Short Stories

Predictive modelling and ground validation of the spatial distribution of the New Zealand long-tailed bat (Chalinolobus tuberculatus)

Description
Predictive modelling and ground validation of the spatial distribution of the New Zealand long-tailed bat (Chalinolobus tuberculatus)
Categories
Published
of 12
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Share
Transcript
  See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/222909468 Predictive modelling and ground validation of the spatial distribution of the New Zealandlong-tailed bat (Chalinolobus...  Article   in  Biological Conservation · May 2006 DOI: 10.1016/j.biocon.2006.04.016 CITATIONS 28 READS 110 3 authors , including: Some of the authors of this publication are also working on these related projects: The Tawaki Project - Studying the marine ecology of New Zealand's forest penguin View projectEcological and Socio-Economic Aspects of Selecting Translocation Sites for Engandered Species   ViewprojectRenaud MathieuCouncil for Scientific and Industrial Research… 92   PUBLICATIONS   1,448   CITATIONS   SEE PROFILE Philip J SeddonUniversity of Otago 239   PUBLICATIONS   3,797   CITATIONS   SEE PROFILE All content following this page was uploaded by Renaud Mathieu on 31 March 2015. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the srcinal documentand are linked to publications on ResearchGate, letting you access and read them immediately.  Predictive modelling and ground validation of thespatial distribution of the New Zealand long-tailed bat (  Chalinolobus tuberculatus  ) Glen J. Greaves a , Renaud Mathieu b, *, Philip J. Seddon a a Department of Zoology, University of Otago, P.O. Box 56, Dunedin, New Zealand b School of Surveying, University of Otago, P.O. Box 56, Dunedin, New Zealand A R T I C L E I N F O Article history: Received 26 March 2005Received in revised form27 March 2006Accepted 2 April 2006Available online 12 June 2006 Keywords: Bat conservationHabitat selection Chalinolobus tuberculatus GIS modelling New ZealandA B S T R A C TThe use of predictive models is continually increasing, but few models are subsequentlyfield-checked and evaluated. This study evaluates the statistical strength and usefulnessfor conservation purposes of a predictive habitat use model developed for  Chalinolobustuberculatus , a threatened microchiropteran bat species found in the temperate rainforestsof New Zealand. The relationship between various environmental variables and the pres-ence/absence of the species was investigated using generalised linear modelling. Themodel developed was coupled with GIS data to develop maps of predicted occurrencewithin the West Coast region of New Zealand’s South Island. It was found that distanceto forest boundary, slope, presence of   Nothofagus , general land cover, variability in meanannual solar radiation, and mean ambient winter minimum temperaturewere significantlyassociated with the occurrence of the species. Evaluation of the statistical strength of thedistribution model with independent data of species’ occurrence collected at 152 sitesfound that the  C. tuberculatus  model showed a moderate ability to predict both species pres-ence and absence ( s ( b ) coefficient = 0.37). The field detection rate (0.45) using this modelwas significantly higher than that of historical surveys (0.12). The value of the species hab-itat model and the need to evaluate its utility in the development of conservation strategiesis discussed.   2006 Elsevier Ltd. All rights reserved. 1. Introduction Recognition of the factors that influence zoogeographical pat-terns is undoubtedly critical for the success of conservationmanagement (Wang et al., 2003). Hence, mapping of spe-cies–habitat relationships has long been a key provider of graphical information for wildlife conservation projects. Yet,the applicability of traditional map products is limited bytheir two-dimensional, ‘static’ nature. The introduction of digital maps and Geographic Information Systems (GIS) tech-nology dramatically changed the capacity of spatial data rep-resentation, as such enhancing their utility in predicting thedistribution of species (Greenberg et al., 2002). Its develop-ment allowed wildlife biologists to analyse variables at largerspatial scales, with greater precision and accuracy than previ-ous methods of spatial analysis.GIS-derived modelling of species distribution has beensuccessfully applied to many animal and plant groups (Martı´-nez-Salvador et al., 2005; Powell et al., 2005), including insec-tivorous bats ( Jaberg and Guisan, 2001; Wang et al., 2003). 0006-3207/$ - see front matter    2006 Elsevier Ltd. All rights reserved.doi:10.1016/j.biocon.2006.04.016*  Corresponding author:  Tel.: +64 3 479 7698; fax: +64 3 479 7586.E-mail addresses: glen_greaves@yahoo.com.au (G.J. Greaves), renaud.mathieu@stonebow.otago.ac.nz (R. Mathieu), philip.seddon@ stonebow.otago.ac.nz (P.J. Seddon). B I O L O G I C A L C O N S E RVAT I O N  132 (2006) 211  –  221 available at www.sciencedirect.comjournal homepage: www.elsevier.com/locate/biocon  However, a conspicuous feature of many of these earliermodels is their lack of independent assessment of model per-formance. Instead, researchers have used alternative tech-niques based on resampling of input data (Seoane et al.,2004) (e.g. bootstrapping, resubstitution, randomisation, orpartitioning), or often undertaken no substantiative testing at all (Fielding and Bell, 1997; Johnson et al., 2004). Although computer-intensive validation procedures have some meritin simulating species occurrence, they fail to provide thesame degree of confidence as using an independent dataset(Pearce and Ferrier, 2000). In our view, the best means of  objectively assessing model performance is to use an inde-pendent set of locality records and quantitative accuracymeasures. Many recent studies have achieved this by with-holding a sub-sample of the srcinal dataset, to which valida-tion tests are applied (Boyce et al., 2002). However, when theoriginal dataset is limited in size and sufficient resourcesare available, validation of the model through comprehensivefield survey is considered the best method.New Zealand’s unique endemic flora and fauna has suf-fered greatly following the arrival of humans and their asso-ciated exotic species introductions (Atkinson, 2001; Dowding and Murphy, 2001; Fitzgerald and Gibb, 2001). This has re-sulted in a highly disproportionate numberof threatened spe-cies, extending widely across plant and animal groups(Atkinson, 2001; Dowding and Murphy, 2001; Nugent et al., 2001; Towns et al., 2001). It is only the relatively recent dateof this colonisation and the success of intensive manage-ment, particularly the relocation of remnant populations topredator-free offshore islands, that has saved the majorityof faunal species from extinction (Armstrong and Ewen,2001; Towns and Ferreira, 2001). The country’s only endemic terrestrial mammals, three species of microchiropteran bat,have not escaped the onslaught, but until recently did not re-ceive the attention afforded to the suite of threatened bird,reptile, and amphibian species. For the Greater Short-tailedbat ( Mystacina robusta ), this attention has likely come too late(the last recorded sighting was 1967). Despite being fully pro-tectedunder the Wildlife Act (1953)in response to their signif-icant post-settlement reduction in abundance anddistribution (Daniel, 1990; O’Donnell and Sedgeley, 1994; Mol-loy, 1995; O’Donnell, 2000), New Zealand’s two remaining batspecies – Long-tailed bat ( Chalinolobus tuberculatus ) and LesserShort-tailed bat ( Mystacina tuberculata ) – continue to decline.The cause of this decline is uncertain, but is likely to be acombination of habitat loss, predation by introduced mam-mals, and competition for food and roosting sites by intro-duced mammals, birds, and wasps (O’Donnell, 2000, 2001a).Both  C. tuberculatus and  M. tuberculata  are under serious threatof following   M. robusta  into extinction in the medium term(O’Donnell, 2000).Before the future management of New Zealand’s bat spe-cies can be effectively planned, a thorough evaluation of theirstatus is required. Improved knowledge of the distribution of bats will enable sites to be selected for ongoing management,monitoring, and advocacy (Molloy, 1995). However, the exist-ing methods of surveying for New Zealand bats have a limitedefficiency, with a detection rate of only 12% for  C. tuberculatus surveys within the current study area (West Coast Bat Data-base, Department of Conservation, DOC). The primary pur-pose of this research was to improve this efficiency bymodelling the likely spatial distribution of bats, thus allowing concentration of surveys within areas where, on the basis of habitat suitability, bats aremore likely to occur. Historical sur-veys have also focused on sites believed to be suitable, but thedefinition of suitability has been anecdotal. There is a needfor more rigorous assessment of habitat suitability, using anobjective measure of resource use.The implicit assumption with animal distribution surveysis that suitable habitat is saturated with individuals, i.e. thehabitat is at carrying capacity. Therefore, the first reasonwhy we may not detect a bat in suitable habitat relates tothe non-detection of bats at a site despite their presence (TypeII error), giving a probability of detection that is less than one.Without specific modelling the actual probability of detectionis unknown, however we can assume that it is constant forboth historical and validation surveys. The second reasonmay be attributed to an imperfect knowledge of what consti-tutes suitable habitat forabat, so surveying for batswould oc-cur in placeswhere they havechosen not to be. This is the keyerror as it dictates the improvement of an objective spatialmodel over a subjective historical method, and as such, pre-sents the two key questions to be answered by this research:(1) is  C. tuberculatus  more likely to be detected in sites con-sidered suitable by the predictive distribution model,compared with sites the model deems unsuitable?(2) is the predictive distribution model a better basis foridentifying suitable bat habitat for surveys than the his-torical approach?This second question is particularly critical as it providesthe justification for the construction of, and is therefore a di-rect measure of the success of, the predictive distributionmodel. 2. Materials and methods 2.1. Study area This research took place within the New Zealand Departmentof Conservation’s West Coast  Tai Poutini  Conservancy (Fig. 1).Several minor and localized extensions to the eastern bound-ary were added to incorporate a few significant bat surveysthat bestride the conservancy boundary. Located on the westcoast of the South Island of New Zealand, the West CoastConservancy manages more than 1.9 million hectares of land.The conservancy contains within its boundaries two nationalparks, parts of three others and a World Heritage Area. Span-ning over 600 km in length, the West Coast region can be re-garded as an ‘ecological island’, confined by the SouthernAlps to the east, and the Tasman Sea to the west. At its wid-est, the distance from the sea to the mountain tops (whichregularly peak over 3000 m) is 90 km, and is often much less.As such, there is markeddiversity in habitat type from coastaltemperate rainforest through to a true alpine zone, and shorttransition distances between these distinct biomes. Indige-nous forest covers about three quarters of the region, charac-terised by a mixture of Podocarp/hardwood at low altitudes,gradually giving way to  Nothofagus  dominant forest from 212  B I O L O G I C A L C O N S E RVAT I O N  132 (2006) 211  –  221  about 500 m above sea level (McGlone et al., 1996; Leathwick,1998). 2.2. Development of the predictive distribution map 2.2.1. Bat field data The database used for this research was sourced from theDepartment of Conservation and comprised bat presence/ab-sence records obtained between 1994 and 2003, derived froma combination of stationary and transect surveying. Wheremore than one survey was done at a particular site, the sur-plus surveys were removed from the analysis to avoid pseudoreplication. To reduce the influence of false negative (Type II)error on the bat presence analysis, all ‘absence’ sites within100 m of a ‘presence’ site were removed. Likewise, presencesites within 100 m of another presence site were removed.Due to the mobility of   C. tuberculatus , it was considered highlylikely that bats would also be located in habitat within the100 m radius around a presence site, and may be detected gi-ven further surveying effort. Moreover, measuring habitatselection at smaller scales than this is beyond the scope of this study. In addition to the removal of these replicate sites,a numberof datawereexcludeddue to recording inaccuraciesfound following assessment of geographical co-ordinates. Asa result, of the over 2500 records available, a total of 1033 re-cords were deemed suitable for input into the model. 2.2.2. Environmental data Environmental descriptors were derived from three databases– Land Environments New Zealand (Ministry for the Environ-ment),NewZealandForestryServiceMaps(Series6)(NewZea-land Forestry Service), and Land Cover Database Version 1(Ministry for the Environment). While the Land EnvironmentsNew Zealand database was provided in raster format(100  ·  100 m grid), the Land Cover Database and New ZealandForestry Service database were provided in vector format andtherefore were converted to raster for ease of analysis, with adiscrete code allocated to each variable. A total of 14 potentialenvironmental predictors were selected following analysis of current knowledge of   C. tuberculatus  ecology (O’Donnell, 1999,2001a) and habitat data availability. Table 1 provides a sum- mary of the variables investigated. Twelve of the 14 variableswereextracteddirectlyfromthedatabases,however,‘distanceto forest boundary’ and ‘presence of beech ( Nothofagus )’descriptors were created within ArcGIS  (Environmental Sci-enceResearchInstituteInc.)usingdatafromLandCoverData-base Version 1 and New Zealand Forestry Service Maps (Series6) respectively. The environmental value allocated to each bat Fig. 1 – Study area, West Coast Conservancy, South Island, New Zealand. The general focal areas for the predictive model validation are displayed (Karamea, Paparoa, Maruia, Shenandoah). B I O L O G I C A L C O N S E RVAT I O N  132 (2006) 211  –  221  213  sample was calculated as the average of values containedwithin a 100 m radius buffer where the measured variablewas continuous, and the value which contributed the largestareawithin the same buffer where thevariablewas discrete. 2.2.3. Spatial accuracy of data Spatial accuracy of the bat location data, both in the sourcedataset and the field validation phase of this research, maybe compromised by error in Global Positioning System (GPS)plotting, or in data transcription. GPS error is associated withfactors such as signal degradation, signal multipath, poor sig-nal reception, or atmospheric delays. Prior to 2000 GPS signalswere artificially degraded through United States Defence ap-plied Selective Availability (SA), resulting in an error of up to100 m. As the majority of the data used in the model wassourced between 1996 and 2000, prior to the removal of SA,a 100 m buffer to all data sites was applied. Transcriptionerror was effectively removed by cross-validation of all coor-dinates with in-depth site location description.The two potential errors associated with categorical envi-ronmental datasets such as those used in this research maybe related to either: (1) positional accuracy, or (2) semanticaccuracy. For Land Cover Database Version 1, the satelliteimagery was orthorectified to produce a positional accuracyof ±25 m (Terralink International Ltd.). The semantic accuracyof this dataset was stated as 90% by Terralink InternationalLtd. In an accuracy assessment undertaken by New ZealandForest Research in 2000, overall map accuracy was estimatedat 93.9% using a simple accuracy percentage statistic.To minimise errors in the New Zealand Forestry Servicedigitisation process, unused and unfolded source maps wereutilised, and any significant linear differences and incorrectpolygon tagging were noted and corrected before the finaldatabase was released. The map producers (New Zealand For-estry Service), and therefore the authors of this paper, con-sider the error value to be insignificant.The Land Environments New Zealand database was thesource for the seven underlying climate and physical habitatcharacteristicsused in thisresearch. As this databasepredictsthe variation in continuous environmental variables acrossspace (interpolated from a number of measuring stations),eachofwhichisdisplayedasdiscretelayers,theassociateder-rors occur within each independent variable and not over thedatabase as a whole. For the majority of layers, the potentialstandard error, for the whole of New Zealand, is stated by thedatabase source (Ministry for the Environment). The two tem-perature variables (Table 1) use a total of 300 meteorologicalstations nationwide from which the layers are derived, of whichonly14arelocatedwithinthestudyarea.Thislownum-ber for the West Coast study area is reflected in the standarderror,which,rangingfrom0.30to0.40   C(meanannualtemper-ature; mean = 10.1   C, SD = 2.03   C) and 0.40–0.60   C (meanwinter minimum temperature; mean = 0.86   C, SD = 3.0   C), isslightlyhigherthanthemajorityoftheremainderofNewZea-land (Leathwick et al., 2002). The potential errors associated with the two Land Environments New Zealand variables esti-mating solar radiation are proportionally smaller than thatfor the temperature variables discussed above (mean annualsolarradiationestimatederror=0.210–0.270MJ/m 2  /day,mean=14.2 MJ/m 2  /day,SD = 0.95 MJ/m 2  /day;meanwintersolarradia-tion estimated error = 0.35–0.55 MJ/m 2  /day, mean = 5.6 MJ/m 2  /day,SD = 0.92 MJ/m 2  /day)(Leathwicketal.,2002),representing 0.017and0.080forannualsolarradiationandwintersolarradi-ation measures, respectively. Standard errors associated withthepredictionsofOctobervapourpressuredeficitarepredom-inantly below 0.05 kPa within the study area (mean = 0.33 kPa,SD = 0.12 kPa), and significantly lower in the mid to low alti-tudesoccupiedby  C. tuberculatus . Thiscorrespondsto apoten-tialerrorof15%aroundthemean,whichishighcomparedwiththe other variables included for analysis. No assessment hasbeen made of the errors associated with the Land Environ-ments New Zealand slope layer (Leathwick et al., 2002). How- ever, mean elevation error for the Digital Elevation Model(DEM) was found to be 0.41 m (standard error: 6.13 m) using over2500independentGPSdatapointstakenfromthroughoutthe South Island (Landcare Research, unpublished data). 2.2.4. Statistical analysis and spatial prediction Thisresearchusedaresourceselectionfunction(RSF)modeltocharacterise the selection of resources by  C. tuberculatus  in the Table 1 – Source, scale, and estimated errors of habitat variables chosen for inclusion into the regression analysis Habitat variable Source Scale Error a Vegetation Variables I Land cover LCDB1 1/50,000 6.1%II Forest type NZFS6 1/50,000 InsignificantIII Presence of beech NZFS6 1/50,000 InsignificantIV Area of indigenous forest patch LCDB1 1/50,000 6.1%V Area of forest patch NZFS6 1/50,000 InsignificantVI Perimeter of forest patch LCDB1 1/50,000 6.1%VII Distance to forest boundary LCDB1 1/50,000 6.1%Underlying Variables Climate VIII Mean annual temperature LENZ 14 stations ±0.3–0.4   CIX Mean winter minimum temp LENZ 14 stations ±0.4–0.6   CX Mean annual solar radiation LENZ 7 stations ±0.21–0.27 MJ/m 2  /dayXI Mean winter solar radiation LENZ 7 stations ±0.15–0.55 MJ/m 2  /dayXII October vapour pressure deficit LENZ 14 stations ±0.04–0.6 kPaXIII Monthly water balance ratio LENZ 65 stations ±0.512Physical XIV Slope LENZ 1/50,000 NALCDB1 = Land-Cover Database Version 1; NZFS6 = New Zealand Forestry Service Maps (Series 6); LENZ = Land Environments New Zealand.a Estimated error (LCDB1, NZFS6) or standard error (LENZ). 214  B I O L O G I C A L C O N S E RVAT I O N  132 (2006) 211  –  221
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x