The ice above the Manson Icefield dropped 148 metres in under 60 days, and there was nobody within hundreds of kilometres to see it happen. Beneath the glacier, a lake holding 4.46 cubic kilometres of water, enough to fill around 1.8 million Olympic swimming pools, had been quietly filling for 14 years before it gave way in late 2020 and emptied through a crack in the bedrock in roughly a month. The flood carved a channel under the ice all the way to the ocean. Satellites caught the whole thing from space, reading the collapse as a sudden sag in the glacier surface. It is the largest subglacial flood recorded anywhere on Earth outside Iceland and Antarctica, and it happened in a region where, until this year, scientists believed almost no subglacial lakes existed at all.
Findings published in The Cryosphere in March 2026 quantify 37 active lakes beneath the glaciers of the Canadian Arctic, 35 of which had never been identified before. Two years ago the confirmed count stood at two, and one of those two turned out not to be a lake at all, just frozen bedrock that had been misread from radar data. The region covers 14 percent of Earth’s glaciers and ice caps by total area, loses 53.6 gigatonnes of ice per year, and ranks second globally for glacier mass loss behind Alaska. Scientists were feeding those loss figures into sea-level projections without accounting for a hidden hydraulic system cycling billions of tonnes of water underneath the ice. The projections are now, at minimum, incomplete.
To find the lakes, the research team stacked more than 23,000 satellite elevation maps acquired between 2011 and 2021 across the entire Canadian Arctic, each one accurate to within about a metre vertically. The physics behind the detection is simple: when a lake fills with water, it pushes the glacier surface up. When it drains, the surface falls. Either movement produces a signal that reads as wildly abnormal against a regional average surface change of less than 1 metre per year. A lake that shifts the ice above it by 10 metres or more in a single year stands out in the data like a sinkhole on a flat road. The team then ran regression models against the stacked readings to pull out the timing, speed, and volume of individual drainage and recharge events, a method precise enough to resolve events that lasted only a few months across a decade of satellite passes.
What came back from that processing was 37 lakes spread across 15 degrees of latitude, from Penny Ice Cap on Baffin Island in the south to the northern tip of Ellesmere Island, 800 kilometres from the North Pole. Their sizes run from 0.3 square kilometres, roughly the footprint of 40 city blocks, to 48.5 square kilometres for the Manson giant. Five of them individually dropped more than 100 metres during their largest recorded drainage events. Greenland’s most active comparable lake managed about 70 metres. Antarctica’s subglacial lakes rarely move the ice surface more than 20 metres. The Canadian Arctic lakes are not outliers on a global scale; they are at the upper end of it, in a region that was previously assumed to have almost none of them.
The drainage events themselves are not slow. Thirty of the 37 lakes follow the same pattern: years of gradual filling, then a rapid collapse. Lakes 4a and 4b on northern Ellesmere Island were captured mid-drain by three separate satellite passes in July 2020. The regression model fitted to those readings puts the total duration at two to three months, the equivalent of a very large river running at full volume before the tap shuts off. The Manson lake took between 30 and 60 days to flush 4.46 cubic kilometres. For comparison, that volume of water is roughly what the Thames carries past London in two full years. It went through a crack under a glacier in the Canadian Arctic in a single month.
What happens to that water once it moves is where the physics gets consequential. Water at the base of a glacier lubricates the contact between ice and bedrock. When a lake fills, it keeps that contact wet and slippery across the area it covers. When it drains, the lubrication vanishes abruptly. Ice can accelerate during a drainage event as the water rushes through subglacial channels, pushing against the bed before it exits. One of the 37 lakes, Lake 17 in the Baffin Island region, recorded ice surface velocities of around 30 metres per year during its 2015 drainage, against a baseline of 0 to 20 metres per year in other years. That acceleration pushes ice mass toward the ocean faster than it otherwise moves, and ice that reaches the ocean melts into sea-level rise. The team did not find clear velocity responses at most of the other lakes, which likely means those drainage events ran through efficient channelised pathways that relieved pressure quickly rather than spreading water widely under the ice. But the potential for acceleration is built into the system, and 37 lakes cycling water repeatedly through that system across a region losing 53.6 gigatonnes of ice per year is a different risk profile than two.
The correlation between lake activity and regional ice loss is the most direct finding in the dataset. In years when the Canadian Arctic loses more ice, more of these lakes drain. The statistical relationship scores at r = minus 0.69, and it strengthens to r = minus 0.74 when the analysis is restricted to the lakes fully enclosed beneath the glacier. Lake 26 makes the mechanism concrete. Through most of the survey period, it drained on a multi-year cycle. In 2019, the regional mass balance flipped from a slight gain of 14 gigatonnes to a loss of 105 gigatonnes, the largest single-year swing in the record. From 2019 onward, Lake 26 drained every year without exception. More meltwater reaching the bed means lakes fill faster, pressure builds faster, and drainage happens more often.
The accounting problem this creates for sea-level science is not trivial. Current satellite-based mass balance methods read the glacier surface height and convert changes in that height to changes in ice volume. When a subglacial lake fills, the surface rises and the models log it as ice gain. When the lake drains, the surface drops and the models log it as ice loss. Neither entry is accurate: the material moving is liquid water, not solid ice, and the numbers feed directly into global sea-level projections. The total water volume shifted by all documented drainage events across the 10-year survey amounts to roughly 1 to 2 percent of the 457 gigatonnes of regional mass change recorded between 2012 and 2021. In a region losing ice at this rate, a 1 to 2 percent systematic misattribution embedded in the baseline data compounds into a meaningful error over the projection timescales that policy decisions actually use.
There are also two structural types of subglacial lake in this dataset that had never been formally categorised before. Terminal subglacial lakes form where two separate glaciers flow into the same valley and press together at a merged terminus. The compression at that junction cracks the ice heavily, opening fracture pathways that let surface meltwater drain down to the bed, where it pools against the valley walls. All 11 terminal lakes in the Canadian Arctic inventory produced at least one sharp drainage event during the survey window. Partial subglacial lakes extend from beneath the ice edge into adjacent open water at the glacier margin, meaning the glacier is effectively floating at one end. Those lakes can refill from rainfall and direct surface runoff, not just from glacier melt, which gives them a recharge rate that does not depend solely on ice temperature. Fifteen of the 37 lakes fall into this category.
No permanent settlements sit near any of these lakes. The direct flood risk to infrastructure is currently low. The concern is not a single catastrophic outburst reaching a populated coastline; it is the cumulative effect of dozens of lakes accelerating ice delivery to the ocean over the coming decades, at volumes and frequencies that current models were not built to include. ICESat-2 passes over its ground tracks every three months and remains the primary active monitoring instrument for these lakes. The SWOT satellite, which tracks surface water changes at hundred-metre resolution, is geometrically limited to targets south of 78 degrees North, which excludes 20 of the 37 lakes from its coverage. The ArcticDEM elevation record used in the inventory ends at 2021, leaving the most recent three years of lake activity unquantified. Lake 7, on northern Ellesmere Island, recorded an anomalous surface drop of approximately 10 metres in the final year of coverage, consistent with the start of a major drainage event. Whether it completed, and where the water went, is not yet in any dataset.
Primary source: Zheng, W., Van Wychen, W., Li, T., and Yamanokuchi, T. (2026). “Active subglacial lakes in the Canadian Arctic identified by multi-annual ice elevation changes.” The Cryosphere, 20, 1699–1714. https://doi.org/10.5194/tc-20-1699-2026
Supporting source (Manson Icefield lake, first documented): Gray, L., Lauzon, B., Copland, L., Van Wychen, W., Dow, C., Kochtitzky, W., and Alley, K.E. (2024). “Tracking the Filling, Outburst Flood and Resulting Subglacial Water Channel From a Large Canadian Arctic Subglacial Lake.” Geophysical Research Letters, 51. https://doi.org/10.1029/2024GL110456
Regional mass loss data cited within the primary source: Zemp, M. et al. (2025). “Community estimate of global glacier mass changes from 2000 to 2023.” Nature. https://doi.org/10.1038/s41586-024-08545-z
