Using Earth observation data to predict bird biodiversity in Central Norway

Abstract Earth observation (EO) data products offer opportunities for data-driven decision making in conservation, but questions remain about their applicability for monitoring and predicting spatial variation in biodiversity. For this, a key element is understanding the relevant spatial scales for species associations with their environment, with important implications from the perspective of land-use management. Here, we test the usage of EO data in predicting spatial occurrence patterns of bird species assemblages. We combine classified and unclassified EO data with stratified sampling of Norwegian breeding bird species occurrences in a joint species distribution modelling approach, comparing different spatial scales for the environmental predictors. We show that the applied approach for predicting bird diversity in areas not used for model training gave reliable estimates of the occurrence of some, but not all species. The most informative spatial scale of environmental predictors varied among bird species but had a minor impact on the aggregated results. We obtained reliable predictions of the ranking of areas in terms of species richness, as well as the spatial turnover of species, but failed to accurately predict the nestedness structure of species assemblages between plots, i.e. locations with few species that should have been represented a subset of a more species-diverse location. The level of processing of EO data had only a minor effect on the predictive performance. Thus, the trade-off between often resource demanding pre-processing of land-cover classes versus ecological interpretability are more important for understanding ecological processes behind biodiversity patterns than obtaining the best predictions. Transparent communication of the outlined uncertainties is crucial for research-based biodiversity conservation.