Managing Water in an Era of Climate Extremes

All it takes is one especially intense storm to turn a peaceful river into a devastating torrent. Snow melting too quickly can force hydroelectric dam managers to release water in a hurry to prevent overflow. With extreme weather events on the rise, water management is as much a question of public safety as it is of energy.

Richard Arsenault, professor at ÉTS and expert in hydrological modelling, is focusing his research in light of this strategic context. His objective: to develop forecasting and decision-support tools to improve hydroelectric reservoir management, mitigate flood risks and anticipate the effects of climate change.

Managing Uncertainty in Real Time

Hydropower relies on the optimal use of reservoirs. Too much water can lead to flooding, too little can compromise hydropower production. To maintain balance, dam managers have to make day-to-day decisions based on weather and water forecasts. However, these forecasts are sometimes inaccurate or too uncertain in the medium term, complicating risk management.

The models developed by Richard Arsenault and his students at the HC3 Laboratory simulate watershed behaviours under different scenarios. They take into account soil moisture, snow cover, precipitation history and other parameters essential to forecasting river flow.

Algorithms to Predict, Decide… and Adapt

For several years now, the team has been using artificial intelligence models capable of learning from time series. These neural networks known as long short-term memory networks (LSTMs), provide better integration of past conditions to estimate future water volumes. This approach represents a paradigm shift in hydrology: models no longer simply reproduce behaviours observed in nature; they now learn to predict them using large quantities of data.

In addition to weather and water forecasting, researchers are also developing decision-support tools using reinforcement learning. These systems propose water management strategies based on expected risks, such as the preventive emptying of a reservoir before a rainfall event. These are decision support tools: they propose possible interventions based on data, while leaving the final decision to humans.

Practical Collaborations

The research carried out by Richard Arsenault and his team is currently being applied in the field. As part of a project with Hydro-Météo—a company that supports municipalities and the Quebec government in flood management—their work led to the modernization of a hydrological forecasting system. This type of tool leads to better planning in the event of ice jams or heavy rainfall.

In the context of climate change, this research is also useful in the longer term, particularly for electricity producers. As hydroelectric power plants are designed to last 80 to 100 years, companies need to plan their investments on the basis of expected water quantities over the coming decades. Hydrological forecasts guide decisions such as turbine replacement or modification, and reassure investors and insurers. 

Applications extend beyond the energy sector. In partnership with the Ministry of Natural Resources and Forestry, researchers are helping forecast the geographical evolution of forests based on groundwater availability. These analyses are used to guide management policies and logging authorizations.

Open-Science, a Shared Database

Acknowledging the importance of sharing data and tools, the team also helped create a database including over 14,000 watersheds across North America. This constantly evolving database is used by numerous researchers to train new artificial intelligence models in hydrology.

Water Management Under Pressure

With climate change, flood and drought events will increase. In light of this, anticipation becomes a condition of survival. And that’s what this applied research aims to make possible: better forecasting, better decision-making, and preventing water management from turning into crisis management.