Understanding the dynamics of small lakes (<1 km2) on the Qinghai-Tibetan Plateau (QTP) is critical for preserving its fragile ecosystem. However, long-term, high-resolution mapping of these lakes remains challenging. We developed the Swin-Res-Unet deep learning model to extract and analyze small lake variations on the QTP from 1990 to 2020. Quantitative comparisons confirm that our model outperforms the baseline Swin-Unet, achieving an Intersection over Union of 92.4%. This study successfully mapped 213,142 lakes with an overall accuracy of 98.1% and an F1 score of 96.3% across the plateau. Our analysis reveals a significant expansion trend, with lake number and total area increasing by 57% and 33%, respectively, resulting in a net expansion of 13,774 km2. Small lakes dominated the observed changes, particularly in continuous permafrost regions, where rapid increases in lake density coincide with accelerated permafrost degradation. In contrast, discontinuous permafrost zones exhibit stabilization or shrinkage. These contrasting responses suggest a threshold relationship where early-stage degradation promotes lake formation, while deeper thaw and talik penetration enhance drainage to trigger shrinkage or disappearance. This study provides the first long-term, high-resolution assessment of small lake dynamics across the QTP and offers new insights into cryosphere–hydrology interactions in a rapidly warming alpine environment.
Yin et al. (Wed,) studied this question.