Conservation assessment and prioritization of areas in Northeast India: Priorities for amphibians and reptiles

Please download to get full document.

View again

of 16
2 views
PDF
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.
Document Description
Conservation assessment and prioritization of areas in Northeast India: Priorities for amphibians and reptiles
Document Share
Document Transcript
  Conservation assessment and prioritization of areasin Northeast India: Priorities for amphibians and reptiles Samraat Pawar a , Michelle S. Koo b,1 , Chris Kelley a , M. Firoz Ahmed c ,Sujoy Chaudhuri d,e , Sahotra Sarkar a, * a Biodiversity and Biocultural Conservation Laboratory, Section of Integrative Biology, University of Texas, Austin, TX 78712, USA b California Academy of Sciences, 875 Howard Street, San Francisco, CA 94103, USA c Aaranyak, 50 Samanwoy Path Survey, Basistha Road, P.O. Beltola, Guwahati 781028, Assam, India d Center for Applied Biodiversity Science, Conservation International, 1919 M Street, NW, Suite 600, Washington, DC 20036, USA e Ecollage, 11A, Rajiv Nagar (S) Viman Nagar, Pune 411014, India A R T I C L E I N F O Article history: Received 12 September 2006Received in revised form30 November 2006Available online 2 February 2007 Keywords: AmphibiansArea prioritizationConservation planning Northeast IndiaReptilesSpecies distribution modelsA B S T R A C TThis study combines niche modeling and systematic area prioritization using distributiondata for 131 species of amphibians and reptiles from Northeast India and Burma, with twoobjectives: (i) to evaluate the performance of the current conservation area network inNortheast India with respect to the representation of amphibians and reptiles, and (ii) toidentify potential areas for expanding the current conservation area network. In a two-stepprotocol, maximum entropy niche modeling was used to project species’ potential geo-graphic occurrences, and the resulting probabilistic distribution data were used to priori-tize areas with algorithms that maximize the representation of all species in minimaltotal area. The results provided a critical assessment of conservation priorities in thisdata-deficient region, and indicate the utility of combining niche modeling with systematicarea prioritization in such situations. Many areas that had been overlooked in previousassessments were identified. Although the existing protected areas were found to be inad-equate for representation of amphibian and reptile diversity, the prioritization results showthat by targeting a minimal representation of 5% of the current total area suitable for eachspecies, the gaps can be filled with a relatively modest (0.41%) increase in the current totalarea covered by the network. Extended analyses were also performed to assess the effectsof putatively rare species on reserve selection, which showed that the inclusion of thesetaxa can change the prioritization solutions significantly. The prioritization results alsohighlight areas of Northeast India that warrant attention from future surveys.   2007 Elsevier Ltd. All rights reserved. 1. Introduction The selection and management of conservation areas in bio-diversity rich tropical regions poses many challenges. On onehand, the tropics have some of the fastest rates of degrada-tion of natural land cover, while on the other, they are gener-ally data-poor and cash-strapped (Myers et al., 2000;Mittermeier et al., 2004). Resource constraints generally pre-clude systematic data collection for multiple taxonomicgroups, making it difficult to ensure maximal representation 0006-3207/$ - see front matter    2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.biocon.2006.12.012*  Corresponding author:  Tel.: +1 512 232 7122; fax: +1 512 471 4806.E-mail addresses: samraat@mail.utexas.edu (S. Pawar), sarkar@mail.utexas.edu (S. Sarkar). 1 Present address: Museum of Vertebrate Zoology, 3101 Valley Life Sciences Building, University of California, Berkeley, CA 94720-3160,USA. B I O L O G I C A L C O N S E RVAT I O N  136 (2007) 346  –  361 available at www.sciencedirect.comjournal homepage: www.elsevier.com/locate/biocon  of overall biodiversity of the region in a set of conservationareas. This situation therefore requires the development of methods that make maximal use of the information contentin available biological data. In the last few years, the integra-tion of two different techniques has shown great promise to-wards achieving this goal: ecological niche modeling combined with systematic area prioritization (Sanchez-Cor-dero et al., 2005; Fuller et al., 2006).Niche modeling (also called habitat modeling; Kearney,2006) aims at predicting species’ geographic distributionsusing data on observed presence records and associated envi-ronmental variables (which may be biotic or abiotic). It isbased on an assumption of niche conservatism, that is, thetendency of species to retain ancestral ecological niches (Pet-erson et al., 1999; Wiens and Graham, 2005). Niche modelsidentify areas that are ecologically suitable for the presenceof a species based upon samples of its realized niche (typi-cally by establishing environmental correlates of observedgeographical occurrences). Systematic area prioritizationaims at selecting conservation area networks using algo-rithms that seek to maximize biodiversity representation inas little land as possible,often also incorporating other spatialcriteria such as size or compactness of each individual area(Margules et al., 2002; Sarkar et al., 2002).Combining these two methods, this study performs a sys-tematic area prioritization for the Northeast (NE) India region.For area prioritization, multiple biological groups should ide-ally be included to maximize overall biodiversity coverage.However, sufficient data to do this are rarely available, andsuch analyses frequently have to make do with available tax-onomic groups. The taxonomic groups used in this study areamphibians and reptiles. Thus, what this study directlyestab-lishes is a network of priority areas for conservation of thesegroups. Amphibians and reptiles are important componentsof biodiversity which are often under-represented in conser-vation planning. NE India and Burma are known to harbor adiverse amphibian and reptile fauna with a high percentageof endemic, threatened species (see study area descriptionbelow). Moreover, amphibians have emerged as a major con-servation concern in many areas of the world because of theglobal declines in their numbers in recent decades (Alford andRichards, 1999). Area prioritization with these groups in NEIndia will therefore offer an important preliminary evaluationof the conservation status of the Himalaya and Indo-Burmaglobal biodiversity hotspots (the region forms a significantmajor part of both these hotspots; see below), both of whichhave hitherto been subjected to negligible conservation eval-uation or area prioritization (Conservation International,2006).Many studies make the explicit or implicit assumptionthat the taxa being used for area prioritization are adequatesurrogates for overall biodiversity in the region. There ismounting evidence however, that this assumption is oftennot valid (Flather et al., 1997; Moritz et al., 2001; Lund andRahbek, 2002; Kati et al., 2004; Sarkar et al., 2005). Recent workon cross-taxon biogeographical concordance in this regionalso indicates that amphibians and reptiles may be reason-able surrogates for certain subgroups of birds, but probablynot for birds as a whole (Pawar et al., 2006). The results of thisstudy, therefore, should not be taken to identify priority areasfor other taxonomic groups. To conserve all of biodiversityadequately in NE India will require the incorporation of fur-ther analyses using other groups to augment the results re-ported here.This study has two main objectives: (i) evaluation of theexisting set of protected areas with respect to its performancein representing the chosen biotic groups (amphibian and rep-tile species) adequately, and (ii) identifying priority areas forexpanding the existing protected areas. The prioritization isbasedupon projecteddistributions generated by niche model-ing of amphibian and reptile occurrence data collected fromextensive surveys in India and Burma. 2. Methods 2.1. Study area NE India lies between 29.46  – 21.96  N and 97.39  –89.87  E, cov-ering a total area of about 255,168 km 2 that includes easternand western parts of the Himalaya and Indo-Burma globalbiodiversity hotspots, respectively (ca. 8.2% of the 3,114,763sq. km occupied by both the hotspots; Mittermeier et al.,2004). The region covers seven Indian states: Arunachal Pra-desh, Assam, Manipur, Meghalaya, Mizoram, Nagaland, andTripura. Physiognomically and biogeographically, the regioncan be broadly differentiated into the Eastern Himalayas inthe north, the Northeast Hills in the south, and the Brah-maputra River Basin in between (Mani, 1974) (Fig. 1). Of these, the Eastern Himalayas and Northeast Hills areprimarily mon-tane with differing geological srcin and morphology, whilethe Brahmaputra River Basin consists of the flood plainsand lower catchments of the Brahmaputra river (Mani,1974). NE India harbors exceptional biological diversity andhas a relatively complex biogeography due to a combinationof factors, including its age, unique plate tectonic and palae-oclimatic history, location at the confluence of distinct realms(Afrotropic, Palearctic, and Indo-Malay; Olson et al., 2001),wide physiognomic range (e.g., altitude ranging from about100 to >7000 m above sea level) and vegetation diversity (fromtropical to alpine) (Mani, 1974).Although both the Himalaya and the Indo-Burma hotspotsare data-deficient, existing knowledge about plant and verte-brate diversity provide clear indications of their biodiversityvalues (Mittermeier et al., 2004). In case of the Himalaya hot-spot, 32% of the 10,000 known vascular plant species, 40% of the 105 known amphibian species, and 27.3% of the 176known reptile species are endemic (Conservation Interna-tional, 2006). In the case of the Indo-Burma hotspot, 52% of the 13,500 known plant species, 53.8% of the 286 knownamphibian species, and 39.1% of the 522 known reptile spe-cies are endemic (Conservation International, 2006). A signif-icant proportion of these species are considered threatened(although the precise numbers are poorly known; Conserva-tion International, 2006). 2.2. Species’ distribution datasets Species’ distribution data were collected from NE India,Burma, and Yunnan Province, China. A majority of the datacame from the Myanmar Herpetological Survey Project, a B I O L O G I C A L C O N S E RVAT I O N  136 (2007) 346  –  361  347  collaborative effort between the Nature and Wildlife Conser-vation Division (a division of the Forest Department of Bur-ma), the Department of Herpetology at the CaliforniaAcademy of Sciences, and the Division of Amphibians andReptiles at the National Museum of Natural History, Smithso-nian Institution. The surveys were conducted from November1999 through May 2004. All data from within NE India werecollected by the Biodiversity and Biocultural ConservationLaboratory at the University of Texas at Austin and collabora-tors based in India. Field techniques : Amphibians and reptiles were located bynocturnal and diurnal opportunistic searching in appropriateterrestrial and freshwater habitats. All animals retained asvoucher specimens were humanely killed according to theguidelines established by the American Society of Ichthyolo-gists and Herpetologists, The Herpetologists’ League, andthe Society for the Study of Amphibians and Reptiles (Ameri-can Society of Ichthyologists and Herpetologists et al., 1987).Specimens were fixed with 10% neutral-buffered formalin.In the case of the Myanmar Herpetological Survey Project,all amphibian and reptile taxa encountered were collected,except those listed in CITES Appendix 1 or on the US Endan-gered Species List; instead, photographic vouchers of thesespecies were taken and deposited at California Academy of Sciences. Detailed microhabitat data and GPS locations wererecorded for all specimens encountered. As far as possible,each specimen was identified to species in the field and veri-fied in the laboratory. The protocol for the data collection inNE India was similar, except that specimens were collectedaccording to area-specific permits given by the Ministry of Environment and Forests, India (Pawar and Birand, 2001). Secondary data : A small proportion (2.4%) of the data wasretrospectively georeferenced from historical museum data.Procedures for retrospectively georeferencing written localitydescriptions follow the protocol of HerpNET (www.herp-net.org ; last accessed on August 14, 2006) and the methodol-ogy followed by Wieczorek et al. (2004). Place names werelocated with gazetteers from NIMA-GeoNET Names Serveror with paper topographic maps. Offsets (unless specified)were assumed to take place along a road or trail, and likelypaths were thus sought out. If none were available, air dis-tances were assumed and measured out. In either case, these Fig. 1 – Map showing digital elevation model of Northeast India (shaded region in the inset map of South Asia) with majorsubregions (delineated by white lines) (modified from Olson and Dinerstein, 2002; GIS data from World Wildlife Fund, 2006 ). The subregions are: [1] Eastern Himalaya (East Himalayan forest zones of  Olson and Dinerstein, 2002 ), [2] Brahmaputra River basin (the Brahmaputra valley forest zone of  Olson and Dinerstein, 2002 ), and [3] Northeast Hills (Meghalaya and Mizoram–Manipur–Kachin forest zones of  Olson and Dinerstein, 2002 ). 348  B I O L O G I C A L C O N S E RVAT I O N  136 (2007) 346  –  361  measurements were calculated in a GIS application (ESRI Arc-View  3.3) or digitized from paper maps. In all cases, maxi-mum distance error values were calculated using the online java Applet: http://elib.cs.berkeley.edu/manis/gc.html (lastaccessed on August 14, 2006). Commonly used base mapsources included: Digital Chart of the World (for roads andhydrology, at 1:1,000,000 scale; available from http://www.maproom.psu.edu/dcw/dcw_about.shtml#DCW, last ac-cessed on August 14, 2006); Australian Centre of the AsianSpatial Information and Analysis Network (for secondaryand tertiary administrative boundaries; available from http://www.asian.gu.edu.au/; last accessed on August 14, 2006);GTOPO (30 minute Digital Elevation Models, available fromhttp://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html;last accessed on August 14, 2006). Georeferencing was verifiedby comparing the historical localities against basic adminis-trative boundaries (primary, secondary, tertiary, etc.) orknown ranges by experts to search for obvious outliers thatmay indicate transcription or georeferencing errors. Data processing : The unprocessed data set consisted of 10,685 records (10,154 from Burma, 369 from Yunnan Prov-ince, China, and 223 from NE India). Of these, 98.2% of thedata were collected between 1997 and 2005. The seasonal(monthly) distribution of sampling effort is summarized inFig. 2. Much of the data included taxa that are either undergo-ing taxonomic revision, or can only be identified tentatively.To address this problem, expert opinion and identificationswere solicited from researchers (see Acknowledgments). Taxathat could not be assigned species-level identification, butclearly belonged to a monophyletic clade, were placed undera pseudotaxon with the generic epithet (typically by adding ‘‘sp.’’ to the genus name). Taxa known to be particularly prob-lematic in terms of their taxonomic status or were unlikely tobe monophyletic were then removed. Taxa with only singlerecords were also removed.The processed dataset consisted of 184 species (63amphibian and 121 reptile species) represented by a total of 2908 records, with a range of 2–208, and a median of 8 (Fig. 2). 2.3. Niche modeling Niche modeling was performed using the Maxent softwarepackage (version 2.2) (Phillips et al., 2004, 2006). Maxent usesthe maximum entropy principle: in the estimation of an un-known probability distribution (over some space), the leastbiased solution is the one that maximizes its entropy, subjectto some constraints that reflect available information. In thecase of ecological niche modeling, the target is to calculatea probability distribution (the niche model) for a species overthe given geographical space (the pixels), using informationfrom the observed association of the species’ localities withenvironmental variables (e.g., climatic layers). Maxent calcu-lates this maximum entropy distribution, using the observedassociation between the species’ and environmental layers toset the following constraint: the expected value (expectation)of each ‘‘feature’’ (which is either an independent variable it-self, or one derived from it) under the estimated distributionmust be similar to its observed average over sample locations.The resulting model is a probability distribution over all gridcells in the chosen geographical space (the values across allthe cells add to 1), and expresses the suitability of each gridcell as a function of the environmental variables in it. A highvalue of the function at a particular grid cell indicates that itis predicted to havesuitableconditionsfor thattaxon. Maxenthas been shown to be robust for modeling presence-onlyoccurrence data, outperforming many other traditional tech-niques such as GARP (Elith et al., 2006).Ninteen environmental variables, each at a resolution of 30 00 (0.008333  ·  0.008333  , or   1 sq. km), were obtained fromthe WorldClim database (interpolated from global climatedatasets; Hijmans et al., 2005). An additional spatial layer of elevation from 30 00 SRTM data was also used. Table 1 providesa list of the environmental variables. All layers were clippedto an area bounded by 29  34 0 25.56 00 N by 87  49 0 20.77 00 E and8  43 0 39.94 00 N by 101  19 0 36.44 00 E, a rectangular box containing all of NE India and Burma, and covering all the sample datapoints (many of which lay in adjoining regions). Additionalenvironmental layers were not included in the niche model-ing primarily because GIS data at 1 km resolution were notavailable for the region. A coarsening of spatial resolutionwould have allowed the use of soil and vegetation type data,but given the relatively small spatial extent of the NE India re-gion, preference was given to a fine-scale analysis, albeit withfewer environmental variables. In addition, a coarsening of resolution would have excluded many taxa from the analysisbecause Maxent was run without duplicates; at most one 0510152025 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 0204060801000 50 100 150 200 250 Occurences    N  u  m   b  e  r  o   f  s  p  e  c   i  e  s   ' Month    P  e  r  c  e  n   t  a  g  e  o   f  r  e  c  o  r   d  s Fig. 2 – (a) Temporal distribution and (b) frequency distribution of occurrence data for amphibians and reptiles from NE Indiaand Burma. B I O L O G I C A L C O N S E RVAT I O N  136 (2007) 346  –  361  349  sample (presence record of each taxon) was allowed per pixel.This was necessary because given the opportunistic nature of the survey effort, presence of multiple records within a pixelcould not be interpreted as abundance.Maxent was run using the linear, quadratic and productfeatures (Phillips et al., 2006). The product feature capturesthe effect of interactions (covariances) between pairs of fea-tures (Phillips et al., 2006). Threshold and hinge features (bothof which involve imposing cut-offs on continuous-valuedenvironmental data) were not used because the theoreticalbasis for their use is unclear (Phillips et al., 2004, 2006), andexploratory models with the Indo-Burma data using thesefeatures were seen to result in spatially clustered predictions.Maxent is a relatively new method, and there have been fewevaluations of how the results vary with changes in the algo-rithm convergence threshold across different taxonomicgroups. Hence a convergence threshold of 10  5 was chosen,as it has been used in previous studies with Maxent (Phillipset al., 2004, 2006).In Maxent, the stringency of the constraint that the fea-tures’ observed averages and expectations must be similar,is controlled by a ‘regularization’ parameter beta  b  (Phillipset al., 2006). Due to biases associated with museum and sur-vey data, empirical means of features will at best be crudeapproximations of the true mean of the environmental datathat comprise a taxon’s realized niche.Forexample, if surveysare biased towards certain areas, the sample data will be spa-tially clustered, and the sample mean will be a poor estima-tor. For small sample sizes, this predisposes the proceduretowards overfitting, resulting in the modeled distributionstending to be clustered around sample localities. Maxent 2.2attempts to minimize this problem by adjusting   b  according to each taxon’s sample size (achieved by setting   b  to ‘‘auto’’in the graphic user interface) and the types of features used(Phillips et al., 2006). Fig. 3 shows a plot of   b  values versussample sizes for the full set of 184 species. If this approach to-wards regularization is sufficient to reduce overfitting, a lin-ear increase in the number of iterations to convergence isexpected across all taxa. To test this, a test run of Maxentwas conducted with 1000 iterations with a subset of the datathat had relatively small sample sizes (<50) for the NE India–Burma dataset. However, no significant increase was notedand for a number of taxa with <20 samples, convergence isnot achieved even after a large number of iterations (Fig. 4).An examination of the geographical distribution of the dataalso suggested that spatial clustering of sample occurrencesmight be an additional factor resulting in overfitting. Thus,despite adjustment of the regularization parameter  b , overfit-ting would remain a problem for a large part of the dataset.Hence, an additional techniquewas used to reduce overfitting by lowering the number of iterations (350). This number waschosen on the basis of convergence pattern in the samplerun; a number of small sample-sized taxa (samples <30) haditeration values above this threshold (Fig. 4).To gauge the accuracy of the niche models, Maxent per-forms threshold-dependent as well as threshold-independenttests (‘‘threshold’’ here refers to the Maxent output, and not tothe derived environmental features mentioned above). Bothmethods require random subsets of the data to be set asidefor testing. For threshold dependent testing, the continuousnumerical value of each pixel is converted to binary pres-ence/absence using some threshold. A one-tailed binomialtest is then used to determine whether a the model predictsthe test localities significantly better than random (Phillipset al., 2006). Maxent 2.2 implements 10 different thresholding methods and reports one-tailed  p -values for the binomial testof each. The threshold-independent test consists of the recei-ver operating characteristic analysis procedure modified forpresence-only data (Phillips et al., 2006). In this method, a re-ceiver operating characteristic curve is obtained by plotting sensitivity (fraction of correctly predicted presences) on the y -axis and 1 – specificity (specificity is fraction of all absencescorrectly predicted as such) on the  x -axis for all possiblethresholds (all possible cutoffs for probability of presence).The area under the curve then provides a measure of modelperformance that is independent of any particular choice of threshold. Values for area under the curve >0.5 indicate a Table 1 – Environmental data used in distributionmodeling Environmental datum Annual mean temperatureMean diurnal rangeIsothermalityTemperature seasonalityMaximum temperature of warmest monthMinimum temperature of coldest monthTemperature annual rangeMean temperature of wettest quarterMean temperature of driest quarterMean temperature of warmest quarterMean temperature of coldest quarterAnnual precipitationPrecipitation of wettest monthPrecipitation of driest monthPrecipitation seasonalityPrecipitation of wettest quarterPrecipitation of driest quarterPrecipitation of warmest quarterPrecipitation of coldest quarterAltitude 00.511.522.530 50 100 150 200 250 Number of Samples Fig. 3 – The relationship between sample size and theregularization parameter  b  of MaxEnt for the initial set of 184 taxa used in the niche modeling. See the text fordiscussion. 350  B I O L O G I C A L C O N S E RVAT I O N  136 (2007) 346  –  361
Similar documents
View more...
Search Related
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