Deep learning–based downscaling of summer monsoon rainfall data over Indian region

Downscaling is necessary to generate high-resolution observation data validate the climate model forecast or monitor rainfall at micro-regional level operationally. Available observations generated by automated weather stations meteorological observatories are often limited in spatial resolution resulting misrepresentation absence of information these levels. Dynamical and statistical downscaling models used get gridded over larger domains. As variability dependent on complex spatio-temporal process leading non-linear chaotic variations, no single method can be considered efficient enough. In domains dominated topographies, quasi-periodicities, non-linearities, deep learning (DL)–based methods provide an solution for regional forecasting real-time high resolutions. We employed three learning-based algorithms derived from super-resolution convolutional neural network (SRCNN) this work. Summer monsoon season India Meteorological Department (IMD) tropical measuring mission (TRMM) set were downscaled up 4 times higher using methods. High-resolution better results than linear interpolation resolution. Among algorithms, namely, SRCNN, stacked DeepSD, here, best distribution amplitude minimum root-mean-square error produced DeepSD-based downscaling. Hence, use DeepSD algorithm advocated future use. found that discontinuity intensity patterns main obstacle precipitation. Furthermore, we applied post-processing, particular, ERA5 reanalysis data. Downscaled show a much covariance temporal variance when compared with observation. This study first step towards developing Indian summer