ADTM: An Adaptive Dual-Threshold Method for Effective Signal Extraction From ICESat-2 Without Manual Thresholding

ICESat-2 provides high-precision photon data with significant potential for bathymetric mapping in shallow waters. However, extracting seafloor signal photons is challenging due to background photon interference and reliance on manual threshold settings. To address these limitations, this study proposes the Adaptive Dual-Threshold Method (ADTM) for seafloor signal photon extraction, which integrates a grid-based photon counting strategy with adaptive elevation difference thresholds. ADTM effectively separates seafloor signal photons from background photons, improving extraction accuracy. We evaluated ADTM under daytime, nighttime, and complex terrain conditions, comparing results with DEM data. ADTM demonstrated superior performance, withR2values of 0.95, 0.97, and 0.70, respectively, and RMSE values of 1.02 m, 0.62 m, and 1.43 m, outperforming existing methods, including the ATL03 official algorithm, the adaptive elevation difference thresholding (AEDTA), DBSCAN, and OPTICS. Quantitative metrics, including recall, precision, F1-score and runtime further verified its stable and reliable performance across different environments. The method also showed stable performance across various water depth ranges, with low errors in RMSE, MAE, and MRE metrics. Additionally, grid size had negligible impact on extraction results, confirming ADTM’s advantage of not relying on manually set thresholds. For practical application, ADTM was combined with the U-Net deep learning network and Sentinel-2 imagery to estimate the water depth of Anegada Island and Barbuda Island (2019-2023). Results demonstrated high accuracy, withR2values from 0.90 to 0.95 and RMSE ranging from 0.69–0.97 m for Anegada Island, andR2greater than 0.9 with RMSE of 0.98–1.12 m for Barbuda Island. These findings confirm the reliability of ADTM in both photon extraction and water depth estimation.