Paleo-Europe

The lost lake of Polesie: a model of the Late-Glacial lake in southern Belarus

45 seconds, from 17,000 years ago to today: a lake in the Pripyat lowland whose level is set by the land itself, the breach of the Mazyr (Mozyr) Ridge, the floods that still come back, and the drainage and dikes that helped shorten them. It is a model reconstruction, and whether such a lake existed at all is debated; the limits are listed below.

What you are seeing

The model lake holds 6,060–6,531 km² of open water around 1,385–1,575 islands (every patch of dry ground it encloses, down to a single cell of the terrain grid; the film draws the larger ones, 407–472 of them). Before the breach, at every modelled moment, it is more than twice the size of Luxembourg: Luxembourg covers 2,586 km² (total area), the lake at least 2.3 times as much. It lies entirely within today's Belarus, north of 51.74° N, in the Pripyat lowland of Pinsk, Turaŭ and Mazyr.

Its level, 147–159 m above today's sea level, is not a dated shoreline. It is the highest level the landscape of each moment can hold: the height of the lowest point of the lake's rim, where it would spill over toward the Dnieper. The ground itself moved: in the ICE-6G_C model of the ice sheets and the Earth beneath them, Polesie stood 16–27 m higher than today, so the rim, and with it the level, rose from 147 m to 159 m about 14,000 years ago and then fell to 158 m.

About 13,300 years ago (a scenario; see the limits) the Mazyr Ridge gives way at Yuravichy (Yurovichi), and the lake drains down the Pripyat valley toward the Dnieper. The scenario date lies inside the published estimates. The film shows this twice, the second time close up at the gate. The drainage is a model of the level falling, from 158 m to 143 m, until only 5 % of its area is left; the route follows the relief. It is not a flood simulation.

On the ridge above the gate, people camped at Yuravichy about 30,600 years ago, long before the lake. Animal remains from the camp are dated to 26,470 ± 420 (LU-125) and 25,660 +160/−150 (GrA-38919) radiocarbon years; the year shown is the middle of the calibrated range of LU-125 (IntCal20). The camp is 1.9 km from the gate and stays dry in every frame; at the closest, the shore is 172 m away.

After the lake

The caption says “After the lake, the water keeps coming back”. From the moment the lake has drained, the film draws today's water over the lowland as satellites saw it from 1984 to 2024, in JRC Global Surface Water v1.5 (2024): how often Landsat saw water at each spot (Pekel et al. 2016). It is not a reconstruction of the past: while the year counter runs on to today, the water, the hatch and the marshes on screen are today's data. Only water connected to the rivers is drawn; ponds and reservoirs away from the rivers are left out. The film sweeps the threshold of that frequency from permanent water, seen in at least 75 % of the observations, to any water seen at all, in at least 1 %, and back, every 5 seconds. The pulse is not a season or a year: at each step it shows where water was seen at least that often. Of the river-connected water in the area measured, 24–31.5° E × 51–53.5° N (the Pripyat, the Dnieper and the other rivers), 564 km² is permanent and 2,198 km² comes and goes; on the old lake floor, water that comes and goes covers 7 % (491 km²).

The satellite probably shows less flooding than there was; that is our reading, not the dataset's. Water under trees and other vegetation cannot be seen by an optical satellite (a European Commission factsheet on the dataset), and wet seasons have fewer cloud-free Landsat scenes (the JRC users' guide; the month-by-month averaging of the record is meant to make up for it). River-side fish ponds and reservoirs can pass the test for water connected to the rivers. On the grid used here, a 3-arc-second cell counts as wet if any 30 m pixel overlapping it is wet: areas are upper bounds (of what the satellite recorded, not of the floods themselves).

Hatched: a model of the floodplain, not a measurement. It marks land up to 5 m above the Pripyat along the paths the water flows (the height above the river, computed with WhiteboxTools on FABDEM at 3 arc-seconds with its depressions filled). The height comes from «Блакітная кніга Беларусі» (1994): the spring rise of the Pripyat reaches up to 5 m in its middle and lower course, and up to 7 m above the lowest low water in the narrow floodplain near Mazyr; the same entry gives a floodplain 16–18 km wide at the mouths of the Pina and the Haryn (Goryn). The hatch covers 43 % of the old lake floor (3,011 of 7,039 km²: the model lake at its largest, with its islands counted as floor). It holds most of the seasonal water the satellite sees: of that water within 9 km of the Pripyat, half of the widest floodplain, 93 % lies inside the hatch.

At the end the film says “Since the 1960s, drainage and dikes have helped shorten the floods”. The largest-scale drainage of Polesie was carried out in 1966–1994. Rusetskii (2010) gives 1.6 million ha of drained farmland in Belarusian Polesie, and about 2 million ha of former mires drained in all, counting peat workings, dams and canals. Comparing 1945–1970 with 1980–2009 on the Kachanovichi–Lyubansky most stretch of the Pripyat (the early levels at Lyubansky most are reconstructed from the Pinsk gauge), he finds that the lowest floodplain there, where the water first leaves the channel, was flooded 18–58 days less in high-water springs and up to 64 days less in low-water ones; in the driest springs it is not flooded at all. He names the drainage of the catchment as one of the factors, and models of the river give shorter flooding for the dikes alone. Hence “helped shorten”: the floods became shorter; they did not stop.

The lines of the last seconds are ditches, drains and canals as mapped in OpenStreetMap on 2026-09-27 (extracts by Geofabrik): 203,794 lines, 108,540 km in all (whole lines, some running beyond the map), 8,911 km of them on the old lake floor. The canals include navigation canals, not only drainage canals. The map shows what has been mapped, not when it was dug or whether it still works.

The marshes are drawn as OpenStreetMap maps them today (areas over 8 km²); their past extent is not modelled. Only 14 % of the mapped marshes in the area measured, 24–30.7° E × 51.2–53.35° N, lie inside the model lake, although its former bed is 2.4 times as marshy as the rest of that area.

From the moment the lake has drained to the end, a dotted line keeps the model lake at its largest, 16,950 years ago, 6,531 km² of open water.

The vegetation line follows regional pollen records; it is not a map:

How it was made

Terrain: FABDEM, a bare-earth model of the land with forests and buildings removed, at a lighter resolution for the wide shots and at its native resolution in the close-up. Earth movement: the solid-surface displacement of ICE-6G_C (VM5a), interpolated in time. For each of 151 moments, every 25 years from 17,000 years ago down to the breach, the model raises the level step by step until the water would escape; the last level that holds is the stage. A model barrier restores the Mazyr Ridge where the river cut through it; every stage is set by a col inside the data, not by the edge of the map, and this is checked. The breach date is one published radiocarbon date, calibrated with IntCal20. The outflow route is traced downhill on the terrain of the breach (WhiteboxTools fill and steepest descent) and checked against today's Pripyat. The satellite water is the occurrence layer of JRC Global Surface Water v1.5 (2024) on the terrain grid, and the hatch is the height above the Pripyat along flow paths on the depression-filled terrain (WhiteboxTools). The ditches are OpenStreetMap ways tagged as ditches, drains or canals; in the final frame there are more mapped ditches per km² on the Belarusian side (0.96 km) than on the Ukrainian side (0.38 km), but OpenStreetMap's completeness differs from country to country, so this is not a measure of drainage. Every number in the film and on this page is computed from the model or copied from a checked source, not typed.

Limits of the model

The interactive viewer is in private preview.

Video licence: CC BY-NC-SA 4.0, because it is derived from FABDEM. Share it with a link to this page; no commercial use. Satellite water: Source: EC JRC/Google. Marshes and ditches: © OpenStreetMap contributors, ODbL; the ditches from extracts by Geofabrik.

Sources

Model: 151 stages · rivers 202609190655 · borders Natural Earth 5.1.2