Spatial downscaling of satellite rainfall retrieval (TMPA 3B43) using best subset regression model in the Cauvery River Delta region, India

Accuracy of the results of climate change and climate variability studies depends solely on the reliability of rainfall data. Rain gauges are traditional sources, generally associated with many practical flaws and missing qualities. In recent times, satellite data products are recognized worldwide for their spatial and temporal coverage over any region. Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) 3B43 is a rainfall dataset with 0.25° × 0.25° spatial resolution used in many research works. In this investigation, a multivariable best subset regression (BSR) downscaling model was developed to produce 1-Km high-resolution rainfall data over Cauvery River Delta (CRD) region of Tamil Nadu state, India. BSR model performed well than widely used geographically weighted regression (GWR) model in producing high spatial resolution and data accuracy. BSR model decreased root mean square error (RMSE) from 37.96 to 22.24mm for monthly scale with significant null BIAS (at 95% confidence interval) on comparing with TMPA 3B43 data. The seasonal estimates through BSR showed a good relation for northeast monsoon and winter seasons with coefficient of determination (R2) as 0.93 and 0.94. BSR model used land use land cover (LULC) as a significant factor along with environmental and topographical variables. Research findings of this investigation showed that BSR model improved the accuracy and spatial resolution of TMPA 3B43 products considerably for CRD region. © 2021, Saudi Society for Geosciences.