Arctic permafrost regions are undergoing drastic disturbances as a result of climate variability, leading to widespread and unprecedented landscape disturbances . Common Arctic permafrost landscape transformations, such as retrogressive thaw slumps, coastal erosion and drained lake basins, are of a gradual and abrupt nature and range from the plot scale to the landscape scale, which is a regional to circumpolar phenomenon . Warming induces the thawing of ice-rich permafrost landscapes, triggering the release of critical greenhouse gases such as methane and carbon dioxide into the atmosphere through decomposition . Additionally, warming permafrost triggers the formation of characteristic landforms (i.e. polygon-shaped features) and can be defined as a process of the disturbance of the thermal equilibrium of the ground . Closed depressions filled with water, formed by the settling of the ground caused by the thermal-induced process, are called thermokarst lakes, and their formation and drainage cycles can span thousands of years. These are common features covering 50 % to 75 % of permafrost lowlands in parts of Alaska, Siberia and Canada . The distribution of thermokarst lakes, drained lake basins (DLBs) and the associated cycles (formation and drainage) differs between areas of different ground-ice content and between regions (subarctic, Arctic and high Arctic). The resulting mosaic provides a unique habitat for both flora and fauna. The lakes are crucial for wildlife and reindeer herders, both when full and drained . When full, they provide essential water sources and support diverse ecosystems . However, late freezing of lakes can delay reindeer migration, causing herders to move reindeer to winter pastures later than in previous decades . Early snowmelt and unsafe ice cover can further complicate migration and increase the risk of avalanches and floods. When drained, the lakes create new grazing areas with abundant and diverse vegetation, attracting reindeer in the summer. This vegetation succession offers rich pastures but may pose health risks to reindeer due to unfamiliar plant species . Arctic landscapes are widely influenced by DLBs and their role in the geomorphological, hydrological and ecological development in these regions . The drainage of thermokarst lakes is a long-existing phenomenon in many lowland permafrost regions prone to thermokarst landforms, with DLBs of several thousand years of age present among existing lakes and recently drained basins. Studies suggest that the frequency of drainage events has been increasing with climate change .

Lake changes and drainage processes can be observed using remote sensing methods. For example, the change in water area due to shrinking, expanding and disappearing of the surface water has been observed using a multitude of different satellite-based sensors as multiple studies showed . Monitoring lake drainage requires separation of open water from land area. Detecting previously drained lakes requires spatial data analyses to identify relevant depressions and gain knowledge about land surface hydrology changes and vegetation succession trajectories. Disappearing lakes were previously reported specifically for the southern part . Central Yamal is known for rising temperatures and changes associated with unusually warm summers (thaw slumps, active layer deepening, etc.; ). The whole region has been shown to be a hot spot of thaw-lake change .

The land cover in lake basins changes due to the decrease in open water area and the establishment of vegetation communities. Drained lake basins become sites for plant communities with high biomass and productivity, peat accumulation, and permafrost aggradation, and they are critical areas for the carbon cycle . Permafrost thaw often increases soil moisture and lake extent, generally enhancing CH₄ emissions . In contrast, lake drainage significantly alters permafrost and greenhouse gas emissions by reducing soil moisture and changing hydrology, leading to decreased CH₄ emissions . However, predicting CO₂ and CH₄ fluxes is complex due to spatial variability in vegetation, soil carbon stocks and geomorphology across the permafrost domain. Additionally, distinguishing between wetlands and lakes in remote sensing data remains challenging, risking double counting of emission sources . and have used the Normalized Difference Vegetation Index (NDVI) to detect vegetation productivity and changes in DLBs. This popular spectral index is calculated by using the ratios between the red and near-infrared parts of the electromagnetic spectrum. An analysis of land cover succession in recently drained lake basins has not been addressed yet. One of the reasons is that the generation of land cover data requires a specialized approach tailored to the unique characteristics of Arctic landscapes and vegetation communities . Land cover classifications in permafrost regions are used, especially for upscaling of soil properties and fluxes . The Circumpolar Arctic Vegetation Map (CAVM; ) provides detailed information for vegetation communities but at a comparably coarse spatial resolution (1 km). presented a novel approach of land cover characterization (10 m nominal resolution), with representation of moisture gradients and vegetation physiognomy, making it uniquely suitable for the analysis of DLBs. Actual land cover information can be potentially used for flux upscaling and studies that focus on habitat changes. The approach depends, however, on the fusion of Sentinel-1 and Sentinel-2 data, both Copernicus satellite missions, with availability limited to 2016 onwards. Several regions with frequent lake drainage have been identified previously. For example, investigated regions in Alaska, Eastern Siberia and Western Siberia. The last region includes the Yamal Peninsula, which is subject to a multitude of different environmental changes over the past 50 years, caused by both climate change and direct impacts of anthropogenic activity . Abundant lakes and DLBs in parts of Yamal create terrain which is important both as habitat and unique grazing ground for the local reindeer-herding community. Changes in the landscape influence those habitats and affect the local biodiversity .

Overall, NDVI analyses have been shown to provide valuable information on vegetation recovery after lake drainage. However, an approach representing changes in wetness is lacking. New land cover description schemes using recent satellite observations can provide relevant information and may thus complement NDVI analyses.

The aim of this study was to investigate the impacts of lake drainage on land cover change across the Yamal Peninsula, specifically focusing on the change in wetness gradients and vegetation composition. As vegetation growth is climate dependent, different bioclimate subzones were considered.

focused on a static land cover product based on Sentinel-1 and Sentinel-2, targeting consistency at the circumpolar scale. Here, we focused on time series of selected DLBs for 2016 onwards. To tackle the issue of data availability in the years before, a space-for-time concept was applied. In a first step, a database of recently drained lakes was created based on the Landsat trend product . The trends were used for identification of potentially drained lakes before the availability of Sentinel-1 and Sentinel-2, between 1999 and 2014, to facilitate the space-for-time assessment. The basin ages were determined using the change in the water area and the corresponding land cover year. In a second step, cloud-free Sentinel-2 images were used to identify drainage events from 2016 onwards and the calculation for the corresponding basin ages for events after 2015 (basin ages younger than 2015) until the year 2022. Land cover units were derived using Sentinel-1 and Sentinel-2 data, adapting the method described by . The land cover units were grouped by vegetation type as well as wetness gradients. In situ data collected within DLBs were utilized to justify the grouping of the wetness gradients. NDVI was obtained from Sentinel-2 in addition. In a final step, the land cover and NDVI data were combined with the drainage age database covering the years before 2015 to enable the space-for-time assessment.

We quantified the succession of land cover and consistent patterns in proportion of moisture groups during the years after lake drainage. In addition, our results demonstrate the capability of the used land cover unit approach to capture relevant post-drainage land cover change processes on the basin scale. Furthermore, we could show that the detected changes cannot be identified with an analysis focusing solely on NDVI time series. Additional information on succession stages and landscape change trajectories could be derived using the land cover units. For example, changes in land cover dominated by “Wetlands” or “Lichen and Moss” are not represented by the changes in NDVI over time (see Figs.  and ). The NDVI values did not show any correlation with the “Dry” group (R2 below 0.1), implying that an NDVI-only survey on DLBs misses important dryness–wetness information during the first 10 years after a drainage event takes place.

The utility of the consideration of bioclimate subzones for land cover change analyses in DLBs is particularly evident for the wetness gradients “Dry” and “Moist”. A higher fraction for the wetness group “Dry” occurs in the northern bioclimate subzones (B and C) compared to the subzones (D and E) located in the south (Fig. ). The comparatively high fraction of (partially) barren area for all bioclimate subzones and all age groups agrees with previous observations of non-vegetated areas in recently drained basins . Our results also showed that “Barren” areas account for a larger proportion of recently drained area in northern subzones. A higher fraction of “Moist” conditions was found in the basins in the south of the peninsula (Fig. ). concluded that the productivity of individual ecosystems can be limited by nutrient availability. We hypothesize that lakes in the south might have higher nutrition-rich lake sediment amounts compared to the north. This influences the colonizing of the basin in addition to different climate variables. Our results underline that the vegetation colonizing the basins exhibited differences between the southern bioclimate subzones (D and E) and the northern subzones (B and C) (Fig. ). This agrees with previous in situ studies which have shown a strong dependence of vegetation establishment on local climate and other factors such as differences in local species abundance . The land cover product is not designed to differentiate between various plant species. The grouping into the wetness gradients “Dry”, “Moist” and “Wet” is supported by field vegetation data, despite limitations. Enhanced precision in these comparisons could be achieved through increased in situ data availability. Additionally, distributing sample plots evenly across the basin, rather than along a transect with varying intervals, may improve the analysis. We suggest one plot on a 10×10 m scale to cover each pixel from the land cover data product. However, the wetness separation is further justified by indicator plants for our in situ data (see Fig.  and Table ). For example, horsetail Equisetum occupies the ecotope with the lowest soil fertility in wet areas with sandy sediments. Moist ecotopes are occupied by sedge meadows (Carex) in the elevated parts as also reported by .

DLBs have been previously found to be sites of wetland formation ; however, they have been mostly described as wetlands in older basins on centennial or millennial timescales. Following drainage, land cover of the area has a tendency to become more heterogeneous over time (Fig. ). Our study showed a higher-than-average percentage of “Shallow water” (no. 2) and “Permanent wetland” area (no. 3) for DLBs of all age groups compared to the percentage of the same land cover unit in the respective bioclimate subzones (Fig. ). Small wetlands formed after the drainage in the basin (Fig. ). They might be covered with a thick (exceeding 10 cm) moss–grass layer and are located in wet depressions . Methane fluxes differ between lakes, wet ecotopes and upland tundra (e.g. ). Therefore, a change in flux patterns can be expected after lake drainage, but the overall magnitude remains to be quantified.

We observed a higher ratio for “Dry to moist tundra, partially barren” (no. 15) and “Moist tundra, abundant forbs, dwarf to tall shrubs” (no. 16), which indicates the successive colonization of the basin by vegetation during the following years. The presence of “Partially barren” (no. 19) indicates that the area is still disturbed, and vegetation succession is still an ongoing process after 10 years of drainage for the northern subzones compared to the southern part of the peninsula where barren areas are less present in the basin. DLBs are known to have dry to aquatic tundra areas with the distributions of these different conditions depending on a multitude of factors such as time passed since drainage, local climate and basin topography . In particular, the high fraction of “Dry to aquatic tundra, dwarf shrubs” (land cover unit no. 8) compared to the surrounding average of that unit might indicate a recently drained basin for the bioclimate subzones C, D and E within the first 10 years (Fig. ).

Drainage events frequently lead to only partial drainage, causing water areas to persist within the drained basin. Remaining “Water” fraction in our study sites dropped below 1 % after 10 years for all bioclimate subzones with water area decreasing over time after drainage. Our results show that more water remained in lakes after drainage in southern subzones within the first 10 years. This can, however, be biased since additional factors may also contribute to further changes: inundation related to floodplain dynamics or basin depth. Due to the lack of related data, we approach the interpretation of water fractions cautiously. Previously published work has shown a continued change in the open water fraction within DLBs over time, but it might vary for different basin age stages, confirming . This trend may reverse in time. For older basins, an increase in open water fraction in the form of secondary lake development has been reported .

The difference between DLBs and their surrounding area in land cover unit distribution is evident for all bioclimate subzones (Fig. ). This supports the generally held assumption that lake drainage introduces land cover heterogeneity into the landscape . used Advanced Very High Resolution Radiometer (AVHRR) surface temperature data from 1982 to 2003 for determination of the mean July temperature values to derive the bioclimate subzones. Due to the changing climate, those mean values are not true any longer (compare with Fig. ). This is supported by recent studies; for example, showed that the July isotherms of 6 °C are changing northwards. This may have an impact on zone-specific results, and updated boundaries should be considered for future studies.

The dependence of surface composition in DLBs on time passed since drainage is a phenomenon which has been observed for permafrost regions across the Arctic . Previous research focused on C-band SAR data to distinguish between wetness gradients . It was highlighted that data availability was a problem and that their approach works only when there is no scattering, e.g. due to tree trunks. The use of HH (horizontally emitted and horizontally received) polarization is beneficial in this case, but it is of limited availability from Sentinel-1 across the terrestrial Arctic. In our land cover approach, we combine two different satellite sensors (SAR and multi-spectral data) to address this issue. The analysis presented here is limited to a time period when satellite data were available and focuses only on a specific region. Longer time series and extending the study area would allow for an analysis stretching beyond the 10-year mark, leading to additional insights into land cover change trajectories within DLBs. Increasing the number of study sites would provide additional insights for these trajectories over a larger number of diverse basins. The timing of a lake drainage event was determined through visual inspection of available satellite imagery. This can lead to inconsistencies, and an automatic approach would need to be considered when expanding the study area to a circumpolar scale. Our approach focuses on DLBs on the Yamal Peninsula that drained recently (between 1998–2021). interpreted trends of indices for change from 1999–2014, and identified DLBs on the North Slope of Alaska, but the question of when the lakes drained (abrupt change) remains. Geological records showed that abrupt changes in the earth system can arise from slow changes in one component . Palaeorecords would be needed to understand the full cycles of DLBs on longer timescales.

To constrain the analysis and to better enable surface change assessment, basins were carefully selected, omitting those which were located in active floodplains or drained due to suspected anthropogenic influence. To further study the complex dynamics of post drainage land cover development, basins of different characteristics, including those connected to floodplains and with partial drainage of less than 50 %, should be considered.

Improving our knowledge of processes following lake drainage events in Arctic environments associated with permafrost conditions and surrounding terrain is crucial for climate change impact assessments. Understanding the landscape changes associated with drainage, as described quantitatively through changes in land cover units in this study, is needed for representing consequences of permafrost thaw-lake change in land surface and landscape models. These models depend on precise quantitative data. We showed that the land cover succession of recently drained basins follows a certain sequence of land cover changes as the ecosystem transitions from a uniform to a heterogeneous land cover. It differs with location on Yamal Peninsula. The bioclimate subzones (north–south climatic gradient) play an important role in the succession progression. DLBs can be seen as hot spots of greening, but important additional information regarding the development of wetlands was also shown. Lake drainage leads to the development of wet ecotopes, which is reflected in the land cover units but not directly represented through conventional NDVI analyses. Differentiation considering wetness is of relevance for methane fluxes. A realistic representation of wetness gradients in land-surface models may allow for improved descriptions of associated changes in future fluxes. In addition to carbon cycle impacts, the corresponding change in biodiversity needs to be addressed since drainage is a common feature across the entire Arctic. The presented approach can facilitate such investigations. In summary, our results advance the understanding of the development for recently drained basins across a range of bioclimate subzones, drying versus wetting and land cover succession. The corresponding change, considering the different identified trajectories of land cover change for the carbon cycle and species abundance (flora and fauna), remains to be addressed.