Ashton Drew 1, Alexa McKerrow 1, Steve Earsom 2

1 Biodiversity and Spatial Information Center, North Carolina State University, Raleigh, NC
2 U.S. Fish and Wildlife Service, National Wildlife Refuge System, Raleigh., NC

Introduction

International, national, and regional conservation plans set broad scale habitat and population objectives that potentially impact local US Fish and Wildlife Service (USFWS) wildlife refuge management practices. For example, Partners in Flight’s South Atlantic Coastal Plain Bird Conservation Plan (Hunter et al. 2001) recommends that eastern North Carolina and southeastern Virginia support a minimum of 1000 pairs of Henslow’s sparrow and the plan sets associated regional grassland and longleaf pine habitat management recommendations. The South Atlantic Migratory Bird Initiative Implementation Plan (Watson & McWilliams 2005) sets a conservation target of 100 piping plover pairs in eastern North Carolina within the next 50 years. Yet, how do these and other, less specific, regional population objectives step down to provide measures of success within individual refuges? How many individuals or breeding pairs would represent a viable population within each refuge and how much variation should be expected from year to year? At an individual refuge, which habitat management strategies or land acquisitions would most benefit the targeted populations? Answering such questions requires models that move beyond species occurrence mapping to represent predicted gradients in species density or abundance (Guisan and Thuiller 2005).

The recently completed Roanoke-Tar-Neuse-Cape-Fear (RTNCF) Ecosystem Biological Review Report (USFWS 2002) considered national and international conservation plans when setting specific habitat management objectives designed to protect and enhance key management species within the Ecosystem’s eleven National Wildlife Refuges. The report stops short, however, of setting specific population objectives for each refuge. Our pilot project will use SE-GAP landcover data and species occurrence models as a foundation to (1) develop habitat-specific population objectives for the RTNCF ecosystem and for the individual refuges and (2) provide quantitative measures of success by which the refuges can monitor the effectiveness of their management actions and their contribution to broader scale population objectives.

We will develop examples of three quantitative species distribution modeling scenarios (inductive, deductive, and aquatic), with confidence intervals, for select USFWS Trust Species within the RTNCF ecosystem. Trust Species are those for which the Service has legislative mandate, and include federally-listed threatened and endangered species, interjurisdictional fishes, and migratory birds (USFWS 1999). Inductive models will be developed for terrestrial vertebrate species represented by sufficient point count data within the RTNCF ecosystem region to quantify habitat-specific species density and independently validate the models. This approach is being developed through another pilot project at SE-GAP (Williams & McKerrow 2005, Laurent et al. 2006), which is exploring the data requirements and limitations of various quantitative modeling techniques to define spatial gradients in habitat suitability for six forest bird species in North Carolina. Deductive models will be developed for selected terrestrial vertebrate species that are well-studied, yet for which there are limited point count data within the RTNCF region. The aquatic models will also follow a deductive approach, but will require different spatial datasets and different spatial analyses than the terrestrial deductive models. Expert opinion, literature review, and a Bayesian belief network will provide the foundation for these deductive models.

Once the three example modeling scenarios have been developed and reviewed, the pilot project will expand by (1) developing the models for additional species of interest according to available data and time, and, for each species, (2) stepping-down the models to explore habitat and species objectives for the ten individual USFWS refuges within the RTNCF ecosystem (Figure 1). The models will be designed to guide future survey work to inform model refinement (e.g. by identifying and quantifying knowledge gaps and sources of model error and uncertainty) so that they may serve as useful, adaptive tools for setting and evaluating stepped-down, refuge specific population and habitat conservation objectives. They must also enable refuge managers to contrast the potential influence of alternative management scenarios, thereby facilitating the prioritization of candidate sites for various management actions such as land acquisition or habitat manipulation.

For all modeling scenarios, significant challenges addressed by this pilot project arise from the USFWS’s need for a product that will work across multiple spatial scales. We describe two of these challenges in this article: (1) How to effectively integrate habitat data compiled for regional applications with locally derived expert knowledge to provide quantitative population objectives at both the Ecosystem and the Refuge scale; and (2) How to effectively utilize expert opinion while distinguishing the uncertainty arising from natural variability from uncertainty arising from applying experts’ locally-derived knowledge, information, and opinion over broader spatial scales.

Matching SE-GAP data to USFWS step-down objectives

Species respond to and impact the environment at multiple spatial scales. Therefore the abundance of a species at any given site reflects that species’ environmental tolerances and preferences in relation to broad climatic gradients, land cover type, and microhabitat characteristics, as well as complex interactions across these scales (e.g. Poiani et al 2000). The use of SE-GAP data for this project will focus attention on processes and patterns that occur, and can be resolved, at a scale of 30 m (the grain of GAP data) to 109,530 km 2 (the extent of the RTNCF ecosystem). Some features of ecological significance will not be mapped in GAP (e.g. snags required by some bird species nesting in bottomland hardwood), and although multiple refuges may offer similar acreage of a desired habitat types (e.g. bottomland hardwood), not all refuges will have habitat of equal suitability (e.g. differing density of snags). While such variation may “average out” at the larger regional scale, as population goals are stepped down to the refuge scale, these fine-scale habitat features become increasingly important. Where USFWS defines a subclass of a mapped ecological system (Comer et al. 2003) from the GAP land cover map, or recognizes the importance of a particular microhabitat feature that varies greatly between refuges, we will have to call upon expert opinion to estimate these characteristics and summarize the uncertainty associated with these estimates.

In addition to the inability to regionally map some very specific habitat features, differences between the general concepts applied to define habitat types become increasingly important when a study addresses applications at multiple spatial scales. A cross-walk between the list of habitat types used throughout the Ecosystem Biological Review Report (USFWS 2002) and the SE-GAP land cover map is crucial for applying these data (Table 1). The SE-GAP map legend was developed with regional habitat modeling as a primary objective. At the same time, the target classes reflect previous experience with mapping based on Landsat satellite imagery and the existing ancillary datasets. The cross-walk shows some of the inherent uncertainty introduced when attempting to apply a landscape scale map to a habitat type classification that has been independently derived based on local knowledge and management practices within a much smaller subregion. That uncertainty is reflected in the many-to-many relationships. For example, the refuge biologists recognize four types of fresh water pools, ponds, and lakes (interdune ponds, vernal pools, oxbow lakes, Coastal Plain semi-permanent impoundment), whereas the SE-GAP map legend currently has a single class: open water fresh.

For two of the refuge classifications, interdunal ponds and vernal pools, these habitats will be found as inclusions in more general SE-GAP classes, similar to the snags needed by birds in bottomland hardwoods. With additional resources, oxbow lakes and Coastal Plain Semi-permanent impoundment habitats could be separated from the general open water cover class, but the question for the biologists will be, “How much will that added detail change the predictions?” In contrast, for bottomland hardwoods, the SE-GAP land cover map offers greater detail than the refuge biologists have included in their habitat classification (due to the relative rarity of this habitat within the existing refuge lands). The SE-GAP classifications account for variation in the stream/river size and blackwater versus brownwater subtypes. Habitat modelers have found these details important for regional and statewide applications, but it will be interesting to see if these distinctions, absent from the refuge documentation, will be applied locally.

Finally, because habitat types in the Ecosystem Biological Review Document (USFWS 2002) are specific to the USFWS Refuge lands and those lands are concentrated along the coast (Figure 1), the habitat types specific to the Outer Coastal Plain and Piedmont region are not represented. Therefore, in order to scale down habitat objectives, we will have to develop an understanding of the habitats available outside the existing network of refuges. To do this we will need to recruit additional experts in the development of habitat models. For those lesser studied regions, the SE-GAP map legend can be used as the template for general habitat types, the experts can help guide the development of additional data necessary to refine the models.

Predicting species abundance despite uncertainty

Conservation science is often described as a crisis science, where specific actions must be recommended despite uncertainty in the outcome (Soulé 1985). For management purposes, uncertainty has primarily been addressed indirectly through the use of qualitative and comparative measures. For example, the use of habitat suitability index models to determine the relative suitability of different sites for a given species has a long history within the USFWS (Schamberger et al. 1982). There is increasing pressure, however, for managers to base their management practices on quantitative, rather than qualitative, predictions of habitat suitability (e.g. the National Wildlife Refuge System Improvement Act, U.S. House 1997). In particular, management practices (and funding requests) must be justified by quantifying how the proposed actions will contribute to population objectives defined at broader scales.

While such quantitative estimates might seem unrealistic given current ecological knowledge, ecologists increasingly recognize the positive role that uncertainty can play when it is quantified and reported along with expert knowledge. Advances in the use of expert opinion allow experts’ knowledge and experience to be elicited in ways that provide the necessary statistics to support quantitative, deductive habitat suitability modeling with confidence intervals (Johnson & Gillingham 2004). One such approach uses an interactive graphical computer program in which expert opinions are transformed to alternative prior distributions for use in Bayesian analyses (Al-Awadhi & Garthwaite 2006). This approach also allows the modeler to explore potential error introduced by any systematic bias in expert’s opinions. Delphi surveys (e.g. Crance 1987), which use a series of questionnaires to build consensus among experts, can also be adapted to quantify uncertainty in expert opinion and thereby quantify uncertainty in model predictions. When uncertainty is explicitly considered, a manager predicts not just a population outcome, but also (1) the confidence intervals surrounding that prediction, (2) the potential consequences of over- or under-estimating the outcome, and (3) the model components for which erroneous values would have the greatest influence on the model outcome (Johnson & Gillingham 2004).

In essence, this pilot project requires a model that can be applied at two spatial scales, yet both (1) the availability of species and habitat data, and (2) the applicability of experts’ experience and knowledge, will differ greatly between the ecosystem and refuge scale. One component of this project will be to evaluate the sources and role of uncertainty in the species distribution models. For example, an expert’s ability to precisely and accurately estimate species and habitat characteristics is expected to be highest within the spatial boundaries of their local work experience, and their uncertainty should increase as they are asked to apply locally derived knowledge over ever-broader regions (Figure 2). Differences in experts’ opinions between refuges at the refuge scale (Figure 2, A versus B) might be more likely to reflect natural variability in species and habitat characteristics, than experts’ uncertainty or inexperience. Differences between their opinions at the ecosystem scale (Figure 2, A and B), however, might reflect greater uncertainty, in addition to any natural variability. Again using the example of snags in bottomland hardwoods, SE-GAP data document the location and extent of the bottomland hardwood habitat type, but would not provide the locations of suitable snags. Expert opinion and literature review would have to provide an estimate of expected snag density. At the refuge scale, individual experts (refuge biologists) may have quite precise and accurate knowledge of snag locations and density within their refuge, and snag density may vary between similar habitats at two refuges. As experts are asked to apply their knowledge to the broader, ecosystem area, however, the uncertainty of their estimates would be expected to increase.

Conclusion

At this point, we have compiled a list of the Key Management Species with their associated habitats, as identified in the RTNCF Ecosystem Biological Review Report (USFWS 2002). This list has been compared to the species and habitat data presented for the same region by (1) the North Carolina Wildlife Action Plan (North Carolina Wildlife Resources Commission 2005), and (2) NC-GAP (McKerrow et al. 2006). For several of the Key Management Species, the USFWS has already developed Habitat Suitability Indices, which offer graphical representations of species-habitat associations that could serve as a starting point for a graphical analysis of uncertainty. An initial crosswalk to match the USFWS and NC-GAP habitat classifications has also been completed (Table 1) and will need to be reviewed by the refuge biologists. Therefore, the next step will be to meet with wildlife biologists to identify a short-list of candidate species for modeling and confirm the classification crosswalk. Short-listed species will be selected based on (1) species data availability (including expert knowledge), (2) habitat data availability (including expert knowledge), (3) suitability of NC-GAP data (habitat classes, resolution) to model species-habitat associations, and (4) species management potential based on presumed threats.

Literature cited

Al-Awadhi , S.A. , and P.H. Garthwaite. 2006. Quantifying expert opinion for modelling fauna habitat distributions. Computational Statistics 21(1): 121-140.

Comer, P., D., Faber-Langendoen, R. Evans, S. Gawler, C. Josse, G. Kittel, S. Menard, M. Pyne, M. Reid, K. Schulz, K. Snow, and J. Teague. 2003. Ecological Systems of the United States: A Working Classification of U.S. Terrestrial Systems. NatureServe, Arlington, VA.

Crance, J.H. 1987. Guidelines for using the Delphi technique to develop habitat suitability index curves. US Department of the Interior Fish and Wildlife Service, Biological Report 82(10.134). 21 pp.

Guisan, A., and W. Thuiller. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8(9): 993-1009.

Hunter,W.C., L. Peoples, and J.A. Collazo. 2001. South Atlantic Coastal Plain Partners in Flight Bird Conservation Plan (Physiographic Area #03), Version 1.0. US Department of the Interior Fish and Wildlife Service, Atlanta, Georgia. 158 pp.

Johnson, C.J., and M.P. Gillingham. 2004. Mapping uncertainty : sensitivity of wildlife habitat ratings to expert opinion. Journal of Applied Ecology 41(6) : 1032-1041.

Laurent, E.J., S.G. Williams, and A.J. McKerrow. 2006. Developing a scientifically rigorous framework for enhancing and evaluating vertebrate models. Gap Analysis Bulletin 15.

McKerrow, A. J., S.G. Williams, and J.A. Collazo. 2006. The North Carolina Gap Analysis Project: Final Report. North Carolina Cooperative Fish and Wildlife Research Unit. Submitted to: The National Gap Analysis Program. U.S. Geological Survey, Biological Resources Division.

North Carolina Wildlife Resources Commission. 2005. North Carolina Wildlife Action Plan. Raleigh, NC.

Poiani, K.A., B.D. Richter, M.G. Anderson, and H.E. Richter. 2000. Biodiversity conservation at multiple scales: functional sites, landscapes, and networks. BioScience 50(2): 133-146.

Schamberger, M., A.H. Farmer, and J.W. Terrell. 1982. Habitat suitability index models: introduction. US Department of the Interior Fish and Wildlife Service. FWS/OBS-82/10. 2 pp.

Soulé , M.E. 1985. What is conservation biology? BioScience 35:727-734.

U.S. Fish and Wildlife Service. 1999. Fullfilling the Promise: The National Wildlife Refuge System, Visions for Wildlife, Habitat, People, and Leadership. 94 pp.

U.S. Fish and Wildlife Service. 2002. Biological Review of National Wildlife Refuges of the Roanoke-Tar-Neuse-Cape Fear Ecosystem in Northeastern North Carolina and Southeastern Virginia. 138 pp.

U.S. House. 1997. National Wildlife Refuge System Improvement Act. H.R. 1420 (Young). 105 th Congress.

Watson, C., and K. McWilliams. 2005. The South Atlantic Migratory Bird Initiative Implementation Plan: An Integrated Approach to Conservation of “All Birds Across All Habitats”, Version 3.1. Atlantic Coast Joint Venture. 88 pp.

Williams, S.G., and A.J. McKerrow. 2005. Refining Southeast regional GAP models for use in regional bird conservation planning: a pilot project. Gap Analysis Bulletin 13: 8-9.

Return to Table of Contents