An ICESat-2 photon cloud classification model coupling slope and complex canopy structure in forest areas

Abstract

The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) is widely used in forest ecosystem research. Photon cloud classification is crucial for ICESat-2's application in sub-canopy topography and forest structure parameter estimation. However, steep topography, complex canopy structure, and dense canopy cover are important error factors affecting the results of photon cloud classification in forest areas. The existing basic photon cloud classification methods classify ground and canopy photons based on the spatial distribution of photons in the elevation direction. However, in areas of steep topography, it is difficult for the existing photon cloud classification methods to distinguish ground photons from canopy photons because ground photons and some canopy photons will have the same elevation, making their characteristics unclear. In addition, the complex canopy structure and dense canopy cover cause the distribution of photons in the elevation direction to have complex multi-peak characteristics, increasing the difficulty of distinguishing ground photons from canopy photons. In this paper, we propose a novel photon cloud classification method coupling slope and complex canopy structure to account for the abovementioned error factors in photon cloud classification. The proposed model describes the spatial distribution of photons under various slopes and canopy structures based on the probability distribution function (PDF) of the photon cloud elevation. The proposed model generates PDFs with simple or complex canopy structures under every slope to characterize the photons' spatial distribution in the elevation direction. A slope lookup table is then introduced to find the most appropriate PDF based on the constructed essential criteria with physical meaning. Finally, the coupling of the proposed model is solved by the most appropriate PDF, and the slope and photon cloud classification results can be obtained. The proposed model was tested in 130 forest plots covering various topographies and canopy structures. The results show that the root-mean-square error (RMSE) of the retrieved slopes, the RMSE of the classified ground photons, and the F-value of the ground photons classified by the proposed model reach 1.95°, 1.4 m, and 0.82, respectively. These accuracy indicators illustrate that the proposed model significantly outperforms the existing basic models in the case of steep topography and complex forest structure. This study will substantially improve the accuracy of ground elevation and forest structure parameter estimation through the use of ICESat-2 data worldwide.