Assign a suitability score to each of the different classes within every factor (such as desert scrub or grassland within landcover). Set biologically meaningful thresholds to divide habitat suitability scores into categories, paying particular attention to the suitability threshold required to support breeding habitat. Assign a score of zero only when the species absolutely won't use a particular class.
Whenever possible, recruit an expert biologist knowledgeable of the focal species to parameterize the model for each species. When this is not possible, we recommend recruiting several non-expert biologists to review all relevant literature for the species, parameterize the model independently, then compare and discuss differences and assumptions in parameterizations before averaging them into the final model.
A fundamental assumption is that habitat suitability and permeability are synonyms, and that both are the inverse of ecological cost of travel. Estimating suitability values (this section) and factor weights is the link between the behavior of the focal species as it moves through the landscape and non-ecological GIS data. Virtually all the relevant literature concerns habitat use, not animal movement, so we find it easier to estimate habitat suitability rather than habitat permeability to movement.
In least-cost modeling, habitat suitability and permeability are synonyms. Big numbers indicate good habitat suitability and high permeability, while small numbers indicate poor suitability. In the literature on corridor modeling, the term travel cost is used more commonly than suitability or permeability, and least-cost modeling is the generic term for the most common corridor models. Throughout this tutorial, just remember that cost and suitability are just opposite sides of the same coin, such that cost plus suitability = 100 (or other maximum value). When we get into corridor modeling, we will shift into cost terminology. But for now, it is easier to write and read this discussion using suitability terminology.
We recommend (and CorridorDesigner requires) using a scale with fixed end points (such as 0 to 100) rather than a scale with no upper limit. We also recommend that you use verbal descriptions of threshold values, such as those given below. As you will see in 2.5 Modeling habitat patches, providing biological interpretations to habitat suitability scores provide a rational basis for modeling habitat patches. An arbitrary scale does not have this virtue.
A score of zero should only be assigned to a class when the animal would not use the class, even if the other factors were optimal. For instance, a score of zero for elevations above 7,000 feet means that “the animal won't use this, even if the vegetation, topography, and road density are otherwise ideal.” This would be appropriate if 7,000 feet is the upper elevation limit of the species distribution.
Sometimes you will have to parameterize models yourself. But whenever possible, we recommend recruiting a biologist who is an expert on the species, especially if he or she has worked in or near the linkage analysis area. Even if they have published papers on habitat use, experts have reams of unpublished data and field experience that you can't get by reading papers.
In addition to scoring habitat preferences based on GIS variables, you should also ask the species expert to provide estimates of uncertainty, estimates of factor weights, and estimates of the sizes of areas needed to support a single breeding event and a breeding population.
It has been demonstrated that a species expert will do a better job parameterizing the model if they refer to the scientific literature while they do so. Even if the expert published most of those papers, he or she has forgotten a lot of it, and filling out the form without referencing the literature will result in a poorer model.
Sometimes you will not be able to find a species expert. When this happened to us, we assigned the task to 3 persons on our team. We provided each scorer with copies of the relevant literature, and we each independently filled out the spreadsheet. If our scores differed by ‹20 (on the 100-point scale described above), we used the average. We discussed each score that differed by 20 or more until we reached a consensus score.
Land cover is a factor in every corridor model we've seen, and it tends to come in distinct flavors, so that it makes sense to assign a permeability score to each class . But elevation and distance to road are inherently continuous variables. How do we estimate permeability as a function of a continuous variable?
We don't. Instead we define a few classes of elevation, and a few classes of distance to road, and estimate suitability for each class. But you need not be constrained by our lack of imagination, and there is nothing inherently wrong with developing a function that relates permeability to a continuous variable. All the issues discussed above still apply.
Assuming you are covering this material in order, you now have several sets of permeability scores that look something like this:
|Land cover||Topographic position||Distance to roads||Elevation|
|Pine forest||60||Canyon bottom||80||0-50 m||30||0-300 m||0|
|Grassland||30||Ridgetop||20||50-200 m||50||300-500 m||40|
|Urban||00||Steep slope||20||›200 m||90||500-1000 m||100|
|Agriculture||30||Gentle slopes||50||1000-1200 m||40|
|Riparian||100||› 1200 m||0|
As you can see, each pixel will have 4 suitability scores—one for each of the four factors. To estimate the overall permeability of a pixel, you must combine these four scores. To do this, you must assign a weight for each factor, and choose an arithmetic operation to apply these weights.