A Comprehensive Investigation into the Accuracy of Soft Computing Tools for Downscaling Precipitation Data Extracted from Climate Models

This study will provide the audience with an understanding of the capabilities of soft tools such as Artificial Neural Networks (ANN), Support Vector Regression (SVR), Model Trees (MT), and Multi-Gene Genetic Programming (MGGP) as a statistical downscaling tool. Many projects are underway around the world to downscale the data from Global Climate Models (GCM). The majority of the statistical tools have a lengthy downscaling pipeline to follow. To improve its accuracy, the GCM data is re-gridded according to the grid points of the observed data, standardized, and, sometimes, bias-removal is required. The current work suggests that future precipitation can be predicted by using precipitation data from the nearest four grid points as input to soft tools and observed precipitation as output. This research aims to estimate precipitation trends in the near future (2021-2050), using 5 GCMs, for Pune, in the state of Maharashtra, India. The findings indicate that each one of the soft tools can model the precipitation with excellent accuracy as compared to the traditional method of Distribution Based Scaling (DBS). The results show that ANN models appear to give the best results, followed by MT, then MGGP, and finally SVR. This work is one of a kind in that it provides insights into the changing monsoon season in Pune. The anticipated average precipitation levels depict a rise of 300–500% in January, along with increases of 200-300% in February and March, and a 100-150% increase for April and December. In contrast, rainfall appears to be decreasing by 20-30% between June and September.