A Novel Hierarchical Denoising Framework for ICESat‐2: Reconstructing High‐Accuracy Lake Level Series and Disentangling Climatic Drivers at Qinghai Lake

Quantifying the interactions between climate change, anthropogenic activities and surface water dynamics is critical for sustainable water resource management. However, bridging the data gap in remote regions remains challenging due to the coarse resolution and signal noise inherent in satellite altimetry. Addressing the specific limitation where conventional DBSCAN clustering fails to distinguish near‐surface noise from primary signals, this study introduces a novel adaptive elevation statistical framework (termed the Ahsjeba algorithm) optimised for ICESat‐2 photon data. Instead of relying on a single‐step process, we developed a hierarchical denoising strategy that integrates coarse filtering with fine‐scale statistical refinement to accurately extract water‐level photons. This approach enabled the reconstruction of a dense, high‐precision time series for Qinghai Lake from 2018 to 2024. Validation against in situ data confirms that this refined framework significantly suppresses noise, elevating the correlation coefficient from 0.88 to 0.99 and reducing the RMSE from 0.12 to 0.02 m, thereby overcoming the precision barriers of traditional methods. Crucially, utilising this high‐quality dataset, we employed a random forest model to disentangle the complex driving factors of lake level fluctuations. The results effectively quantified the contributions of key environmental variables, providing a robust scientific basis for understanding regional hydrological responses to environmental changes. This study offers a transferable solution for precise monitoring in data‐scarce, high‐altitude regions.