Statistical Learning-Based Spatial Downscaling Models for Precipitation Distribution

Abstract

The downscaling technique produces high spatial resolution precipitation distribution in order to analyze impacts of climate change in data-scarce regions or local scales. In this study, based on three statistical learning algorithms, such as support vector machine (SVM), random forest regression (RF), and gradient boosting regressor (GBR), we proposed an efficient downscaling approach to produce high spatial resolution precipitation. In order to demonstrate efficiency and accuracy of our models over traditional multilinear regression (MLR) downscaling models, we did a downscaling analysis for daily observed precipitation data from 34 monitoring sites in Bangladesh. Validation revealed that $R^2$ of GBR could reach 0.98, compared with RF (0.94), SVM (0.88), and multilinear regression (MLR) (0.69) models, so the GBR-based downscaling model had the best performance among all four downscaling models. We suggest that the GBR-based downscaling models should be used to replace traditional MLR downscaling models to produce a more accurate map of high-resolution precipitation for flood disaster management, drought forecasting, and long-term planning of land and water resources.