New Accuracy Indicators for Assessing ICESat-2 Photon Cloud Filtering in Forested Areas

Accurate photon cloud filtering is a critical preprocessing step for deriving forest parameters from ICESat-2 data. As filtering quality directly affects the reliability of subsequent applications, selecting an appropriate filtering method for a forest area is essential. Therefore, it is necessary to objectively and accurately evaluate filtering performance before further applying ICESat-2 data. However, conventional filtering accuracy indicators such as recall, precision, and F-score often fall short in objectively reflecting filtering performance, as they do not adequately capture the preservation of ground and canopy photons, nor the resulting terrain and canopy height accuracy. To address these limitations, this study proposed four new filtering accuracy indicators tailored to forest environments: Ground Signal Retention Rate (GSRR), Ground Noise Residual Rate (GNRR), Canopy Signal Retention Rate (CSRR), and Canopy Noise Residual Rate (CNRR). These indicators enable a more detailed and application-oriented evaluation of filtering performance across both ground and canopy layers. Five commonly used filtering algorithms were assessed across six representative test datasets with varying canopy cover and terrain complexity. Experimental results show that the proposed indicators exhibit a strong and statistically significant correlation with terrain RMSE, withGSRRachieving a correlation coefficient of −0.93 and ap-value below 0.001, whereas the conventional indicators exhibit absolute correlation coefficients below 0.40 andp-values greater than 0.05.