Downscaling of precipitation in Mahanadi Basin, India using Support Vector Machine, K-Nearest Neighbor and Hybrid of Support Vector Machine with K-Nearest Neighbor

Abstract

The climate impact studies in hydrology often rely on climate change information at fine spatial resolution. Downscaling is a practice for obtaining local-scale hydrological variables from regional-scale atmospheric data that are provided by General Circulation Models. Among two downscaling methods, Statistical Downscaling is taken into account, as it offers less computational work as compared to Dynamic Downscaling and also provides us with a platform to use ensemble GCM outputs. In the present study, a Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Hybrid of Support Vector Machine (SVM) with K-Nearest Neighbor (KNN) approaches are proposed for Statistical Downscaling of precipitation at monthly time scale. To reduce the dimensionality of the dataset, the Principal Component Analysis (PCA) is also performed. The CanCM4 simulations are run through the calibrated and validated SVM, KNN and hybrid of SVM with KNN downscaling models to obtain future projections of precipitation values. A comparison is made between the models in this study.