Data driven fuzzy suitability modelling as a method for assessing habitat choice of migratory Red Kites (Milvus, milvus) across Spain

Abstract: As an increasing number of species drift onto the IUCN’s list of endangered species, it has become increasingly important to understand the fundamental aspects that are essential in maintaining the fitness of these threatened species. Using tracking data from 75% of a sample of Swiss juvenile Red Kites during their wintering period in Spain, various bioclimatic predictors (BCMPs) were taken into account to create fuzzy membership functions based on the spatial distribution of the related biogeographic elements within the operating range of the sample set of kites. The aim was to produce a continuous surface of values between 0 and 1, 1 being most suitable and 0 unsuitable for Red Kites across Spain. Using these functions, weighted linear combinations were undertaken to derive the best combination of BCMPs and what weight each carried that would best explain the distribution of the sample set. This was then validated against the remaining 25% to assess how well the most performant WLC would explain their distribution. Although the training data sample managed to explain its own distribution with some success, it failed to explain the distribution of the validation set with any statistical success. This could be partially attributed to the absence of potentially significant predictors, such as a food source, which for the red kite, mostly consists of small mammals. This, along with the generally elevated species plasticity and sample size, could have been responsible for the shortcomings in the model’s predictive abilities. er migration to Spain. During the months of December 2016 and January 2017 the data received was analysed and fuzzy membership functions were created based on 75% of the data, the training set, using different bioclimatic predictors that were deemed potentially important to species habitat choice. These functions were then combined using various weighted linear combinations (WLC), with both the combinations of predictors and their respective weights being altered in to best explain the habitat choice of the training set, the best result then being validated against the remaining 25% of the data, the validation set. Although the training set was well explained by the resulting WLC’s, the model failed to explain the validation set’s habitat choice with any statistical success. The model showed that no single predictor had the ability to explain the habitat choice and could be partially attributed to the absence of potentially significant predictors, such as a food source, which for the red kite, mostly consists of small mammals. This, along with the generally elevated species plasticity, could have been responsible for the shortcomings in the model’s predictive abilities.