Northern Lakes: How Their Glacial Reflection Cools the Hemisphere and Impacts Climate

When temperatures drop below zero in subarctic regions, over a third of snow-covered surfaces come from small lakes whose shiny ice reflects up to 80% of solar radiation back into space. This previously underestimated phenomenon acts like a natural thermostat, mitigating Northern Hemisphere warming during winter months. Researchers found that due to this glacial reflection, lakes contribute to a more positive global radiative balance than surrounding land, which could alter long-term climate predictions.

What the study reveals: an amplified cooling role of northern lakes

Analyzing albedo variations across the entire Nordic basin, scientists found that snow- or ice-covered lakes increase the regional average albedo by 0.02 to 0.04, comparable to that of large areas of permanent snow. This albedo increase translates into a notable reduction in net solar energy absorption, creating local cooling that propagates regionally through atmospheric circulations. Climate models incorporating this mechanism show a seasonal warming attenuation of several tenths of a degree, a modest figure but a significant portion of the hemisphere's energy budget.

Their impact is even more crucial as lakes are often scattered in areas where global models underrepresent water surfaces. Neglecting these small expanses can lead to overestimating heat absorption and thus future warming. The study thus emphasizes that each glacial reflection matters, especially in regions with rapid snow-ice cover changes and high prediction uncertainties.

How the study was conducted: Neural networks, satellite data, and ECMWF models

Researchers combined multiple data sources to quantify lake effects. They first exploited images from the Sentinel-2 mission of the Copernicus program, offering sufficient spatial resolution to distinguish small water bodies even when snow-covered. These satellite data were coupled with the ERA5 reanalysis from ECMWF, providing detailed fields of temperature, humidity, and radiative flux. To integrate these into a predictive framework, they trained a deep neural network using machine learning to correlate surface observations with atmospheric responses.

Among the AI models used, teams compared the performances of GraphCast and Pangu-Weather, two forecasting models based on graph and convolutional neural network architectures. Both systems were fed with the same atmospheric data sets and evaluated on their ability to reproduce observed atmospheric conditions.