Backcasting biodiversity at high spatiotemporal resolution using flexible site-occupancy models for opportunistically sampled citizen science data

Abstract: For many taxonomic groups, online biodiversity portals used by naturalists and citizen scientists constitute the primary source of distributional information. Over the last decade, site-occupancy models have been advanced as a promising framework to analyse such loosely structured, opportunistically collected datasets. Current approaches often ignore important aspects of the detection process and do not fully capitalise on the information present in these datasets, leaving opportunities for fine-grained spatiotemporal backcasting untouched. We propose a flexible Bayesian spatiotemporal site-occupancy model that aims to mimic the data-generating process that underlies common citizen science datasets sourced from public biodiversity portals, and yields rich biological output. We illustrate the use of the model to a dataset containing over 3M butterfly records in Belgium, collected through the citizen science data portal Observations.be. We show that the proposed approach enables retrospective predictions on the occupancy of species through time and space at high resolution, as well as inference on inter-annual distributional trends, range dynamics, habitat preferences, phenological patterns, detection patterns and observer heterogeneity. The proposed model can be used to increase the value of opportunistically collected data by naturalists and citizen scientists, and can aid the understanding of spatiotemporal dynamics of species for which rigorously collected data are absent or too costly to collect.