Integration of Optical Remote Sensing and Laser Point Cloud for Forest Stock Estimation in Karst Mountainous Areas

Abstract

This study addresses the challenges posed by the complex topography and forest structure in karst mountainous areas, as well as the difficulties in estimating forest stock using traditional methods. We propose a method that integrates optical remote sensing data from Sentinel-2 into airborne LiDAR data to estimate forest stock in karst areas. First, an Allometric Growth Model correlating tree height and diameter at breast height (DBH) in karst areas was developed based on field measurements. Tree height information extracted from LiDAR data was then combined with the binary wood volume model specific to fir trees in Guizhou Province to calculate the individual tree biomass of fir trees. In addition, this study evaluated the robustness of three machine learning methods, the Random Forest Regression Model, K-Nearest Neighbors Regression Model, and Backpropagation Neural Network Model, in estimating forest stock in karst mountainous areas. The results indicate the following: (1) The Allometric Growth Model based on field data showed strong predictive power for DBH and can be used for large-scale estimation. (2) The distribution characteristics of individual tree biomass and plot biomass under different site conditions revealed the distribution pattern of fir trees in the study area, providing important information for understanding the growth status of forest stock in the region. (3) The Random Forest Regression Model demonstrated exceptional accuracy, generalization capability, and robustness in the estimation of forest stock within karst mountainous regions. This study provides an effective technical tool for estimating forest stock in karst areas and under complex terrain conditions and has significant scientific value and practical implications for the monitoring and management of forest ecosystem carbon sinks.