Update of model for wild boar abundance based on hunting yield and first models based on occurrence for wild ruminants at European scale

Abstract

In the previous ENETWILD model, the predicted patterns of wild boar abundance based on hunting yield data reached an acceptable reliability when the model was downscaled to higher spatial resolution. This new approach, based on the modelling of hunting yield densities instead of hunting yield counts and the assessment of spatial autocorrelation, was only applied with simulated data and with data from two regions at hunting ground level, the smallest spatial resolution. In this report, (1) we evaluate whether this approach can correct the overpredictions for high-resolution predicted patterns when raw data are present at a different spatial resolution (i.e. the European region). For this purpose, hunting yield densities were incorporated as response variable (one model per bioregion) and predictions reliability at 10x10km and 2x2km spatial resolution were assessed. Internal validations and comparisons with the previous two-step model carried out at European scale were addressed, as well as an evaluation with external data at the same scale at country level. The model presented certain overprediction (much less than the previous model) of the total hunting bags reported per country, although a good correlation in terms of values and linearity between observed and predicted values was achieved. Secondly (2), a generic model framework to predict habitat suitability and likely occurrence for wildlife species using opportunistic presence data was proposed (occurrence records for wild ungulate species from the past 20 years exclusively from the Global Biodiversity Information Facility extracted on 9/12/2020). Across all wild ungulate species (elk (Alces alces), roe deer (Capreolus capreolus), red deer (Cervus elaphus), dam deer (Dama dama), muntjac (Muntiacus reevesi), wild boar (Sus scrofa)) the model framework performs well. For those species where area under the curve is below 0.7 we note lower accuracy in predicting absences, which requires further investigation to understand the root cause; whether a result of underlying assumptions regarding the testing data or due to the model performance itself.