Venkatesh Panchariya, Ph.D. Candidate 39th cycle, University of Trento, DICAM

Of the earth’s total freshwater resources, groundwater comprises around 30% after the frozen polar freshwater resources (around 68%). Groundwater is a major component of our hydrological cycle, and recent studies have proved its significant contribution (global average of around 55%) in sustaining our rivers during dry periods as baseflow. Thus, groundwater becomes vital in making our rivers perennial and sustaining riverine ecosystems during drought years. It also acts as a buffer during deficit soil moisture conditions, i.e. during agricultural drought years. For irrigation and domestic water needs groundwater has become an important source in Italy.

Our attempts to estimate water resources (both surface and sub-surface groundwater resources) are marred with uncertainty (low confidence in estimates compared to the real observations). This uncertainty stems from shortcomings in process-based modeling and inherent complexity (heterogeneity) in our natural environment. This heterogeneity gives rise to the non-linear behavior of various bio-geochemical and physical processes in our environment. Recent advances in data-driven approaches such as machine learning/artificial intelligence (ML/AI) have shown their capability to mimic this non-linear behavior of our environment. However, there lies the risk of ML models drifting from the physical reality in simulating our environment under non-stationary conditions (such as climate change). A hybrid modeling approach attempts to combine the ‘learning’ capabilities of ML models with the process-based models rooted in
physical principles. Thus, it becomes a more robust approach to improving our modeling
capabilities.

The purpose of this study is to follow a hybrid approach to investigate groundwater resources in north-east Italy. Our research group has collated a long period of historical groundwater depth observations for this region of Italy. Using geostatistical techniques and the water-budget approach, we aim to quantify the groundwater storage fluctuations through time. The water budget approach involves estimating recharge and lateral contribution from surrounding mountain aquifers to the plain aquifers, which have been hypothesised to support surface and subsurface water resources in this region during drought years. We intend to use advanced ML
techniques to decipher non-linear relationships between various drivers and groundwater storage that can help predict the future evolution of this critical resource under climate change.

Observations of water resources from satellite missions such as GRACE can be used to constrain and validate regional water budgets. This work can facilitate the sustainable use of groundwater which has become one of the critical natural resources due to increasing human consumption and climate change.