The devil is in the details: The effect of spatial scale of fishery and environmental data on the performance of fish habitat models

Ismael Núñez Riboni, Anna Akimova and Anne Sell, Johann Heinrich von Thünen Institute, Bremerhaven, Germany
Habitat models of commercially exploited fish species are widely used to explore past and predict future climate-driven changes in fish distribution to support sustainable fishery management. To model the relation between climate and fish distribution, matching environmental and fishery data is an unavoidable step. According to our literature review, many modelers choose matching data at the highest possible resolution and favor using in situ over gridded data. Contrary to this, we show how resolution reduction (downsampling) of both fishery and environmental data improves the performance of fish habitat modeling. We use spatially resolved abundances throughout 5 decades from 12 functional groups of North Sea fish, as well as bathymetry and bottom temperature. We match our data at 5 spatial scales (between 1’ and 3°) by downsampling environmental data and mapping irregular fishery data on the same grid. Through combinations of spatial filter scales, we obtain 10 data-sets per functional group yielding 120 data-sets in total. Finally, we fit a realistic habitat model (with mixed effect for spatial autocorrelation) to each data-set and evaluate its performance with various metrics on training and cross-validating data. Our results show that downsampling of both environmental and fishery data improves the vast majority of the metrics in comparison to using in situ fishery and in situ or upsampled (interpolated) environmental data. Furthermore, matching data at high spatial resolutions often yielded unrealistic habitat models. The cross validation indicated that best predictions are made with data of higher resolution than the training data. To our knowledge, the present is the most comprehensive study of the effect of scale when matching environmental and fish abundance data and shows that the choice of an appropriate spatial scale could be critical to correctly model changes in fish distribution under climate change.