Non-uniform spatial downscaling of climate variables

Abstract

The goal of this study is to present a scalable and robust approach to spatial downscaling of climate variables. We explore the ability of artificial neural networks (ANN) to downscale a climate variable to a given location of interest. We illustrate our proposed method in a downscaling application of monthly mean air temperature and precipitations at twelve stations located across the topographically complex province of British Columbia, Canada. Our method generalizes well to different locations and leads to high downscaling accuracy. The performance of the models is measured based on four statistical metrics, including the coefficient of determination, and the root mean square error.