Scaling spatial pattern metrics: impacts of composition and configuration on downscaling accuracy

Context: Reliable scaling relationships allow prediction of unknown quantities at scales not directly observed. Scaling relationships for landscape patterns are commonly estimated following a process of coarse graining, in which pixels are aggregated and a mathematical function is then fit across the multiple resolutions. However, few studies have investigated how aggregation changes the landscape composition and configuration, which in turn alters the fitted scaling function and any predictions made using that function. Objective: Evaluate how changes in composition/configuration produced by coarse graining affect scaling function fit and the accuracy of that function for predicting patterns at finer resolutions than observed. Methods: Neutral model landscapes with controlled composition/configuration were generated in QRule. Each raster was aggregated to seven coarser resolutions, and landscape metrics computed. Five scaling functions were fit to the seven-value set for each landscape and metric. The best-performing function for each landscape/metric was downscaled to predict the metric at a fine resolution not used in the scale-fitting process. Downscaling accuracy was measured through relative error. Results: (1) The power law is the only function that consistently fit scalograms across all compositions/configurations. (2) Downscaling accuracy is more sensitive to configuration than composition. (3) Majority rules aggregation preserves information better with dispersed land covers. (4) Downscaling accuracies were highest around power law exponents of − 0.5 suggesting the structure of the power law may affect results. Conclusions: Aggregation impacts the fit and use of scaling functions for downscaling/prediction. Research to differentiate aggregation effects from physical processes in complex, real-world landscapes is needed.