

eqn 1

(Fig. 1a; after Swets 1986), and bounded at (0,0) and (1,1) as required. For this equation, the area under the curve takes a simple form, AUC = a ⁄ (a + 1), which asymptotically approaches one as a increases. Therefore a higher value of a implies a more reliable map (a = 1 essentially implies no knowledge and random searches).

SEARCH STRATEGY AND SUCCESS

We assume that a pest is present at some density / within a region of area A (we list symbols and parameters in Table 1). In a map, A can be measured as the number of cells in a grid covering the region (i.e.

number of sites in which the pest may potentially occur), and / as the proportion of those cells that are infested. The map directing our searches will produce sites labelled as occupied, either correctly (in eqn 1, proportion y of occupied sites) or incorrectly (proportion x of empty sites). When we use the map to choose sA sites to search (a proportion s of the region), eqn 1 allows us to express this in terms of the sites searched that are either occupied (/Ay) or unoccupied ([1–/]Ax): sA ¼ /Ay þ ð1 /ÞAx ) s ¼ /y þ ð1 /Þya :

The value of s increases with y, A and / (Fig. 1b). We can use eqn 2 to find the proportion of a region, s1, needed to be searched in order to visit some proportion y1 of the occupied sites. Alternatively we could search fewer sites (proportion s2 < s1), spending longer in each, giving:

0·8

800

Number of sites to search

(b) 1000

Proportion of true positives

(a) 1·0

0·6

0·4 a = 25 a = 10 a= 5 a= 2 a= 1

0·2

0

0

0·2

0·4

0·6

0·8

600

400

200

0

1·0

φ = 0·1; a = 2 φ = 0·1; a = 5 φ = 0·1; a = 8 φ = 0·3; a = 2 φ = 0·3; a = 5 φ = 0·3; a = 8 φ = 0·5; a = 2 φ = 0·5; a = 5 φ = 0·5; a = 8

0

0·2

Search success; a = 2

Expected no· colonies found

Expected no· colonies found

200

150

100

50

0

0

0·2

0·4

0·6

0·6

0·8

1·0

Search success; a = 8 (d) 300

$50 000; cursory $100 000; cursory $200 000; cursory $50 000; focussed $100 000; focussed $200 000; focussed

250

0·4

Proportion of true positives required

Proportion of false positives

(c) 300

eqn 2

0·8

Regional pest density (φ)

1·0

$50 000; cursory $100 000; cursory $200 000; cursory $50 000; focussed $100 000; focussed $200 000; focussed

250

200

150

100

50

0

0

0·2

0·4

0·6

0·8

1·0

Regional pest density (φ)

Fig. 1. Relationships underpinning the framework of active searching vs. knowledge acquisition. (a) Family of theoretical Receiver Operating Characteristic curves, described by y = x1 ⁄ a, where higher values of a reflect better predictive capability. Two example search strategies are indicated for a habitat map quality of a = 5. The cursory-widespread strategy (‘+’) attempts to find 95% of the infestation (380 sites) and the focussed strategy (‘o’), visits half that number, allowing double the search time per site. (b) The total number of sites (empty and occupied) required to search, to visit given proportions of occupied sites, depending on regional infestation density, /, and quality of habitat map, a. (c, d) Expected number of colonies detected in one time step as a function of regional pest density, for three different budgets and the two search strategies (cursory and focussed). Two levels of map quality are shown, (c) a = 2 and (d) a = 8. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 86–95

Optimizing search strategies for invasives 89 Table 1. Parameters used, with their symbols and values (where applicable; otherwise indicated as a variable or a function f(…) of other parameters)

Symbol

Description

Value

/ /0 kmax kd ku A B Di S95 a a0 d k s v x

proportion of region infested initial proportion of region infested in simulations maximum rate of invasion spread rate of spread due to detected colonies rate of spread due to uncontrolled colonies area of region (#grid cells in species distribution model) budget per time-step expected proportion of colonies detected with action i value of s giving 95% worthwhile searches ROC curve exponent initial ROC curve exponent in simulations probability of detecting a colony if present search effort at which d = 0Æ5 proportion of region to be searched search effort per site ($) proportion of all futile sites searched (‘false positive proportion’) proportion of all worthwhile sites searched (‘true positive proportion’)

variable 0Æ01 1Æ19 f(/), £ 1Æ09 f(/), £ 1Æ19 1000 $100,000 f(/,A,B,k,si,yi) f(/,a) variable 2 f(v,k) $500 f(y,/,a) f(A,B,s) 0–1

y

0–1, f(x,a)

ROC, Receiver Operating Characteristic.

s1 ¼ /y1 þ ð1 /Þy1 a and s2 ¼ /y2 þ ð1 /Þy2 a ;

d ¼ v=ðk þ vÞ; eqn 3

with y1 > y2 when s1 > s2 (Fig. 1b). For example we can set s1 = S95, which we define as the proportion of sites needed to include 95% of all occupied sites (y1 = 0Æ95): s1 ¼ S95 ¼ 095/ þ ð1 /Þ095a

eqn 4

(Fig. 1a, ‘+’). If we want to find the ‘hit’ rate y2 for searching half those sites we can use the equation s2 ¼ 05S95 ¼ /y2 þ ð1 /Þy2 a ;

eqn 5

and (knowing S95 and estimating a and /) we can find y2 numerically (Fig. 1a, ‘o’). For example, for a regional infestation density of / = 0Æ4, and map quality a = 5, we get S95 = 0Æ844, and y2 = 0Æ735 (giving s2 = 0Æ422 = 0Æ5S95). This means that, for an area of 1000 sites, in order to search in 380 sites that are occupied (95% of the 400 infestations), we need to search in 844 sites in total, as our map will misdirect us to 464 sites (so 55% of searches are futile – still better than random, which would lead to 60% of searches being futile). By halving the number of sites searched (422), our map would send us to 0Æ4(0Æ735) = 294 occupied sites, and leave us with 128 futile searches (30% of all searches). Therefore when we employ the map, reducing the total search coverage also reduces the proportion of searches that are futile. Furthermore, searching half the sites allows us to double our per-site search effort, increasing the detection probability at occupied sites. Nonetheless, the increase in proportion of worthwhile searches and detection probability must be traded off against the reduction in overall search coverage.

PROBABILITY OF DETECTION

Detecting a pest at an occupied site is more likely if we expend more effort searching the site. Assume the probability of pest detection (conditional on its presence), d, is a saturating function of search effort (v):

eqn 6

where we have a 50% chance of finding an infestation if we search with effort v = k. Search effort could reflect, for example, time spent at a site, number of fieldworkers and different search methods used. It is convenient therefore to measure effort in terms of its total cost, which gives the effort per site searched as v ¼ B=sA;

eqn 7

where the budget for the management period is $B. Combining eqns 6 and 7 gives the detection probability in an occupied site as a function of budget, proportion of areas searched, and the area of the region: d ¼ B=ðksA þ BÞ:

eqn 8

IMPROVING THE SPECIES DISTRIBUTION MODEL

Another possible strategy is to defer searching to concentrate resources on improving the predictive accuracy of the map (increasing a in eqn 1). These improvements could come from updating habitatpredictive algorithms, or acquiring more or better environmental data relating to pest habitat, including development of novel techniques to do so.

DECISION TRADE-OFFS

In this example, therefore, we choose from three actions (i) at each time step: i = 1: search proportion s1 = S95 of the region, visiting 95% of all occupied sites, 0Æ95/A. In each we have a probability B ⁄ (ks1A + B) of detecting the pest, but we also search A(1 – /)0Æ95a sites in vain, and leave (1 – S95)A sites unsearched; i = 2: search proportion s2 = 0Æ5S95 of the region, which gives us /y2A occupied sites in which we have a probability B ⁄ (ks2A + B) of detecting the pest. We will also search ð1 /Þya2 A empty sites and leave (1 – 0Æ5S95)A sites unsearched; or i = 3: postpone searching (s3 = 0) and develop a better map (increase a).


90 P. W. J. Baxter & H. P. Possingham For the first two options, the expected proportion of colonies detected and destroyed is Di ¼

/yi AB ksi A þ B

i ¼ 1; 2

eqn 9 PERFORMANCE EVALUATION

(Fig. 1 c,d). For the third option (i = 3) we assume that all /A colonies remain undetected (D3 = 0).

PARAMETERIZATION AND OPTIMIZATION

Cost parameters For purposes of illustration, we assume that the map covers A = 1000 sites, or grid-cells, each of which may or may not contain the pest species; that the budget per time step is B = $100 000; and that the cost of the effort required to have 50% chance of detecting a pest present on an infested site is k = $500. These values give a detection probability per occupied site of d = 0Æ17, if all sites are searched, increasing to d = 0Æ67 if only 10% of the region is searched.

The spread of the infestation will depend on both the organism’s biology and search-and-control success. In this example we use simplified fire ant population dynamics to demonstrate our approach. Scanlan & Vanderwoude (2006) modelled the spread of fire ants in Australia at two spatial scales, and assumed that invasion extent doubled every 2–4 years when measured at the broader scale (10 000-km2 blocks), with faster dynamics at local scales. We compromise between these two scales and assume a maximum doubling period, in the absence of control or density-dependence, of 24 months. Thus, in a 6-month management period, the maximum rate of spread is kmax = 2(6 ⁄ 24) 1Æ19. We assume that the increase in regional density follows logistic growth (Scanlan & Vanderwoude 2006; see also Shryock et al. 2008), giving a rate of increase of ku ð/Þ ¼ 1 þ ðkmax 1Þð/ 1Þ=1

To test the performance of employing the optimization (SDP) results vs. three simpler management regimes, we simulated 20-year management of a pest invasion with dynamics as above (Table 1), beginning at 1% regional infestation; this level would in practice be dependent on both timeliness of detection and the spatial resolution of the map. We assumed an initial knowledge level of a = 2, reflecting a reasonable lower-end AUC value (Latimer et al. 2009; Vaćlavı´ k & Meentemeyer 2009). The alternative management regimes were based on those used for the SDP formulation and comprise: always search S95 sites; always search S95 ⁄ 2 sites (doubling effort-per-site); or rotate between search and learning modes. The ‘rotating strategy’ iteratively followed the sequence: widespread control (S95) - upgrade map (increase a) - focussed control (S95 ⁄ 2).

Results

Invasion parameters

eqn 10

if colonies are uncontrolled (the denominator of 1 indicates the ‘carrying capacity’ of / = 1 in a fully colonized area). Even if a colony is detected and destroyed, it may already have reproduced. Assuming that colonies are discovered on average halfway through their reproductive cycle, the rate of increase of detected-and-removed colonies is kd ð/Þ ¼ ku ð/Þ1=2 :

(Fraser et al. 2006), avoided costs of ongoing management, or even societal values).

eqn 11

For ku = 1Æ19 at its maximum value, this implies that at most 9% of detected ant colonies will have spread prior to their destruction (1Æ19½ 1Æ09). For the optimization (Appendix S1, Supporting Information), we describe the system state by the combination of map quality (a in eqn 1) and regional pest density /. The system undergoes transitions between states with probabilities governed by the outcome of each management option. Expected future pest density depends on the rates of spread from controlled and uncontrolled sites, and we assume that learning improves the map quality a by one (with probability 0Æ2) or two (probability 0Æ8) units, giving diminishing increases in AUC. We accord a ‘utility’ value to each state depending on the pest density only (map quality has no utility apart from improving future searches): utility increases linearly as pest density decreases, with a 100-fold bonus if the pest is eradicated (this high bonus could reflect renewed access to export markets in the case of agricultural pests

STATE-DEPENDENT OPTIMIZATION

The optimal strategies for learning about and controlling an invasion (i.e. the SDP solution; Fig. 2) depend on the system state (pest infestation density / and map quality a), and on the management time horizon T. We first consider long-term management recommendations (e.g. more than a decade; Fig. 2a–c). For the longest management horizon considered, 20 years, improving the predictive quality of the map initially takes precedence over either search method for most of the system states. Exceptions are when our map is already excellent (a > 15, or AUC > 0Æ9375; not unrealistically high values; Zurell et al. 2009) or when the infestation is at moderate densities. If the infestation is at very low densities we can afford to delay searching until we have a better map to improve targeting of future searches. On the other hand, if the infestation is widespread then searching with a restricted budget will have little effect on pest density and so we again delay searching until we have a better map. If we strive for shorter-term success (Fig. 2d–f) the value of improving the map diminishes and the optimal strategy is usually immediate search-and-control. Generally, if the pest is already widespread, then cursory widespread searching is optimal (i = 1 above), as even undirected searches will have high success. If the pest is at lower densities then we should use targeted site-intensive searching (i = 2). This strategy depends on having a reasonable-quality map to exploit: if the map is very poor, we should still use cursory searches (close in effect to random searches: a 1). Nonetheless, at all but the shortest management timeframes, it is always recommended to improve a very poor map if the infestation is still at low levels (e.g. / = 0Æ01; bottom-left of Fig. 2a–e).

PERFORMANCE EVALUATION

As expected, the state-dependent optimization performed better in our simulations than the alternative strategies (Fig. 3a,


Optimizing search strategies for invasives 91 (c)

15-year horizon

25

25

20

20

20

15

15

15

10

10

10

5

5

5

0

0·5

1

(d)

5-year horizon

0

0·5

(e)

2-year horizon

1

0·5

(f)

6-month horizon

25

25

20

20

20

15

15

15

10

10

10

5

5

5

0

0·5

1

0

0·5

Pest density (φ)

10-year horizon

0

25

(a)

1

0

Pest density (φ)

0·5

1

1

Pest density (φ)

(b)

0·035

Cursory Focussed Optimal Rotating

0·030 0·025 0·020 0·015 0·010 0·005 0·000

(b)

20-year horizon

0

5

10

15

20

Time (years)

Mean model quality (a)

Mean infestation density (φ)

Fig. 2. Optimal strategies depending on regional infestation density (/) and current quality of species distribution model (a), over a selection of management time horizons. The optimal strategies for each (/, a) state are indicated as: white = cursory widespread searching (s1 = S95); grey = fewer, more focussed and intensive searches (s2 = S95 ⁄ 2; v2 = 2v1); black = re-direct funding towards improving the species distribution model.

Map quality (a)

Map quality (a)

(a) 25

30

Optimal Rotating

25 20 15 10 5 0

0

5

10

15

20

Time (years)

Fig. 3. Simulated performance of pest management over 20 years under default model assumptions. (a) Comparison of four management strategies: cursory widespread searches; intensive focussed searches; optimal state-dependent strategy recommended by stochastic dynamic programming; and continual rotating between cursory search, model-improvement and focussed search. (b) Acquisition of knowledge when the optimal and rotational strategies are implemented, showing mean (±SD) values of the Receiver Operating Characteristic curve exponent a. The two non-learning strategies (cursory and focussed searching) remain at the initial level of a = 2 (dotted line).

mean trajectories shown), usually achieving eradication by year 20. This eventual success is dependent on tolerating an initial increase in pest density as funds are initially devoted to improving the map (Fig. 3b). Other strategies avoid the initial ‘spike’ in pest density but fail to achieve eradication over the long-term. The cursory-widespread search strategy performs worst, with the site occupancy steadily increasing. The ‘focussed’ strategy results in a steady, but slow, decrease in density. The ‘rotating’ strategy allows some map improvement as well as search-and-control efforts, and thus performs comparatively well. Nonetheless, in terms of achieving eradication the optimal strategy performed considerably better, eradicating the pest in 97% of simulations (compared to 59% for the rotating strategy, and never for the two non-learning strategies). We also investigated departures from our default assumptions and parameter values (Fig. 4). Increasing the budget by 50% (Fig. 4a) improved the performance of all strategies (unsurprisingly), with most strategies achieving high levels of

suppression. Starting with a better map (a = 5; Fig. 4b) also led to improvements in all strategies, particularly the ‘focussed’ strategy, which particularly depends on reliable site selection for its success. We also examined our assumption that research increases the value of a by 2 units with probability pa = 0Æ8, however, this appeared to have little overall effect on results (Figs 4c,d). The spike in pest density when the optimization solution is applied decreases with pa, because managers will be less willing to allow temporary population explosions if the scope for improving the map is reduced, and so switch to search ⁄ control operations sooner. Similar effects resulted from eliminating the eradication bonus (Fig. 4e) or incorporating a 3% discount rate (Fig. 4f): these scenarios reduce the emphasis on eradication vs. containment (Odom et al. 2003; Fraser et al. 2006) either in the longer-term (discounting) or permanently (no bonus), leading to decreased incentive to improve knowledge. All strategies had more success controlling slower invasions (Fig. 4g), while rapidly-spreading invasions (Fig. 4h) were only able to be suppressed by the optimal and rotating


Larger management budget (a)

0·035

Cursory Focussed Optimal Rotating

0·030 0·025 0·020 0·015 0·010 0·005 0·000 0

5

10

15

20



92 P. W. J. Baxter & H. P. Possingham Better initial knowledge (b)

0·035 0·030 0·025 0·020 0·015 0·010 0·005

0·000 0

5

0·030 0·025 0.020 0·015 0·010 0·005 5

10

15

20



Moderate knowledge gain (c)

0·035

0·000 0

0·030 0·025 0·020 0·015 0·010 0·005 10

15

20



No eradication bonus (e)

5

0·025 0·020 0·015 0·010 0·005 5

10

15

20



Slower lnvasion speed (g)

0

20

15

20

0·030 0·025 0·020 0·015 0·010 0·005 0·000 0

5

10

Discounting of benefits (f)

0·035 0·030 0·025 0·020 0·015 0·010 0·005 0·000

0

5

10

Time (years)

0·030

0·000

15

Slower knowledge gain (d)

Time (years)

0·035

20

Time (years)

0·035

0

15

0·035

Time (years)

0·000

10

Time (years)

Time (years)

0·09

Faster lnvasion speed (h)

0·08 0·07 0·06 0·05 0·04 0·03 0·02 0·01 0·00 0

5

Time (years)

strategies. Overall, the ranking of performances was robust to changes in assumptions (including others not shown; e.g. higher initial pest density). The tendency of the SDP solution to allocate approximately 3–4 years’ initial funding to map improvement was similarly consistent.

Discussion We have introduced and demonstrated an approach for trading-off actions that search for and remove a pest, against an action that only gains knowledge. Our optimizations indicate

10

Time (years)

15

20

Fig. 4. Mean performances of four management strategies (as Fig. 3a) under alternative parameter values and assumptions. (a) Budget per time-step of $150 000; (b) initial good quality species distribution model (a = 5); (c) 50% and (d) 20% probability of increasing a by 2 units (otherwise a increases by 1 unit); (e) no eradication bonus given; (f) applying a 3% discount rate to the performance benefits; and maximum doubling time of invasion set to (g) 30 months (kmax = 0Æ15) and (h) 18 months (kmax = 1Æ26; note different vertical scale).

that spending time improving knowledge about the pest’s habitat preferences, before searching for it, is optimal. We were surprised to discover this, as deciding to improve knowledge while a pest incursion grows exponentially seems like fiddling while Rome burns. Nevertheless this highlights the value of learning even when at the expense of control operations under seemingly urgent conditions. With a long-term perspective, it is optimal to learn rather than take direct action at the start of an invasion (Fig. 2), but exactly for how long we should delay action and learn depends on many factors. These include the cost of learning relative to


Optimizing search strategies for invasives 93 on-ground search and control operations, the desire for eradication vs. containment, the likely improvement in our knowledge of habitat preferences and the biological characteristics of the pest species (Fig. 4). Once the initial learning phase is over, searching should be focussed and intensive, rather than widespread and cursory, when we have less widespread infestation or intermediate predictive capabilities (Fig. 2d–f). Simulation of alternative plausible management strategies confirmed the expected superiority of the optimization approach (Fig. 4). The overall similarity of trajectories under different assumptions also highlighted that at the initial stages of pest incursion (when / is still small), allowing some spread of the infestation may be acceptable so long as we can improve our map to better predict and control the pest in future. This is partly because our management objective gives more weight to eventual eradication than to merely suppressing the population (cf. Fig. 4e), and increased knowledge is ultimately required to improve searching for and removing the invader. This result raises the interesting question of whether the best emergency response to a new incursion is actually to ‘do nothing’ – take no immediate direct action but concentrate funds into developing high-quality predictive maps to maximize efficacy of future management. By taking a long-term focus (and deferring control), it could be argued that managers are being more pragmatic, trading off the apparent urgency of a new incursion against the strategic allocation of resources to knowledge acquisition and better chance of success further in the future. This however must also be evaluated in the light of the structure and assumptions presented here. For example, the framework can be expanded to include many different strategies (e.g. more search areas si and their corresponding search intensities). More refined budgetary options in the optimization could allow both searching and habitat modelling simultaneously by selecting the proportion of funds allocated to learning vs. control, instead of the all-or-nothing choice presented here. Alternatively some funding could be allocated to improving detection probability at occupied sites by developing enhanced on-site search techniques. Pursuing this research will provide further insights into knowledge ⁄ action trade-offs. The optimization (Fig. 2) indicates cases where we should search and control straight away rather than improving a map. The most obvious case is when we already have a reasonable quality map. The simulation results indicate that the switch from knowledge acquisition to active control takes place after about 3–4 years’ research (average map quality of a = 16Æ4, AUC = 0Æ94), but of course this depends on other factors such as the severity of the incursion and the management time horizon. Nonetheless, the shorter the management timeframe, the less the relative merits of acquiring additional information and the more likely control action will be taken. This diminishing return on investment in data acquisition has recently been shown for conservation of South African fynbos flora (Grantham et al. 2009), in which case optimal decisions on choosing patches to reserve could be made after a relatively small initial data-gathering effort. Other cases demanding quicker action are when the pest population spreads quickly, or the management timeframe imposes too tight a deadline to be able to act

on the research results. These two aspects reflect increased urgency in countering the invasion, either in terms of spatial spread or the wish to produce positive outcomes quickly. Another factor to favour immediate action is having a large budget, which allows greater search effort and better success even with poor predictive ability. The size of the budget may itself reflect an urgent desire to control an invasion. This was the case in the Queensland fire-ant incursion, for which a large budget was available from the outset (AU$123 million; George 2007). Such well-funded programmes allow us the luxury of immediate action as well as simultaneous production of a predictive map; however many conservation efforts operate on much tighter budgets, making trade-offs between knowledge gain and control efforts unavoidable. Even in the fire-ant case success has proven elusive despite more than AU$200 million invested (Williams 2010). While we deliberately used simple models to demonstrate the learn-or-act trade-off, the assumptions made in developing our framework should be noted, particularly if being applied to a real-world situation. For example, we have assumed that diversion of funds into research will definitely have a (measureable) positive outcome, that a will increase precisely by either 1 or 2 (with probability 0Æ2 and 0Æ8 respectively); while altering the relative values of these probabilities had little effect on simulated performance (Fig. 4c,d), the possibilities of no map improvement, or even perverse disimprovement, were not considered. We have assumed that this improvement costs the same per time-step as searching for the pest; the actual costs and benefits of map improvement would need to be estimated based on acquiring suitable personnel and infrastructure (software, data layers etc.), and anticipating the projected map improvements (reduction in level of false positives with an improved map) – these estimates, while uncertain, may provide sufficient insight into whether greater weight should be given to searching or knowledge improvement. The current value of a could be estimated from the map’s performance in predicting the species’ native range (with caution), or from previous search results if available. We have disregarded the spatio-temporal dynamics of the infestation. In reality dispersing ants may not find all suitable available habitat in which to found a new colony, so our assumptions also (conservatively) overestimate doubling speed. Just as a searched site may be ideal habitat, but not yet colonized by a dispersing pest, a site deemed to be of marginal suitability may become colonized if the surrounding area is already saturated with colonies. We have also assumed that the map extends to all possible areas of spread – judgement may be required to trade its spatial extent off against sufficient grain to provide meaningful information on finer scales. We could use more sophisticated optimization methods to address some of our assumptions. D’Evelyn et al. (2008) demonstrate the value of incorporating search results directly into estimates of pest population density, in order to choose the optimal effort of control in later years, emphasizing, as here, the value of early learning (in their case via control). Our optimization is dependent on both infestation density and map accuracy, both of which may be imprecisely known (Vaćlavı´ k


94 P. W. J. Baxter & H. P. Possingham & Meentemeyer 2009). We could therefore re-express the problem in terms of our belief of what the current values of / and a are, in a partially-observable Markov decision process (POMDP), to optimize our future ‘belief state’ rather than the actual, but unknowable, state of the system (for ecological examples see Lane 1989; White 2005; Chade`s et al. 2008). In terms of economic simplifications, we have ignored the costs of travelling between sites; so that the selection of many sites to search may incur extra costs if extensive travel is involved. McDonald-Madden, Baxter & Possingham (2008) demonstrate a succinct approach to this problem when the locations of the populations are known. We have also assumed that the cost of pest removal, once discovered in a site, is negligible, or integrated into the search costs. Similarly we ignore the possibility of multiple pest occurrences on one site which would influence both the search success and removal costs; or other aspects of spatial contagion that also affect costs and strategy performance. We have made these assumptions and simplifications so that we could get to the heart of the act-or-learn problem in invasion management; some of these assumptions will be relaxed in future work. Nonetheless, striking results emerge, particularly the consistent recommendation that learning first, and looking (more successfully) later on, is the long-term optimal approach to new pest incursions.

Acknowledgements We thank Iadine Chade`s, Hedley Grantham, Cindy Hauser, Dane Panetta, the editor and two anonymous reviewers for helpful discussions and comments. Financial support was provided by the Australian Centre of Excellence for Risk Analysis (ACERA) and the Applied Environmental Decision Analysis (AEDA) research hub, a Commonwealth Environmental Research Facility.

References Baret, S., Rouget, M., Richardson, D.M., Lavergne, C., Egoh, B., Dupont, J. & Strasberg, D. (2006) Current distribution and potential extent of the most invasive alien plant species on La Reúnion (Indian Ocean, Mascarene islands). Austral Ecology, 31, 747–758. Baxter, P.W.J., Wilcox, C., McCarthy, M.A. & Possingham, H.P. (2007) Optimal management of an annual weed: a stochastic dynamic programming approach. MODSIM 2007 International Congress on Modelling and Simulation (eds L. Oxley & D. Kulasiri), pp. 2223–2229. Modelling and Simulation Society of Australia and New Zealand, http://www.mssanz.org.au/MOD SIM07/papers/41_s34/OptimalManagement_s34_Baxter_.pdf. Bellman, R.E. (1957) Dynamic Programming. Princeton University Press, Princeton, NJ. Bogich, T. & Shea, K. (2008) A state-dependent model for the optimal management of an invasive metapopulation. Ecological Applications, 18, 748–761. Cacho, O.J., Hester, S. & Spring, D. (2007) Applying search theory to determine the feasibility of eradicating an invasive population in natural environments. Australian Journal of Agricultural and Resource Economics, 51, 425–443. Callcott, A.M.A. & Collins, H.L. (1996) Invasion and range expansion of imported fire ants (Hymenoptera: Formicidae) in North America from 19181995. Florida Entomologist, 79, 240–251. Chade`s, I., McDonald-Madden, E., McCarthy, M.A., Wintle, B., Linkie, M. & Possingham, H.P. (2008) When to stop managing or surveying cryptic threatened species. Proceedings of the National Academy of Sciences, 105, 13936– 13940. D’Evelyn, S.T., Tarui, N., Burnett, K. & Roumasset, J.A. (2008) Learningby-catching: Uncertain invasive-species populations and the value of information. Journal of Environmental Management, 89, 284–292.

Eliassen, S., Jørgensen, C., Mangel, M. & Giske, J. (2009) Quantifying the adaptive value of learning in foraging behavior. The American Naturalist, 174, 478–489. Fortin, D. (2002) Optimal searching behaviour: the value of sampling information. Ecological Modelling, 153, 279–290. Fraser, R.W., Cook, D.C., Mumford, J.D., Wilby, A. & Waage, J.K. (2006) Managing outbreaks of invasive species: eradication versus suppression. International Journal of Pest Management, 52, 261–268. George, C.R.R. (2007) Establishment risk modelling of invasive organisms: Solenopsis invicta as a case study. PhD Thesis, University of Queensland, Brisbane. Grantham, H.S., Wilson, K.A., Moilanen, A., Rebelo, T. & Possingham, H.P. (2009) Delaying conservation actions for improved knowledge: how long should we wait? Ecology Letters, 12, 293–301. Hauser, C.E. & McCarthy, M.A. (2009) Streamlining ‘search and destroy’: cost-effective surveillance for invasive species management. Ecology Letters, 12, 683–692. Jennings, C. & McCubbin, K. (2004) The National Red Imported Fire Ant Eradication Program Overview – October 2004. Fire Ant Control Centre, Queensland Department of Primary Industries, Brisbane, Australia. King, J.R. & Tschinkel, W.R. (2008) Experimental evidence that human impacts drive fire ant invasions and ecological change. Proceedings of the National Academy of Sciences, 105, 20339–20343. Lane, D.E. (1989) A partially observable model of decision-making by fishermen. Operations Research, 37, 240–254. Latimer, A.M., Banerjee, S., Sang Jr, H., Mosher, E.S. & Silander Jr, J.A. (2009) Hierarchical models facilitate spatial analysis of large data sets: a case study on invasive plant species in the northeastern United States. Ecology Letters, 12, 144–154. Lobo, J.M., Jimeńez-Valverde, A. & Real, R. (2008) AUC: a misleading measure of the performance of predictive distribution models. Global Ecology and Biogeography, 17, 145–151. Mangel, M. & Clark, C.W. (1988) Dynamic Modeling in Behavioral Ecology. Princeton University Press, Princeton, NJ. McCarthy, M.A., Possingham, H.P. & Gill, A.M. (2001) Using stochastic dynamic programming to determine optimal fire management for Banksia ornata. Journal of Applied Ecology, 38, 585–592. McDonald-Madden, E., Baxter, P.W.J. & Possingham, H.P. (2008) Subpopulation triage: how to allocate conservation effort among populations. Conservation Biology, 22, 656–665. McNamara, J.M. & Houston, A.I. (1996) State-dependent life histories. Nature, 380, 215–221. Metz, C.E. (1986) ROC methodology in radiologic imaging. Investigative Radiology, 21, 720–733. Moloney, S. & Vanderwoude, C. (2002) Red imported fire ants: a threat to eastern Australia’s threatened wildlife? Ecological Management and Restoration, 3, 167–175. Moody, M.E. & Mack, R.N. (1988) Controlling the spread of plant invasions the importance of nascent foci. Journal of Applied Ecology, 25, 1009–1021. Murray, J.V., Goldizen, A.W., O’Leary, R.A., McAlpine, C.A., Possingham, H.P. & Choy, S.L. (2009) How useful is expert opinion for predicting the distribution of a species within and beyond the region of expertise? A case study using brush-tailed rock-wallabies Petrogale penicillata. Journal of Applied Ecology, 46, 842–851. Odom, D.I.S., Cacho, O.J., Sinden, J.A. & Griffith, G.R. (2003) Policies for the management of weeds in natural ecosystems: the case of scotch broom (Cytisus scoparius, L.) in an Australian national park. Ecological Economics, 44, 119–135. Pearce, J. & Ferrier, S. (2000) Evaluating the predictive performance of habitat models developed using logistic regression. Ecological Modelling, 133, 225– 245. Pimentel, D., McNair, S., Janecka, J., Wightman, J., Simmonds, C., O’Connell, C., Wong, E., Russel, L., Zern, J., Aquino, T. & Tsomondo, T. (2001) Economic and environmental threats of alien plant, animal, and microbe invasions. Agriculture Ecosystems & Environment, 84, 1–20. Regan, T.J., McCarthy, M.A., Baxter, P.W.J., Panetta, D.F. & Possingham, H.P. (2006) Optimal eradication: when to stop looking for an invasive plant. Ecology Letters, 9, 759–766. Richards, S.A., Possingham, H.P. & Tizard, J. (1999) Optimal fire management for maintaining community diversity. Ecological Applications, 9, 880–892. Sala, O.E., Chapin, F.S., Armesto, J.J. et al. (2000) Global biodiversity scenarios for the year 2100. Science, 287, 1770–1774. Scanlan, J.C. & Vanderwoude, C. (2006) Modelling the potential spread of Solenopsis invicta Buren (Hymenoptera: Formicidae) (red imported fire ant) in Australia. Australian Journal of Entomology, 45, 1–9.


Optimizing search strategies for invasives 95 Shryock, K.A., Brown, S.L., Sanders, N.J. & Burroughs, E. (2008) A reactiondiffusion equation modeling the invasion of the argentine ant population, Linepithema humile, at Jasper Ridge Biological Preserve. Natural Resource Modeling, 21, 330–342. Stephens, D.W. & Krebs, J.R. (1986) Foraging Theory. Princeton University Press, Princeton, N.J. Sutherst, R.W. & Maywald, G. (2005) A climate model of the red imported fire ant, Solenopsis invicta Buren (Hymenoptera: Formicidae): Implications for invasion of new regions, particularly Oceania. Environmental Entomology, 34, 317–335. Swets, J.A. (1986) Indexes of discrimination or diagnostic-accuracy – their ROCs and implied models. Psychological Bulletin, 99, 100–117. Swets, J.A. (1988) Measuring the accuracy of diagnostic systems. Science, 240, 1285–1293. Swets, J.A. & Pickett, R.M. (1982) Evaluation of Diagnostic Systems: Methods From Signal Detection Theory. Academic Press, New York. Taylor, C.M. & Hastings, A. (2004) Finding optimal control strategies for invasive species: a density-structured model for Spartina alterniflora. Journal of Applied Ecology, 41, 1049–1057. Vaćlavı´ k, T. & Meentemeyer, R.K. (2009) Invasive species distribution modeling (iSDM): Are absence data and dispersal constraints needed to predict actual distributions? Ecological Modelling, 220, 3248–3258. Walters, C.J. (1986) Adaptive Management of Renewable Resources. Macmillan, New York. White, B. (2005) An economic analysis of ecological monitoring. Ecological Modelling, 189, 241–250. Whittle, A.J., Lenhart, S. & Gross, L.J. (2007) Optimal control for management of an invasive plant species. Mathematical Biosciences and Engineering, 4, 101–112. Williams, D.F. (1994) Control of the introduced pest Solenopsis invicta in the United States. Exotic Ants: Biology, Impact, and Control of Intro-

duced Species (ed. D.F. Williams), pp. 282–292. Westview Press, Boulder, CO. Williams, B. (2010) Fire ants advance. Courier Mail 31 March 2010. News Limited, Brisbane, pp 1–2. Wintle, B.A., Elith, J. & Potts, J.M. (2005) Fauna habitat modelling and mapping: a review and case study in the Lower Hunter Central Coast region of NSW. Austral Ecology, 30, 719–738. Zurell, D., Jeltsch, F., Dormann, C.F. & Schro¨der, B. (2009) Static species distribution models in dynamically changing systems: how good can predictions really be? Ecography, 32, 733–744. Received 27 January 2010; accepted 30 September 2010 Handling Editor: Andy Sheppard

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. Details of the optimization. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.



doi: 10.1111/j.1365-2664.2010.01894.x

REVIEW

The frequency and magnitude of non-additive responses to multiple nutrient enrichment Jacob E. Allgeier1*, Amy D. Rosemond1 and Craig A. Layman2 1

Odum School of Ecology, University of Georgia, Athens, GA 30602, USA; and 2Marine Sciences Program, Department of Biological Sciences, Florida International University, 3000 NE 151st Street, North Miami, FL 33181, USA

Summary 1. Anthropogenic eutrophication is among the greatest threats to ecosystem functioning globally, often occurring via enrichment of both nitrogen (N) and phosphorus (P). As such, recent attention has focused on the implications of non-additive responses to dual nutrient enrichment and the inherent difficulty associated with predicting their combined effects. 2. We used a simple metric to quantify the frequency and magnitude of non-additive responses to enrichment by N, P and N + P in 653 experiments conducted across multiple ecosystem types and locations. 3. Non-additive responses were found to be common in all systems. Freshwater ecosystems and temperate latitudes tended to have frequent synergistic responses to dual nutrient enrichment, i.e. the response was greater than predicted by an additive model. Terrestrial and arctic systems were dominated by antagonistic responses (responses to N + P that were less than additive). 4. The mean of all experiments was synergistic because despite being less common, synergistic responses were generally of greater magnitude than antagonistic ones. 5. Synthesis and applications. Our study highlights the ubiquity of non-additive effects in response to dual nutrient enrichment and further elucidates the complex ways in which ecosystems respond to human impacts. Our results suggest how alternative nutrient limitation scenarios can be used to guide approaches to conservation and management of nutrient loading to ecosystems. This review provides the first published summary of non-additive responses by primary producers. Key-words: Antagonism, co-limitation, eutrophication, interaction, nitrogen, nutrient loading, phosphorus, primary production, synergy

Introduction The ecological impacts of excessive nutrient loading are substantial, driving losses of ecosystem services world-wide (Vitousek et al. 1997; Smith & Schindler 2009) and stimulating debate over how to most effectively regulate anthropogenic nutrient inputs (Conley et al. 2009). At the crux of the debate is whether controlling nitrogen (N), phosphorus (P) or both, should frame conservation initiatives (Carpenter 2008; Conley et al. 2009). The underpinning research that has informed this debate is generally based on quantifying the primary producer response to enrichment by these key nutrients. Most notably, measuring the production response to multiple nutrients, i.e. both N and P, has received much attention because many anthropogenic stressors tend to alter concentrations of both *Correspondence author. E-mail: [email protected]

nutrients simultaneously (Sala & Knowlton 2006; Halpern et al. 2008). A recent study by Elser et al. (2007) demonstrated the prevalence of nutrient co-limitation across ecosystems. Here we define nutrient co-limitation as a greater response to simultaneous enrichment by both nutrients than enrichment by either nutrient individually. Some interpretations of these findings have suggested that they likewise imply a dominance of synergy in ecosystems, assuming that co-limitation is necessarily synergistic (Davidson & Howarth 2007; Elser et al. 2007). However, a synergism only occurs when the response is greater than additive, whereas co-limitation can also be an equal to or less than additive response. Understanding these different outcomes forms the basis of our ability to predict how an ecosystem will respond to nutrient enrichment and, therefore, our ability to develop effective management strategies.


Non-additive responses to nutrient enrichment 97 We developed a simple metric to quantify the relative response to additions of both N and P compared to predicted additive effects in plant production to: (i) quantitatively assess the generalities of non-additive responses to nutrient enrichment and (ii) distinguish different types of co-limitation across ecosystem types and latitudinal gradients. We also test the hypothesis that the distribution of these data is consistent with null distributions based on random values. Our results suggest how alternative nutrient limitation scenarios can be used to guide approaches to conservation and management of nutrient loading to ecosystems.

Materials and methods We developed the interaction effect index (IEI) to quantify the response of primary producers to N and P additions: IEI ¼ ln½response NP=ðresponse N + response PÞ:

eqn 1

Where response NP is the primary producer biomass (and in some cases the change in mass) reported for N + P treatments (hereafter NP) and response N and response P are primary producer biomass responses in those treatments. Taking the natural log of the quotient proportionally centers the IEI values around zero. For example, an IEI value generated from an experiment, where response NP is two times greater than response N + response P (i.e., ln(2)) is equal to the absolute value of an experiment, where response N + response P is two times greater than response NP (i.e., ln(0Æ5)). We applied the IEI to 653 experiments from marine, freshwater and terrestrial ecosystems that tested for primary producer responses to enrichment in all three treatments: N, P and NP (compiled in Elser et al. 2007; obtained via the National Center for Ecological Analysis and Synthesis). Experiments that used the metric of biomass per unit area or volume were included, but proxy variables for biomass were also allowed (e.g. chlorophyll a concentration, ash-free dry mass, carbon mass, biovolume, per cent cover; Elser et al. 2007). We included only studies that reported mean community-level biomass responses to nutrient enrichment. Thus, the only single species responses that

(a)

(b)

were included were drawn from communities dominated by single species. One hundred and twenty-nine studies were conducted in laboratory settings; the rest of the experiments were conducted in situ. A total of 39 of the 653 experiments included additional manipulations (e.g. grazer exclusion), but only data from unmanipulated controls (e.g. grazers at natural densities) were included. Because of the nature of our categories, all experiments were classified simultaneously in two categories (based on ecosystem type and latitudinal zone) A simple prediction regarding dual nutrient enrichment is that NP response would be equal to the sum of individual N and P responses (i.e. an additive response; Fig. 1b). Our metric provides a continuous measure to assess the relative departure from additivity. IEI values close to zero, either positive or negative, can be characterized by additive co-limitation (AD; Fig. 1b). As IEI increases or decreases, the non-additive effect becomes more pronounced and can be classified into one of three response categories: synergistic co-limitation (SC), antagonistic co-limitation (AC) and absolute antagonism (AA; Fig. 1a,c,d). Co-limitation implies that the producer is limited by both nutrients (Arrigo 2005; Davidson & Howarth 2007), and is demonstrated when the response to both nutrients is greater than either nutrient individually. Synergistic co-limitation results when there is a positive nonadditive response, whereby NP response is greater than the sum of N and P responses (Fig. 1a). Antagonistic co-limitation is a less than additive response that occurs when NP response is less than the sum of N and P responses, but is still greater than response to either single nutrient. Absolute antagonisms occur where NP response is less than at least one of the single nutrient enrichments. The relative strength of the non-additive effect (i.e. SC, AC, AA) increases as the IEI value deviates from zero, either positively (SC) or negatively (AC, AA). The term nutrient co-limitation has been subject to various interpretations and requires specific clarification (Arrigo 2005; Lewis & Wurtsbaugh 2008). According to Liebig’s law of the minimum, only one nutrient can functionally limit primary production at a given point in time. However, with dual nutrient enrichment, an individual (or producer assemblage with similar physiological nutrient demands) may oscillate between single nutrient limitation of two nutrients (here N and P). In this case, the supply of one nutrient is

(c)

(d)

Fig. 1. A conceptual diagram of possible responses from enrichment by N, P and NP. An additive response is indicated in each panel by summing the individual N (yellow) and P (blue) responses. (a) Synergistic co-limitation (SC) such that the biomass or production response to dual enrichment (NP) is greater than the additive response of both single nutrient treatments (N and P alone). (b) Additive co-limitation (AD), whereby the response to NP is equal to that of the sum of N alone and P alone. (c) Antagonistic co-limitation (AC), whereby the response to NP is greater than that of either N or P alone, but not their sum. (d) Absolute antagonism (AA), whereby NP results in less biomass or production than either N or P alone. 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

98 J. E. Allgeier, A. D. Rosemond & C. A. Layman sufficient to shift demand towards that of the other, next most limiting nutrient. This interplay continues until either another factor becomes limiting or a saturation state is reached (Davidson and Howarth 2007). As such, over the course of time, e.g. an experimental time period, an individual producer (or producer assemblage) may be considered functionally co-limited, even though a single factor may always be limiting at any instantaneous point. We test the hypotheses that the distribution of the data from each category (e.g. freshwater) was consistent with null distributions based on random numbers. To do this we compared the distribution of a given category (e.g., marine; n = 105) with the distribution of a randomly sampled data set of the same size, using Kolmogorov–Smirnov tests for 1000 permutations. Then we pooled the P-values from these permutations to determine the proportion of the model runs that showed statistical difference between the randomly generated and the observed distributions (a = 0Æ05). The data set of artificial IEI values from which the null distributions for each category was sampled, was generated by randomizing each response variables (N, P and NP) from the original data set and recalculating IEI values based on these numbers. The null distribution for each permutation was then sampled from this data set.

Results Synergistic co-limitation, AC and AA occurred in all ecosystem types and latitudinal zones (Fig. 2). When comparing the frequency of each response for all experiments combined, 37% were SC, 40% were AC and 23% were AA (Fig. 2). Across all six subcategories (marine, freshwater, terrestrial, arctic, temperate and tropical), SC occurred more frequently in all but terrestrial and arctic ecosystems, in which AC occurred 64% and 71% of the time, respectively (Fig. 2). AA occurred more frequently than SC in arctic (8% SC, 21% AA) and terrestrial

systems (18% SC and 18% AA), but never occurred more frequently than AC (Fig. 2). Across all categories, SC occurred substantially less frequently than antagonistic responses (i.e. AC and AA combined). A study that incorporates multiple experimental units can be considered additive if the mean of all experiments does not significantly differ from zero (i.e. the 95% confidence intervals overlap zero). Because of the complex nature of our data set, applying such confidence intervals to individual studies was inappropriate. Thus to provide perspective as to the number of studies that were characterized by values close to additive (i.e. zero), we chose an arbitrary positive and negative interval of 10% from perfect additivity (0Æ095 > IEI > )0Æ095). Under these conditions, we found only 5% of experiments yielded additive responses (AD). Extending the interval to 15% (0Æ139 > IEI > )0Æ139), the frequency of such responses increased to only 11%. All experiments combined reflect a mean SC response (IEI = 0Æ12, P < 0Æ001 for t-test of IEI = 0). Freshwater, temperate and tropical subcategories had mean net SC IEI values [P < 0Æ005 for t-test of IEI = 0 for freshwater and temperate; tropical did not differ from zero, P = 0Æ43 (see Appendix S1 in Supporting Information)]. Marine, terrestrial and arctic subcategories had mean AC IEI values [P < 0Æ001 for t-test of IEI = 0 for terrestrial and arctic, marine did not differ from zero P = 0Æ83 (Appendix S1)] (red lines; Fig. 3). SC values were on average of greater magnitude than AC or AA values in most subcategories (coloured bars; Fig. 3). Freshwater ecosystems had the greatest mean SC value (IEI = 1Æ23 ± 0Æ07, NP responses 3Æ4· greater than additivity). Tropical and marine systems demonstrated the lowest IEI

Fig. 2. Frequency of IEI values within each subcategory. In each plot, the white background bars indicate the frequency of IEI values for all experiments combined. A positive value represents synergistic co-limitation, a negative value indicates either antagonistic co-limitation or absolute antagonism and zero represents additive co-limitation. Categories are not orthogonal, thus experiments can be within multiple categories (i.e., temperate and marine). 2010 The Authors. Journal of Applied Ecology 2010 British Ecological Society, Journal of Applied Ecology, 48, 96–101

Non-additive responses to nutrient enrichment 99 of random organization of data, or some underlying pattern driving these trends. Over 99% of random permutations of the data set differed from the observed distribution of values from all the experiments combined. With the exception of marine and tropical categories, >95% of the random permutations of the data set differed from the observed distribution of values in every category. These findings provide evidence that the distribution of these data is a product of underlying patterns that emerge from each subcategory.

Discussion

Fig. 3. Full range of all values (grey bars) and mean values for each response type (as indicated by the height of coloured bars; e.g. SC) for different ecosystem types and latitudinal zones. Positive and negative values as in Fig. 2. The red line indicates the net mean IEI value for the respective category. For context, an absolute IEI value of 0Æ69 or 1Æ09 indicate a 100% or 200% increase or decrease from additivity, respectively. The coloured bars indicate mean values for each category: yellow bars for synergistic co-limitation (SC), green for antagonistic co-limitation (AC) and blue for absolute antagonism (AA). Categories include fundamental ecosystem types (Mar = marine, Fresh = freshwater and Terr = terrestrial) and well as categories based on latitudinal zones [Arct = arctic (latitudes >66Æ5), Temp = temperate (latitudes 23Æ5–66Æ5) and Trop (latitudes 23Æ5N to 23Æ5S)].

values (IEI = )0Æ88 ± 0Æ5, )0Æ92 ± 0Æ12; NP responses 2Æ4· and 2Æ5· less than additivity, respectively). Terrestrial ecosystems and arctic latitudes were the categories that had greater absolute mean AA than SC values. The highest IEI value (IEI = 5Æ01; NP response 150· greater than additivity) was from a benthic freshwater stream (Chessman, Hutton & Burch 1992). However, of the top 50 highest IEI values, all but two (both in benthic freshwater environments) experiments were conducted in pelagic freshwater and marine environments. The lowest IEI value (IEI = )2Æ81; NP response 16· less than additivity) was conducted on the benthos of a temperate marine estuary (Taylor et al. 1995). Unlike the positive IEI values, the lower IEI values were not dominated by experiments from any category. A bimodal trend is apparent in freshwater, marine, temperate and tropical categories, whereby there is a secondary mode centred around IEI 2 (Fig. 2). Examination of the data showed that this trend was strongly driven by a single set of experiments in temperate lakes (62 of the 82 studies) (Maberly et al. 2002). Of the 82 experiments that fall within the range of 1Æ5 < IEI < 2Æ5, we found that all but two were conducted in the pelagic zone of freshwater or marine environments, emphasizing that pelagic environments may tend towards relatively strong synergistic response to dual nutrient enrichment. Comparing the distribution of the data within each category with that of a randomly generated null distribution allows inference as to the probability that these data were the product

Synergies have garnered much attention in the ecological literature, often under the assumption that they occur frequently and with great magnitude (Myers 1995; Sala & Knowlton 2006; Halpern et al. 2008). Our findings provide more detail to this broad generalization. Though synergistic responses (SC) were often demonstrated, they occurred less frequently than antagonistic responses (the combination of AC and AA). However, where they occurred, SC tended to be of greater magnitude than antagonisms, as is supported by the bimodal distribution of the data with the second mode occurring approximately around 2. Thus, although the distribution of experiments is skewed towards negative IEI values (Fig. 2), the overall mean IEI is positive. The presumed mechanism for synergisms results from primary production that is limited by both nutrients to such a degree that little production occurs under enrichment by a single nutrient. SC is generally a result of oscillating nutrient limitation, whereby ambient availability of nutrients is minimal, and given supply of one nutrient, limitation shifts towards limitation by the other (Davidson & Howarth 2007). Thus, limitation oscillates between nutrients (if supply rate of both nutrients is constant relative to demand) until either production is maximized or another factor becomes limiting. These conditions are often prevalent in extremely nutrient poor ecosystems (Arrigo 2005). Antagonistic co-limitation, the most common response type, can be explained by a third (or additional) limiting factor. Other micronutrients (e.g. iron, magnesium, molybdate, sylica), as well as physical factors (e.g. light, water), can limit production (Howarth, Marino & Cole 1988; Arrigo 2005; Davidson & Howarth 2007). Thus, stimulating production beyond a certain level may incur limitation by a resource(s) besides N or P. Another mechanism may derive from physiological and ⁄ or environmentally related limitations (e.g. maximum physical size, disturbance or grazing), whereby the upper bound of community or individual primary production is constrained in mass or size irrespective of nutrient resources (Rosemond 1993). An additional plausible mechanism for AC may occur if increased supply of one nutrient concomitantly decreases the need for another. An example is the requirement of N for the anabolism of phosphatase enzymes which can be used to process organic P at low availability of inorganic or bioavailable P (Chrost 1991). In this case, enrichment of N can enhance net primary production (via increased production of phosphatase,


100 J. E. Allgeier, A. D. Rosemond & C. A. Layman and thus increased access to inorganic P). However, under conditions of enrichment with N and P, the availability of inorganic P can simultaneously inhibit production of phosphatase resulting in potentially similar or only slightly higher production than with N additions alone. The net response to NP would then be less than additive, but still greater than the response to N or P alone (Ivancic et al. 2009; Rees et al. 2009; Scott et al. 2009). Absolute antagonisms, perhaps the most counterintuitive response, were the most infrequent response category. The effect of grazing could lead to AA, whereby the grazer could selectively feed on the resource with the highest production rate, or potentially with the highest nutrient content from enrichment (DeMott, Gulati & Siewertsen 1998; Heck et al. 2006). However, there are multiple examples that suggest that antagonisms could simply be experimental artefacts. For example, Taylor et al. (1995) reports a strong AA response (IEI = )2Æ81) by eelgrass to enrichment by NP. The enrichment study was conducted in mesocosms where, under enriched conditions, phytoplankton, which was growing simultaneously with eelgrass, responded synergistically to nutrient addition (Taylor et al. 1995). This experiment was characterized by a large algal bloom, causing light limitation and thus reducing seagrass biomass. These findings are consistent with the widely predicted response of seagrass to nutrient enrichment at an ecosystem scale (Deegan et al. 2002), and arose due to complex interactions involving two different producer assemblages. The experiments compiled in this study measured the biomass response to enrichment by monocultures (e.g. a stand of a single tree species) as well as entire assemblages of different producer species (e.g. a phytoplankton assemblage). The differences in response to nutrient enrichment between an individual species and a community of different species can be substantial. For example, a diverse assemblage of producers probably consists of organisms with varying physiological requirements (e.g. N limited or P limited) and growth potential (e.g. greater size ⁄ growth rate). As such, under various magnitudes and time duration of nutrient enrichment, differential non-additive responses may be expected, and knowledge of the existing community is required to fully understand the mechanisms behind these responses. These differences may help explain the disparity in findings between terrestrial and aquatic (freshwater and marine together) systems, whereby aquatic systems are characterized by a greater range in IEI values with notably greater frequency and magnitude of SC. Many aquatic studies were conducted on assemblages of producers, whereas the majority of studies conducted on monocultures were from terrestrial ecosystems. These findings are consistent with the fact that pelagic environments with mixed species assemblages (e.g. phytoplankton) tend to be particularly susceptible to large production responses (e.g. algal blooms) from multiple nutrient enrichment (Conley et al. 2009). Distributional trends that emerge from these data appear to be the product of underlying ecological patterns as opposed to randomness within the data. Yet, isolating specific factors that determine the frequency of the type of non-additive effects are

difficult given the biological complexity (i.e. species life history, physical conditions, etc.) associated with interaction of multiple nutrients. A notable finding from our study was the dominance of antagonistic responses (AC and AA combined) in terrestrial and arctic subcategories. One explanation for terrestrial ecosystems may be that the growth rate is typically slower and generation time of terrestrial producers is typically greater than for producers in aquatic systems due to the greater requirement of structural and supporting tissue (Cebrian 1999; Chapin 2002). Thus, even given adequate experimental time frames, physiological constraints may hinder synergistic responses. Consistent with this observation, the strongest synergistic effects tended to occur in aquatic ecosystems, particularly in the pelagic zone, occurring among more speciose assemblages with relatively minimal structural demands (see Appendix S2). As for arctic regions, a less than additive response to nutrient enrichment may reflect the fact that producer growth rates are positively correlated with temperature and thus temperature could be a physical factor limiting synergistic responses (Chapin 2002). However, despite the similarity in frequency of response types between terrestrial and arctic subcategories, arctic experiments were primarily conducted in freshwater ecosystems (S2). Our findings have important implications for management of nutrient loading to aquatic ecosystems. The prevalence of non-additive effects across all systems suggests that when possible, both nutrients should be controlled in conservation and management because the ecological repercussion of simultaneous nutrient enrichment is relatively unpredictable. This is particularly relevant in ecosystems where IEI is close to zero, as they are often characterized by a relatively large response to at least one, but more often both, nutrients individually (Fig. 1b,c). As the IEI value deviates from zero, positively or negatively, it may indicate the potential for effective control of nutrient loading by focusing on the single most limiting nutrient. For example, a large IEI value (i.e. a synergistic response) generally indicates that both nutrients are critical for enhancement of production, thus controlling the single most limiting nutrient (in the case of Fig. 1a; P is most important to control) may be an effective way to mitigate unwanted ecosystem responses. Likewise, an extremely negative IEI value (i.e. AA) generally indicates that only one nutrient is significantly limiting and thus suggests that controlling the loading rate of this most limiting nutrient may provide a significant reduction in ecosystem-scale responses. In a perfect world, all stressors that negatively affect ecosystems would be carefully managed. Yet, conservation efforts are constrained by cost, time and societal will to manage ecosystems. Our findings show frequent and strong non-additive responses to nutrient enrichment across ecosystem types and locations. We emphasize that a single conservation model for mitigating nutrients is not appropriate and stress that future efforts need to account for the complex nature of dual nutrient limitation. We further highlight the importance of incorporating all treatments (N, P and NP) into enrichment experiments in conjunction with quantitatively assessing the nature of the interaction on a system-specific basis. These data are critical


Non-additive responses to nutrient enrichment 101 for building effective predictive models needed to inform conservation and management decision-making regarding nutrient control.

Acknowledgements We would like to thank D.S. Gruner, J.J. Elser, W.S. Harpole and their colleagues for allowing access to their data and Cynthia Tant, Ashley Helton, Andrew Mehring and William Lewis for comments that improved the manuscript. We also thank an anonymous reviewer and the editor for helpful comments. Funding was provided by a University-wide Graduate Student Fellowship, University of Georgia and National Science Foundation Grant OCE#0746164.

References Arrigo, K.R. (2005) Marine microorganisms and global nutrient cycles. Nature, 437, 349–355. Carpenter, S.R. (2008) Phosphorus control is critical to mitigating eutrophication. Proceedings of the National Academy of Sciences of the United States of America, 105, 11039–11040. Cebrian, J. (1999) Patterns in the fate of production in plant communities. American Naturalist, 154, 449–468. Chapin, F.S.I. (2002) Principles of Terrestrial Ecosystem Ecology. Springer Science. Chessman, B.C., Hutton, P.E. & Burch, J.M. (1992) Limiting nutrients for periphyton growth in sub-alpine, forest, agricultural and urban streams. Freshwater Biology, 28, 349–361. Chrost, R.J. (1991) Environmental Control of Synthesis and Activity of Aquatic Microbial Ectoenzymes. Springer-Verlag, New York. Conley, D.J., Paerl, H.W., Howarth, R.W., Boesch, D.F., Seitzinger, S.P., Havens, K.E., Lancelot, C. & Likens, G.E. (2009) Controlling eutrophication: nitrogen and phosphorus. Science, 323, 1014–1015. Davidson, E.A. & Howarth, R.W. (2007) Environmental science – nutrients in synergy. Nature, 449, 1000–1001. Deegan, L.A., Wright, A., Ayvazian, S.G., Finn, J.T., Golden, H., Merson, R.R. & Harrison, J. (2002) Nitrogen loading alters seagrass ecosystem structure and support of higher trophic levels. Aquatic Conservation-Marine and Freshwater Ecosystems, 12, 193–212. DeMott, W.R., Gulati, R.D. & Siewertsen, K. (1998) Effects of phosphorusdeficient diets on the carbon and phosphorus balance of Daphnia magna. Limnology and Oceanography, 43, 1147–1161. Elser, J.J., Bracken, M.E.S., Cleland, E.E., Gruner, D.S., Harpole, W.S., Hillebrand, H., Ngai, J.T., Seabloom, E.W., Shurin, J.B. & Smith, J.E. (2007) Global analysis of nitrogen and phosphorus limitation of primary producers in freshwater, marine and terrestrial ecosystems. Ecology Letters, 10, 1135–1142. Halpern, B.S., Walbridge, S., Selkoe, K.A., Kappel, C.V., Micheli, F., D’Agrosa, C., Bruno, J.F., Casey, K.S., Ebert, C., Fox, H.E., Fujita, R., Heinemann, D., Lenihan, H.S., Madin, E.M.P., Perry, M.T., Selig, E.R., Spalding, M., Steneck, R. & Watson, R. (2008) A global map of human impact on marine ecosystems. Science, 319, 948–952. Heck, K.L., Valentine, J.F., Pennock, J.R., Chaplin, G. & Spitzer, P.M. (2006) Effects of nutrient enrichment and grazing on shoalgrass Halodule wrightii and its epiphytes: results of a field experiment. Marine Ecology-Progress Series, 326, 145–156. Howarth, R.W., Marino, R. & Cole, J.J. (1988) Nitrogen fixation in freshwater, estuarine and marine ecosystems: biogeochemical controls. Limnology and Oceanography, 33, 688–701.

Ivancic, I., Radic, T., Lyons, D.M., Fuks, D., Precali, R. & Kraus, R. (2009) Alkaline phosphatase activity in relation to nutrient status in the northern Adriatic Sea. Marine Ecology-Progress Series, 378, 27–35. Lewis, W.M. & Wurtsbaugh, W.A. (2008) Control of lacustrine phytoplankton by nutrients: erosion of the phosphorus paradigm. International Review of Hydrobiology, 93, 446–465. Maberly, S.C., King, L., Dent, M.M., Jones, R.I. & Gibson, C.E. (2002) Nutrient limitation of phytoplankton and periphyton growth in upland lakes. Freshwater Biology, 47, 2136–2152. Myers, N. (1995) Environmental unknowns. Science, 269, 358–360. Rees, A.P., Hope, S.B., Widdicombe, C.E., Dixon, J.L., Woodward, E.M.S. & Fitzsimons, M.F. (2009) Alkaline phosphatase activity in the western English Channel: elevations induced by high summertime rainfall. Estuarine Coastal and Shelf Science, 81, 569–574. Rosemond, A.D. (1993) Interactions among irradiance, nutrients, and herbivores constrain a stream algal community. Oecologia, 94, 585–594. Sala, E. & Knowlton, N. (2006) Global marine biodiversity trends. Annual Review of Environment and Resources, 31, 93–122. Scott, J.T., Lang, D.A., King, R.S. & Doyle, R.D. (2009) Nitrogen fixation and phosphatase activity in periphyton growing on nutrient diffusing substrata: evidence for differential nutrient limitation in stream periphyton. Journal of the North American Benthological Society, 28, 57–68. Smith, V.H. & Schindler, D.W. (2009) Eutrophication science: where do we go from here? Trends in Ecology & Evolution, 24, 201–207. Taylor, D., Nixon, S., Granger, S. & Buckley, B. (1995) Nutrient limitation and the eutrophication of coastal lagoons. Marine Ecology-Progress Series, 127, 235–244. Vitousek, P.M., Mooney, H.A., Lubchenco, J. & Melillo, J.M. (1997) Human domination of Earth’s ecosystems. Science, 277, 494–499. Received 8 April 2010; accepted 6 October 2010 Handling Editor: Marc Cadotte

Supporting Information Additional Supporting Information may be found in the online version of this article. Appendix S1. T-test results and confidence intervals for all designated categories (ecosystem type and latitudinal zone). Appendix S2. Frequency of IEI values for each latitudinal zone (arctic, temperate, tropical) within each ecosystem type (freshwater, marine, terrestrial). In each plot, the white background bars indicate the frequency of IEI values for all experiments within that given ecosystem type (e.g. the first row the white bars indicate the IEI values for all experiments in freshwater ecosystems). As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.



doi: 10.1111/j.1365-2664.2010.01896.x

Long-term impact of changes in sheep Ovis aries densities on the breeding output of the hen harrier Circus cyaneus Arjun Amar1*, Jacob Davies2, Eric Meek3, Jim Williams4, Andy Knight3 and Steve Redpath5 1

RSPB–Scotland, Dunedin House, 25 Ravelston Terrace, Edinburgh, EH4 3TP, UK; 2Banks, Northside, Birsay, Orkney KW17 2LU, UK; 3RSPB–Scotland, 12 ⁄ 14 North End Rd., Stromness, Orkney KW16 3AG, UK; 4Fairholm, Finstown, Orkney KW17 2EQ, UK; and 5Aberdeen Centre for Environmental Sustainability, Aberdeen University & Macaulay Institute, Tillydrone Avenue, Aberdeen AB24 2TZ, UK

Summary 1. Livestock grazing is an important form of land use across the globe and changes in grazing pressure can have profound effects on vertebrate populations. 2. In Scotland, over the last decade sheep numbers in many areas have declined from historically high levels, providing an opportunity to explore the implications of these declines for biodiversity. 3. The hen harrier Circus cyaneus is a bird of high conservation importance in the UK, and a species that may be heavily influenced by the indirect effects of sheep on habitat and prey. The hen harrier population on the Orkney Islands in Scotland has been monitored since 1975 and represents an ideal case study for considering the impact of sheep de-stocking on a key predator. 4. Declines in the harrier population were associated with a doubling in sheep numbers between the early 1980s and the late 1990s. Subsequently, as sheep numbers have fallen the harrier population has recovered. These changes indicate an association but no clear mechanism, so we tested whether reductions in sheep numbers have led to increases in harrier prey or preferred foraging habitat. We then tested whether breeding output over the last 33 years correlates with sheep stocking levels or variation in weather conditions (rainfall and temperature). 5. Orkney sheep numbers declined by about 20% between 1998 and 2008. Surveys in 1999 ⁄ 2000 and repeated in 2008 showed increases in rough grassland, the preferred harrier foraging habitat, and increases in a key prey species, the Orkney vole Microtus arvalis orcadensis. 6. Overall, hen harrier breeding output over the last 33 years was significantly negatively correlated to both sheep abundance and spring rainfall. 7. Synthesis and application. This study provides strong evidence for the consequences of changes in sheep numbers on a top predator. Our results indicate that reductions in sheep numbers are likely to prove beneficial for some upland species, particularly small mammals and their predators. Key-words: agriculture, grasslands, grazing, grouse moors, Orkney, predation, rainfall, voles

Introduction The impacts of grazing on vegetation structure and composition, and on ecosystem processes have received considerable attention (Milchunas & Lauenroth 1993; Augustine & McNaughton 1998; Perevolotsky & Seligman 1998; Knapp et al. 1999; Cote et al. 2004; Hanley et al. 2008). In Britain, sheep Ovis aries are the principal domestic grazing species in the uplands, and between 1950 and 1990 their numbers rose *Correspondence author. E-mail: [email protected]

from 19Æ7 million to 41Æ2 million with particularly dramatic increases apparent during the 1980s and the early 1990s (Fuller & Gough 1999; Amar & Redpath 2005; Condliffe 2009). These increases stem from changes to the subsidy and support systems operated through the Common Agricultural Policy (CAP) (Fuller & Gough 1999; Hanley et al. 2008). Increased sheep abundance dramatically affected some bird species, particularly in the uplands, and almost certainly reduced the habitat quality for some ground nesting bird species (Fuller & Gough 1999; Thirgood et al. 2000). In contrast,


Grazing impacts on hen harriers 221 increased grazing may have benefited those species preferring shorter or more grass-dominated vegetation (Smith et al. 2001; Pearce-Higgins & Grant 2006) and their predators, as well as carrion feeders which would have benefited from the increase in sheep carcasses (Watson, Rae & Stillman 1992; Ratcliffe 1997; Fuller & Gough 1999; Redpath & Thirgood 1999). However, since the late 1990s sheep numbers in Scotland have declined quite dramatically (SAC 2008). The outcome for biodiversity of these declines will inevitably vary between vegetation types and between bird species (Pearce-Higgins et al. 2009). There is an urgent need to understand the biodiversity responses to spatial and temporal changes in grazing patterns, indeed understanding the consequences of changes in upland grazing regimes for biodiversity is currently viewed as one of the most important ecological questions of high policy relevance for the UK (Sutherland et al. 2006). Changes to vegetation structure or communities caused by high levels of grazing can have a direct impact on vertebrate and invertebrate species that rely on ungrazed or lightly grazed habitats (Jepson-Innes & Bock 1989; Steen, Mysterud & Austrheim 2005; Evans et al. 2006). Reductions in the abundance of these species can in turn influence the abundance of the predators that feed on them, although links between these different trophic guilds have been poorly studied due to their inherent complexities (Duff 1979; Douglass & Frisina 1993; Steenhof et al. 1999; Johnson & Horn 2008). For example, small mammal populations are known to be affected by grazing levels (Steen, Mysterud & Austrheim 2005; Evans et al. 2006) and avian predators of small mammals, such as hen harriers Circus cyaneus whose populations can be strongly influenced by vole abundance (Redpath, Thirgood & Clarke 2002a), may therefore be influenced by changes in grazing densities. The Orkney Islands in North Scotland are an important breeding area for hen harriers in the UK (Sim et al. 2007). Most hen harriers on Orkney breed in the west of the island of Mainland (hereafter referred to as West Mainland) and this population has been monitored on the same area annually from 1975 (Amar et al. 2005). A doubling in sheep numbers is believed to have been responsible for a decline in hen harriers between the 1970s and 1990s when the numbers of chicks produced each year declined by 73% (Amar & Redpath 2005; Amar et al. 2005). The mechanism for this was thought to have been a reduction in rough grassland (the preferred habitat for foraging harriers) and Orkney voles, reducing the amount of prey available to harriers during the critical pre-laying period (Amar & Redpath 2002, 2005; Amar, Redpath & Thirgood 2003a). However, since the end of the 1990s the hen harrier population on Orkney has recovered. Between 1998 and 2004, the numbers of breeding females on Orkney increased by 118% from 34 to 74 (Sim et al. 2001, 2007), which contributed to an overall population increase for the UK, the Orkney population representing 12% of the Scottish population in 2004 (Sim et al. 2007). In this paper, we describe this population recovery on Orkney to 2008, and quantify the changes in sheep abundance

within their breeding and foraging areas. Secondly, we repeat vegetation and prey surveys first undertaken in the late 1990s and test the hypothesis that a reduction in sheep numbers has allowed the amount of rough grassland and the abundance of key prey species to recover. We also test if hen harrier breeding output on Orkney correlates with weather variables because previous work has shown that hen harrier breeding success is influenced by the effect of weather on prey delivery and nestling mortality (Picozzi 1984; Redpath et al. 2002b). Lastly, we explore if changes in sheep abundance and ⁄ or weather can account for the changes in breeding output of this harrier population over the last 33 years.

Materials and methods HARRIER DATA

Hen harriers have been monitored on Orkney (5910’ N, 312’ W) to a varying degree since 1953 (Amar et al. 2005). Since 1975 the same areas of moorland on West Mainland have been systematically monitored, with the total number of broods and total number of young produced being the minimum data recorded each year. This population decline dramatically between the 1970s and 1990s and intensive monitoring revealed that the key demographic change during this decline was a reduction in the proportion of breeding females (linked to lower levels of polygny) and a reduction in breeding success of secondary females, with little variation apparent in the brood size of successful nests (Amar et al. 2005). Both these factors led to a reduction in the number of broods fledging and a lower number of young produced. Due to the labour intensive methods, data on the proportion of breeding females and levels of polygny are unavailable throughout the whole study period. For this study, we instead used the total number of young produced each year between 1975 and 2008. This measure combines several variables into a single productivity estimate, including numbers of breeding females and breeding success rate, together with the small variation in brood size at fledging. During 1999 and 2000, a supplementary feeding experiment was undertaken and improved the population’s breeding performance (Amar & Redpath 2002). Therefore, data from these 2 years are excluded from our analyses examining productivity.

SHEEP AND WEATHER DATA

We obtained the total annual number of sheep from the seven regional areas or parishes (Birsay, Harray, Evie, Rendall, Firth, Orphir and Stenness) with breeding harriers on West Mainland from 1975 to 2008 from the June Agricultural Census data. Weather data for the same period was extracted from the Met Office MIDAS Land Surface Observation Stations dataset, held by the British Atmospheric Data Centre. All data came from the Kirkwall weather station (58Æ9N, 2Æ9W), situated about 20 km from the main breeding areas for harriers on West Mainland. Using the same divisions as Redpath et al.’s (2002b) analysis of weather on harrier breeding success, we summarized data from March and April as ‘spring’ and June and July as ‘summer’. Rainfall data were the sum of rainfall (mm) in both months in each season. Mean spring and summer maximum and minimum temperatures ( C) were derived from the average minimum and maximum temperatures from both months in each season. Temperature data were missing for a small number of seasons (max. spring = 4; max. summer = 2; min. spring = 4; min. summer = 2), so we used a fuller dataset from Lerwick, Shetland


222 A. Amar et al. (60Æ1N, 1Æ1W), about 100 miles north of West Mainland, to complete the Kirkwall data using predictive linear regressions (Whitfield, Fielding & Whitehead 2008) (R2 for all correlations>0Æ81).

study), along two 1-km transects, giving 100 quadrats per square. Data were collected in July or August, and although this was later than our surveys, previous work in Orkney suggests that vole indices change little between these two time periods (Amar 2001).

PREY AND HABITAT SURVEYS

Surveys of Orkney voles, lagomorphs (rabbit Oryctolagus cuniculus and brown hare Lepus capensis), and meadow pipits Anthus pratensis, and the area of rough grassland were carried out using line transects within 18 1-km squares. Squares were selected (non-randomly) to allow unobstructed observation of hunting harriers (as part of another study), and were all sited within 5 km of active harrier territories, the locations of which remained similar from year to year (Amar & Redpath 2005; Amar et al. 2008). Within each square, surveys were undertaken along two parallel transects, located at 250 m and 750 m from a randomly selected side of each square. Surveys of voles, lagomorphs and rough grassland were carried out in spring, when voles formed the largest component of the hen harrier’s diet; surveys of meadow pipits took place in summer, when meadow pipits become an important part of the diet (Amar 2001). Rough grassland measures, and vole and lagomorph sign indices were recorded simultaneously. Quadrats (25 · 25 cm) were placed every 40 m along the two transect lines within each square, giving 50 quadrats per square. Vole and lagomorph abundance was indexed using the presence or absence of fresh vole or lagomorph droppings in each quadrat, and we recorded the number of quadrats dominated by unmanaged grass. Fresh vole droppings provided the most reliable index of relative vole abundance on Orkney as estimated by simultaneous snap trapping (Oates 1996). Rough grassland was defined by a build up of dead vegetation forming a litter mat under the living vegetation, adequate to conceal a moving vole (Hewson 1982). Original surveys of voles, lagomorphs and rough grassland were undertaken in nine of the 18 squares between 23 and 31 March 1999, with the remaining nine surveyed between 2 and 24 March 2000. Repeat surveys were undertaken between 1 and 10 April 2008. Meadow pipits were surveyed using standard passerine transects along the same transect routes as the vole and vegetation surveys, between 06:00 and 09:00 h, as this period provided the highest repeatability in estimating passerine numbers (Thirgood, Leckie & Redpath 1995). Numbers of individuals seen within 200 m on either side of the transect line were recorded. Counts were undertaken between 2 July and 6 August in either 1998 or 1999 (all but two counts occurring by 20 July), with the repeat surveys undertaken slightly earlier between 24 June and 9 July 2008. This discrepancy in survey date (average of 10 days between surveys) is unlikely to have influenced our comparison of meadow pipit counts between surveys because we found no correlation between meadow pipit abundance and date (F1,34 = 0Æ29 P = 0Æ59).

LONGER-TERM VOLE MONITORING

We used longer-term vole abundance data from a separate study (Royal Society of the Protection of Birds unpublished data) to examine whether any changes found between our original and repeat surveys were likely to reflect real changes or larger-scale temporal fluctuations. Vole abundance data collected annually from 1999 to 2008 were available from three 1-km squares in West Mainland. Habitat within these squares was principally heather moorland with some rough grassland with little or no grazing. These surveys used comparable methods to our study, recording the presence of fresh vole droppings within 25 · 25 cm quadrats placed every 20 m (cf. 40 m in our

STATISTICAL ANALYSIS

Trends in the numbers of young fledged and the numbers of sheep were analysed using linear regression with a normal error structure. Changes in the abundance of prey (count of vole signs or meadow pipits) and in the number of rough grassland dominated quadrats were analysed using a Generalised Linear Mixed Model (GLMM), with a unique identifier for ‘square’ as a random factor and survey period (original or resurvey) as a categorical fixed effect. For meadow pipit abundance, the GLMM would not converge, so for this analysis we used a Generalised Linear Model with ‘square’ and survey period as fixed effects. Models were fitted with a Poisson error structures and a log link function and were corrected for over-dispersion. Denominator degrees of freedom for the GLMM were estimated using Satterthwaite’s formula (Littell et al. 1996). To examine the influence of weather variables and sheep abundance on the number of young fledged, we used a General Linear Model, with a normal error structure and an identify link function. All analyses were carried out in sas version 9.1 (SAS Institute Inc. 2004).

Results TRENDS IN HARRIER BREEDING OUTPUT AND SHEEP ABUNDANCE

The number of young hen harriers fledged declined from the end of the 1970s to a low during the 1990s before rising again at the start of the 2000s (Fig. 1). The number of young fledged declined by 79% between 1975 and 1997 (F1,.21 = 32Æ73, P < 0Æ001) and then increased by 92% between 1998 and 2008 (F1,.9 = 6Æ53, P = 0Æ03). In contrast, sheep numbers increased by 140% between 1975 and 1997 (1975 – c. 20 000, 1997 – c. 48 000; F1,.21 = 418Æ96, P < 0Æ001); and declined by about 20% between 1998 and 2008 (1998 – c. 50 000, 2008 – c. 40 000; F1,.9 = 59Æ70, P < 0Æ001) (Fig. 1). CHANGES IN HABITAT AND PREY ABUNDANCE

We found an increase in the number of quadrats dominated by rough grassland (F1,15 = 4Æ97, P = 0Æ04) and an increase in quadrats with vole signs (F1,13 = 7Æ25, P = 0Æ01) between spring 1999 ⁄ 2000 and spring 2008 (Fig. 2). However, no differences in the number of lagomorph signs were found between the two surveys (F1,20 = 0Æ60; P = 0Æ44). We also found no difference in the number of meadow pipits counted between summer 1998 ⁄ 1999 and summer 2008 (F1,17 = 21Æ25, P = 0Æ26). Longer-term data on vole abundance from three 1-km squares subject to relatively constant low levels of grazing suggested that a wider pattern of temporal fluctuations in vole abundance was unlikely to explain the increase in voles detected from the 18 squares surveyed in 2008. Indeed, these longer-term data displaying the annual fluctuations, suggested that vole abundance was actually higher in 1999 and 2000 than it was in 2008 (Fig. 3).


Grazing impacts on hen harriers 223 60 000

90 80

50 000

50 30 000 40 20 000

30

Total sheep numbers

40 000

60

20 10 000 10 0 2007

2005

2003

2001

1999

1997

1995

1993

1991

1989

1985

1987

1983

1981

1979

1977

1975

0

Year

8

12

7

10

6

Vole signs

Number of quadrats

Fig. 1. Graph showing the total number of young hen harriers fledged from West Mainland Orkney (open circles–the two close circles show the years when a diversionary feeding experiment took place) between 1975 and 2008 together with the 3-year running mean (dashed line). Also shown are the total numbers of sheep (closed squares) recorded between 1975 and 2008 from the June Agricultural Census in the seven parishes with breeding harriers.

Total young fledged

70

5 4 3

8 6 4

2

2

1

0 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Year

0 Unmanaged grass

Voles

Fig. 2. Changes between 1999 ⁄ 2000 (shaded bars) and 2008 (clear bars) in rough grass abundance and fresh vole signs (per 50 quadrats) within eighteen 1-km squares distributed throughout West Mainland, Orkney. Data are mean (±1 SE) model estimates from the GLMM fitted with survey period as a fixed effect and square as a random term in the model.

Fig. 3. Graph showing the inter-annual changes in vole signs (per 50 quadrats) between 1999 and 2008 on three 1 km squares. Squares were additional to the main study and were largely un-grazed throughout the period. Data shown are the mean counts taken in July ⁄ August across the three squares ± 1 SE. Data indicate that changes in vole abundance between 1999 and 2008 in the main study were not explained by broad-scale between year fluctuations in vole abundance.

RELATIONSHIP BETWEEN CLIMATE, SHEEP DENSITIES AND HARRIER BREEDING SUCCESS

No relationship was found between total young fledged and either spring or summer temperature or summer rainfall (Table 1a, Fig. 4). However, there was a significant negative association between spring rainfall and numbers of young fledged (Table 1a, Fig. 4). Between 1975 and 1997, spring rainfall increased (F1,.21 = 5Æ03, P = 0Æ04), although there was no trend between 1998 and 2008 (F1,.9 = 0Æ05, P = 0Æ83). We also found a highly significant relationship between the abundance of sheep and the numbers of young fledged between 1975 and 2008 (Table 1a). Combining these significant terms in a full model, both the abundance of sheep (Fig. 5) and the spring rainfall (Fig. 6) remained significant (Table 1b), with no significant interaction (F1,.28 = 1Æ03, P = 0Æ31). These two terms in the final model accounted for nearly 40% of the variation in the numbers of young fledged between years (Table 1b).

Discussion This study provides convincing evidence that the decline and subsequent recovery of the hen harrier population on Orkney was due to changes in sheep abundance. It is important to note that this conclusion is not based on just a one way relationship auto-correlated with time, but with numbers of sheep both increasing and decreasing and harriers following the converse trend. Increases in sheep are thought to have reduced the amount of rough grassland, the habitat preferred by foraging male harriers (Amar et al. 2003b; Amar & Redpath 2005) which in turn led to a decrease in the abundance of voles, an important prey species which is heavily dependent on this habitat type (Amar & Redpath 2005). During the period of harrier recovery, sheep numbers declined in the main hen harrier breeding areas on Orkney by over 20%, with the loss of around


224 A. Amar et al. Table 1. Relationships between the number of young hen harriers fledged on West Mainland, Orkney and the spring and summer climate variables (rainfall and min. and max. temperature) and sheep abundance recorded on Orkney from 1975 to 2008 (omitting 1999 and 2000), (a) univariate relationships for each term, (b) final full GLM model including only terms significant at the univariate stage. Significant terms are presented in bold Variable

Intercept

Estimate

d.f.

(a) Spring rainfall Summer rainfall Spring max temp Summer max temp Spring min temp Summer min temp Total sheep

67Æ05 53Æ31 36Æ83 10Æ05 30Æ76 )2Æ40 77Æ28

)0Æ200 )0Æ160 )0Æ123 1Æ88 1Æ469 3Æ901 )0Æ001

1,30 1,30 1,30 1,30 1,30 1,30 1,30

(b) Spring rainfall Total sheep

)0Æ149 )0Æ001

96Æ16

Residual young fledged per year

Rainfall (mm)

0Æ01 0Æ08 0Æ98 0Æ67 0Æ74 0Æ47 90% of the trees of the same species, even-aged and consisting of one layer) were considered negative for biodiversity, and plots with these registrations were excluded from the selection (1005 plots) since it would not be realistic to establish a reserve in an area containing a plantation.

There were large differences in points between indicators; therefore a method that could accommodate these variations was essential. When using ordinary linear programming (LP) it is necessary to find weights for each indicator to include into one single objective function. We used a goal programming approach to allow impartial treatment of all indicators and avoid manual weight determination. In a two-phase approach, we first found the best possible outcome for each indicator, which became a goal. In the second phase, we searched for a solution that was as close as possible to each individual indicator goal but that considered all indicators at the same time. Each of the models used are described below (see Table 2 for parameters and decision variables). The first LP problem can be formulated as follows: XXX ½P1 max z ¼ we pite xit eqn 1 i2I t2T e2E

subject to XX cit xit b

eqn 2

i2I t2T

XX i2I t2T

xit q

XX

ait

eqn 3

i2I t2T

xit ait ; 8i 2 I; t 2 T

eqn 4

xit 0; 8i 2 I; t 2 T

eqn 5

The objective function, eqn 1, maximizes the sum of the points from the biodiversity indicators in the selected areas (hereafter referred to as the biodiversity indicator score). Constraint set in eqn 2 is the budget constraint preventing the total cost of the selected areas to exceed the available budget (b). Constraint set in eqn 3 is the area constraint, preventing the total selected area from exceeding a certain proportion (q) of the total area. Constraint set in eqn 4 ensures that the area selected from each 50 · 50 km plot and age class is smaller than its existing area, and set in eqn 5 is the non-negativity constraints on the decision variables. It is difficult to establish weights we that can be considered to be appropriate (Polasky, Camm & Garber-Yonts 2001). Typically, an indicator with large value will dominate and the solution tends to


136 J. Lundstro¨m et al. Table 1. List of biodiversity indicators and criteria for points Indicator

100 points

50 points

0 points

Uneven age1 Gaps2 Stand character3 Tree layer4 Ground structure5 Large pine Large spruce Large birch Large aspen Large deciduous tree (other than aspen or birch) Dead conifer tree lying Dead deciduous tree lying Dead conifer tree standing Dead deciduous tree standing Presence of rowan Affected by water (moving water ⁄ spring ⁄ temporarily flooded) Volume of dead wood

Not even-aged Several gaps Pristine Fully layered ⁄ several layers Very uneven ⁄ fairly uneven >40 cm dbh >40 cm dbh >40 cm dbh >40 cm dbh >40 cm dbh Tree > 20 cm dbh Tree > 20 cm dbh Tree > 20 cm dbh Tree > 20 cm dbh Present Yes >20 m3 ha)1

Fairly even-aged Some gaps

Completely even-aged No gaps Normal One layer ⁄ no layer Very even Not present Not present Not present Not present Not present Not present Not present Not present Not present Not present No

Two layers Fairly even >30 cm dbh >30 cm dbh >30 cm dbh >30 cm dbh >30 cm dbh

£ 20 m3 ha)1*

*Normalized point according to the volume of dead wood ha)1, from 0 to 100. 1 Totally even-aged: > 95% of the volume within an age interval of 5 years, fairly even-aged: > 80% of the volume within an age interval of 20 years; remaining stands classed as uneven aged. 2 Gap: an area without main crop seedlings ⁄ main trees larger than a square with a length of 2Æ5 times the average distance between main crop seedlings ⁄ main trees, but at least 5 m. Several gaps: at least 4 gaps within a 20 m radius from the centre of the plot, some gaps: 2–3 gaps; remaining stands are classed as no gaps. 3 Pristine character: presence of coarse (> 25 cm diameter) dead wood and no trace of management actions during the last 25 years. 4 Tree layer: group of trees amongst which the height is approximately the same, but their mean height differs from other layers. Fully layered: all diameter classes represented, the biggest tree > 20 cm in diameter, the number of stems increasing with increasing diameter class, and the volume density (relationship between the actual volume in the stand and the potential volume) > 0Æ5. 5 Ground structure: Classification based on height and frequency of irregularities (rocks, small hills and holes) on the ground.

Table 2. Parameters and decision variables for the model Notation

Description

Parameters I T E pite ait cit we q b

Set of 50 · 50 km plots (i = 1,...,n) Set of age classes (t = 1,...,m) Set of biodiversity indicators (e = 1,...,o) Point of biodiversity indicator e in plot i and age class t Area (ha) of plot i in age class t Cost ha)1 of plot i and age class t Weight of biodiversity indicator e Maximum proportion that can be selected Available budget (SEK)

Decision variables xit

Area (ha) selected in plot i and age class t

select areas with high values for one (or maybe a few) indicators. Goal programming is an approach that includes several objectives (expressed as goals) in the same objective function and still allows a trade-off that is considered impartial. In goal programming, we establish goals in phase 1. In our case, we simply solved the problem [P1] as many times as we had different indicators. When we solved [P1] for indicator e, we set we to 1 for that indicator and 0 for all other indicator weights. This means that we independently searched for the best possible value for each indicator. We let those 17 values be denoted as ze. In the second phase, we wanted to find a solution in which all indicators were as close as possible to their goal. Since it would not be

possible to reach the goal of each indicator, a quadratic deviation from these goals was minimized. The goal programming model in phase 2 can be formulated as !2 X XX ½P2 min w ¼ ðze pite xit Þ=ze eqn 1b e2E

i2I t2T

subject to eqns 2–5 In this objective function, eqn 1b, the squared difference between the goal and the actual biodiversity indicator score for each indicator is minimized. The difference is scaled with the goal value and hence we measure the deviation as a percentage deviation. Problem [P2] is a quadratic programming problem. The problem is convex (Lundgren, Ro¨nnqvist & Va¨rbrand 2010) so a global optimal solution is guaranteed. The models were formulated in the modelling language AMPL and solved using the software CPLEX 11Æ2 (ILOG 2006). All tests were conducted on a standard PC with 2Æ99 GHz and 3Æ25 GB of internal memory. The number of variables (in both models) was 9520 (112*5*17) and the number of constraints was 562 (1 + 1 + 112*5). The solution time for each problem was within a fraction of a second.

Results BIODIVERSITY INDICATORS

In the original NFI data used for analysis, the biodiversity indicators were distributed unevenly over the five age classes, but with all indicators represented in all age classes (Table 3). The


Cost-effective boreal forest reserves 137 Table 3. Biodiversity indicator data from NFI (based on individual plots) with mean points ± standard deviation, as well as total area and total cost for the five age classes Age class 0–14 Biodiversity indicator Uneven age Gaps Stand character Tree layer Ground structure Large pine Large spruce Large birch Large aspen Large deciduous tree (other than aspen or birch) Dead conifer tree lying Dead deciduous tree lying Dead conifer tree standing Dead deciduous tree standing Presence of rowan Affected by water (moving water ⁄ spring ⁄ temporarily flooded) Volume of dead wood Total area (1000 ha) Total cost (billion SEK)

7Æ8 15Æ7 0Æ03 24Æ5 35Æ3 5Æ8 0Æ2 0Æ4 0Æ3 0Æ1 13Æ2 3Æ6 4Æ9 1Æ0 34Æ4 0Æ8

15–39

± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±


23Æ2 ± 34Æ1 2346 21Æ6


± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

40–69


10Æ3 ± 23Æ1 3396 67Æ8


± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±

‡ 100

70–99


17Æ1 ± 29Æ3 2975 87Æ8


± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±


29Æ6 ± 37Æ5 2021 74Æ7


± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±


37Æ1 ± 40Æ1 3550 141Æ6

The points for the indicator ‘Volume dead wood’ were given proportionally to the volume ha)1, with volumes > 20 m3 ha)1 given 100 points. The actual volumes per 1000 ha are shown.

relative magnitude of the mean points should not be interpreted as a sign of importance since the optimization models neutralized the advantage of common indicators. Instead differences between age classes are of interest. More than half of the 17 indicators peaked at ages > 100 years. Several of the large-tree indicators and all deadwood indicators had a higher mean point in the youngest forests (0–14 years) than in the subsequent age class. All deadwood indicators had the lowest values at intermediate age classes. ‘Uneven age’ increased over time, whereas ‘rowan’ decreased.

OPTIMAL AGE DISTRIBUTIONS

To investigate the question of whether the optimal combination of forest ages differed when a budget constraint or an area constraint were used, two versions of the stated model were solved. In the first version, budget was limiting (i.e. the area constraint, eqn 3, was omitted). This model was solved 100 times with an incremental increase in budget, starting at 1% of the total cost of all forest, up to 100%, with automatic registration of age distribution in each stage, resulting in 1800 ((17 + 1)*100) optimizations. In the second version, area was limiting and the budget constraint, eqn 2, was omitted. This model was also solved 100 times, with an incremental increase in area limit, starting at 1% of the total area, up to 100%, also with automatic registration of age distribution in each stage. The optimal age distributions when cost was limiting differed markedly from the optimal age distributions when area was limiting. With a budget-constrained approach, a large proportion of young forest was chosen at small budgets (Fig. 2a)

whereas forests in the 40- to 99-year age class were selected to a lower extent. However, when the selection was made with an area-constrained approach, the proportion of old forest was clearly more dominant (Fig. 2b). Forests in the 15- to 39-year age class were selected the least when costs were not considered, but younger forests were also selected in small proportions at low area limits. In general, the area-constrained approach covered less area but with higher biodiversity indicator scores, whereas the budget-constrained approach covered more area, but with lower biodiversity indicator scores (Fig. 3). The budget-constrained approach achieved a higher biodiversity indicator score compared to the area-constrained approach at any given cost (Fig. 3a).

GEOGRAPHICAL DISTRIBUTION

Two scenarios were used to analyse the differences in geographical distribution of selected forests between a budget-constrained model and an area-constrained model. In the first scenario, a budget was set to 10 billion SEK (2Æ5% of the cost for the total area). The limit was chosen based on the current political targets in Sweden for nature reserve establishment, with 6 billion SEK allocated to forest protection for the years 1998–2008 (Swedish Government 2009). The area limit was set to 714 000 ha (5% of the total area) since this scenario gave approximately the same biodiversity indicator score. The 10 billion SEK budget scenario led to a reserve area of 1Æ2 million ha with a strong bias for selection of areas in the north-western section of boreal Sweden (Fig. 4a). The scenario with an area limit of 714 000 ha corresponded to a cost


138 J. Lundstro¨m et al. (a)

(a)

(b)

(b)

Fig. 3. The biodiversity indicator score plotted as a function of (a) cost and (b) area for the budget-constrained and area-constrained model.

Discussion Fig. 2. Optimal age distributions of forest reserves in boreal Sweden plotted as a function of (a) cost and (b) area. The age distributions toward the left in the graphs are most relevant for the actual situation in Sweden, with about 6 billion SEK allocated to forest protection during the last 10-year period (Swedish Government. 2009), and with an environmental target of protecting an additional 900 000 ha. When the limits increase and approach the total area or total cost (the right hand side of the graphs) the age distribution equals the original distribution in the data set.

of 41 billion SEK, and was strikingly different with strong representation in the south-eastern part of the boreal region (Fig. 4b). As in the analysis on optimal age-distributions, the forests chosen in the budget-constrained scenario were mostly young, whereas the forests chosen in the area-constrained scenario were mostly old (Table 4). The biodiversity indicator score in the scenarios with a budget constraint and an area constraint were both based on contributions from all indicators (Table 5). Indicators on dead wood were more represented in the budget-constrained scenario whereas large trees were much more represented in the area-constrained scenario. The goal programming approach led to a biodiversity indicator score in which the contributions of all indicators were higher, in some cases substantially, than the mean of their contribution when each indicator was maximized separately (Table 5).

The results clearly show that it is more cost-effective to protect young forests and more area-effective to protect old forests, but that a combination of age classes always gives the highest biodiversity indicator score. This indicates that all age classes have a value to biodiversity, and that a reserve network ideally should consist of forests of different ages regardless of whether the selection is limited by budget or by area. This is also challenging since it demonstrates that there is a need to reorient current boreal forest conservation strategies, which almost exclusively target the oldest forests. It was evident that it is more cost-effective to use a budgetconstrained model compared to an area-constrained model when selecting reserves. Previous studies have shown that when land prices vary and area is used as a limitation, more money than necessary is spent, which is unfortunate since conservation is always restricted by scarce resources (Ando et al. 1998; Polasky, Camm & Garber-Yonts 2001). When using an area-constrained model the same biodiversity indicator score can be obtained in a smaller total area. Decision makers, therefore, have to integrate ecological and economic data and balance short- and long-term constraints in terms of cost and area in order to design cost-effective conservation strategies (Polasky, Camm & Garber-Yonts 2001; Juutinen et al. 2004; Messer 2006; Naidoo et al. 2006).


Cost-effective boreal forest reserves 139 (a)

(b)

Fig. 4. The proportion of the total area selected in each 50 · 50 km plot with (a) a budget limit of 10 billion SEK (9% of total area) or (b) an area limit of 715, 000 ha (5% of total area). The biodiversity indicator score in both scenarios was approximately the same (51 million in the budget-constrained scenario and 50 million in the areaconstrained scenario). For names of geographical regions see Fig. 1.

The analyses show that if costs are considered, large areas of young forests in north-western Sweden are selected. There are already numerous large nature reserves in this section of the boreal region, and more than 75% of all protected forests in the country are found here. However, the present reserves are overwhelmingly old, and by setting aside young, structurally rich forests nearby, dispersal and colonization of some species might be facilitated. One risk of concentrating reserves in the northwest region is that rare species confined to more eastern regions would not be protected When costs are not considered, as in the area-constrained model, old forests are mostly selected and these are primarily located in the south-eastern section of the study area. One reason for this southern dominance could be a higher productivity, leading to higher representation of large trees, tentatively a subject to be further scrutinized in subsequent studies along with complementarity analyses between our model selections and existing reserves. The NFI data showed that a majority of the structure-based variables of importance to biodiversity are most common in the oldest age classes, but that there are substantial amounts also found in younger forests. For the youngest age class, 0–14 years, this mainly reflects the practice of tree retention, (leaving trees for biodiversity at clear cutting) introduced in Sweden and other countries a few decades ago. However, at least in terms of dead wood the same pattern is also found following natural disturbances such as forest fires or storms, with plenty of dead wood present in the early successional stages, whilst amounts are lower at intermediate ages and then higher again in old-growth stages (Siitonen 2001). Much of the high conservation values of young forests reported here are thus likely to prevail both in young forests originating from management practices and in those originating from natural disturbances, although the latter would probably host higher amounts of dead wood and large living trees (Uotila et al. 2001). The biodiversity indicators chosen were to some extent biassed towards features most common in old-growth forests (e.g. uneven age, multi-layered canopy and presence of large trees). A more unbiassed list of indicators would also include those that are important for rare species that prefer young age classes, such as sun-exposed dead trees (Kaila, Martikainen & Punttila 1997; Jonsell, Weslien & Ehnstro¨m 1998). Including a larger proportion of such indicators would further strengthen the result that young forests host important biodiversity potential. There are some weaknesses in the structure-based approach. The same structures can be present in young and old forests, but support different species compositions, mainly due to


140 J. Lundstro¨m et al. Table 4. Area distribution (proportion of selected area, %) in the six geographical regions and five age classes selected under a budget constraint of 10 billion SEK (9% of the total area) and an area constraint of 714 000 ha (5% of the total area)

15–39 years

40–69 years

70–99 years

‡ 100 years

All age classes

Budget constraint 10 billion SEK (area 1Æ2 million ha) Norrbotten 22Æ6 Va¨sterbotten 10Æ4 Ja¨mtland 14Æ8 Va¨sternorrland and Ga¨vleborg 14Æ6 Dalarna 6Æ3 Va¨rmland and O¨rebro 2Æ3 All regions 70Æ8

9Æ8 3Æ7 1Æ2 0 1Æ2 0 15Æ8

0 0Æ8 0Æ2 0 0Æ8 0 1Æ8

1Æ2 0 0Æ5 0 0 0 1Æ8

5Æ6 2Æ4 1Æ8 0 0 0 9Æ8

39Æ2 17Æ2 18Æ4 14Æ6 8Æ3 2Æ3 100

Area constraint 714 000 ha (cost 41 billion SEK) Norrbotten 0 Va¨sterbotten 0 Ja¨mtland 0 Va¨sternorrland and Ga¨vleborg 1Æ5 Dalarna 0 Va¨rmland and O¨rebro 0 All regions 1Æ5

0 0 0 0 0 0 0

0 0 0Æ3 0 0 16Æ7 17Æ0

0 0 0Æ7 15Æ8 3Æ2 13Æ4 33Æ0

0Æ6 4Æ2 10Æ6 9Æ0 7Æ8 16Æ3 48Æ5

0Æ6 4Æ2 11Æ6 26Æ2 11Æ0 46Æ4 100

0–14 years

Table 5. The contribution to the biodiversity indicator score of each indicator from both phases in the goal programming under a budget constraint and an area constraint Budget scenario

Area scenario

Biodiversity indicator

Max1

Mean2

Min3

% of mean4

Max

Mean

Min

% of mean

Uneven age Gaps Stand character Tree layer Ground structure Large pine Large spruce Large birch Large aspen Large deciduous tree Dead conifer tree lying Dead deciduous tree lying Dead conifer tree standing Dead deciduous standing Presence of rowan Affected by water Volume of dead wood5

56986 48442 7306 59801 69991 14910 9682 2944 1953 1574 32103 11012 12124 5090 62552 5789 66677

26488 24171 1265 32115 32317 6005 2258 542 313 234 14262 3907 4746 1150 27191 1106 33714

15502 8801 0 14258 9415 2559 139 149 36 49 4814 1199 2351 199 7429 301 13509

145 143 131 182 211 109 138 138 176 166 149 123 140 127 261 125 150

79266 43714 9832 58258 61775 38861 30494 6176 3780 2498 24202 10475 19116 7990 57944 8756 70850

52006 24332 2122 39889 35602 16751 13372 1873 860 510 11366 3240 8924 2429 24975 1934 39003

25247 18587 100 28864 21916 8570 4019 398 150 149 4995 1804 4108 700 10306 700 18945

112 110 133 164 201 120 115 106 224 202 179 141 173 103 179 111 127

1 The biodiversity indicator score when maximizing each indicator separately (the goal) gives a maximal sum of points that each indicator can obtain (when the optimization is made only considering that specific indicator). 2 A mean sum of points from all 17 goal optimizations. 3 The lowest sum of points that the indicator gets from another indicator’s goal optimization. 4 The minimization of all indicator’s quadratic percentage deviation from their goal (phase 2 in the goal programming) gives a contribution from each indicator to the biodiversity indicator score shown here as the percentage of its mean value. 5 Shown in units of 1000 points.

differences in microclimatic conditions and colonization opportunities. Further, the presence of structures is no guarantee of the presence of associated species since other aspects, such as forest history and connectivity might be decisive for species occurrences (Nilsson, Hedin & Niklasson 2001). Therefore, a further development of our study would be to repeat the analyses for species distribution data and compare the results with those based on structural data. There are detailed data on

occurrences of red-listed species from organism groups such as vascular plants, birds, bryophytes, lichens and fungi in boreal Sweden which could be used for such a comparison. The proposed goal programming approach provides impartial and objective weighting. We used the same weighting for analysis of the deviation from the goals, but this can be easily modified if desired. If weights are decided manually, care must be taken since the results can greatly disadvantage some


Cost-effective boreal forest reserves 141 variables (Table 5, ‘Min’ column). We note that the importance of different indicators are indirectly weighted when deciding criteria for points, but those decisions are based on existing knowledge of which features are important for biodiversity. A further development of the model could potentially be to add specific requirements regarding geographical distribution or a minimum amount of different indicators. This, however, needs to be substantiated by high-quality ecological studies on the critical requirements of different species. In northern Europe, with its long history of relatively intense forest use, there are so few old-growth forests left that areas strongly impacted by humans need to be included when new forests are protected. Consequently, it is vital to prioritize the few high-quality old-growth remnants that still exist, although our analyses indicate that it is more cost-effective to include young forests in reserve networks. The young forests that we propose for protection have a decidedly different character than those normally regenerating after clear-cutting, even with tree retention. A careful selection will be needed for sites especially rich in dead wood, remnant live trees and other qualities of importance to biodiversity. Protection of young forest allows much more land to be set aside than protection of oldgrowth forest, due to lower net present values. A shift towards more protection of young forest might therefore eventually cause a reduction in timber volumes available for forest industry. A novel conservation strategy, and a future research challenge, is to analyse if some reserves with old forests could be systematically replaced by younger forests without causing biodiversity decline at the landscape level. Possibly, such a dynamic reserve scheme could benefit both timber production and biodiversity protection. In general, early-successional stages are overlooked in forest conservation. Thus, our approach with protection of different age classes has a general interest also for other biomes where biodiversity is adapted to frequent disturbances and where early successional stages are common in natural forest landscapes.

Acknowledgements We thank Per Nilsson for help with the NFI-data and Peder Wikstro¨m and Torgny Lind for assistance during the Heureka calculations. We also thank Viktor Johansson and three anonymous reviewers for constructive comments on the manuscript. The study was financially supported by FORMAS.

References Ahti, T., Ha¨met-Ahti, L. & Jalas, J. (1968) Vegetation zones and their sections in northwestern Europe. Annales Botanici Fennici, 5, 169–211. Ando, A., Camm, J., Polasky, S. & Solow, A. (1998) Species distributions, land values, and efficient conservation. Science, 279, 2126–2128. Angelstam, P. (1998) Maintaining and restoring biodiversity in European boreal forests by developing natural disturbance regimes. Journal of Vegetation Science, 9, 593–602. Anonymous. (1932) Uppskattning av Sveriges skogstillga˚ngar. Jordbruksdepartementet, Stockholm, Sweden (in Swedish). Anonymous. (2005) RIS Riksinventeringen av skog. Fa¨ltinstruktion 2005. Institutionen fo¨r skoglig resurshusha˚llning och geomatik SLU, Umea˚, Sweden (in Swedish). Anonymous. (2007a) Skogsstatistisk a˚rsbok 2007. Sveriges officiella statistik, Skogsstyrelsen, Jo¨nko¨ping, Sweden (in Swedish).

Anonymous. (2007b) Skogsdata 2007. Sveriges officiella statistik, Institutionen fo¨r skoglig resurshusha˚llning, SLU, Umea˚ (in Swedish). Esseen, P.A., Ehnstro¨m, B., Ericson, L. & Sjo¨berg, K. (1997) Boreal forests. Ecological Bulletins, 46, 16–47. FAO (Food and Agriculture Organization of the United Nations). (2006) Global Forest Resources Assessment 2005: Progress towards sustainable forest management. FAO, Rome, Italy. Ferris, R. & Humphrey, J.W. (1999) A review of potential biodiversity indicators for application in British forests. Forestry, 72, 313–328. Ga¨rdenfors, U. (Ed.). (2005) The 2005 Red List of Swedish Species. Swedish Species Information Centre, Swedish University of Agricultural Sciences, Uppsala, Sweden. Gustafsson, L. & Ahleń, I. (1996) The National Atlas of Sweden. Geography of Plants and Animals. SNA, Stockholm, Sweden. Hagar, J.C. (2007) Wildlife species associated with non-coniferous vegetation in Pacific Northwest conifer forests: a review. Forest Ecology and Management, 246, 108–122. Ibbotson, R.G. & Sinquefield, R.A. (1986) Stocks, Bonds, Bills and Inflation: 1986 Yearbook. Ibbotson Associates, Chicago. ILOG (2008) CPLEX system 11.2. Gentilly, France. Jonsell, M., Weslien, J. & Ehnstro¨m, B. (1998) Substrate requirements of redlisted saproxylic invertebrates in Sweden. Biodiversity and Conservation, 7, 749–764. Junninen, K., Simila, M., Kouki, J. & Kotiranta, H. (2006) Assemblages of wood-inhabiting fungi along the gradients of succession and naturalness in boreal pine-dominated forests in Fennoscandia. Ecography, 29, 75–83. Juutinen, A., Mantymaa, E., Mo¨nkko¨nen, M. & Salmi, J. (2004) A cost-efficient approach to selecting forest stands for conserving species: a case study from northern Fennoscandia. Forest Science, 50, 527–539. Kaila, L., Martikainen, P. & Punttila, P. (1997) Dead trees left in clear-cuts benefit saproxylic Coleoptera adapted to natural disturbances in boreal forest. Biodiversity and Conservation, 6, 1–18. Kouki, J., Lo¨fman, S., Martikainen, P., Rouvinen, S. & Uotila, A. (2001) Forest fragmentation in Fennoscandia: linking habitat requirements of woodassociated threatened species to landscape and habitat changes. Scandinavian Journal of Forest Research, 16, 27–37. Kuuluvainen, T. (2002) Natural variability of forests as a reference for restoring and managing biological diversity in boreal Fennoscandia. Silva Fennica, 36, 97–125. Kuuluvainen, T. (2009) Forest management and biodiversity conservation based on natural ecosystem dynamics in Northern Europe: the complexity challenge. AMBIO: A Journal of the Human Environment, 38, 309–315. La¨ma˚s, T., Dahlin, B., Olsson, H., Sallna¨s, O., Stenlid, J. & Sta˚hl, G. (2006) Preface. Scandinavian Journal of Forest Research, 21, 3–4. Lindenmayer, D.B. & Franklin, J.F. (2003) Towards Forest Sustainability. CSIRO Publishing, Melbourne, Australia. Lindenmayer, D.B., Margules, C.R. & Botkin, D.B. (2000) Indicators of biodiversity for ecologically sustainable forest management. Conservation Biology, 14, 941–950. Lo¨fman, S. & Kouki, J. (2001) Fifty years of landscape transformation in managed forests of Southern Finland. Scandinavian Journal of Forest Research, 16, 44–53. Lundgren, J., Ro¨nnqvist, M. & Va¨rbrand, P. (2010) Optimization. Studentlitteratur, Lund, Sweden. Margules, C.R. & Pressey, R.L. (2000) Systematic conservation planning. Nature, 405, 243–253. Messer, K.D. (2006) The conservation benefits of cost-effective land acquisition: a case study in Maryland. Journal of Environmental Management, 79, 305–315. Naidoo, R., Balmford, A., Ferraro, P.J., Polasky, S., Ricketts, T.H. & Rouget, M. (2006) Integrating economic costs into conservation planning. Trends in Ecology & Evolution, 21, 681–687. Nilsson, S.G., Hedin, J. & Niklasson, M. (2001) Biodiversity and its assessment in boreal and nemoral forests. Scandinavian Journal of Forest Research, 16, 10–26. O¨stlund, L., Zackrisson, O. & Axelsson, A.L. (1997) The history and transformation of a Scandinavian boreal forest landscape since the 19th century. Canadian Journal of Forest Research, 27, 1198–1206. Polasky, S., Camm, J.D. & Garber-Yonts, B. (2001) Selecting biological reserves cost-effectively: an application to terrestrial vertebrate conservation in Oregon. Land Economics, 77, 68–78. Ranneby, B., Cruse, T., Ha¨gglund, B., Jonasson, H. & Swa¨rd, J. (1987) Designing a New National Forest Survey for Sweden. Swedish University of Agricultural Sciences, Uppsala, Sweden.


142 J. Lundstro¨m et al. Siitonen, J. (2001) Forest management, coarse woody debris and saproxylic organisms: Fennoscandian boreal forests as an example. Ecological Bulletins, 49, 11–41. Spanos, K. & Feest, A. (2007) A review of the assessment of biodiversity in forest ecosystems. Management of Environmental Quality: An International Journal, 18, 475–486. Spies, T.A. & Franklin, J.F. (1991) The Structure of Natural Young, Mature, and Old-growth Douglas-fir Forests in Oregon and Washington. U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station, Portland, OR, USA. Swanson, M.E., Franklin, J.F., Beschta, R.L., Crisafulli, C.M., DellaSala, D.A., Hutto, R.L., Lindenmayer, D.B. & Swanson, F.J. (2010) The forgotten stage of forest succession: early-successional ecosystems on forest sites. Frontiers in Ecology and the Environment, in press. doi: 10.1890/090157. Swedish Government. (2005) Svenska miljo¨ma˚l – ett gemensamt uppdrag. Proposition 2004 ⁄ 2005:150. Swedish Government, Stockholm, Sweden (in Swedish).

Swedish Government. (2009) Ha˚llbart skydd av naturomra˚den. Proposition 2008 ⁄ 09:214. Swedish Government, Stockholm, Sweden (in Swedish). Tikkanen, O., Martikainen, P., Hyvarinen, E., Junninen, K. & Kouki, J. (2006) Red-listed boreal forest species of Finland: associations with forest structure, tree species, and decaying wood. Annales Zoologici Fennici, 43, 373–383. Uotila, A., Maltamo, M., Uuttera, J. & Isoma¨ki, A. (2001) Stand structure in semi-natural and managed forests in eastern Finland and Russian Karelia. Ecological Bulletins, 49, 149–158. Zackrisson, O. (1977) Influence of forest fires on the North Swedish boreal forest. Oikos, 29, 22–32. Received 9 March 2010; accepted 20 October 2010 Handling Editor: Harald Bugman



doi: 10.1111/j.1365-2664.2010.01898.x

Variation partitioning in canonical ordination reveals no effect of soil but an effect of co-occurring species on translocation success in Iris atrofusca Sergei Volis1*, Michael Dorman1, Michael Blecher2, Yuval Sapir3 and Lev Burdeniy1 1

Life Sciences Department, Ben-Gurion University of the Negev, Beer Sheva, 84105 Israel; 2Ein Gedi Nature Reserve, Israel Nature and Parks Authority, Dead Sea, 86980 Israel; and 3The Botanical Garden, Department of Plant Sciences, Tel Aviv University, Tel Aviv, 69978 Israel

Summary 1. Despite being expensive, complicated and less successful than the conservation of primary habitat, translocation is rapidly gaining importance as a conservation approach due to accelerated loss of natural environment. Finding the optimal abiotic and biotic conditions needed for successful translocation of plants can be difficult for species with limited information on prior distribution. Unfortunately, this is often the case with endangered plant species, including those urgently needing action. 2. We present a method of evaluating the relative importance of multiple environmental parameters in translocation success. This method is based on the application of variation partitioning in canonical ordination and it allows usage of not only multiple independent biotic and abiotic variables, but also multiple dependent variables for fitness estimates. 3. In this study, six soil parameters together with the abundance of 61 plant species and their total biomass were used to explain the variation in translocation success of Iris atrofusca plants among 22 microsites. The relative importance of each of the three factors was estimated using ordination techniques. 4. Soil characteristics and total biomass of other plants did not significantly affect the performance of translocated irises, but the species composition of the surrounding vegetation did have a significant effect. The abundance of relatively rare species was closely correlated with iris performance. It is likely that these species do not affect the irises directly but instead represent environmental conditions not measured in this study, which are necessary for the survival of irises. 5. Synthesis and applications. Variation partitioning appears to be a highly promising method for planning the translocation of plants and evaluating success due to its ability to estimate the unique contribution of each of two or more sets of environmental factors. It can be used to monitor success, and to identify the key contributory factors, in experimental translocations preceding actual introduction of plants in conservation programmes. Key-words: canonical correspondence analysis, endangered plant species, habitat-suitability, niche space, redundancy analysis, soil characteristics, species abundance, variation partitioning

Introduction Translocation of plants refers to their accidental or deliberate movement, within or beyond their natural range, by humans. The goal of translocations for conservation purposes is to establish new populations of rare and endangered species in order to increase the survival of the species as a whole (Hey-

*Correspondence author. E-mail: [email protected]

wood & Iriondo 2003). This practice may become more prevalent as human impact on the natural environment increases, even though it is more expensive, complicated and less successful than the conservation of primary habitat (e.g. Gordon 1996; Milton et al. 1999). Currently, the evidence for the long-term success of translocations is limited (Maunder 1992; Seddon, Armstrong & Maloney 2007) and the reasons for success or failure can be difficult to determine (e.g. Morgan 1999). In the past, many translocation projects were performed without scientific rigour


266 S. Volis et al. or hypothesis testing. Today, it is recognized that the factors determining success or failure of a translocation have to be studied using a scientifically based approach combining ecological theory and empirical tests (Sarrazin & Barbault 1996; Seddon, Armstrong & Maloney 2007; Menges 2008). When plants are moved to a site outside the known native range (termed ‘introduction’; IUCN 1987) one of the most important decisions is selecting the relocation site, and then the best microsites within the relocation site (e.g. Adamec & Lev 1999; Jusaitis 2005; Maschinski & Duquesnel 2006; Colas et al. 2008). By definition, the location where a population can be established has to be within the ecological niche of the species, i.e. the set of abiotic and biotic conditions under which the species can maintain populations without immigration (Grinnell 1917). However, determination of the species niche is not an easy task because the actual distribution of the species can be limited due to interspecific interactions (the difference between fundamental and realized niches, MacArthur 1972), as well as limited colonization ability and local extinction (Burkey 1995; Peterson, Soberoń & Sańchez-Cordero 1999). The method currently used for identifying the most suitable habitats for species is to search for a quantitative relationship between ecological and environmental features in the landscape and either (i) species occurrence in space prior to relocation or (ii) population establishment after relocation. Much progress has been made in the last two decades in the development of techniques predicting species distribution and then estimating potential site suitability for establishment (reviewed in Guisan & Zimmermann 2000; Stauffer 2002; Guisan & Thuiller 2005; Richards, Carstens & Lacey Knowlrs 2007; Elith & Leathwick 2009). These techniques generally involve the use of spatially explicit data through geographic information systems (GIS) and modelling a species’ ecological niche. Particular examples include BIOCLIM (Busby 1991), HABITAT (Walker & Cocks 1991), DOMAIN (Carpenter, Gillison & Winter 1993), Genetic Algorithm for Rule-set Prediction (GARP) (Stockwell & Peters 1999) and Ecological-Niche Factor Analysis (ENFA) (Hirzel et al. 2002; Engler, Guisan & Rechsteiner 2004; Basille et al. 2008). These methods require detailed maps of species presence and environmental parameters, which limit the application of species distribution modelling to (i) large geographic scale of kilometres and (ii) species with documented and relatively wide distributions. Species that occupy only a limited number of locations including a patchy distribution at the fine geographic scale (tens or hundreds of metres) or where there is little information on prior distribution, require another approach. In these cases, an alternative approach is to analyse the success of an experimentally translocated population. As the environmental variables affecting performance of relocated organisms can be complex, they need to be identified and differentiated using multivariate statistical methods. One approach is to use stepwise multiple regression to develop a predictive equation for success of translocation in which independent variables will be ranked by their importance (Griffith et al. 1989). However, this approach requires a single dependent variable for translocation success, which can be binary

(success vs. failure), ordinal (varying degrees of success) or continuous (e.g. population growth rate, percentage of survived plants or populations that became self-sustained). A requirement for a single variable summarizing the existing information on relocated plants ⁄ populations limits the application of this approach because many estimates of fitness are stage specific and vary in time. A potentially more efficient approach is one using not only multiple independent variables representing different environmental biotic and abiotic effects, but also multiple dependent variables for fitness estimates. Canonical correspondence analysis (CCA) and redundancy analysis (RDA) are constrained multivariate ordination techniques widely used in ecology and vegetation science to extract the major gradients in response (dependent, usually biotic) variables attributed to the explanatory (independent, usually environmental) variables (e.g. Pivello, Shida & Meirelles 1999; Clarke, Latz & Albrecht 2005; Svenning & Skov 2005). A particular strength of CCA and RDA is their ability to remove the effect of undesirable variables (covariables) by regressionbased covariance analysis prior to the analysis itself. This procedure is called partial canonical correspondence analysis (pCCA) or partial redundancy analysis (pRDA; ter Braak 1986). It is possible to measure the fraction of the variation in the dependent variables explained by each set of environmental variables alone as well as the fraction of the variation shared by the sets of variables by using CCA and pCCA (or RDA and pRDA) (Borcard, Legendre & Drapeau 1992; Noe & Zedler 2001; Volis et al. 2004). High analytical power and the ability to efficiently reduce large data sets of both independent and dependent variables into only a few canonical axes make CCA and RDA attractive for finding a range of suitable environmental conditions for successful relocation. Specifically, CCA ⁄ RDA can help to find which environmental factors have the highest effect on individual performance and also the unique contribution of each of two or more sets of environmental factors to explaining the variation in individual performance. For example, species composition may significantly affect the performance of introduced plants (Elmendorf & Moore 2007). However, a hypothetical situation is possible when the vegetation effect may be indirect and reflecting other effects, such as differences in soil properties. Thus, both vegetation and soil will appear important for successful relocation, while in fact only the soil is important. Iris atrofusca Baker is a highly endangered species in Israel (Shmida & Pollak 2007), with habitats in the Northern Negev being the most vulnerable throughout its distribution. Rapid destruction of the natural habitat of I. atrofusca due to land clearing and a lack of nature reserves containing populations of I. atrofusca in the Negev leave very limited conservation options for this species. Thus there is no alternative to translocation, i.e. introduction of the species into seemingly suitable protected areas with no record of prior occupancy. The habitat characteristics needed for success are not known for this species or similar endangered irises in Israel. For example, translocation of Iris hermona Dinsmore in the Golan Heights was unsuccessful, even though translocation was into a very


Variation partitioning in translocation 267 similar habitat to the one that had been destroyed (Y. Sapir, unpublished data). In order to investigate species habitat preferences for I. atrofusca, we set up a translocation experiment, using rhizomes rescued from a site under threat of destruction. We applied CCA ⁄ RDA for (i) partitioning the variation in performance of I. atrofusca into components due to vegetation composition, total plant biomass and soil properties and their shared effects and (ii) for determining the environmental factor(s) directly affecting species introduction success.

70

Produced fruits

60

Produced flowers

50

Non-reproducing

40 30 20 10 0 70

Materials and methods The study area was Lahav North Nature Reserve which is 1Æ15 km in size and located in the semi-arid climatic zone of Israel (ca. 300 mm annual rainfall; Shachak et al. 2008). The reserve is typified by low hills less than 500 m above sea level, with plant formation typical for the transitional zone between Mediterranean and desert vegetation (known as batha), dominated by Sarcopoterium spinosum (L.) Spach, Phlomis brachyodon (Boiss.) Zohary, Asphodelus ramosus L. and Gundelia tournefortii L. (Tsoar & Ramon 2002; Fig. 1). There are no records of I. atrofusca ever occupying this site. However, it is within the discontinuous distribution range of I. atrofusca, with the nearest population found 9 km south-west of the Reserve at the Dudaim forest. Rhizomes rescued in spring 2006 in the nearby Goral Hills region (road-building strip for new railroad tracks) were planted in autumn 2006 in sets of 62 rhizomes in each of 22 microhabitats in Lahav North Nature Reserve (Fig. 1). Each set comprised the following size classes (number in parentheses): 40 g (2). In spring 2007, 2008 and 2009 we counted the number of plants that emerged, flowered and set fruit at each site, making a total of nine performance variables (Fig. 2). In addition, in the second year (spring 2008) the plant community at each site was sampled using random quadrats of either 1 m2 (one per site) or 0Æ125 m2 (two per site). Quadrat size was chosen according to the vegetation density and homogeneity of the site. All plants from a quadrat were harvested, brought to the laboratory, identified to species level, counted and dried to constant weight to determine the total plant biomass at each site. The vegetation data set had, therefore, two parts: the number of individuals per species (Table S1, Supporting information) and total plant biomass (Table S2, Supporting

Number of plants

60 2

50 40 30 20 10 0 70 60 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

Site Fig. 2. Survival and reproduction of I. atrofusca (number of plants) 1, 2 and 3 years after experimental introduction at Lahav North Reserve in the 22 sites (from top to bottom). 62 rhizomes of Goral origin with equal representation of different size classes were introduced at each site in fall of 2006, counting was done in spring of 2007, 2008 and 2009. information) per 1 m2 per site (when 0Æ125 m2 quadrats were used, the data were standardized per 1 m2). In total, 75 plant species were found across the sites. During the same year (spring 2008), soil samples were taken at each site and analysed for six soil characteristics (Tables 1 and S3, Supporting information). Therefore, we had four multivariate data sets: species abundance (75 variables), biomass (one variable), soil (six variables) and iris performance (nine variables) data for each of 22 sites.

DATA ANALYSIS

Fig. 1. Example of five relocation sites within the Lahav North Nature Reserve.

The effects of the three sets of environmental variables (species abundance, biomass and soil) on variation in plant performance were examined with ordination techniques, using canoco (ver. 4.02; ter Braak & Smilauer 2002). Since a relationship between each of the three factors and performance of relocated iris plants may be caused by partial redundancy with the other factors, we applied the method of variation partitioning (Borcard, Legendre & Drapeau 1992). We estimated the following components of variation: ‘pure’ effects of abundance, biomass and soil (i.e. variation that can be explained by


268 S. Volis et al. Table 1. Qualitative soil characteristics, their description and categorization Soil characteristics

Description

Categories

Depth

Depth of soil profile to parent or subjacent material

60 cm

Humus

Humus of root-inhabited horizon, according to Munsell Soil Colour Charts

Low (10 YR 4 ⁄ 4, 5 ⁄ 3, 5 ⁄ 4); intermediate (10 YR 4 ⁄ 3, 5 ⁄ 2); high (10YR 3 ⁄ 3, 4 ⁄ 2; 7Æ5YR 4 ⁄ 2)

Presence of Bk horizon

The subsoil layer (horizon B) with large accumulation of carbonates (k)

Absent; low (Bk fragmentary, weak or unstable sub-angular structure); intermediate (Bk visible, moderate sub-angular structure); high (Bk clear, sub-angular to cubic or angular structure)

Alluvium

Depth of crumb or granular structure

Absent; low (0Æ5), dark grey cells – areas with medium suitability (0Æ2–0Æ5), and white with low suitability ( 65 dB > 62 dB > 59 dB > 56 dB > 53 dB > 50 dB > 47 dB > 44 dB

1 km

Fig. 1. Maps of the Buunderkamp area showing nest-boxes, sampling locations and noise levels. Motorway (triple line) and railway (dashed line) are shown. a) nest-box distribution (small dots). Only breeding data from nest-boxes within the rectangle was used. b) sampling locations (filled rectangles) along 10 transects (open rectangles, 2 of them shown). Numbers refer to locations of example recordings used in Fig. 2. c) GIS-map showing spatial variation in sound levels. Traffic noise shows a strong decrease with distance from the motorway (absolute range at sampling locations 46Æ5–67Æ8 dB SPL, A-weighted), but there is substantial spatial variation in this decline. that would have affected the spatial spread of noise coming from the motorway.

NOISE DATA ACQUISITION

We made sound recordings between March and May 2008, before major leafing of the deciduous trees. We sampled sound levels along ten transects perpendicular to the motorway (Fig. 1b), with automatic SongMeter recorders (16 bit, 24 kHz sample rate; Wildlife Acoustics Inc., Concorde, MA, USA). Exact sampling locations were determined with a GPS (Garmin 60CSx, Olathe, KS, USA). The sampling transects started 100 m from the mid-line of the motorway and six sampling locations at approximately 100 m intervals were chosen within each transect. The transects were spaced 80–100 m apart and two transects were sampled simultaneously for 3–5 consecutive days. Transects were each sampled twice in a random order, once between 8th and 30th of March, and once between 31st of March and 1st of May. The sampling grid encompassed most

of the area, but we used two additional SongMeters to monitor the remaining area. Recorders were attached to large trees (>40 cm in diameter) at 2 m above the ground with the recording microphone directed towards the motorway. Recording levels for the microphones were adjusted to a sensitivity ranging from 0Æ0 to 1Æ5 dBV pa)1 (reaching full scale between 92Æ5 and 94Æ0 dB SPL) and amplitude levels were adjusted according to the effective sensitivity of each individual Song Meter recorder. Recorders were randomly swapped between sampling locations to control for any remaining variation in recording levels. Recorders were scheduled to record for 30 s at 30 min intervals, day and night. We analysed sound recordings in the computer program Matlab (Mathworks Inc., Natick, MA, USA). We measured overall sound levels (using an A-weighted filter), and also sound levels in four adjacent octave-bands, centred at frequencies of 0Æ5, 1Æ0, 2Æ0, and 4Æ0 kHz. Sound measurements were averaged over either 30-min or 24-h intervals, and ⁄ or sampling locations, depending on the type of analysis.


Traffic noise and avian reproductive success 213 We used 76 sampling locations to visualize spatial variation in noise levels for the Buunderkamp in the computer program ArcGis (version 9Æ0, ESRI). Sixty locations from the sampling transects and 16 additional sampling locations were plotted onto a geo-annotated reference map from which noise maps were derived with the Spatial Analyst toolbox. Spatial resolution was set at 5 m and raster values between sampling locations were calculated with a weighted distance interpolation tool (IDW). Additionally we calculated distances for all nest boxes and sampling locations to the nearest mid-point on the motorway. We assessed the temporal overlap between traffic noise and vocal bird activity throughout the season and at different times of day. At our study site most of the non-anthropogenic sound comes from vocalizing birds with the majority of acoustic energy in the range of 2–8 kHz. We selected a subset of sampling locations at distances over 400 m from the motorway where there is little traffic noise present in the 4 kHz octave band so temporal variation in sound levels was mainly related to the vocal activity by birds. For these locations, we compared sound levels, averaged over 1 or 24 h intervals, in the 1 kHz band (mainly due to traffic noise) with those in the 4 kHz band (mainly due to bird activity, including great tits).

LONG-TERM BREEDING DATA

Great tit breeding data were collected between 1995 and 2009 by the Netherlands Institute of Ecology (NIOO-KNAW). We used data from both large and small nest-boxes within the sampling grid (Fig 1a) on laying date, clutch size, number of hatchlings, number of fledglings and fledging mass (average weight of chicks for the brood when chicks are 15 days old) for all first great tit clutches over this period, except for 2007 and 2008 when data were excluded because of an unrelated experiment. Additional data on female identity, female age and fledging mass were only available for 1995–1999, 2001 and 2009. For analysis of breeding performance we used only first clutches (categorised using female identity or because laying date was within 30 days of the first laying date for a given year). For analyses of laying date we used only clutches for which this could be reliably calculated. We were interested in the mechanisms underlying breeding success and therefore focused on life history traits that reflected decisions made by the birds. For the analysis of clutch size we therefore excluded clutches that were not incubated, because including nests that were abandoned (either through a decision by the parents, or predation of the parents) would introduce unwanted heterogeneity in the data. Similarly, we excluded nests where no chicks hatched or fledged from the analyses of the number of hatchlings and fledglings, respectively, because it was usually unknown whether failure was caused by death of all the embryos or chicks, abandonment by the parents or predation of the parents (away from the nest).

WEATHER DATA AND HABITAT MEASUREMENTS

We assessed habitat characteristics, including tree density, tree diameter and species composition, at the level of woodland plots (0Æ2– 1Æ0 ha). We measured tree density and diameter and noted tree species at each of the 60 sampling locations, and at the 2 nest boxes nearest to these locations. We calculated the percentage of deciduous trees per plot and averaged tree density and diameter over all locations within a plot. We used weather data on daily wind direction and speed, and temperature, recorded by the Royal Netherlands Meteorological

Institute (KNMI) at de Bilt (situated ±50 km to the west of the Buunderkamp).

STATISTICAL ANALYSIS

We analysed all data using SPSS (version 17Æ0) and log-transformed variables when necessary to meet model assumptions. Temporal variation in daily and seasonal sound levels were explored using repeated measures anovas with sound level grouped by sampling location as the dependent variable and time of day or date as an explanatory variable. Additionally, we compared recordings made on weekdays with recordings from weekends with type of day as a fixed factor. We examined the effect of daily weather conditions on the propagation of noise with full factorial linear mixed models. To test for the effect of wind direction we discriminated between days with northerly (coming from the direction of the motorway) and southerly winds (going towards the direction of the motorway). Wind direction was included as a fixed factor, and sample location as a random factor. Distance to the motorway, wind speed, and daily temperature were included as covariates. We constructed a set of linear mixed models for each life history trait and compared them using a model selection approach based on Akaike’s information criterion (Burnham & Anderson 2002). Models always included nest-box type (large or small), sampling location and breeding year as random factors. Depending on the model, we also included other reproductive traits as explanatory variables (cf. Wilkin et al. 2006). For instance, clutch size can correlate with laying date and an effect of noise on clutch size could be indirectly caused by an effect of noise on laying date. Including laying date in the clutch size model therefore allows us to test for a direct effect of noise. For the number of hatchlings we included clutch size and for the number of fledglings we included number of hatchlings in the models. For the fledging mass model we included both clutch size and laying date as these factors are known to have a large effect on fledging mass (e.g. Wilkin et al. 2006). In a first analysis we compared models that included overall noise levels, distance to the motorway, tree density, tree diameter and percentage deciduous trees as explanatory factors only. Models contained single factors or in combination with other factors as main effects as we had no a priori knowledge that interactions among factors would be of importance. The total set contained 32 models to be compared for each trait, including the Null model. We calculated for each explanatory factor the probability that it would be in the best approximating model using Akaike weights (see e.g. Whittingham et al. 2005; Garamszegi et al. 2009). We used the subset of models with a delta-AIC < 4Æ0 from the top model to get model-averaged estimates and standard errors for each factor (cf. Burnham & Anderson 2002). In a second analysis we focused on temporal overlap between noise sampling period and breeding stage. We used the models with delta-AIC < 4Æ0 from the previous analysis and only exchanged the overall noise with noise levels sampled either in March or in April. In a third analysis, we repeated this procedure, but focused on the spectral overlap with song and explored whether noise in a certain frequency range (0Æ5, 1Æ0, 2Æ0, or 4Æ0 octave band, or overall noise) better explained variation of the data. Breeding performance is known to be age-dependent (Kluyver 1951; Wilkin et al. 2006) and we therefore re-ran analyses for which we found strong support using the subset of data for which female age was known. Female identity was added as a random factor and female age (first year or older) as a fixed factor.


214 W. Halfwerk et al. distances from the motorway. For instance, at 700 m from the motorway, sound levels below 1 kHz could increase by over 10 dB SPL on cold days or days with northerly winds (Fig. 2c,d).

Results SPATIAL PATTERNS IN NOISE LEVELS

Overall sound levels gradually decreased with distance from the motorway (F5,54 = 200Æ5, P < 0Æ001) with an average drop of 20 dB SPL (A-weighted) over less than 500 m (Fig. 1c). Furthermore, high frequencies attenuated faster than low frequencies (F3,59 = 12Æ03, P < 0Æ001; Fig. 2a). There was substantial spatial variation in traffic noise, independent of distance to the motorway (Fig. 1c): different locations at medium (>300 m) to large (>700 m) distances from the motorway differed by more than 9 dB SPL (A-weighted) in noise level (Fig. 1c). Train noise can be very loud (see e.g. Fig. 2b) but, in contrast to motorway noise, is transient and average daily noise levels near the railway line were among the lowest (Fig. 1c).

TEMPORAL FLUCTUATIONS IN TRAFFIC NOISE LEVELS AND THE OVERLAP WITH BIRD ACTIVITY

Traffic noise levels changed throughout the season (F1,59 = 7Æ57 P = 0Æ008) with March being noisier and more variable than April (Fig. 3a). Additionally, noise levels on weekdays were significantly higher than at the weekend (F1,59 = 4Æ87 P = 0Æ032; Fig. 3). Noise levels showed a strong daily pattern (F1,59 = 8Æ776 P = 0Æ005), with a clear drop between 0:00 and 4:00 AM, but no distinct rush-hour peaks (Fig. 3b). Screening of recordings revealed that, at distances over 400 m from the motorway, variation in sound levels in the 4 kHz band was indeed mainly influenced by bird vocal activity, and we therefore used recordings at these distances to assess seasonal and daily overlap of traffic noise and bird vocal behaviour. Bird vocal activity as measured at the peak of the dawn chorus increased throughout the season (4 kHz-band; F1,59 = 7Æ88, P < 0Æ001) whereas traffic noise during this time period decreased (1 kHz-band; F1,59 = 5Æ13, P < 0Æ001; Fig. 3a). Bird vocal behaviour showed a temporal shift between early March and late April due to changes in the time of sunrise, but despite this, the temporal overlap with traffic noise remained remarkably high on weekdays (Fig. 3b), probably due to the change from winter to summer time (i.e. clock

WEATHER-DEPENDENT NOISE LEVELS

Wind direction, wind speed and daily temperature all had an effect on overall sound levels (see Table 1). Furthermore, wind direction and temperature interacted with distance to the motorway (Table 1). We reanalysed a subset of recordings made at distances of 400–700 m from the motorway to explore the effect of weather conditions on sounds in different octave bands. Both temperature (F1,59 = 27Æ78; P < 0Æ0001) and wind direction (F1,59 = 5Æ27; P = 0Æ001) interacted with frequency, with the strongest effect at lower frequencies and large

70

70 (b)

50

40

40

30

30

20

20

10

60

1

2

3

4

5

6

Cold

(c)

1

April 16, 2008 16:00;6·4 °C

50

April 25, 2008 16:00;12·6 °C

60

20

20

10

10

0 2

3

4

5

2

3

6

0

4

5

Northern

(d)

April 18, 2008 16:00

6 2 C

Southern

40

E

30

1

1

50

30

0

0

1

Warm

40

B

10

4

4 D April 21, 2008 19:30

60

50

0

Train No train

1

April 21, 2008 D 20:00

60

Noise level (dB SPL)

3

Near Far

(a)

March 31, 2008 16:00

0

Frequency (kHz)

1

2

3

4

5

2 A

6

Fig. 2. Variation in sound profiles across different environmental conditions. a) powerspectrographic example comparing sound profiles near to (±100 m), and far from (±700 m), the motorway. At larger distances, the high-frequency components of traffic noise are more attenuated and even disappear above ±3 kHz. b) recordings made near the railway (±100 m from the track and ±1 km from the motorway) shortly before and during the passage of a train. c) comparison of sound profiles on days with different temperatures, but similar wind conditions illustrates large effect of weather conditions on noise levels. d) comparison of sound profiles on days with opposite wind directions, but similar temperature and wind speed. Small numbers refer to locations illustrated in Fig. 1. Capital letters refer to recording days illustrated in Fig. 3.


Traffic noise and avian reproductive success 215 and 3) and virtually no support in the remaining life history models. Overall noise levels had an independent negative effect on clutch size, with females laying on average about 10% fewer eggs across a noise gradient of 20 dB SPL (A-weighted) (Table 3). Reanalysing the top clutch size model to include female identity and age confirmed the effect of noise (F1,268 = 7Æ82, P = 0Æ007), but failed to show an effect of female age on clutch size (F1,268 = 0Æ20, P = 0Æ82). Noise levels had a negative effect on fledging mass (Table 3), but in none of the top models was the effect significant (all P > 0Æ2).

Table 1. Results from mixed model showing effect of weather condition on overall noise levels. Sampling location (N = 60) was added as random factor. Only first order interactions are reported Source

d.f.

F

P

Distance Wind direction (N vs. S) Daily temperature Wind speed* Distance · Wind direction Distance · Daily temperature Distance · Wind speed* Wind direction · Daily temperature Wind direction · Wind speed* Daily temperature · Wind speed*

5 1 1 1 5 5 5 1 1 1

6Æ61 10Æ92 9Æ65 29Æ30 3Æ81 2Æ73 1Æ75 1Æ32 10Æ38 11Æ26

Journal of Applied Ecology volume 48 tahun 2011

Advances in Applied Microbiology, Volume 48

October journal No.48 Spring (1989)

48

48

‘48

Journal of Biblical Literature Winter 2011

Woodworker's Journal - March-April 2011

Tussenspel op Tahun

Tussenspel op Tahun

Advances in Agronomy, Volume 48

Advances in Immunology Volume 48

Advances in Catalysis, Volume 48

Progress in Optics, Volume 48

Applied Ecology & Environmental Management

International Review of Neurobiology, Volume 48

Animal Models of Dementia (Neuromethods, Volume 48)

Asian Music - Journal Volume 42, Number 1, Winter - Spring 2011

Zwischenspiel auf Tahun

Tussenspel Op Tahun

Cell Journal, January, 21, 2011, Volume 144, No. 2

American Heart Journal, Volume 161, Issue 2, Feb 2011

48

'48

48

Microwave Journal, February 2011, Volume. 54, Issue No.2

Cell Journal, Volume 144, Number 1, January 7, 2011

Zwischenspiel auf Tahun

'48

'48

'48

Journal of Applied Ecology volume 48 tahun 2011

Advances in Applied Microbiology, Volume 48

October journal No.48 Spring (1989)

48

48

‘48

Journal of Biblical Literature Winter 2011

Woodworker's Journal - March-April 2011

Tussenspel op Tahun

Tussenspel op Tahun

Advances in Agronomy, Volume 48

Advances in Immunology Volume 48

Advances in Catalysis, Volume 48

Progress in Optics, Volume 48

Applied Ecology & Environmental Management

International Review of Neurobiology, Volume 48

Animal Models of Dementia (Neuromethods, Volume 48)

Asian Music - Journal Volume 42, Number 1, Winter - Spring 2011

Zwischenspiel auf Tahun

Tussenspel Op Tahun

Cell Journal, January, 21, 2011, Volume 144, No. 2

American Heart Journal, Volume 161, Issue 2, Feb 2011

48

'48

48

Microwave Journal, February 2011, Volume. 54, Issue No.2

Cell Journal, Volume 144, Number 1, January 7, 2011

Zwischenspiel auf Tahun

'48

'48

'48

Recommend Documents