LAUSR

Management of groundwater resources needs continuous and efficient monitoring networks. Sparsity of in situ measurements both spatially and temporally creates hindrance in framing groundwater management policies. Remotely sensed data can be a possible alternative. GRACE satellites can trace groundwater changes globally. Moreover, gridded rainfall (RF) and soil moisture (SM) data can shed some light on the hydrologic system. The present study attempts to use GRACE, RF and SM data at a local scale to predict groundwater level. Ground referencing of satellite data were done by using three machine learning techniques- Support Vector Regression (SVR), Random Forest Method (RFM) and Gradient Boosting Mechanism (GBM). The performance of the developed methodology was tested on a part of the Indo-Gangetic basin. The analyses were carried out for nine GRACE pixels to identify relationship between individual well measurements and satellite-derived data. These nine pixels are classified on the basis of presence or absence of hydrological features. Pixels with the presence of perennial streams showed reasonably good results. However, pixels with wells located mostly near the stream gave relatively poorer predictions. These results help in identifying wells which can reasonably represent the regional shallow groundwater dynamics.