USGS - science for a changing world

South Florida Information Access (SOFIA)

projects > climate envelope modeling for evaluating anticipated effects of climate change on threatened and endangered species in the greater everglades > work plan

Project Work Plan

Department of Interior USGS GE PES

Fiscal Year 2011 Study Work Plan

Study Title: Climate envelope modeling for evaluating anticipated effects of climate change on threatened and endangered species in the Greater Everglades
Study Start Date: 01 October 2010
Study End Date: 30 September 2011, with possibility of future funding tied to progress
Duration: 12 months
Location (Subregions, Counties, Park or Refuge): Southern Florida, entire range of species being considered including parts of the Southeastern US, and Central and South America
Funding Source: GE PES
Other Complementary Funding Sources: USFWS science hold back award
Funding History: FWS to University of Florida partners, CESI to University of Florida partners, GE PES
Principal Investigators: Stephanie S. Romañach (USGS), James Watling (University of Florida), Laura A. Brandt (USFWS), Leonard G. Pearlstine (NPS), Don DeAngelis (USGS), Ikuko Fujisaki (University of Florida), Frank J. Mazzotti (University of Florida)
Study Personnel: Emily Pifer (University of Florida), Yesenia Escribano (University of Florida), Carolina Cabal (University of Florida), Michelle Curtis (USFWS)
Supporting Organizations: FWS, NPS, University of Florida
Associated / Linked Studies: 'Climate envelope modeling for evaluating anticipated effects of climate change on threatened and endangered species' (funded by CESI); 'Refinement of species-habitat models and development of climate envelope models for evaluation of potential effects of climate change on threatened and endangered species' (funded by USFWS); 'Scientific and Technical Support for a Joint Ecosystem Modeling Laboratory' (funded by PES)

Overview & Objective(s): Climate change is poised to induce a cascade of direct and indirect effects on biodiversity that will require novel, data-driven approaches to management. To provide robust estimates of climate change effects on species of greatest conservation concern, we will create climate envelope models for 21 species of non-marine threatened and endangered (T&E) vertebrates that occur in south Florida. In addition to the initial 21 species targeted in FY10, in FY11 we will be incorporating five additional T&E species within the Peninsular Florida Landscape Conservation Cooperative (LCC) boundary, the Florida Bonneted Bat (Eumops floridianus), Florida Salt Marsh Vole (Microtus pennsylvanicus dukecampbelli), Anastasia Island Beach Mouse (Peromyscus poliontus phasmus), Whooping Crane (Grus americana), and the Flatwoods Salamander (Ambystoma cingulatum; Table 1). We will compile the most up-to-date information on T&E vertebrates in the study area and use these data to select a subset of climate variables with known and quantifiable relationships to the ecology of focal species. Our analytical approach will emphasize an ensemble modeling approach to projecting future distributions of climate niche space for T&E vertebrates. Because our research group includes representatives of major agencies involved in biodiversity conservation, our results will be made available to managers in a timely and accessible manner that maximizes the utility of our models for conservation decision-making.

Table 1. Focal species for climate envelope modeling
Common name Scientific name
Florida Bonneted Bat Eumops floridianus
Florida Salt Marsh Vole Microtus pennsylvanicus dukecampbelli
Key Largo Woodrat Neotoma floridana smalli
Key Deer Odocoileus virginianus clavium
Silver Rice Rat Oryzomys palustris natator
Key Largo Cottom Mouse Peromyscus gossypinus allapaticola
Southeastern Beach Mouse Peromyscus polionotus niveiventris
Anastasia Island Beach Mouse Peromyscus polionotus phasmus
Florida Panther Puma concolor coryi
Lower Keys Marsh Rabbit Sylvilagus palustris hefneri
Cape Sable Seaside Sparrow Ammodramus maritimus mirabilis
Florida Grasshopper Sparrow Ammodramus savannarum floridanus
Florida Scrub Jay Aphelocoma coerulescens
Piping Plover Charadrius melodus
Whooping crane Grus americana
Wood Stork Mycteria americana
Red-cockaded woodpecker Picoides borealis
Audobon Crested Caracara Polyborus plancus audobonii
Everglades Snail Kite Rostrhamus sociabilis plumbeus
Roseate Tern Sterna dougallii dougallii
Amphibians and Reptiles
Flatwoods Salamander Ambystoma cingulatum
American Crocodile Crocodylus actutus
Eastern Indigo Snake Drymarchon corais couperi
Bluetail Mole Skink Eumeces egregius lividus
Sand Skink Neoseps reynoldsi
Atlantic Salt Marsh Snake Nerodia clarkii taeniata

Specific Relevance to Major Unanswered Questions and Information Needs Identified:

Status: Many conservation and resource management groups have recognized the need to have an integrated, adaptive, landscape approach to conservation. To more effectively address the growing threats to fish and wildlife conservation in the 21st century, the FWS in partnership with USGS adopted a landscape approach to conservation termed Strategic Habitat Conservation. This adaptive approach links site specific actions to population and landscape sustainability through the use of biological planning, conservation design, conservation delivery, decision-based monitoring, and assumption-driven research.

The methods and tools we develop will allow resource managers to examine potential effects of climate change on species under their stewardship in the context of ecosystem and landscape planning. We will work to close the gap between managers and scientists who produce data necessary for making key decisions in the face of climate change. Locally, resource managers within South Florida (including NPS, FWS, and the South Florida Water Management District) have indicated the need to run predictive models and view model outputs on their desktop computers, the ability to adjust model parameters when assessing alternatives, and for a spatially-explicit visualization environment for comparing alternatives. The Greater Everglades Priority Ecosystem Science program provided seed funding to develop a prototype application, called EverVIEW, to address the needs of resource managers. EverVIEW has been met with enthusiasm by the aforementioned agencies and further development should serve our need of making data, information, and model output available to resource managers.

In our effort to develop appropriate models for resource managers' needs, we will use the best and most up-to-date information and methods available. For each of the T&E species under consideration, we are developing bioclimatic models, also called 'climate envelope' models or 'niche' models. These models allow us to relate species' geographic distributions to climate factors. Determining the niche of a species allows its potential geographic range to be forecast through projection of the estimated niche boundary on a spatial domain (Drake & Bossenbroek 2009). Predicted future climate variables are used to predict future species distributions. Bioclimatic models are widely used because they can effectively predict climate-induced range shifts for large numbers of species (Beaumont et al. 2007, Jiguet et al. 2007, Lawler et al. 2006, Huntley et al. 2004, Thomas et al. 2004, Thuiller et al. 2005, Pearson et al. 2002) and provide a first step that can address issues and needs at different spatial and temporal scales. For example, Lawler et al. (2006) included land cover as a variable in their models because climate-induced changes in vegetation should be able to make more accurate future projections.

Planned Products: Through the course of the project, we plan to deliver:


In FY11 we will finish building a relational database for the first 21 species. We also will finish acquiring references and build a database with trait data for the five additional species. All literature searches and efforts will be documented. We will continue to interact with potential users of the database within FWS and Florida Fish and Wildlife Conservation Commission to identify key information needs.

Compile map layers of current species' geographic ranges

Work completed during FY10

Range information has been acquired and documented for the initial 21 taxa. Spatial data collection for four species has been completed. These species are: Crocodylus acutus, Puma concolor, Aphelocoma coerulescens, and Rostrhamus sociabilis. Occurrence data was collected from web-based biodiversity databases and numerous georeferenced museum collections. Additional georeferenced locations were extracted from the 'in-house' literature database. All occurrence data for these four species has been processed, evaluated for validity, and is now being pilot tested in climate change models. Proper metadata also has been created for said species data. Occurrence data for the remaining 17 species and the five new species added to the project taxa also have been collected from web-based biodiversity data base and geoferenced museum collections.

Work to be completed during FY11

In the following months we will finish acquiring occurrence data for the remaining species from georeferenced literature, provide documentation, and input the data into our climate models.

Land cover/habitat information will be compiled from a variety of sources and then associations/linkages will be made among classes in the various layers (i.e., cross-walking)

Work completed during FY10

We have acquired land cover/habitat information for four sources of land cover data that range from local to continental in extent: (1) land cover developed for the Comprehensive Everglades Restoration Plan system-wide Monitoring and Assessment Plan (South Florida Water Management District); (2) a synthesis of the Florida Gap Map for Everglades National Park, and land cover maps produced for above for WCA1, WCA2 and WCA3; (3) Florida Fish and Wildlife Conservation Commission 2003 Florida vegetation and land cover raster dataset; and (4) the MODIS Terra Land Cover Type Yearly L3 Global 1 km SIN Grid.

Work to be completed during FY11

We will crosswalk the four classifications in the coming months.

Begin compiling database of climate information important for species geographic range limits (e.g., temperature, precipitation)

Work completed during FY10

We have acquired climate data describing current conditions (e.g., roughly 1950—2000, inclusive) for temperature and rainfall from two global climate datasets (the WorldClim dataset, available at, and the Climate Research Unit (CRU) dataset, available at All data have been converted to both ESRI raster and ASCII format for ease of use across modeling platforms, and are stored on a shared network drive at Fort Lauderdale Research and Education Center.

plots showing model performance (mean values) and uncertainty (standard errors) for three species, two climate datasets and four algorithms
Figure 1. Model performance (mean values) and uncertainty (standard errors) for three species, two climate datasets and four algorithms. Response variables are area under the receiver-operator curve (AUC, a measure of the ability of the model to discriminate between sites occupied by a species and sites for which occupancy data are unavailable) and spatial correlation among predictive maps. WorldClim and CRU are two alternative datasets describing contemporary climate conditions; the same suite of climate predictors were extracted from each dataset and incorporated into models for each species based on an independent variable selection procedure (see text for more details). Maxent, Generalized Linear Models (GLM), Random Forests (RF) and Support Vector Machines (SVM) are four algorithms used to model species-climate relationships. [larger image]
Although selection of data inputs is critical to model performance (Araújo & New 2007), the criteria used to select contemporary climate data inputs is usually unreported, and we are not aware of any studies that systematically compare performance of models constructed on the basis of alternative climate inputs. Therefore, we have initiated a quantitative assessment of data input effects on model performance, using three species (Florida panther, Puma concolor coryi, American crocodile, Crocodylus acutus, and Florida scrub jay, Aphelocoma coerulescens) selected for the methods comparison (see below) as case studies. Briefly, we have used an independent variable selection approach (Ecological Niche Factor Analysis, ENFA, Hirzel et al. 2002) to select a subset of candidate climate variables most associated with occurrence of the three target species. The same suite of variables was extracted from the WorldClim and CRU datasets, and model performance was assessed based on 100 replicate runs of each species (N=3) by dataset (N=2) by method (N=4) combination (a total of 24*100 randomizations). Further details on the modeling approaches and assessment are described under the methods comparison below. Results to date indicate substantial variation in model performance associated with alternative climate datasets (Figure 1), and effect of climate inputs appeared to be most pronounced when model performance (AUC) was relatively low. These results suggest that no single climate dataset is associated with consistently highly-performing models, and argue for adoption of an ensemble approach to defining initial climate conditions.

Work to be completed during FY11

We will continue to assess performance of alternative climate datasets, expanding the pool of species and, as appropriate, additional climate datasets. The preliminary data presented here will form the basis of a manuscript to be submitted to a scientific journal. The goal for submission of the manuscript describing our quantitative assessment of climate conditions on model performance will be the first quarter of 2011. The target journal is Ecography. In addition, the information will be used in a technical guidebook that will be developed in the next two years with funding from FWS (National Office 2010 science hold back money) and NPS (CESI). The guidebook will describe construction and interpretation of climate envelope models. There will be sections that discuss data needs, assumptions, and uncertainty in models using our target species as case studies to illustrate the use of climate envelope models as decision support tools.

Comparison of potential modeling methods (e.g., support vector machines, random forest models) using 1—3 species as test cases

Work completed during FY10

As presented above, we have completed a preliminary methods assessment based on three species (Florida panther, Puma concolor coryi, American crocodile, Crocodylus acutus, and Florida scrub jay, Aphelocoma coerulescens), two climate datasets and four algorithms. In consultation with colleagues from USGS and NPS, we have selected a subset of four modeling algorithms that we will use to construct climate envelope models: maximum entropy (Maxent, Phillips et al. 2006; Phillips & Dudík 2008), generalized linear models (GLM; McCullugh & Nelder 1989), random forest models (RF; Cutler et al. 2007) and support vector machines (SVM; Schölkopf & Smola 2001). Our selection of this subset of algorithms was based on a combination of high model performance in published methods comparisons, high potential for future widespread adoption in the species distribution modeling (SDM) world, and methodological flexibility that will allow us to readily and rapidly conduct in-house evaluation of model performance as a function of data inputs. Importantly, many of the methods we have selected are robust to small sample sizes (Elith et al. 2006; Hernandez et al. 2006) and produce useful models even when few observations of species occurrence are available (as is the case for many T&E species). Each method has particular strengths: Maxent produces robust output even when few species observations are available, GLM allows for transparent tracking of species-environment relationships, RF allows the user to model highly non-linear relationships, and SVM can be constructed on the basis of presence-only data.

Our preliminary assessment focuses on two metrics of model performance: AUC and spatial correlation among prediction maps. The AUC statistic describes the tendency for random occupied points to have a higher predicted suitability than random points for which occupancy is unknown (e.g., a random pseudoabsence point); models with high AUC values (generally greater than 0.9; Manel et al. 2001) better differentiate climate conditions at points occupied by a species from random points. The spatial correspondence in predicted climate suitability between prediction maps can be estimated using the Pearson correlation coefficient (Syphard & Franklin 2009), and provides a spatially-explicit metric of model performance. We developed code to execute bootstrap analyses in program R for GLM, RF and SVM; random subsets of species occurrences were repeatedly appended to a database of 'pseudo-absence' data where the species has not been observed. Model performance was assessed on the basis of average values across 100 random partitions of the occurrence data, and uncertainty was expressed as the standard deviation in performance. A similar approach was utilized based on user-defined selections in the program Maxent (note that Maxent software does not allow for direct comparison of prediction maps, so the spatial correspondence statistics are not included for the Maxent algorithm). In general, Maxent was the single best-performing algorithm (Figure 1), although there was considerable variation in performance of models describing climate envelopes for the Florida panther. GLM also produced generally high-performance models (Figure 1). Random forest and SVM were mode variable, performing well under some circumstances, but lower than Maxent and GLM for two of three focal species (Figure 1). SVM, in particular also tended to produce the most variable spatial predictions of the algorithms (Figure 1). Although a consensus is emerging that ensemble models produce more robust results than any single approach along (Araújo & New 2007), it may be reasonable to consider a weighted-average approach to ensemble modeling in recognition of the apparent superior performance of Maxent relative to other algorithms.

Work to be completed during FY11

We will continue to refine our methods assessment, including additional performance criteria (e.g., Cohen's Kappa) and target species. We will also more explore differences in performance and the spatial signature of predictions for sub-species versus nominal species (note that data presented here for the Florida panther are for nominal species Puma concolor rather than the Florida subspecies Puma concolor coryi).

Fact sheet(s) describing the project and potential impacts of climate change on threatened and endangered species

Work completed during FY10

The fact sheet for the project and a general fact sheet on science support for climate change were completed and published by UF-EDIS (see Appendix I).

Work to be completed during FY11

None planned.

Literature Cited

Araújo, M. B. and M. New. 2007. Ensemble forecasting of species distributions. TRENDS in Ecology and Evolution 22:42—47.

Beaumont L. J., A. J. Pitman, M. Poulsen, and L. Hughes. 2007. Where will species go? Incorporating new advance in climate modelling into projections of species distributions. Global Ecology and Biogeography, 13:1369-1385.

Breiman, L., 2001. Random Forests. Machine Learning 45: 5–32.

Cox, J., Kautz, R., MacLaughlin, M. and Gilbert, T., 1994. Closing the Gaps in Florida's Wildlife Habitat Conservation System. Office of Environmental Services, Florida Game and Fresh Water Fish Commission, Tallahassee.

Cutler, D. R., T. C. Edwards Jr., K. H. Beard, A. Cutler, K. T. Hess, J. Gibson and J. J. Lawler. 2007. Random forests for classification in ecology. Ecology 88:2783—2792.

Drake, J.M. & J.M. Bossenbroek. 2009. Profiling ecosystem vulnerability to invasion by zebra mussels with support vector machines. Theoretical Ecology 4: 189-198.

Elith, J., C. H. Graham, R. P. Anderson, M. Dudík, S. Ferrier, A. Guisan, R. Hijmans, F. Huettmann, J. R. Leathwick, A. Lehmann, J. Li, L. G. Lohmann, B. A. Loiselle, G. Manion, C. Moritz, M. Nakamura, Y. Nakazawa, J. McC. Overton, A. Townsend Peterson, S. J. Phillips, K. Richardson, R. Scachetti-Pereira, R. E. Schapire, J. Soberón, S. Williams, M. S. Wisz and N. E. Zimmermann. 2006. Novel methods improve prediction of species' distribution from occurrence data. Ecography 29:129—151.

Hernandez, P. A., C. H. Graham, L. L. Master and D. L. Albert. 2006. The effect of sample size and species characteristics on performance of different species distribution modeling methods. Ecography 29:773—785.

Hirzel, A. H., J. Hausser, D. Chessel and N. Perrin. 2002. Ecological nighe-factor analysis: how to compute habitat-suitability maps without absence data? Ecology 83:2027—2036.

Huntley, B. R. E. Green, Y. C. Collingham, J. K. Hill, S. G. Willis, P. J. Bartlein, W. Cramer, W. J. M. Hagemeijer, and C. J. Thomas. 2004. The performance of models relating species geographical distributions to climate is independent of trophic level. Ecology Letters, 7: 417–426.

Jiguet F., A. Gadot, R. Julliard, S. Newson, and D. Couvet. 2007. Climate envelope, life history traits and the resilience of birds facing global change. Global Change Biology, 13:1673-1685.

Lawler J. J., D. White, R. P. Neilson, and A. R. Blaustein. 2006. Predicting climate-induced range shifts: model differences and model reliability. Global Change Biology, 1584.

Manel, S., H. C. Williams and S. J. Omerod. 2001. Evaluating presence-absence models in ecology: the need to account for prevalence. Journal of Applied Ecology 38:921—931.

McCullugh, P. and J. A. Nelder. 1989. Generalized Linear Models. London: Chapman and Hall.

Phillips, S. J. and M. Dudík. 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31:161—175.

Phillips, S. J., R. P. Anderson and R. E. Schapire. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling 190:231—259.

Pearson, R.G. and T.P. Dawson. 2003. Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology & Biogeography 12:361–371.

Schölkopf, B. and A. Smola. 2001. Learning with Kernals: Support Vector Machines, Regularization, Optimization, and Beyond. Cambridge: MIT Press.

Syphard, A. D. and J. Franklin. 2009. Differences in spatial predictions among species distribution modeling methods vary with species traits and environmental predictors. Ecography 32:907—918.

Thomas, C.D., A. Cameron, R.E. Green, M. Bakkenes, L.J. Beaumont, Y.C. Collingham, B.F.N. Erasmus, M. Ferreira de Siqueira, A. Grainger, L. Hannah, L. Hughes, B. Huntley, A.S. van Jaarsveld, G.F. Midgley, L. Miles, M.A. Ortega-Huerta, A.T. Peterson, O.L. Phillips, S.E. Williams, 2004. Extinction risk from climate change. Nature 427:145–148.

Thuiller, W., S. Lavorel, M.B. Araujo, 2005. Niche properties and geographical extent as predictors of species sensitivity to climate change. Global Ecology and Biogeography 14:347–357.

U.S. Fish and Wildlife Service. 1999. South Florida Multi-Species Recovery Plan. Atlanta, Georgia.

Appendix I

Abstracts & Presentations

Climate-based distribution models for the American Crocodile, Crocodylus acutus: Illustration of methodological challenges and management opportunities. Watling, James, I., Laura A. Brandt, Stephanie S. Romañach, Ikuko Fujisaki, Yesenia Escribano, Emily Pifer, Michelle J. Curtis, Frank J. Mazzotti, Don DeAngelis, and Leonard G. Pearlstine. Abstract and poster presentation at Ecological Society of America Meeting August, 2010, Pittsburgh, PA.

Fact Sheets

Planning for climate change in south Florida. Climate envelope modeling for threatened and endangered species. IFAS Publication WEC-282.

Science Support for Climate Change Adaptation in South Florida. IFAS Publication WEC-286.


Climate envelope modeling to forecast climate change effects on terrestrial vertebrates within Peninsular Florida. Submitted to FWS/USGS Science Support Project call for proposals and South Florida Ecological Services Office 2010 call for proposals.

Developing a framework for robust climate envelope models to forecast climate change effects on wildlife. Submitted to NASA Research Opportunities for Space and Earth Sciences 2010 call for proposals.

Climate Envelope Models in Support of Landscape Conservation. Submitted to FWS National office for consideration for funding with FY10 science hold back.