Improving Rainfall Forecasting in San Diego County: A Hybrid Machine Learning and Bias-Correction Downscaling Framework

Downscaling climate data bridges the gap between coarse-resolution global climate models (GCMs) and finer-scale regional projections. Statistical downscaling, a popular technique, uses global climate data to achieve localized insights. This study evaluates machine learning algorithms for downscaling and bias correction, focusing on rainfall prediction in San Diego County. We incorporated data from the four nearest latitude and longitude points around gauges and combined it with local neighborhood precipitation data to improve granularity. A hybrid approach was proposed, integrating machine learning with bias correction to enhance the accuracy and reliability of downscaled precipitation data. Random forest and support vector regression were utilized for downscaling, while Linear regression, probability density function adjustment, and quantile mapping were applied for bias correction. Among these, the random forest with linear regression scheme performed best, demonstrating the potential of combining machine learning and statistical techniques to refine climate projections at regional and local scales.