[A-HW17] Near Surface Investigation and Modeling for Groundwater Resources Assessment and Conservation

[AHW17-05] Transforming Coarse Remote Sensing Observations for Decision-Scale Groundwater Management: GRACE Downscaling via Machine Learning in a data-scarce Semi-Arid Aquifer

*Abba Ibrahim^1,5, Aimrun Wayayok^1,2,3, Helmi Zulhaidi Mohd Shafri⁴, Noorellimia Mat Toridi¹ (1.Department of Biological and Agricultural Engineering, Faculty of Engineering, Universiti Putra Malaysia. 43400 UPM Serdang, Selangor DE, Malaysia, 2.SMART Farming Technology Research Center (SFTRC), Faculty of Engineering, Universiti Putra Malaysia, 43400 UPM Serdang, Selangor DE, Malaysia., 3.International Institute of Aquaculture and Aquatic Sciences (I-AQUAS), Universiti Putra Malaysia, Mile 7, Kemang Rd. 6. Kemang Bay, Si Rusa, Port Dickson, Negeri Sembilan 71050, Malaysia., 4.Department of Civil Engineering, Faculty of Engineering, Universiti Putra Malaysia. 43400 UPM Serdang, Selangor DE, Malaysia, 5.Department of Agricultural and Environmental Engineering, Faculty of Engineering, Bayero University, Kano, Nigeria.)

Keywords:Spatial resolution, Data fusion, Groundwater, Hydrology, Decision Support

The coarse spatial resolution of the Gravity Recovery and Climate Experiment (GRACE) satellite data limits its utility for localised hydrological analysis. This is problematic for many regions, particularly in the semi-arid aquifer systems of sub-Saharan Africa which rely on groundwater; however, data on water storage dynamics are lacking. Downscaling methods are needed to integrate GRACE's insights of GRACE with the factors driving local variability. This study presents a novel downscaling approach applying machine learning within the GRACE Downscaler framework. The integration of GRACE Terrestrial Water Storage (TWS) anomalies with climate and precipitation data established relationships for a boosted resolution. The semi-arid Hadejia Jama’are River Basin (HJRB) aquifer in Nigeria, with limited hydrological observation history yet growing abstraction reliance, underscores improving water storage knowledge constraints. This study aims to downscale the GRACE TWS to 1 km resolution estimates using total TWS, precipitation, and auxiliary sources to reveal the local variability. This study uniquely applies the GRACE Downscaler workflow to the HJRB. Following input data generation and training, advanced XGBoost and Random Forest machine learning algorithms downscale the GRACE TWS data by linking anomalies to corresponding rainfall and climate drivers. The downscaled high-resolution outputs were statistically evaluated and hydrologically interpreted; the 1 km estimates provide enhanced GRACE data resolution for further processing to study groundwater storage anomalies in the region. The successful framework adoption and model performances showcase the potential of machine learning for augmenting GRACE data in climate-vulnerable regions. As the pioneering GRACE downscaling for the region, data-driven products constitute an important monitoring advancement to inform local decision-makers.