Satellite imagery and machine learning used for the mapping of six seminatural open wetland types and land-use disturbance in European countries shows that wetlands are highly fragmented and have uneven restoration needs.
requires at least 30% of wetland ecosystems not in ‘good condition’ to be restored by 2030, yet spatially consistent information on wetland types and condition remains scarce.
Using 10-m satellite imagery and machine learning, we map six seminatural open wetland types and land-use disturbance across 38 European countries. Wetlands are highly fragmented, with an estimated 27–33% of wetland area occurring in map-defined patches <25 ha and 7–11% in patches <1 ha, exposing many small sites missed by coarser products.
We estimate that human activities affect 20.4 ± 3.4% of wetland areas , with inland wetland types most affected, and up to 5 Gt CO-eq of soil carbon potentially lost relative to an undisturbed baseline. Translating disturbed area into NRL restoration targets, we find that several countries’ pledges are broadly consistent with our 2030 estimates, whereas others lack quantified commitments despite substantial candidate areas identified by our maps.
The resulting standardized, high-resolution products provide an EU-wide baseline tailored to the NRL and a reproducible template for linking satellite mapping to restoration targets, supporting progress tracking.. Although they cover only 5–8% of Earth’s land surface, they hold 20–30% of global soil carbon . Article 4 covers terrestrial, coastal and freshwater ecosystems and defines restoration as actions that move systems towards their original condition to target Europe’s main natural and seminatural open wetland classes , while excluding densely forested wetlands that the CLC framework does not distinguish from other forests.
Under present EU/Intergovernmental Panel on Climate Change land-use classifications, drained peat soils managed as cropland or grassland fall under NRL Article 11 and tree-covered organic soils under Article 12 ; these deeply drained former wetlands are therefore outside our scope here to avoid double counting with Article 4, which targets natural and seminatural wetlands.but its manual interpretation, multiyear update cycle and 25-ha minimum-mapping unit hinder timely detection and bias against small, managed mosaics, propagating errors in distribution and carbon estimates. Meeting restoration objectives thus requires high-resolution, standardized, regularly updated mapping of wetland extent, type and disturbance.
Here we map six open wetland types at 10-m resolution across Europe by combining optical and radar satellite data in a scalable machine-learning framework anchored to the CLC framework. The resulting geodataset refines carbon-stock estimates and quantifies anthropogenic impacts at national to subcontinental scales, providing the spatial detail needed to guide Article 4 restoration for 2030 and beyond.
We mapped six natural and seminatural open wetland types across large parts of Europe at 10-m resolution with an overall accuracy of 96.6 ± 0.3%, covering an estimated 413,504 ± 17,782 kmmachine-learning algorithm trained on 101,000 CLC-labelled sample points and applied to a multisensor satellite data cube spanning 2017–2019. The resulting 49-variable predictor set combined optical and radar observations from more than 470,000 Sentinel-1 and Sentinel-2 scenes with auxiliary geospatial layers on topography, climate, biomes and soil across 38 European countries . European-scale wetland areas and map accuracies were estimated directly from the independent stratified reference sample using a design-based estimator.
Country-level wetland areas were estimated using a hierarchical framework that combines calibration weighting with empirical-Bayes partial pooling to stabilize estimates in which national sample sizes are limited = 76%; user’s accuracy = 96%). It also includes, by definition, several Arctic, Alpine and maritime non-wetland habitats.
‘Moors and heathland’ are concentrated in high-latitude regions or high-altitude zones , in which lower temperatures and persistent moisture dampen vegetation growth, evapotranspiration and decomposition. Although differently classified as ‘shrubland’ or ‘wetland’ in land-cover products, are principally found in northern Atlantic areas, in which postglacial landforms and abundant precipitation support peat accumulation and bog development.
‘Inland marshes’, totalling 39,698 ± 6,975 km, encompass a variety of wetlands with herbaceous vegetation, including fens and transitional mires, often concentrated in river valleys, floodplains, deltas and lakeside areas. Smaller coastal wetlands occupy 20,819 ± 4,154 km; 0.2 ± 0.1%; PA = 94%; UA = 75%) are scattered mostly along Atlantic and North Sea coasts, the western and central Mediterranean shores and parts of the Black Sea margin, shaped by salt-tolerant vegetation, sedimentary processes and tidal flow.
‘Intertidal flats’ , occurring along the Atlantic, are continually shaped by coastal dynamics, whereas ‘salines’ , the smallest class, represent both active and formerly exploited salt-evaporation basins. ) holds the largest extent, followed by the United Kingdom, Iceland, Sweden, Ireland and Finland, a pattern consistent with wet, cool climates and low-relief postglacial terrains that sustain waterlogging, peat formation and long-term carbon storage .
Calculating the patch-size distribution shows that European wetlands are fine-grained and highly fragmented, with an estimated 27–33% of wetland area occurring in map-defined patches <25 ha and 7–11% in patches <1 ha . Small-patch prevalence is greatest for inland marshes , followed by salt marshes and lowest for intertidal flats , consistent with long-term fragmentation in intensively used landscapes.
By contrast, widely used land-cover and wetland products under-represent this class and size diversity: wetlands <25 ha and <1 ha are commonly missed by coarser inventories. Together, these results show that fine spatial detail is essential to represent wetland extent and heterogeneity across Europe. They also provide the robust spatial basis needed for continental-scale carbon stock estimates, which cannot be consistently derived from coarser-resolution products.
Building on this high-resolution wetland extent, we next estimated associated soil carbon stocks across Europe. Carbon stock calculations used sample-based wetland area estimates, with stratified estimates at the European scale and calibrated hierarchical estimates at the country level .
These density ranges synthesize measurements from several studies across Europe sampled over different depth intervals , from surface soils to full peat profiles , depending on wetland type , capturing depth-integrated variability in soil carbon. Our estimates exclude carbon stocks from drained and forested wetlands.
The resulting total wetland carbon storage potential spans 2.8–14.3 Gt C, broadly consistent with earlier estimates At the national level, the respective carbon stocks peak at 3.5 Gt C in the United Kingdom, 2.7 Gt C in Norway and 2.5 Gt C in Sweden, reflecting values at the upper end of our range . By CLC wetland class, ‘peatbogs’, although covering only about 2% of the study area, store on average 3.3 Gt C .
‘Moors and heathland’ account for 1.8 Gt C , whereas ‘inland marshes’, despite limited area, contain 0.9 Gt C , making them the most carbon-dense type in countries such as Turkey, Romania, Hungary and Poland. Notably, large portions of areas mapped as ‘inland marshes’ in the Central European Lowland and the Danube Delta correspond to fen peatlands, Geometric-mean estimates of potential carbon stocks in total wetlands for 2018, showcasing contributions per type at the country level and across Europe, with ‘±’ values denoting 95% confidence intervals, calculated as ±1.96 × standard error from the stratified, area-weighted sample-based estimator.
For each country , the 2018 mean estimate is calculated as the geometric mean of the lower and upper bounds of the total carbon-stock range obtained by aggregating class-specific minimum and maximum stock estimates. , Cross-country comparison of total carbon storage in wetlands, showing geometric-mean estimates , the corresponding stock ranges and type proportions.
Owing to high burial rates, coastal wetlands also contribute disproportionately: ‘salt marshes’ cover just 0.2% of the study area yet store 0.3 Gt C and also have low CHand their carbon storage potential is not well understood. We therefore highlight the need for targeted in situ measurements of ‘tidal flats’ in Europe. In the absence of such data, both ‘intertidal flats’ and ‘salines’ are excluded from our carbon accounting analyses.
Although undisturbed wetlands typically act as carbon sinks, human disturbances in wetlands can lead to a shift in their carbon balance, causing a net carbon loss from wetland soil to the atmosphere. Wetlands located within or near agricultural or urban areas are probably influenced by human activities such as land development, pollution, soil compaction, vegetation removal and drainage.
Consequently, wetlands under agricultural or urban land use are expected to fail to meet the NRL’s ‘good condition‘ thresholdsTo distinguish natural wetland areas under anthropogenic pressure and characterize wetland condition at the continental scale, we used agricultural and urban land-use classes as proxies for physical, chemical and biological stressors. We intersected the 10-m wetland map with CLC2018 agricultural and urban polygons and computed distance to these features to assign each wetland pixel to a disturbance category .
Wetlands overlapping these land-use classes were classified as ‘most disturbed’, those within 150 m as ‘intermediately disturbed’ and all others as ‘least disturbed’. At the European scale, disturbance-domain areas were obtained using the design-based stratified estimator. At the country scale, disturbance areas were derived by calibrated allocation from the pooled country-level wetland-class totals, with uncertainty propagated by Monte Carlo sampling . Regional patterns indicate that undisturbed wetland areas are predominantly situated in high-altitude and high-latitude regions, such as the Alps, the Scandinavian mountains, the Scottish Highlands and Iceland.
These undisturbed areas align with ‘moors and heathland’, confirmed as the least disturbed CLC wetland class in Europe, probably because of their location in agriculturally unfavourable terrains with difficult access. By contrast, ‘inland marshes’ represent the most disturbed wetland ecosystems, with nearly half of their area classified as most disturbed.
This is followed by ‘salt marshes’, in which roughly one-third of the area shows signs of disturbance, and ‘peatbogs’, in which around one-fifth is affected by human activity. ). In absolute terms, the United Kingdom , Ireland , Turkey , Romania and France host the largest disturbed extents.
However, the relative share of disturbed wetlands differs substantially, ranging from 23.7% in the United Kingdom and 21.6% in Spain to 59.5% in Turkey and 49.7% in Romania. The highest proportional disturbance occurs in Bulgaria , Serbia , Hungary and Poland , despite their smaller absolute areas . By contrast, Iceland , Norway , Sweden and Finland retain the highest shares of undisturbed wetlands.
At the other end of the gradient, Norway , Iceland , Finland and Sweden retain the lowest disturbed shares, with most wetland extent remaining intact at northern latitudes. Although disturbed areas in these countries range from about 0.1 to 0.5 Mha, the predominance of undisturbed wetlands may partly reflect the exclusion of deep-drained and forested wetlands from our accounting.to the baseline carbon stocks described above.
We evaluated these factors against approximately 22,000 SOC measurements from the LUCAS 2018 topsoil dataset across Europe and found that the magnitude of the observed decline is broadly consistent with our reduction factors and generally lies towards their lower bound . This near-surface focus for carbon exchanges is consistent with IPCC guidelines and peatland flux studies, which identify the upper approximately 30 cm as the main aerobic decomposition zone, in which repeated oxygen exposure drives most COFor each wetland type and country, the potential carbon storage loss is defined as the difference between the undisturbed baseline and the disturbance-adjusted stock, expressed in CO).
Summed across wetland classes and countries, these estimates indicate a disturbance-associated stock difference observed under 2018 conditions, rather than a realized emission trajectory over a specified time period. Land-use disturbances correspond to potential losses of 0.8–4.9 Gt CO). By inheriting the variability of the underlying carbon-density values, these potential loss estimates reflect heterogeneity in topography, peat depth, land use and management.
‘Peatbogs’ show the widest spread and therefore the broadest ranges of potential losses . All estimates represent 2018 conditions and quantify potential carbon storage losses associated with contemporary land-use disturbance, rather than annual greenhouse-gas emission rates, and are restricted to natural and seminatural open wetlands, excluding historically lost wetlands, forested wetlands and heavily drained peatlands Our findings reveal the highest potential carbon-storage losses in the United Kingdom, Norway, Sweden and Ireland, reflecting the combination of large wetland extents and substantial disturbance mean loss of carbon storage potential owing to disturbances across countries in 2018, with relative country-level contributions by open wetland type.
Potential carbon storage loss is estimated relative to an undisturbed baseline mean for reference. Estimates represent differences in potential soil carbon stock under the 2018 disturbance state relative to the undisturbed baseline and do not imply a specific temporal rate of carbon loss.
, which encompass distribution, structural integrity and characteristic vegetation, we interpreted land-use disturbance in wetlands as a signal of non-compliance and translated it into scenario-based wetland restoration target areas . At the country level, we define the restoration target as the area of natural open wetlands under land-use disturbance that may be eligible for restoration under these criteria.
These country-level restoration target areas should be interpreted as stylized scenarios, assuming that a fixed share of disturbed seminatural open wetlands is restored to meet the Article 4 percentage targets, rather than as exact legal obligations or implementation plans.
For each Article 4 scenario, the restoration area was calculated as the mandated percentage of the estimated disturbed wetland area for each wetland type and country and 30% of the total disturbed ‘peatbogs’ area , Ireland , Turkey and Romania together account for a large share of the total.
The United Kingdom’s eligible restoration area is nearly four times that of Germany . France , Denmark and Norway also stand well above the regional average, whereas Sweden , Poland and Hungary cluster closer to the mean despite contrasting land areas. In northern Europe, restoration targets are strongly influenced by peatbog distribution.
Peatbog-specific targets reach 234,000 ± 71,300 ha in the United Kingdom and 256,900 ± 64,900 ha in Ireland, with Sweden and Norway also showing substantial peatland components. By contrast, Turkey and Romania, in which peatbog extent is limited, show large overall targets , driven mainly by inland marsh systems.
Taken together, these country-level patterns show that restoration demand reflects the distribution of seminatural wetland types and their disturbance within each country, underscoring the need for harmonized, spatially detailed wetland datasets to enable robust cross-border comparison of restoration commitments. To place our estimates in context, we next compared our 10-m wetland map with existing continental-scale wetland products. Our cross-product assessment revealed that many existing wetland maps and reporting systems are limited by incompatible typologies.
Broader class definitions are often misaligned with specific policy aims such as the NRL. For example, some inventories group forested wetlands and drained peat soils into broad classes, whereas others exclude inland types altogether. Compiled layers often mix heterogeneous national legends, so class definitions can drift across borders. Coarser pixels or large minimum-mapping units tend to systematically omit or merge small wetland patches, biasing area metrics.
Time adds another axis of inconsistency: many compilations integrate legacy geodata from different decades, so ‘extent’ may reflect historical distributions and can retain wetlands that are now degraded or lost without stating this explicitly. When explicitly distinguished from present wetland extent rather than conflated with it, however, such legacy layers can still be valuable for locating degraded or lost wetlands now under agriculture, particularly drained peat soils relevant to NRL Article 11.
A further constraint of existing products is the limited quantification of uncertainty: accuracies, when reported, are often presented without confidence intervals or an explicit sampling design, restricting their use in planning, model-based assessments and applications requiring uncertainty propagation. By contrast, our 10-m European wetland map applies a single, continent-wide methodology grounded in standardized wetland definitions and resolves sub-hectare structure, reducing omission of small patches.
We report confidence-bounded accuracies and sample-based area estimates and the workflow is reproducible and updatable for change assessment, providing a comparable baseline for restoration planning across Europe. Combined with explicit legal targets, this framework supports operational comparison and tracking of restoration pathways under the EU NRL.
To anchor the magnitude of our restoration scenario estimates in existing policy contexts, we compared our 2030 target areas for all wetlands against national pledges and found that some countries are well on track to meet their restoration objectives . These comparisons use national total areas only and do not assume spatial overlap between our estimated restoration targets and pledged sites, as national definitions of areas allocated to restoration may differ from the wetland categories used here.
Among these frontrunners,is in near one-to-one agreement with our 2030 estimate . In the United Kingdom, peat-only commitments lie near the upper bound of our peatbog-only 2030 estimate, exceeding the mean by about 31% and falling only roughly 1% below the upper 95% confidence limit.
Ireland’s confirmed projects correspond to about 47% of our peatbog-only 2030 estimate and about 42% of our all-wetlands 2030 estimate, with further sites under preparationfrom peat soils by 2030) milestone corresponds to about 89% of our 2030 estimate, reflecting a substantial rewetting pipeline already in motionin Austria, Hungary, Slovenia, Croatia, Serbia, Romania and Bulgaria could contribute towards meeting each country’s 2030 NRL obligations.
Romania’s long-term objective exceeds our 2040 estimate , whereas Hungary’s 2030 pledge represents only about 40% of our estimate, implying restoration potential three to five times higher than the present commitment, consistent with previous assessments) equals about 175% of our 2030 estimate and about 87% of our 2040 figure, probably reflecting broader definitions beyond our ‘peatbogs’ class. However, restoration of forested wetlands typically requires tree removal and is therefore strictly capped under the NRL, limiting how such areas can be counted towards NRL targets.
Nevertheless, the success of restoration is measured not only in hectares restored but also in the protection and retention of soil carbon in disturbed wetlands. To discuss restoration in a broader climatic context, we compared the magnitude of potential soil carbon loss associated with land-use disturbance in Europe’s wetlands with realized or observed wetland carbon losses elsewhere.
Our estimates are grounded in continent-scale LUCAS Soil observations and indicate a potential loss of 0.84–4.87 Gt CO. However, substantial uncertainty remains about the timescales over which these carbon losses have occurred or may continue, reflecting a central limitation: Europe lacks continent-scale, temporally resolved monitoring of wetland carbon stock change. For comparison, wetlands in the conterminous United States experienced a realized carbon loss of 1.06 Pg C between 2011 and 2016 are under-represented in ‘peatbogs’, in which seasonal wetness and woody cover blur separation from ‘inland marshes’. Coastal optical effects can also produce false positives in ‘salines’, in which algal blooms and green water resemble evaporative pans despite contextual screening.
Carbon-stock estimates span wide ranges because depth-resolved SOC measurements are sparse, spatially clustered and rarely capture land-use and management effects at depth. Regional variation in carbon densities within classes is likely but further subdivision beyond standardized CLC types would reduce cross-country comparability. Country-level area estimation with hierarchical pooling provides a practical compromise between continent-wide summaries and direct country-domain estimation.
Because validation support within individual countries is sparse for some wetland types, and especially for disturbance-specific subdomains, direct country-level estimates can be unstable. A nationally stratified validation design with adequate support for every wetland type and disturbance category in every country would require substantially larger and more costly reference-data collection. Our approach therefore preserves the continental probability-sampling basis while stabilizing sparse country-level estimates through hierarchical pooling towards regional mean compositions.
Unlike the Europe-wide design-based estimates, country-level estimates are model-based and their 95% intervals are posterior credible intervals. Even with explicit restoration area targets to 2030, implementation is constrained by logistical, political and financial factors , and the law’s reach is uneven: non-EU partners and candidates participate only voluntarily, whereas the main wetland holders outside the EU are not bound by EU rules or supported by the Common Agricultural Policy .
Although the reformed CAP allocates resources to environmental measures, wetland-specific criteria are limited and alignment with the NRL remains weak. Even in some member states , CAP-linked financing for wetland restoration was unavailable before 2025, constraining early implementation. With National Restoration Plans due by 2026, the window for demonstrable progress is narrow, raising the risk that rapid delivery gives priority to area over ecological quality.
Even so, the 2030 target may serve as a catalyst, mobilizing institutions, aligning finance and setting a credible trajectory towards extensive wetland restoration by 2040 and beyond. Wetlands occupy only a small fraction of Europe’s land surface. Our results show that seminatural open wetlands are rare, highly fragmented, heavily modified and important for carbon storage and thus for the global carbon budget.
Disturbance is concentrated in inland, carbon-rich systems; restoration needs are unevenly distributed among countries; and many wetlands occur in small patches embedded in Europe’s mosaic landscapes, reflecting centuries of land conversion. Viewed through the lens of the EU NRL, our blueprint shows how harmonized, pixel-scale satellite observations can be translated into restoration policy insights across countries.
Until truly continental or global field campaigns can routinely characterize wetland condition and carbon stocks, such large-scale, remotely sensed proxies will remain central to supporting wetland restoration and conservation pledges. In the context of historic wetland loss, aligning legal commitments with ecological understanding and high-resolution, data-driven observation offers a rare opportunity to move from a legacy of loss to an era of targeted restoration, in Europe and globally.
) multispectral imagery at 10-m resolution in the visible and near-infrared bands and 20 m in the shortwave-infrared bands . Sensor-provided quality bands were used to mask clouds and cirrus. From these bands, we derived spectral indices including the normalized difference vegetation index$${\rm{M}}{\rm{N}}{\rm{D}}{\rm{W}}{\rm{I}}=\frac{{\rm{B}}3-{\rm{B}}11}{{\rm{B}}3+{\rm{B}}11},$$ as well as a grey-level co-occurrence matrix contrast texture from the near-infrared band using a 2-pixel window.
) synthetic aperture radar imagery at 10-m resolution in VV and VH polarizations, together with the local incidence angle. From these, we derived the VV/VH backscatter ratio and its temporal statistics. To provide further thermal context, we also included Landsat-8 , so that for each band$${\tilde{x}}_{b}={{\rm{median}}}_{t},\,{\sigma }_{b}={{\rm{sd}}}_{t},\,{x}_{b}^{\text{max}}=\mathop{\text{max}}\limits_{t},$$. This yielded, per 10-m pixel, a multisensor predictor image stack summarizing central tendency and intra-annual variability in optical and radar signals. As well as optical and radar imagery, we incorporated several ancillary variables. Land-surface temperature was derived from a MODIS daily land-surface temperature product, using the mean daytime temperature over 2000–2020 .
Total precipitation was obtained from a global atmospheric reanalysis using the 2000–2020 mean of daily total precipitation to provide a long-term hydroclimatic baseline. Topography was represented by the EU-DEM v1.1 , from which we computed the slope using a terrain operator; both elevation and slope were included as predictors.
We further derived a distance-to-coast layer as the Euclidean distance to the nearest coastline , capturing coastal–inland gradients relevant for salt-influenced wetlands. Following a preliminary soil analysis based on the European Soil Database v2.0 , we included categorical layers for Food and Agriculture Organization of the United Nations soil units and parent materials , as well as terrestrial biomes from the RESOLVE Ecoregions dataset.
These categorical maps were represented as indicator variables for use in the machine-learning model. In total, the resulting feature stack comprised 49 predictor variables per 10-m pixel, spanning spectral, radar, thermal, topographic, climatic, edaphic and biome controls. Our workflow combines cloud-based preprocessing with high-performance local computing for mosaic creation.
All primary predictors and ancillary variables were first computed as 10-m, cloud-free composites and long-term summaries within Google Earth Engine. These multiband images were then exported as tiled GeoTIFFs. The European study area was partitioned into country polygons and further subdivided into a regular 5-km grid, aligned with the corresponding local UTM zone to ensure equal-area representation.
For each 5-km grid cell, we assembled all predictors into a single multiband mosaic at 10-m resolution, producing local feature stacks as multiband GeoTIFFs in local UTM coordinates, suitable for supervised learning. , covering all 27 EU member states together with extra European Environment Agency countries , hereafter referred to as the EEA38.
Throughout, ‘Europe’ and ‘the continent’ refer to this reporting extent. CLC is the most comprehensive continental-scale dataset for wetland types but its minimum-mapping unit of 25 ha omits many small wetlands and introduces uncertainty around land-cover boundaries. Such noisy labels are problematic for convolutional neural networks trained on high-resolution satellite imagery.
By contrast, pixel-based approaches that use point labels, and are less sensitive to polygon boundaries, have shown strong performance when trained on coarse or noisy annotations To construct a supervised training dataset from CLC2018, we selected 101,000 training sample points. Each training location was required to be at least 100 m from CLC class boundaries, reducing label noise from neighbouring land-cover types.
We defined seven target wetland classes based on their corresponding CLC classes. For each wetland class$${N}_{{\rm{wet}}}=\mathop{\sum }\limits_{c=1}^{7}{n}_{c}=7\times \mathrm{5,000}=\mathrm{35,000}$$ wetland training locations.
To represent non-wetland land cover, we drew training sample points from 30 diverse non-wetland and non-water CLC classes, with 2,000 sample points per class, and extra background locations to capture further variability, yielding in total= 35,000 + 66,000 = 101,000 training sample points, of which 35,000 represented target wetland classes and 66,000 represented various background classes. This stratified design imposed equal sampling effort across wetland classes while also drawing a large and diverse set of background points from many non-wetland land-cover types.
At each of the 101,000 training sample locations, we extracted the selected image features using Google Earth Engine∈ })+\mathop{\sum }\limits_{m=1}^{M}\Omega ,$$are mislabelled. In gradient boosting, instance weights are implicitly updated by means of the gradient of the lossprevent overfitting to noisy labels.
In practice, this means that persistently inconsistent or mislabelled points contribute less to the final decision function, which mitigates the effects of label noise on the CLC-based training set. $$\hat{{\mathcal{L}}}=\frac{1}{K}\mathop{\sum }\limits_{k=1}^{K}{{\mathcal{L}}}^{},$$across 25 candidates . Cross-validation averages over different training–validation splits, reducing sensitivity to both sampling variability and label noise in the validation folds.
The final model used GPU-accelerated boosting with tuned values for the learning rateAfter predicting wetland classes at 10-m resolution, we applied a 3 × 3 median filter to the class map to remove isolated speckle and smooth class boundaries, replacing each pixel label by the most frequent class within its 3 × 3 neighbourhoodWe constructed a validation set drawn independently of the training data and not used in model fitting using stratified random sampling with disproportionate allocation, with a minimum of). All non-target strata were merged into a single background stratum.
Stratum areas were derived by intersecting CLC2018 polygons with the 10-m European Forest Type 2018 dataset, excluding forested regions from the target classes. Within each stratum, we then generated validation sample locations at least 10 m away from all training sample points to ensure spatial independence.
Of the planned roughly 15,000 locations, 3,691 could not be placed: the placement algorithm could not find positions satisfying the ≥10-m criterion, with the shortfall concentrated in the geometrically narrow salines stratum . A further 326 fell outside the mapped image extent and were removed and eight were excluded during final reference-label quality control because no reliable land-cover label could be assigned, givingEach validation sample point was visually interpreted on-screen by an expert using high-resolution Google satellite imagery, following the CLC illustrated nomenclature guidelines, which provide class descriptions, surface-pattern diagrams and example photos to support consistent labelling.
Uncertain or ambiguous points were jointly reviewed by two more interpreters until consensus was reached. As contextual information, we consulted the Global Lakes and Wetlands Database v2 . These ancillary datasets informed interpretation only; final reference labels were assigned solely by expert visual assessment.
The sample point retained its original sampling stratum and associated inclusion probability. = 10,975 using design-based stratified estimation. Area estimation followed a stratified estimator using sampling strata defined by collapsed CLC classes =\sum _{h}{A}_{h}^{2}\left\frac{{s}_{{hk}}^{2}}{{n}_{h}},\,{s}_{{hk}}^{2}=\frac{{n}_{h}}{{n}_{h}-1}{\hat{p}}_{{hk}},$$\\)$${\hat{A}}_{{cr}}=\sum _{h}{A}_{h}{\hat{p}}_{{hcr}},\,{\hat{p}}_{{hcr}}=\frac{1}{{n}_{h}}\sum _{i\in h}I,$$$${\rm{OA}}=\frac{\sum _{k}{\hat{A}}_{{kk}}}{\sum _{c}\sum _{r}{\hat{A}}_{{cr}}},$$$${{\rm{PA}}}_{k}=\frac{{\hat{A}}_{{kk}}}{\sum _{c}{\hat{A}}_{{ck}}},\,{{\rm{UA}}}_{k}=\frac{{\hat{A}}_{{kk}}}{\sum _{r}{\hat{A}}_{{kr}}}.
$$ To characterize wetland fragmentation from the mapped patch structure, map-defined patch-size bins were derived from contiguous clusters of 10-m cells within each class. The wetland area in each bin was then estimated using the same stratified-design-based indicator estimator as for class-area estimation, with cumulative categories such as <25 ha estimated directly.
), direct country-domain estimation alone was often too unstable for reliable reporting and country-level wetland areas were therefore estimated using a calibrated country-level estimator with hierarchical pooling built on the continental stratified probability sample. The domain-estimation motivation follows standard survey-sampling logic for sparse domainsthe reporting countries. To preserve the continental sampling basis, initial expansion weights were defined asis the number of validation sample points in that stratum.
To align this continental sample with country-level reporting, these base weights were then calibrated by generalized raking, implemented through iterative proportional fitting, so that the weighted sample matched both the known country frame totals). In this setting, calibration estimation refers to the adjustment of design-based expansion weights using known auxiliary totals, thereby linking the continental probability sample to the country reporting domains.
This calibration step assumes that, after adjusting to known country-frame and stratum-area totals, the calibrated pseudo-counts are conditionally representative of each country’s true class composition. After calibration, weights were normalized within country and normalized weighted class pseudo-counts were computed for each wetland class an indicator for the final reference class. To stabilize country-specific class compositions when these weighted counts were sparse, countries were assigned to fixed macro-regions and region-level mean class compositions were estimated from pooled weighted country counts.
We then used a two-level hierarchical empirical-Bayes Dirichlet formulation, in which each country-level class-composition vector). In this hierarchy, country-level compositions are pooled to a shared regional mean for all countries belonging to the same macro-region, with the degree of shrinkage controlled by. This hierarchical formulation preserves positivity and unit-sum constraints and stabilizes weak country estimates by shrinking them towards macro-regional mean compositions, while allowing countries with more informative weighted counts to remain closer to their own observed class composition.
Country-level wetland areas were summarized by posterior means and 95% posterior credible intervals. ≥ 200 , direct calibration-only estimates of total wetland area fell within the 95% posterior credible interval in all cases ; for countries with fewer sample points, wider credible intervals reflect appropriately increased uncertainty.
For these data-rich countries, low pooling fractions < 0.12) ensure that country-level calibration data dominate the posterior; the hierarchical structure mainly serves to regularize estimates for countries with sparse validation support, for which direct estimation would be unstable. At the macro-regional level, the aggregate of pooled posterior means across all countries in each of the five regions fell within the 95% posterior credible interval in all cases, with deviations of less than 2.2% from the corresponding direct calibration aggregate.
Anthropogenic disturbance was defined by intersecting the wetland map with agricultural and urban classes from CLC2018. Wetlands overlapping these classes were classified as most disturbed, those within 150 m of them as intermediately disturbed and all others as least disturbed. Disturbance labels were assigned after sampling and treated as post-stratification reporting domains. Europe-wide areas for disturbance levels and wetland class–disturbance combinations were estimated within the wetland domain using the same stratified indicator estimator described above.
At the country level, further post-stratification by disturbance further reduced effective sample sizes within individual wetland classes. Country-level disturbance subdomain areas were therefore estimated by calibrated allocation from the pooled country-level wetland class totals. Let$${q}_{k,d}^{{\rm{EEA}}38}=\frac{{\hat{A}}_{k,d}^{{\rm{EEA}}38}}{\sum _{d}{\hat{A}}_{k,d}^{{\rm{EEA}}38}}. $$$${A}_{c,k,d}^{}={A}_{c,k}{p}_{c,k,d}^{{\rm{map}}}.
$$$$\sum _{d}{\tilde{A}}_{c,k,d}={A}_{c,k},\,\sum _{c}{\tilde{A}}_{c,k,d}=\left{q}_{k,d}^{{\rm{EEA}}38}. $$ Calibration was implemented by means of iterative proportional fitting. Uncertainty was propagated by Monte Carlo sampling of pooled country totals and Europe-scale disturbance estimates, with calibration repeated for each draw. Sample-based wetland area estimates formed the basis for carbon-stock calculations.
At the European scale, stratified class-area estimates derived from the stratified validation sample were used. At the country scale, calibrated hierarchical estimates provided country-level wetland areas. Carbon-density ranges ,$$Disturbance-specific wetland areas were estimated directly. At the European scale, disturbance-domain areas were obtained using the design-based stratified estimator described above. At the country scale, disturbance areas were derived by calibrated allocation from the pooled country-level wetland-class totals, with uncertainty propagated by Monte Carlo sampling of posterior area draws.
Carbon densities were adjusted as). For peatbogs, reductions of 0%, 30% and 50% were assumed for the least, intermediate and most disturbed categories; for other wetland types, reductions of 0%, 20% and 25% were applied.
These values were evaluated against LUCAS SOC observations and were conservative relative to observed depletion patterns natural open European wetlands failing the ‘good condition’ criteria owing to anthropogenic disturbance using country-level, sample-based disturbance-area estimates. For each country, disturbed wetland area was defined as the sum of intermediately and most disturbed categories across wetland classes,.
Restoration targets were then computed as fixed fractions of disturbed area,\\). Finally, we benchmarked the estimated 2030 targets against published national wetland and peatland restoration commitments for the 15 countries with the largest disturbed wetland area, compiled from official policy documents of the University of Copenhagen at.
An interactive web viewer for visual exploration of the dataset will be made available through the Global Wetland Center, University of Copenhagen .
Regulation 2024/1991 of the European Parliament and of the Council of 24 June 2024 on nature restoration and amending Regulation 2022/869. . Page, S. E., Rieley, J. O. & Banks, C. J. Global and regional importance of the tropical peatland carbon pool. Hendriks, K., Gubbay, S., Arets, E. & Janssen, J. Carbon storage in European ecosystems; a quick scan for terrestrial and marine EUNIS habitat types.
Taddeo, S., Dronova, I. & Depsky, N. Spectral vegetation indices of wetland greenness: responses to vegetation structure, composition, and spatial distribution. Proc. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data MiningCouncil Directive 92/43/EEC of 21 May 1992 on the conservation of natural habitats and of wild fauna and flora. . Lehner, B. et al.
Mapping the world’s inland surface waters: an update to the Global Lakes and Wetlands Database . Zhang, X. et al. GWL_FCS30: a global 30 m wetland map with a fine classification system using multi-sourced and time-series remote sensing imagery in 2020. Zhang, X. et al.
Global annual wetland dataset at 30 m with a fine classification system from 2000 to 2022. Adams, C. A., Andrews, J. E. & Jickells, T. Nitrous oxide and methane fluxes vs. carbon, nitrogen and phosphorous burial in new intertidal and saltmarsh sediments. He, H. & Roulet, N. T. Improved estimates of carbon dioxide emissions from drained peatlands support a reduction in emission factor.
Orgiazzi, A., Ballabio, C., Panagos, P. & Fernández-Ugalde, O. LUCAS Soil, the largest expandable soil dataset for Europe: a review. . Xu, H. et al. Modification of normalised difference water index to enhance open water features in remotely sensed imagery.
ERA5: fifth generation of ECMWF atmospheric reanalysis of the global climateEuropean Environment Agency. European Digital Elevation Model v.1.1 . European Commission & European Soil Bureau Network. European Soil Database v.2.0 .
Sainte Fare Garnot, V., Landrieu, L., Giordano, S. & Chehata, N. Satellite image time series classification with pixel-set encoders and temporal self-attention. InLanganke, T., Herrmann, D., Ramminger, G., Buzzo. & Berndt, F. Copernicus Land Monitoring Service – High Resolution Layer Forest: Product Specifications Document. Langanke, T., Moran, A., Dulleck, B. & Schleicher, C. Copernicus Land Monitoring Service – High Resolution Layer Water and Wetness: Product Specifications Document.
Stehman, S. V. Estimating area and map accuracy for stratified random sampling when the strata are different from the map classes. Wang, X., Berg, E., Zhu, Z., Sun, D. & Demuth, G. Small area estimation of proportions with constraint for National Resources Inventory survey. We acknowledge the use of Computerome 2.0, the Danish national high-performance computing system.
The views expressed in this article are those of the authors and do not necessarily reflect the views or position of the European Environment Agency . G.M. K. discloses support for the research of this work from Villum Fonden through the DeReEco research grant and from the Novo Nordisk Foundation through the Global Wetland Center .
R.F. discloses support from the Danish National Research Foundation through the Center for Remote Sensing and Deep Learning of Global Tree Resources . F.T. and S.v.d. L. disclose support from the Deutsche Forschungsgemeinschaft through SFB Transregio 410/1, project 531801029, ‘WETSCAPES2.0’.
Department of Geosciences and Natural Resource Management, University of Copenhagen, Copenhagen, Denmark Gyula Mate Kovács, Xiaoye Tong, Dimitri Gominski, Stéphanie Horion, Christin Abel, Guy Schurgers, Bo Elberling, Alexander Prishchepov & Rasmus FensholtGyula Mate Kovács, Stefan Oehmcke, Stéphanie Horion, Guy Schurgers, Bo Elberling & Rasmus FensholtSebastian van der Linden, Alexandra Barthelmes & Franziska TannebergerG. M.K. conceived and led the study, developed the remote sensing workflow, processed the satellite data, served as the primary interpreter for the validation data, wrote the manuscript, led the statistical analyses for area and disturbance estimation and led the revision process.
D.G. and X.T. acted as secondary interpreters for uncertain validation cases, advised on the remote sensing workflow, contributed to data processing and D.G. also revised the statistical analyses during the revision process. S.O. advised on the machine learning workflow and supported data processing. C.A. supported spatial data visualization. A.P. and S.v.d.
L. provided guidance on land-cover classification and validation. S.P. , G.S. and B.E. advised on biogeochemical processes, carbon stock estimation and interpretation of results. A.B. and F.T. contributed expertise on peatlands and peatland mapping and European environmental policy on peatlands and were added to the author list in the first revision.
E.I. advised on policy relevance. S.H. and R.F. supervised the research and contributed to the study design. All authors approved the final version of the manuscript. , The overall ranked total wetland coverage per country and type.
Bars show calibrated hierarchical country-level estimates and error bars denote 95% posterior credible intervals with our estimated potential wetland area to restore by 2030 for countries with explicit targets. Each point represents a country, with symbol size proportional to the total mapped disturbed wetland area and colours distinguishing individual countries.
The dashed line indicates parity between reported commitments and mapped 2030 restoration potential; points above this line represent countries whose stated commitments exceed our estimated 2030 restoration target. Numerical values for national commitments and mapped restoration needs are provided in Supplementary TableOpen Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.
You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material.
If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
Please follow us on Google to support usUnited States Latest News, United States Headlines
Similar News: You can also read news stories similar to this one that we have collected from other news sources.
California fruit giant hit by shocking ‘highly toxic’ accusations and deceptionWatsonville-based berry company Driscoll’s has been accused in a class action lawsuit alleging its strawberries contained forever chemicals that are “highly toxic” and take 1,000 years to break dow…
Read more »
Colorado Four-Star Commit Turns Heads in Front of Highly-Touted RecruitsThe Colorado Buffaloes featured representation at the annual Under Armour Next S7VNS camp. Colorado fans and coach Deion Sanders will be pleased to know a futur
Read more »
Tom Cruise unveils dramatic transformation in first look at Alejandro Iñárritu’s highly-anticipated 'Digger'Tom Cruise is nearly unrecognizable in the 'Digger' trailer, playing an eccentric oil magnate whose venture could trigger an $18 trillion disaster.
Read more »
SES and Airbus to Build Optical Ground Station for Quantum-Secure EAGLE-1 Satellite in the NetherlandsSES and Airbus Netherlands have secured a ground lease in Noordwijk to construct an optical ground station at the NL Space Campus. The facility will communicate with the EAGLE-1 satellite, Europe's first quantum-secure communication satellite, using laser links to deliver quantum-safe keys. The station, featuring advanced atmospheric correction optics, will serve governments and financial institutions. The project, part of an ESA and European Commission partnership, marks a major step toward a European Quantum Key Distribution system.
Read more »