Optimizing ground photons for canopy height extraction from ICESat-2 data in mountainous dense forests
Ruiqi Zhao,Wenjian Ni,Zhiyu Zhang,Huabing Dai,Chengling Yang,Zhen Li,Yao Liang,Qingwang Liu,Yong Pang,Zengyuan Li,Guoqing Sun
DOI: https://doi.org/10.1016/j.rse.2023.113851
IF: 13.5
2023-10-15
Remote Sensing of Environment
Abstract:The ICESat-2 data products work well in extracting terrain elevations and canopy heights in boreal forests. However, identifying ground photons in mountainous dense forests remains a challenge due to high canopy covers, complex terrains, and limited penetrability, particularly in tropical areas. Accurate identification of ground photons is crucial for improving the performance of ICESat-2 data. This study proposes an algorithm called OPIC to optimize ground photons provided by ICESat-2 data products. The OPIC algorithm involves removing outliers of ground photons, reclassifying photons misclassified as canopies or noises, and extracting optimized ground elevations. The OPIC algorithm is evaluated in four distinct forests located in subtropical southern China, tropical Indonesia, temperate Estonia and tropical America. The results show significant improvements in the number of valid segments and the accuracy of canopy height estimations. The ICESat-2 product of Land and Vegetation Height (ATL08) provides canopy heights by segments with an along-track distance resolution of 20 meters. In forests with canopy covers ≥85% on terrains with slopes ≥15°, the ATL08 is only able to provide canopy height estimations for 10 out of 368 segments (i.e., 2.7%) at the Chinese test site, but the OPIC algorithm improves this to 157 segments (i.e., 42.7%). At the Indonesian test site, the OPIC algorithm improves the number of valid segments from 11 to 369 out of 453 segments (i.e., from 2.4% to 81.5%). Similarly, the OPIC algorithm improves the number of valid segments from 353 to 1148 out of 2776 segments (i.e., from 12.7% to 41.4%) at the American test site. The airborne lidar data is utilized as reference data to evaluate the performance of OPIC. For valid segments provided in ATL08, the OPIC can reduce the relative root mean square error (%RMSE) of canopy height estimations from 31.64% m to 24.97% at the Chinese site, from 25.74% to 18.84% at the Indonesian site, from 7.27% to 2.47% at the Estonian site, and from 27.97% to 23.87% at the American site. For all segments provided by ATL08 in each test site disregarding terrains and canopy covers, the OPIC algorithm is able to increase the number of valid segments from 383 to 1244 out of 2252 (i.e., from 17.0% to 55.2%) at the Chinese site, from 173 to 1276 out of 1482 (i.e., from 11.7% to 86.1%) at the Indonesian site, from 8638 to 10,149 out of 10,344 (i.e., from 83.5% to 98.1%) at the Estonian site, and from 6145 to 9383 out of 16,150 (i.e., from 38.0% to 58.1%) at the American site, while maintaining approximately the same estimation accuracy of canopy heights as the valid segments of ICESat-2 data products. These results demonstrate the potential of the OPIC algorithm to provide more valid segments and more accurate estimations of canopy heights in mountainous dense forests.
environmental sciences,imaging science & photographic technology,remote sensing