Mapping global lake dynamics reveals the emerging roles of small lakes

Lakes are vital components of global hydrological and biogeochemical cycles, supporting diverse ecosystem services. However, they are experiencing rapid and widespread changes due to climate change and human activities. These alterations threaten ecosystem services and have far-reaching consequences, including lake desiccation leading to water shortages and international conflicts, and glacier lake expansions caused by snowmelt and glacial melting. A spatially explicit understanding of lake size changes is crucial for assessing these impacts. While satellite imagery offers a potential tool for tracking lake dynamics, the optical similarity between lakes and rivers poses a challenge in accurate differentiation. Existing global lake datasets often rely on snapshots of historical imagery, neglecting seasonal and interannual fluctuations and primarily focusing on larger lakes. Small lakes (<1 km²), despite making up a relatively small portion of the total lake area, are disproportionately important due to their high sensitivity to environmental changes and their outsized contributions to primary productivity, biodiversity, and carbon cycling. However, their size and distribution are often poorly estimated, impacting assessments of ecosystem parameters and biogeochemical cycles. This study addresses this gap by mapping over 3.4 million lakes globally, using deep learning to improve the detection of smaller lakes and analyzing lake area changes over four decades to better understand their contribution to lacustrine carbon emissions.

Several existing datasets provide estimates of global lake extent, but these often rely on single time points and lack detailed information on long-term changes. Previous studies have primarily concentrated on medium-to-large lakes, overlooking the significant role of small lakes. Research has highlighted the importance of small lakes in various ecosystem processes, such as primary productivity, biodiversity, and the carbon cycle, but these assessments often rely on uncertain estimations of lake size and distribution. There is a lack of global-scale studies characterizing lake size changes over time, limiting our understanding of their influence on freshwater biogeochemical cycles. This study builds upon existing research by providing a more comprehensive and detailed assessment of global lake dynamics, particularly considering the role of smaller lakes, and linking these changes to carbon emissions.

The study developed a new global lake dataset, GLAKES, based on the Global Surface Water Occurrence (GSWO) dataset and a deep-learning classification algorithm. GSWO provides the probability of water presence using Landsat satellite observations (1984-2019). A modified U-Net deep learning model was used to differentiate lakes from rivers in GSWO images, allowing for the detection of lakes as small as 0.03 km². The model was trained using a dataset of manually labeled lake polygons created from GSWO, Global River Widths from Landsat (GRWL), OpenStreetMap Water Layer (OSMWL), and HydroLAKES datasets. Two separate models were trained: one for normal lakes and one for floodplains. The models were validated using independent test sets, achieving high accuracy levels (>98.7% overall accuracy and >88.7% MIoU). The resulting GLAKES dataset includes the maximum extents of over 3.4 million lakes. Lake size changes were analyzed across three periods (1984–1999, 2000–2009, 2010–2019) by comparing water probability-weighted areas within lake boundaries. Glacier-fed, permafrost-fed, and reservoir lakes were identified using additional datasets. Global lake carbon emissions were estimated using a previously established method, incorporating the GLAKES data to improve accuracy. Finally, this study analyzed the relationship between lake area changes and population density using the Gridded Population of the World (GPW) dataset.

GLAKES dataset revealed a total lake area of 3.2 × 10⁶ km² (2.2% of global land area). Lakes showed a net increase of +46,278 km² between the 1980s and 2010s, with 56% attributed to reservoirs. Small lakes (<1 km²) showed higher temporal variability in size than larger lakes. Small lakes contributed disproportionately to global lake expansion (46.2% of net areal increase). In half of the global inland lake regions, small lakes dominated the variability in total lake size. Updated calculations of CO₂ and CH₄ emissions, utilizing GLAKES, resulted in lower estimates than previous studies, with small lakes contributing 25% and 37% respectively to total emissions. Small lakes contributed 45% and 59% to net increases in lake CO₂ and CH₄ emissions from the 1980s to the 2010s. Decadal lake variations generally increased with regional population density, with this effect being greater for small lakes. While climate warming-induced lake expansion in glacial and permafrost regions was observed, it was not the primary driver of global lake area changes. A net loss trend in lake area within endorheic basins was observed, but this trend reversed when the Aral Sea was excluded.

This study's findings address the research question by quantifying the global dynamics of lakes over four decades, emphasizing the importance of small lakes. The results demonstrate the substantial contribution of human-made reservoirs to global lake expansion, highlighting the significant impact of human activities on global water resources. The disproportionate role of small lakes in driving lake size variability and carbon emissions underscores their crucial ecological role and suggests the need for more focused management strategies. The updated estimates of lacustrine carbon emissions, utilizing improved lake size data, provide a more accurate assessment of their contribution to global carbon budgets. The observed relationships between lake area changes and population density indicate potential interactions between human activities and lake dynamics, warranting further investigation. The study's findings are highly relevant to the field by offering a more comprehensive and accurate understanding of global lake dynamics and their implications for water resources and carbon emissions. This enhances our ability to better predict and manage the impacts of climate change and human activities on lake ecosystems.

This study presents the GLAKES dataset, a high-resolution, global-scale map of lake extents spanning four decades. The findings reveal the dominant role of reservoirs and small lakes in shaping global lake dynamics and carbon emissions. This underscores the need for improved monitoring and management strategies that consider the disproportionate influence of small lakes. Future research should focus on refining the estimation of carbon emissions from smaller lakes, further investigating the interactions between human activities and lake dynamics, and exploring the potential impacts of lake changes on various ecosystem services.

Several limitations should be considered. The accuracy of lake mapping is affected by factors including lake definition, auxiliary datasets, the limitations of the deep learning model, and the post-processing procedures. The probability-weighted lake areas used may not fully capture the complexities of seasonal lake dynamics. The accuracy of carbon emission estimations depends on the representativeness of mean flux estimates and may not fully capture the impacts of short-term changes and emissions from other pathways. Additionally, the inclusion of some agricultural fields or floodplains in the lake dataset may introduce some uncertainty.