Integrating ICESat-2 altimetry and machine learning to estimate the seasonal water level and storage variations of national-scale lakes in China

Lakes comprise the largest element of terrestrial surface liquid freshwater bodies, playing an indispensable part in the Earth's water cycle and alleviating floods and droughts. The seasonal variation of lake water level (LWLsv) and storage (LWSsv) reflects the periodic hydrologic fluctuations and related driving forces of the water balances at the basin scale. China, a vastly diverse country that descends from the "Roof of the World" to monsoonal coast zones, hosts a wide distribution of lakes with diverse hydro-climatological and topographic features. Most previous studies focused on monitoring long-term changes of lakes in China, while the seasonality (mostly regarding the lake area) of lake hydrologic dynamics was only investigated for typical lakes or local basins/zones due to the limited spatial coverage and temporal resolution of various observation data. Benefiting from the finer footprints and increased beams of ICESat-2 laser altimeter, we examined the seasonal variations of LWLsv of Chinese lakes (3473 lakes historically larger than 1 km2 during the 1980s–2010s) on a national scale from 2019 to 2021. Then, the machine learning algorithm termed the extreme gradient boosting tree was employed to model the LWLsv of the lakes that were not monitored by ICESat-2. We further quantized the national-scale LWSsv by combining the LWLsv estimates and lake area data. Results show that the mean LWLsv of the 1255 lakes observed by ICESat-2 during 2019–2021 is 0.44 ± 0.07 m. Among them, 1167 lakes have the LWLsv <1 m, and 12 lakes have the LWLsv exceeding 3 m. The accuracy evaluation indicates that the XGBoost model performs well in predicting LWLsv results for unobserved lakes, with the coefficient of determination of 0.76, the mean absolute error of 0.14 m, and the root mean square error of 0.03 m. Overall, the predicted LWLsv of unobserved lakes has a similar spatial pattern to that of the observed lakes. The LWSsv is estimated at 77.29 ± 6.87 Gt in total, but exhibits obvious spatial heterogeneity in China. The Middle and Lower Reaches of Yangtze River Basin and the endorheic Qiangtang Plateau Basin rank the first two contributors of the net LWSsv, which were respectively attributed to the most significant LWLsv changes and the largest lake group area. This national-scale quantification of the LWLsv and LWSsv helps advance our scientific understanding of the seasonal hydrologic behaviors for Chinese lakes in regulating the water cycle and providing a valuable reference for regional water resource management.