How disappearing lakes and AI let scientists map Arctic permafrost thaw almost in real time

Why lakes matter for watching permafrost

The team combined historical aerial photos (dating back to the late 1940s), satellite imagery available since the 1970s, and automated image analysis to map lakes across large Arctic regions. Using machine learning to process new imagery quickly, the group behind the Permafrost Discovery Gateway now maps and monitors over 4 million lakes across the Arctic and about 70 million thermokarst ponds in the Alaska tundra.

Why abrupt lake drainage is significant

A lake that forms over centuries can be drained in hours when permafrost thaw creates a drainage channel. Such abrupt disappearance is a strong indicator that ice-rich permafrost has thawed and the ground has collapsed or changed hydraulically. Detecting drained basins and new ponds across seasons and years helps pinpoint active thaw areas.

Recent signals and regional examples

Alaska are highlighted as hot spots. The region has lost many large lakes that existed for thousands of years. In 2018, nearly 200 of about 4,600 local lakes lost more than a quarter of their area after an unusually warm winter. Even though winter 2025–26 was relatively cold for Alaska overall, that region continued to lose lake area, and researchers identified about 30 additional lakes that drained in June and July 2026. These events mirror what happens across hundreds of lakes each summer in many Arctic zones.

Practical uses of near-real-time maps

Up-to-date mapping gives communities, planners and researchers timely information about where ground is ice-rich and likely to subside if it thaws. Maps can indicate areas that are risky for building, and areas where frozen ground is less ice-rich and therefore comparatively safer for infrastructure. For researchers, the continuous stream of data shortens the gap between events on the landscape and scientific analysis.

The combination of long-term imagery archives and automated analysis allows monitoring of lake appearance, disappearance and change across large Arctic areas with a cadence that was not possible when studies relied only on field campaigns and publications released years after observations. This approach provides operationally useful maps rather than only retrospective scientific descriptions.

This work is described by Anna Liljedahl (University of Alaska Fairbanks) and Ingmar Nitze (Alfred Wegener Institute). The monitoring effort is associated with the Permafrost Discovery Gateway and leverages satellite archives and machine learning to observe lake and permafrost dynamics across the Arctic.