Progressive noise photons removal from ICESAT-2 data based on the characteristics of different types of noise

Removing noise photons from Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) data is crucial for various applications of the photon-counting LiDAR system. Existing methods for noise photon removal often struggle with parameter tuning, lack robustness, and may compromise accuracy across different datasets. To address these issues, this study proposes an innovative progressive noise removal method. Unlike conventional approaches that treat all noise photons uniformly, our method first categorizes noise photons into isolated, low-density clustered, and outer clustered types based on their unique spatial distribution characteristics. Each type is then targeted with specific denoising techniques, resulting in higher denoising efficiency and better signal photon preservation. Specifically, isolated noise photons are automatically identified using a multi-thresholding strategy based on the maximum between-clustering variance algorithm without requiring parameter tuning. Low-density clustered noise photons are removed using the ellipse-based photon counting method, where the Douglas-Peucker algorithm is utilized to align the ellipse’s major axis with the locally calculated terrain slope. Outer clustered noise photons are also automatically detected through a box plots analysis technique based on local elevation distributions. The efficacy of the proposed method was evaluated using diverse datasets containing strong and weak signals, as well as various land covers. Experimental results demonstrate that the proposed method outperformed five traditional denoising methods in terms of both denoising effectiveness and signal photon fidelity. Furthermore, testing on datasets with diverse land covers showcased the robustness of the proposed method.