A Progressive Noise Removal Method for ICESat-2 Data Based on Terrain Slope Adaptive Calculation

Abstract. The ICESat-2, the satellite-borne photon-counting laser altimeter, has a wide range of applications. However, the data collected by ICESat-2 are often affected by high levels of noise photons. Therefore, the removal of this noise is a crucial step in processing the ICESat-2 data further. This paper proposes a novel noise removal method based on adaptive terrain slope calculation to address this issue. The method takes advantage of the distinct density distribution differences between signal and noise photons in the vertical dimension. By analyzing filtering windows, the algorithm identifies areas with high-density, low-elevation photons and creates a 50-meter elevation buffer around these points to filter out noise photons that are far from the signal clusters. The Douglas-Peucker algorithm is then used to merge data segments with similar slopes, enabling the adaptive calculation of terrain slopes within local photon regions. Furthermore, clustering based on elliptical density along the primary terrain slope direction is applied to remove discrete noise photons that are in close proximity to signal photons. Lastly, the Local Outlier Factor algorithm is utilized to eliminate residual noise photons located below ground level, in aerial regions, and near tree canopies, effectively separating noise photons from signal photons. To evaluate the effectiveness of the proposed method, experimental data sets from two regions with different geographical characteristics in the United States are selected for testing. The results show an average improvement in F1-score ofabout 4.6% in gentle terrains and 9.5% in rugged terrains, highlighting the method's superior accuracy and efficiency compared totraditional denoising algorithms.