Construction of a climate downscaling neural network

Abstract

Climate studies call for more localized climate scenarios. The scarcity of high-resolution scenarios is a problem, however. To address this problem, we aimed to construct an artificial neural network (ANN) that performs climate downscaling for an area in Luzon, Philippines. In this paper, we present the constructed ANN and an analysis of its performance. Tests done show that our ANN can perform spatial downscaling of precipitation. However, the presence of low correlation, between actual observed values and ANN output, for certain locations, tells us that there is a need to improve the present ANN.