Overview of habitat modeling

In this section, we review the concept of habitat and habitat modeling approaches, and call attention to a critical, untested assumption about the relationship between habitat modeling and corridor modeling—namely, that animals make decisions about how to move across the landscape using the same rules they use to select habitat.

What is habitat?

Habitat is “where an animal lives” or “the living and non-living characteristics of a landscape that an animal uses.” Although habitat is fundamentally a description of what animals use and where animals are found, most ecologists assume that habitat also is what animals need to survive and reproduce. Technically, only experiments can determine what animals need, and wildlife ecologists regularly engage in soul-searching about this slippery concept and whether our habitat studies are properly designed and interpreted. We will not get bogged down in this important and valuable debate, however. We will try to keep the focus on habitat as a description of what animals use, but at times we will slip into the assumption that habitat is what animals need to survive and reproduce—and to move across the matrix between wildland blocks.

Habitat is often broken down into several components, depending on what the animal is doing in a particular area or with a particular element of the landscape. Five components are usually listed as food, water, hiding cover (prey) or ambush cover (predators), thermal cover (against heat or cold or both), and nest sites (or other special needs for reproduction). Some ecologists add a 6th component, namely the minimum amounts and spatial arrangement of the first 5 components. Survival and reproduction require that an animal has enough of each habitat component within the range of its daily, seasonal, or annual activities.

An overview of habitat modeling approaches

Habitat models allow you to assess the quality of habitat for a species within the study area or a modeled corridor, and serve as the required cost layer for least-cost path and corridor analyses. In GIS, habitat suitability models relate suitability to raster-based layers such as land use/land cover, elevation, topographic position, human disturbance (e.g. distance from roads, road density, housing density, etc), or other important factor available as a GIS layer. We refer to these raster layers as factors. Within each factor, there are several to many classes. For instance, the factor land cover may include classes such as juniper woodland, desert scrub, and urban land. There are two common ways to build these models:

Literature review and expert opinion-based habitat suitability models

The most common habitat suitability modeling technique—and that used by CorridorDesigner—is based on literature review and expert opinion, and generally follows the ideas found in the 1981 U.S. Fish and Wildlife Service publication Habitat Evaluation Procedures Handbook. While literature-based models are subject to uncertainty and errors when translating literature-based habitat studies to a habitat suitability score, they are relatively easy to create, do not require new collection of detailed field data for all species in the linkage zone, and can be applied to multiple study areas, allowing for rapid analyses and linkage designs.

The procedure requires a biologist to assign a weight to each factor (section 2.4) and a habitat suitability score to each class within a factor (section 2.3). Suitability scores for all habitat factors are then combined to form a single habitat suitability map with a suitability score for each pixel. The two most common methods of combining factors are arithmetic (or additive) mean and geometric mean models. We elaborate on the differences between these algorithms in 2.4 Combining habitat factors. Further details on these models can be found in the Standards for Development of HSI Models section of the Habitat Evaluation Procedures Handbook.

Empirical and statistical techniques for estimating habitat suitability

Species occurrence

If presence-absence data or abundance is available for the species in the study area, then empirical statistical models can be created by relating the species occurrence data to habitat factors. Statistical techniques such as generalized linear or generalized additive models (e.g. logistic or Poisson regression), artificial neural networks, classification and regression trees (CARTs), and genetic algorithms can all be used to create a map of a species probability of occurrence at any pixel in the landscape.

With these models, data is typically extracted from the GIS layers, assembled into a site by occurrence matrix, analysed with a statistics package such as R, S-Plus, or SAS, then fed back into the GIS software to create a map depicting probability of occurrence. Stand-alone modeling packages such as Biomapper, openModeller, Maxent, or DesktopGARP can also be used.

While empirical models are probably more accurate than rule-based or literature-review based models, they require gathering a good set of field observations for every species in the linkage area, which can take a considerable amount of time.

How is habitat modeling related to corridor modeling?

Our approach has a fundamental, untested assumption—we assume that animals make decisions about how to move across the landscape using the same rules they use to select habitat. It is reasonable to assume that an animal prefers to move through areas that provide food, water, cover, and reproductive opportunities. But it is important to admit that we don't know this for sure. And in one study conducted by Horskins, Mather, and Wilson (Landscape Ecology 21: 641-655) we know this eminently reasonable assumption was false!

Horskins, Mather, and Wilson studied two small mammals which occurred in an 85-year old woodland corridor in Australia and in the woodland blocks it connected, but did not occur in the matrix of grassland and pasture surrounding the corridor. Reproductive individuals were trapped in the corridor, suggesting that the animals bred there, but there was apparently no gene flow between the two woodland blocks for either species! Their genetic divergence was just as extreme as populations in isolated woodland patches.

Given even one counter-example as demoralizing as this one, why do we make the assumption that animals make decisions about how to move across the landscape using the same rules they use to select habitat.? We have no choice. Over 95% of the ecological literature we use to parameterize our habitat models are papers on habitat use. For any single species there will be at most 2 papers on animal movement; typically there are none. And only a small fraction of papers on movement describe the type of movement we are most interested in—namely how animals move between patches of suitable habitat.