Forest Site Quality Evaluation Using UAV Remote Sensing Technology

Accurately assessing the quality of a forested site is essential for sustainable forest management. In forest practices, assessment methods primarily rely on ground meas-urements, but these approaches face challenges such as high costs, low efficiency, and spatial and temporal limitations in data collection. At present, a large number of studies have explored the application of UAV remote sensing technology in forest resource monitoring and have made significant progress in biomass estimation, forest structure analysis and carbon stock assessment. However, existing research still lacks a program to integrate traditional site quality assessment methods with UAV remote sensing data systems. To fill this gap, this paper takes UAV LiDAR acquisition of high-precision point cloud data as the core, and combines UAV data processing with the evaluation model based on potential productivity to realize the accurate extraction and quantitative assessment of stand factors. The study mainly includes three major steps: data acquisition and preprocessing, stand factor extraction, and quantitative assessment of stand quality, and the stand factors include elevation, slope direction, slope gradient, slope position, soil type, and depression. This study was conducted at Chinese fir (Cunninghamia lanceolata) plantations in Guangdong, southern China. High-precision point cloud data were collected from 133 sample plots using drone LiDAR for experimental validation. In this study, the data of 133 sample plots are randomly divided into two groups, one group of 89 sample plots is used to train the model, and one group of 44 sample plots is used for model valida-tion. In particular, the growth models constructed achieved determination coefficients of 0.6745 for stand height, 0.7460 for diameter at breast height, and 0.8071 for volume, indicating high model fitting and demonstrating the method’s good predictive capa-bility. And the validation results of the validation dataset for the model have R2 as low as 0.5634 and as high as 0.7856, with most of them around 0.7, which proves that the model has a fair prediction ability.