Live beings have been continually changing distributions, and they occur with varying frequency across space. Hence, categorical occurrence maps, whether observed or model-derived, are always incomplete and oversimplified representations of species’ actual distributions. Moreover, categorical occurrence derived from model predictions usually depends on largely arbitrary user-specified thresholds. Analyses that build on such categorical information, such as most indices of diversity, overlap, (dis)similarity and change, thus omit important gradations in species occurrence and can be visibly conditioned by threshold choice. Fuzzy logic is a simple tool to eliminate the need for these thresholds and formally incorporate the locational uncertainty and gradual variations that characterize natural biodiversity patterns. This course will show how fuzzy logic can be easily integrated into biogeographical analyses to improve the prediction and combination of species distribution patterns, with applications in (macro)ecological interaction, global change, biodiversity and biotic regionalization studies.
We will use presence-(pseudo)absence models that produce presence probability values (e.g. generalized linear and generalized additive models; tree-based classification and regression methods), which can be mathematically converted to favourability or fuzzy membership values. We will see how these values allow direct comparison and combination of gradual distribution patterns across species, regions and time periods. Finally, we will see how diversity and (dis)similarity indices, which normally require binary presence/absence information, can be generalized to work with fuzzy (degree of) occurrence values. This allows the use of presence probability models without depending on thresholds to convert them into binary predictions, thus avoiding the compounded effects of threshold choice on the results and conclusions.
The course will include both theoretical lessons explaining the concepts behind the described procedures, and practical hands-on sessions where participants will put these procedures into practice using R. In this short version of the course, we will not cover data preparation and model building. We will provide some already-made species distribution models as examples for participants to work with. Participants are also encouraged to bring their own presence-(pseudo)absence models, with their observed and predicted values, for different (ecologically related) species or time periods.