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

Abstract

In a previous report, ENETWILD proposed a generic model framework to predict habitat suitability and likely occurrence for wild ruminant species using opportunistic presence data (occurrence records for wild ungulate species from the Global Biodiversity Information Facility). In this report, for the first time, we develop models based on hunting yield data (HY) for the most widely distributed wild ruminant species in Europe: roe deer (Capreolus capreolus) and red deer (Cervus elaphus). We also update models based on occurrence (roe deer, red deer, fallow deer (Dama dama), European moose (Alces alces) and muntjac (Muntiacus reevesi), evaluate the performance of both approaches, and compare outputs. As for HY models, we could not conduct one model per bioregion as there are not enough data for modelling in some bioregions, and therefore, we calibrated a unique model, including eco-geographical variables as predictors. The calibration plots for HY models showed a good predictive performance for red deer in the Eastern bioregion and roe deer at Eastern and Western. The abundance distribution pattern of red deer HY was widely scattered over all Europe, as expected for a widely distributed species which shows high ecological plasticity, and roe deer presented the highest abundance in Atlantic and Eastern Europe, progressively decreasing towards Northern Mediterranean bioregions. Overall, calibration plot did not perform well in the Northern region, which could be due to the low availability of data for both species in this bioregion. As for occurrence data models, performances using our revised approach for most species showed similarly moderate predictive accuracy. To sum, HY model projections showed good patterns where good quality data was provided, while worst predictions are found in neighbouring countries/bioregions. Two approximations to be explored for next models are: (i) modelling HY per bioregion providing more flexibility to the models, even if data projection is done at lower resolution scales, and (ii), modelling HY by accounting the fact that certain countries provide most data, to avoid that these areas overinform the model. As for occurrence data model, next steps for data acquisition and occurrence data modelling are: (i) review target group definitions for each species, (ii) revise definitions of “true” absence for model testing for better parity with fitting, and (iii) either replace principal component analysis with variance inflation factor analysis to remove co-correlates and model calibration for variable selection or develop post-model analysis to recover environmental dependencies.