Machine learning models can forecast monthly water levels across the Great Lakes with noticeably less error than conventional methods, according to new research that also explains which factors drive those changes. The work points to a practical tool for communities, shipping operators and water managers who have been caught off guard by the lakes' sharp swings.

The swings have been dramatic. Lake Michigan fell to a record low in January 2013, then reached a record high in summer 2020, when flooding and shoreline erosion hit communities along its shores. Water levels moved by close to two metres over that period, and Lake Superior, Lake Erie and Lake Ontario showed similar reversals. Those levels matter for harbours, shorelines, drinking-water intakes and wetlands, so forecasts that arrive too late or miss by a wide margin carry real costs.

The researchers drew on 40 years of data, from 1982 to 2022, and trained eight machine learning algorithms on monthly levels for four of the lakes. The models used nine variables, including air temperature and inflow, and looked at how conditions from up to six months earlier affected the present. The best-performing models reduced forecasting error to about 12 centimetres, compared with 14 to 21 centimetres for the comparison approaches.

Two analytical tools were used to make the models more transparent. One, known as SHAP, ranks how much each variable contributes to monthly changes in lake level. The other, VARS, tests how time lags affect results by altering inputs and measuring the response. Together they showed that the influence of past conditions grows over three to four months, and that winter snowfall helps forecast the following summer's levels.

The study also identified a clear limitation. Forecast errors for Lake Ontario were more than 50 per cent higher than for the other lakes, which the researchers traced to decisions about how water is released through the dams at the Moses-Saunders complex. Monthly averages do not capture those operating rules, so the models struggle where human control of outflow is strong. Future work will look at that gap, and the team is also extending the approach to groundwater recharge in southern Québec in partnership with Ouranos, a Québec climate research consortium.

We rewrote this story on 4 November 2026 after an editorial review.

Credit. Based on reporting by Ehsan Raei, Jan Franklin Adamowski and Rahim Barzegar for The Conversation Science. Is this your work? Write to us.

314Views

Kelsey Sawyer

Author

Society Reporter

Kelsey Sawyer covers public affairs, politics, business, culture and daily news for Science Official. The role focuses on verification, context, and clear explanations for readers.