Spatial Interpolation of Surface Air Temperatures Using Artificial Neural Networks: Evaluating Their Use for Downscaling GCMs

Abstract

Many climate studies need to generate estimates of a climate variable at a given location based on values from other locations. In this research, a new method for the spatial interpolation of daily maximum surface air temperatures is presented. This new method uses artificial neural networks (ANNs) to generate temperature estimates at 11 locations given information from a lattice of surrounding locations. The out-of-sample performance of the ANNs is evaluated relative to a variety of benchmark methods (spatial average, nearest neighbor, and inverse distance methods). The ANN approach is superior both in terms of predictive accuracy and model encompassing. In 94% of case comparisons, the predictive accuracy of the ANN is superior to the benchmark methods. The ANN approach encompasses the benchmark methods in 77% of case comparisons, while benchmark methods encompass the ANN in only 2%. In light of these results, the potential to use this new method of spatial interpolation to downscale GCM temperature simulations is discussed.