Habitat Suitability Modelling for the Model for Nature Policy; Evaluating methods for using crowd-sourced occurrence data to map habitat suitability in the Netherlands

By Max Hoving

Abstract
The Model for Nature Policy (MNP) is a digital model created and maintained by Wageningen University and Research (WUR), that predicts population groups of species in the Netherlands, by first mapping habitat suitability. There is a need to explore the use of occurrence data for modelling habitat suitability in the Model for Nature Policy (MNP), as their current modelling method is based mainly on theory and policy-related variables. Crowd-sourced occurrence data is freely available for this purpose. However, these data are expected to be spatially biased due to preferential sampling. The objective of this research is to explore how bias correction and habitat suitability modelling methods using crowd-sourced occurrence data, may affect MNP outcomes. 3 bias correction methods and 4 habitat suitability modelling methods were tested on crowd-sourced occurrence data of 48 bird species. Each bias correction and modelling method was applied to each species’ occurrence data. Resulting models were assessed using independent non-biased occurrence data with three evaluation methods: The Area Under the Curve (AUC), the Root Mean Square Error (RMSE) and a visual evaluation of prediction maps. The most suitable bias correction and modelling method to use for the MNP were identified and finally applied to a set of 9 different species’ occurrence data. The resulting habitat suitability maps of the 9 species were used as input for the MNP. The output population group predictions were compared to those using the original MNP modelling method. The results show that the most suitable habitat suitability modelling method is MaxEnt, and the best bias correction method in this context is one known as Presence Thinning (PT). However, these results are not conclusive as there are uncertainties regarding the model fitting practices and the validation data. Using MaxEnt on Presence Thinning (PT) corrected data within the MNP, resulted in overpredicted population group distributions and sizes. This was due to the assumption of the MNP that input habitat suitability maps exclude any natural areas that are seen as inherently unsuitable according to policy. Recommendations were given for creating habitat suitability maps that are more relatable to the original MNP method, as well as further study into what bias correction methods are suitable in what scenarios.

Keywords: Habitat Suitability Modelling; MaxEnt; Model for Nature Policy; Crowd-sourced Occurrence Data; Spatial Bias Correction; Population Groups