Using more than 470,000 satellite images, researchers have mapped Europe’s open wetlands in unprecedented detail. Up to one third of the mapped area lies in patches smaller than 25 hectares, around one fifth overlaps with or lies close to farming or urban development, and such disturbance may be associated with up to 4.9 billion tonnes less carbon storage, expressed in CO₂-equivalents.
Wetlands rarely announce themselves. They may be a narrow strip beside a field, a waterlogged hollow or a patch of marsh squeezed between roads and farmland. Many are so small that conventional maps merge them into the surrounding landscape – leaving them out of the evidence used to plan protection and restoration.
Gyula Mate Kovács, a researcher at the Department of Geosciences and Natural Resource Management and the Global Wetland Center at the University of Copenhagen, led a team that used 10-metre satellite imagery and machine learning to map six types of open wetland in 38 European countries.
The researchers estimate that 27–33% of the mapped wetland area lies in patches smaller than 25 hectares, including 7–11% in fragments smaller than a single hectare.
“It is much like moving from a blurred image to a high-resolution screen. Suddenly, you see details that were there all along. We realised that Europe contains far more small wetland patches than the coarser maps had shown,” explains Kovács.
The team then compared the map with agricultural and urban land use. This did not measure ecological condition directly but showed where wetlands overlap with or lie close to likely sources of pressure – and why the same restoration target would create very different tasks from one country to another.
Europe drained its wetlands – now it must decide what to restore
Wetlands form where water remains in or near the soil long enough to shape vegetation, chemistry and biological life. They include peat bogs, inland marshes, salt marshes and tidal flats. Globally, they cover an estimated 5–8% of the land surface but contain 20–30% of the world’s soil carbon.
“A wetland is not simply a wet piece of land. Its hydrology determines whether organic material accumulates or breaks down. When a peatland is drained, oxygen enters the soil and carbon that may have remained stored for centuries can begin to escape,” notes Kovács.
Europe has altered these ecosystems more extensively than most regions. Since 1700, the continent has lost more than half of its wetlands on average, while several countries have lost more than three quarters.
“Europe has spent centuries draining wet landscapes for agriculture and forestry and turning them into land that could be cultivated or planted,” Kovács says.
The European Union (EU) Nature Restoration Law is intended to reverse part of this loss. It requires member states to put restoration measures in place for at least 30% of the area of listed habitat types that is not in good condition by 2030, rising to 60% by 2040 and 90% by 2050.
“Setting a percentage is the easy part. The harder question is which areas should count and whether countries are restoring the places where it will make the greatest difference,” Kovács says.
A common map required a common definition of a wetland
Comparable information has been difficult to obtain because countries and mapping systems define wetlands differently. Even peatlands may be classified using different minimum peat depths, making national inventories hard to combine.
“If definitions change across countries, a model may learn to recognise different things under the same name. We needed categories that were consistent across Europe and linked to the legal framework,” Kovács explains.
The aim was therefore to create a shared baseline that could distinguish wetland types, identify pressure from surrounding land use and make countries easier to compare.
To map wetlands consistently across 38 countries, the researchers combined optical images, which record reflected light, with radar images that can detect surface structure and moisture through clouds. In total, they used data from more than 470,000 Sentinel-1 and Sentinel-2 images collected between 2017 and 2019.
Wetlands can look very different across Europe. A peatland in Scandinavia may appear unlike one in the Mediterranean, even when both belong to the same broad category.
“There are a thousand ways this can go wrong. If the training data are biased or do not capture the variation across climates and landscapes, the model may recognise a wetland in one region and miss the same type elsewhere. That is why traditional statistical sampling was just as important as the machine learning,” Kovács says.
The model had to recognise wetlands throughout Europe
The team trained the model using 101,000 locations from the CORINE Land Cover database, selecting points away from uncertain class boundaries and including a broad range of non-wetland landscapes. It then assessed Europe square by square, each measuring 10 by 10 metres.
For every square, it combined 49 clues – including vegetation, moisture, temperature, rainfall, elevation, slope, soil type, biome and distance from the coast – to decide which type of wetland, if any, was present.
Google Earth Engine assembled the satellite and environmental data, and the calculations ran on the Danish supercomputer Computerome.
“Without consistent 10-metre satellite coverage, continent-wide training data and enough computing power, mapping Europe at this detail would not have been realistic,” Kovács explains.
The model distinguished six types of natural and seminatural open wetland: inland marshes, peat bogs, salt marshes, salines, tidal flats and moors and heathland. The map excludes densely forested wetlands because the CORINE framework used for the harmonised training data does not distinguish them from other forests.
“That limitation fit the policy context quite well. Keeping forested wetlands outside the map allowed us to focus clearly on open wetlands and avoid counting the same carbon stores twice,” Kovács says.
Nearly 11,000 human checks tested the machine-made map
To test the result, Kovács interpreted an independent validation sample of 10,975 locations in high-resolution satellite imagery, while two colleagues reviewed uncertain cases. Overall, the map agreed with the assessments in an estimated 96.6% of cases.
Overall, the map agreed with their assessments in an estimated 96.6% of cases. Existing European and global wetland maps showed a broadly similar distribution but often merged or missed the small patches visible at 10-metre resolution.
“A map can look impressive without us knowing how reliable it really is. We therefore used the human checks not only to test the model but also to correct the area estimates and show how uncertain they are,” Kovács explains.
The researchers next combined the wetland map with agricultural and urban land-use registers. Wetlands overlapping these areas were classified as most disturbed, those within 150 metres as intermediately disturbed and more distant wetlands as least disturbed. These categories indicate likely human pressure; they do not prove that every wetland within them is ecologically degraded.
“A wetland does not have to be ploughed or drained directly to be affected. Fertiliser and nitrogen can run off neighbouring fields into standing water and change the ecosystem. That is why nearby intensive land use also matters,” Kovács says.
The classification therefore does not replace field measurements of water levels, vegetation, pollution or ecological condition. Its strength is that it applies the same pressure indicator throughout the study area.
The restoration challenge looks very different across Europe
The six mapped wetland types cover an estimated 413,500 square kilometres, or 6.8% of the combined land area of the 38 countries. Most lie in northern and northwestern Europe, where cool, wet conditions favour waterlogging and peat formation.
“The European picture is extremely uneven. Some countries still contain very large wetland areas, whereas others have fewer wetlands left and a much larger share under pressure,” Kovács explains.
Moors and heathland form the largest mapped category, followed by peat bogs and inland marshes. Area, however, does not determine ecological importance. Peat bogs cover only around 2% of the study region but account for about 53% of the mean estimated carbon pool.
“The type of wetland matters enormously. A relatively small peatland can store far more carbon than a larger area with mineral soil, so restoration priorities cannot be based on hectares alone,” Kovács says.
Small peatlands can hold enormous carbon reserves
For the four wetland types with sufficient data, the researchers estimate that the soils could hold 2.8–14.3 billion tonnes of carbon. They combined the mapped areas with carbon-density estimates from 34 previous studies . The result overlaps with earlier European estimates of 3.2–8.4 billion tonnes, although the finer map produces a higher upper limit. The broad interval reflects large differences in peat depth and carbon density, which remain unevenly measured across Europe.
“Peat depth and carbon density differ greatly across Europe, and we do not have equally good measurements everywhere. Even with this uncertainty, these soils hold very large carbon reserves. The task is to restore damaged wetlands while preventing further losses from those that still store carbon,” Kovács explains.
The estimated carbon stocks were then adjusted according to the disturbance category assigned to each wetland type. About 22,000 soil measurements from the European LUCAS survey showed a broadly similar decline in topsoil carbon.
“The soil measurements generally fell near the lower end of our estimates, suggesting that the reductions were conservative. But the result still describes how much less carbon may have remained stored under 2018 conditions – not emissions released during one defined period,” Kovács cautions.
Inland marshes were both the most fragmented and the most severely affected category. Almost half their area comprises patches smaller than 25 hectares, around one fifth occurs in fragments smaller than one hectare and nearly half was classified as highly disturbed.
“These wetlands often lie in river valleys, floodplains and flat, fertile landscapes. These are exactly the places people have wanted for farming, settlements and infrastructure, so the pressure is especially intense,” Kovács says.
The same restoration target creates very different national tasks
More than 70% of mapped wetlands were classified as disturbed in Bulgaria, Serbia, Hungary and Poland. In Iceland, Norway, Sweden and Finland, more than 90% fell into the least-disturbed category.
“A country may have a relatively small share of disturbed wetlands but still face a major restoration task because its total wetland area is so large. You need to look at both the percentage and the absolute area,” Kovács explains.
To explore what the EU target could mean in practice, the researchers treated wetlands in the two disturbed categories as possible candidates and calculated 30% of this area. This produced a 2030 scenario of about 2.66 million hectares – not a list of sites that governments are legally required to restore.
The largest estimated targets were in the United Kingdom, Ireland, Turkey and Romania, with France, Denmark and Norway also well above the average.
“These estimates should not be read as exact legal obligations for each country. They are scenarios showing the scale and distribution of restoration if the EU percentage targets are applied consistently to the disturbed wetlands identified in the study,” Kovács explains.
The scenarios were also compared with published national commitments. Denmark’s pledge closely matched the study’s 2030 estimate, while Finland’s exceeded it, probably because its target includes wetland categories outside the map’s scope. Commitments in the United Kingdom, Ireland and Germany also fell within or close to parts of the estimated ranges.
Human inspection tested the map, earlier estimates placed the carbon range in context, soil samples tested the disturbance assumptions and national commitments indicated whether the restoration scenarios were plausible. The checks answered different questions but supported a consistent overall picture.
“The comparison is not perfect because restoration is defined and reported differently from one country to another. But a common baseline makes differences in ambition visible and helps governments ask whether they are restoring the places with the greatest ecological and climatic value,” Kovács says.
The easiest hectares may not deliver the greatest gains
The map gives governments a common basis for comparing restoration needs, not a ready-made list of projects. Field surveys are still needed to identify altered water levels, pollution or broken connections to rivers before deciding whether a site should be rewetted, reconnected, protected or managed in another way.
“It shows where wetlands are, what type they are and where surrounding land use points to pressure. This can support better decisions, but it cannot replace local knowledge or field measurements,” Kovács explains.
As countries turn the EU Nature Restoration Law into national plans, area-based targets may encourage them to choose the easiest hectares rather than those with the greatest ecological or climatic value.
“When countries are under pressure to meet a quota, they may select areas with little carbon value or poor connections to existing habitats. The map gives them a better basis for deciding where restoration can achieve what the policy was intended to do rather than effectively flipping a coin,” Kovács says.
Small wetlands can help nature, farmers and the climate
Small wetlands may be harder to manage individually, but together they form mosaics that provide breeding grounds and stepping stones for species moving through intensively used landscapes.
“One approach is to protect a few large areas and intensively use almost everything outside them. The other is to share the landscape with nature, with small patches forming a mosaic through farmland and settlements. Our map shows that Europe’s wetlands are part of that mosaic,” Kovács explains.
Restoration may also benefit farmers. Wet parts of cropland often flood and produce unreliable yields, making them costly to cultivate.
“Some of these areas may already be poor farmland. Restoration can benefit both nature and farmers, especially if landowners are compensated for changing how the land is used,” Kovács says.
Rewetting can slow or halt further carbon losses by keeping oxygen out of the soil, but rebuilding peat that accumulated over centuries or millennia takes far longer. Protecting the carbon still in the ground may therefore matter more than trying to recover everything already lost.
A baseline can reveal whether restoration is actually working
The map excludes forested wetlands, deeply drained peat soils now used as cropland or grassland and wetlands that have disappeared entirely.
“We kept the scope narrow and consistent so that countries could be compared. The map is one layer in a broader picture and not a complete inventory of every former or degraded wetland,” Kovács says.
Because the workflow can be repeated, future satellite observations could reveal changes in wetland extent and surrounding land use. The map provides a 2018 baseline, although field measurements will still be needed to determine whether the ecosystems themselves are recovering.
“Without a baseline, countries may report how many hectares they have restored without showing whether these wetlands are actually recovering,” Kovács explains.
The researchers now aim to extend the method globally, although regions without Europe’s consistent land-cover records will require different data and more advanced computer vision.
“Europe gave us the training data to develop and test the method. The larger goal is to improve global wetland maps and recalculate how wetlands contribute to the world’s greenhouse-gas and carbon budgets,” Kovács concludes.
