Spatio-temporal modelling of groundwater recharge and low water levels due to climate change using a physics-informed graphical neural network
This project is developing a physics-informed artificial intelligence model to better anticipate changes in groundwater recharge in southern Quebec in the context of climate change.
Project details
Principal(s) investigator(s)
Context
Groundwater recharge and low water conditions play a central role in water availability and ecosystem resilience. In the context of climate change, changing precipitation patterns, reduced snow cover, unstable freeze-thaw cycles and an increase in extreme events are modifying surface and underground hydrological processes. These changes make it more difficult to anticipate periods of water stress and accentuate the uncertainties associated with the modelling tools currently in use. To support planning and adaptation, Quebec depends on the Hydroclimatic Atlas of Southern Québec, which is based on regional simulations of recharge and low water levels. However, new approaches are required to improve the representation of surface water–groundwater interactions and the spatio-temporal variability of hydro(geo)logical processes, and the consideration of climate-related uncertainties. The aim of this project is to respond to this challenge by developing an innovative modelling framework based on physics-informed artificial intelligence in order to enhance the scientific robustness and forecasting capability of hydrological simulations in a changing climate.
This project is part of the QClim’Eau initiative, a collaboration between the Ministry of the Environment, Climate Change, Wildlife, and Parks (MELCCFP) and Ouranos.
Objective(s)
The aim of the project is to develop and operationalize a physics-informed artificial intelligence model to simulate groundwater recharge and low water conditions in southern Quebec in the context of climate change, while assessing uncertainty, in support of the Hydroclimatic Atlas of Southern Québec.
Methodology
The proposed approach is based on the development and application of an innovative modelling framework combining artificial intelligence and hydrological principles. The project’s methodology has three complementary components:
The development of a physics-informed spatio-temporal model with hydrological connectivity between spatial units, daily temporal dynamics (e.g. snow, freeze-thaw, low water levels), and mass and energy conservation constraints
Training, calibration and validation of the model using climate, hydrological and geospatial data representative of southern Quebec, followed by regionalization to allow the model to be applied to ungauged basins
Assessment of the robustness of simulations by quantifying uncertainties and performing sensitivity analyses under different climate scenarios in order to produce interpretable recharge and low water indicators that are useful in decision-making
Expected results
Development and operationalization of a physics-informed artificial intelligence model to simulate groundwater recharge and low water conditions on a regional scale
Improved representation of surface water–groundwater interactions and daily hydrological dynamics in the context of climate change
Production of spatialized recharge and low water indicators suitable for regional planning and water management
Quantification and propagation of uncertainties related to climate forcing, land use and subsurface properties
Benefits for adaptation
Benefits for adaptation
Improving our ability to simulate groundwater recharge and low water conditions will enable us to better anticipate the impacts of climate change on groundwater resources and the ecosystems associated with them.
This will also support decision-making on long-term adaptive water resource management.
Funding
Other participants
Jan Adamowski, Université McGill
Vincent Cloutier, UQAT
Eric Rosa, UQAT
718000
