Recent lake surface dynamics in the Hunshandake sandy land (2017–2022) and their response to climatic factors

Abstract

Lakes, as critical components of the terrestrial water cycle, play an indispensable role in maintaining ecological balance, particularly in arid ecosystems like the Hunshandake (Otindag) Sandy Land (HSDK) of China. Understanding the spatiotemporal dynamics of these lakes is essential for deciphering regional hydrological cycles and predicting their ecological evolution in water-stressed environments. Leveraging the Google Earth Engine (GEE) platform and Sentinel-2 satellite imagery, we mapped the monthly water extents of lakes in the HSDK (2017–2022) at 10 m spatial resolution and analyzed their drivers. Key findings include: (1) Three classification approaches — pixel-based random forest (RF), object-oriented random forest (OB-RF), and support vector machine (SVM) — achieved high accuracy (Overall Accuracy: 98.5 %, 97.4 %, and 98.4 %; Kappa Coefficients: 0.970, 0.946, and 0.967, respectively). Compared with seasonal lakes, permanent lakes exhibited superior extraction accuracy. Notably, the OB-RF method generated clustered artifacts when mapping small fragmented water bodies. (2) The annual maximum lake area in the HSDK fluctuated between 345.61 and 419.42 km² 2017–2022 (mean: 379.55 km²). Though seasonal lakes were more numerous, permanent lakes made up 70 % of the total area. Monthly variations revealed a three-phase pattern: a gradual decline from April to June, a marked expansion in July–September, and subsequent contraction in October. (3) Interannual lake area changes were positively correlated with precipitation (2017–2021, R² = 0.80, p < 0.05), although anomalous expansion occurred in 2022 despite reduced rainfall, suggesting hydrological inertia. At the monthly scale, lake areas exhibited a significant one-month lagged response to precipitation (R² = 0.61, p < 0.001), highlighting delayed hydrological feedback.