Data-Driven Modelling of Large-Scale Groundwater Dynamics

Reliable prediction of groundwater behaviour is essential for sustainable water resource management, particularly as climate variability intensifies pressure on subsurface reserves. Richards' equation provides a physically rigorous description of water movement through both unsaturated and saturated zones, yet its computational demands have long precluded application at continental scales.

Here we present a novel three-dimensional solver for Richards' equation built on Firedrake, a flexible finite element framework. This approach enables simulation of groundwater dynamics across spatial scales ranging from metres to thousands of kilometres, bridging the gap between local process studies and regional water management.

We demonstrate the solver through a case study of the Lower Murrumbidgee basin in Australia, encompassing approximately 3600 km². The simulation assimilates observational data from the Bureau of Meteorology, including basin stratigraphy comprising three layers of varying depth, annual rainfall estimates, and present-day water table depths. Our results capture decadal-scale flow dynamics at spatial resolutions previously unattainable for basins of this size. This work addresses critical limitations in current operational approaches, which do not fully couple unsaturated and saturated flow processes, and offers a pathway toward improved hydrological forecasting and water resource assessment at regional to continental scales.