{
  "id": 12452456,
  "title": "How AI could help predict and explain water levels in the Great Lakes",
  "url": "https://urgent.news/2026/10/06/how-ai-could-help-predict-and-explain-water-levels-in-the-great-lakes",
  "topic": "science",
  "section": "Science",
  "published": "2026-10-06T19:10:01.000Z",
  "source": {
    "name": "Phys.org",
    "slug": "phys-org",
    "url": "https://phys.org/news/2026-10-ai-great-lakes.html"
  },
  "original_language": "en",
  "account": "In January 2013, Lake Michigan hit its lowest water level on record, while seven years later in summer 2020, it experienced widespread flooding with eroded shorelines and closed roads. The water level difference was nearly two meters (6.6 feet), with Lakes Superior, Erie, and Ontario also experiencing similar reversals within a few months. However, predicting these fluctuations has long been a challenge in the field of hydrology.\n\nResearch published in Science of the Total Environment explored the use of artificial intelligence (AI) to predict and explain water level variations in the Great Lakes. Traditional methods treat the hydrologic balance like a bank account, adding up deposits and subtracting withdrawals to determine the water level. Yet, this approach requires extensive calibration and fails to capture unusual climate variations.\n\nMachine learning, on the other hand, fed 40 years of data from Lakes Superior, Michigan, Erie, and Ontario, along with nine variables including air temperature and inflow rates. The model was trained on monthly water levels from 1982 to 2022, allowing it to link the water level in June to snowfall in January. Incorporating SHapley Additive exPlanations (SHAP) values and variogram analysis of response surfaces (VARS), the study identified the variables responsible for water levels and their time lags.\n\nThe results revealed that large lakes like Superior and Michigan, fed by snowmelt, depend on runoff and outflow, while shallow Lake Erie relies on inflow from upstream. Lake Ontario's water level is significantly affected by evaporation. Interestingly, the influence of variables on water levels increases after three or four months, unlike in rivers where the impact of a downpour becomes invisible after a few days. This inertia presents an opportunity for seasonal forecasting, as part of what will determine next summer's water level has already fallen in the form of snow.\n\nHowever, Lake Ontario posed a challenge, with an error more than 50% higher than expected due to human decisions at the Moses-Saunders Power Dam. The model struggled to account for the sequence of human decisions affecting the dam's operations, leading to inaccurate results when water levels were altered artificially. To address this issue, providing operational rules to the model can improve its accuracy, ensuring algorithms make informed decisions based on complete data.",
  "summary": "In January 2013, Lake Michigan reached its lowest water level on record. Seven years later, in the summer of 2020, it broke the opposite record with widespread flooding, eroded shorelines and closed lakeside roads.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}