Can large-scale patterns in insect atlas data predict local occupancy?

Coarse-grain atlas data are economical to collect, but fine-grain data contain more detail about the distribution of species within occupied cells. This information can change our perception of species’ distribution size, rarity and risk of extinction. Species occupancy describes the proportion of grid cells where a focal species is present, but occupancy depends on the spatial grain (grid cell area) of the units used to record species presences. Downscaling models have been developed to describe and extrapolate the relationship between spatial grain and occupancy (the occupancy-area curve, OAR), but have not previously been tested for highly mobile organisms. Here, we use atlas data for 38 British Odonata species. This taxon is highly mobile and also aggregated in the landscape due to a dependence on freshwater bodies for reproduction. Occupancy data at five coarse grains ≥ 100 km2 were used to parameterise 10 downscaling models. Predictive accuracy of the models were compared, using predicted and observed occupancy at the 1, 4 and 25 km2 grains. The Hui model gave the most accurate downscaling predictions across 114 species:grain combinations and gave the best predictions for 15 of the 38 species. Species-level traits were able to explain nearly 60% of the variation in in downscaling predictive error. Species with widespread, localised-aggregated and localised- sparse distributions were better predicted than species with a climatic range limit in Britain. The fine-grain occupancy of species with good dispersal abilities were poorly predicted by downscaling. Habitat generalists and specialists were better predicted than species with intermediate habitat breadth. Our results suggest that downscaling models, using widely available and economical coarse-grain atlas data, can provide sensitive, fine-grain estimates of distribution size, rarity, abundance and range change, even for high mobile taxa with a strong spatial structure.  The code to obtain downscaling predictions for the Hui model is included as a pdf.