Modelling wildlife distributions: Logistic Multiple Regression vs Overlap Analysis

Abstract

We compare the results, benefits and disadvantages of two techniques for modelling wildlife species distribution: Logistic Regression and Overlap Analysis. While Logistic Regression uses mathematic equations to correlate variables with presence/absence of the species. Overlap Analysis simply combine variables with the presence points, eliminating the non‐explanatory variables and recombining the others. Both techniques were performed in a Geographic Information System and we attempted to minimise the spatial autocorrelation of data. The species used was the Schreiber's green lizard Lacerta schreiberi and the study area was Portugal, using 10 X 10 km UTM squares. Both techniques identified the same group of variables as the most important for explaining the distribution of the species. Both techniques gave high average correct classification rates for the squares with presence of the species (79% for Logistic Regression and 92% for Overlap Analysis). Correct absence classification was higher with Logistic Regression (73%) than with Overlap Analysis (32%), Overlap Analysis tends to maximise the potential area of occurrence of the species, which induces a reduced correct classification of absences, since many absences will fall in the potential area. This is because a single presence in a given class of a variable makes all the area of that class to be considered as potential. The technique does not consider that the species may occasionally occupy an unfavourable region. Although, in Logistic Regression, modelling procedures are more complex and time‐consuming, the results are more statistically robust. Moreover. Logistic Regression has the capability of associating probability of occurrence to the potential area. Overlap Analysis is very simple in building procedures and swift in obtaining reliable potential areas. It is a valid technique especially in exploratory analysis of species distributions or in the initial stages of research when data may be scarce.

Statistical distribution models for migratory fish in Jacuí basin, South Brazil

“…= Standard Deviation. the regression models have been broadly used to predict distribution, abundance and habitat preference of species (Brito et al, 1999). However, when this information is joined with Geographic Information Systems (GIS), it is possible to incorporate ecological factors directly obtained from remote images to prediction models.…”

Can Iberian Water RailRallus aquaticusbe sexed reliably using simple morphometrics?

“…We selected this level of significance rather than the usual threshold (0.05) to ensure robust conclusions. Finally, an objective cut-off point (Pereira & Itami 1991, Brito et al 1999 was defined to convert the probability output (values 0 to 1) into dichotomous data (male-female) and the performance of the model for sexing individuals was evaluated. To avoid prevalence problems in the analysis, equal numbers of males and females were included.…”

Assessment of several spectral indices derived from multi-temporal Landsat data for fire occurrence probability modelling

“…In order to simplify further steps, we retained one model per index family (decision was based on the validation results). Optimal cut-off values were calculated for each model (Brito et al, 1999;Pereira & Itami, 1991) to create boolean maps, identifying pixel at risk of fire. For each year, a new integrated model was derived from the sum of the individual boolean maps.…”