Change of support for zero-inflated data: deriving fine-scale species distribution inferences from spatially aggregated data

Abstract In environmental sciences, data are often available at a coarse resolution. This can result in a mismatch between the data resolution and the resolution at which process inference should be made. This misalignment, usually named change of support (COS), can lead to biased inference if not addressed appropriately. Yet, solutions to the COS issue have only been proposed for a limited number of observational distributions (e.g. Poisson or Gaussian), which narrows their range of application. Motivated by a fisheries case study, we introduce a hierarchical method addressing COS for zero-inflated data with highly skewed tails. Such data are common in environmental sciences but are not handled by existing COS methods. Our approach requires the spatial locations of point-level data to be known and considers that aggregated data are convolutions of these point-level data. We assess the accuracy of our method through a simulation study. Subsequently, we apply our model to a real case study, here common sole in the Bay of Biscay. Our findings illustrate that the model provides better estimates and predictions than the ad hoc method used for geoprocessing aggregated data. Specifically, the effect of the environment on the species distribution is better evidenced with our model.