A Parameter-Adaptive Photon Denoising Algorithm for ICESat-2 Glacier Monitoring in Tibet

Accurate extraction of glacier surface elevation from ICESat-2 photon-counting data is fundamental for monitoring glacial morphology and assessing ice mass balance. However, effectively suppressing noise while retaining signal photons remains a significant challenge in complex alpine environments. To address this issue, we propose a Parameter-Adaptive Photon Denoising (PAPD) algorithm for mountain glacier monitoring. The method incorporates an optimized coarse denoising stage to minimize the risk of erroneous signal rejection and introduces an innovative adaptive elliptical neighborhood mechanism that dynamically adjusts based on local photon distributions, thereby enhancing adaptability to terrain variations. We evaluated the algorithm using four ICESat-2 datasets with varying signal-to-noise ratios (SNR) across the southeastern Tibetan Plateau (SETP). Experimental results based on the reference dataset indicate that PAPD achieves a superior average F1-score of 0.980, effectively outperforming benchmark algorithms including DBSCAN (0.966), RANSAC (0.907), ATL03 (0.883), and Adaptive DBSCAN (0.836). Notably, in regions with low SNR or abrupt topographic changes, the proposed method yields elevation profiles with improved continuity and richer local details. Validation against ATL06 elevation products further corroborates its robustness. These findings suggest that PAPD provides a reliable data foundation for high-resolution glacier elevation series, demonstrating significant potential for precise mass balance estimation in challenging alpine glacier environments.