Application of machine learning models in downscaling GRACE derived groundwater storage changes for different hydro-geologic basins of India

Abstract:

The advent of Gravity Recovery and Climate Experiment (GRACE) has opened the doors for remote monitoring of gravitational changes and its derivatives like terrestrial water storage anomalies (TWSA) across the globe. In spite of its unprecedented measurement accuracy (1 μGal), GRACE applications have received less attention among the stakeholders and water managers of India due to poor spatial (~ 110 km) and temporal (1 month) representation. Statistical models with varying complexity are commonly employed to downscale the GRACE datasets for use with local to regional scale applications. This study presents two commonly employed machine learning (ML) models, viz., multi linear regression (MLR) and random forest (RF) in spatially downscaling (from 1^o to 0.25^o) the GRACE-derived TWSA by establishing a correlation with various land surface and hydro-climatic variables. Applicability of the proposed methods was tested on four contrasting hydro-geologic basins of India. For each basin, the significant predictor variables were considered to establish linear (using MLR) and non-linear (using RF) relations. The downscaled TWSA was further converted into groundwater storage anomalies (GWSA) by removing the contributing hydrological fluxes derived from the best performing land surface model. Seasonal groundwater levels observed in 236 wells during 2006-15 were used for method validation and accuracy assessment. We observed a close match between GRACE derived groundwater levels and the measurements for three out of the four basins (r= 0.40 to 0.92, RMSE = 3.6 to 10.5 cm). The mean depletion in GRACE-derived GWSA was ranged from -0.17±0.01 cm yr^-1 (AL basin) to 0.015±0.01 cm yr^-1 (BS basin). Our results conclude that, the predictor variables to downscale TWSA are to be considered cautiously based on the hydro-geologic, topographic, and meteorological characteristics of the basin.