Forests, Journal Year: 2025, Volume and Issue: 16(1), P. 125 - 125
Published: Jan. 11, 2025
Forest canopy height (FCH) is a critical parameter for forest management and ecosystem modeling, but there lack of accurate FCH distribution in large areas. To address this issue, study selected Wuyishan National Park China as case to explore the calibration method mapping complex subtropical mountainous region based on ZiYuan-3 (ZY3) stereo imagery limited Unmanned Aerial Vehicle (UAV) LiDAR data. Pearson’s correlation analysis, Categorical Boosting (CatBoost) feature importance causal effect analysis were used examine major factors causing extraction errors digital surface model (DSM) data from ZY3 imagery. Different machine learning algorithms compared calibrate DSM results. The results indicate that accuracy primarily influenced by slope aspect, elevation, vegetation characteristics. These influences particularly notable areas with topography dense coverage. A Bayesian-optimized CatBoost directly calibrating original (the difference between high-precision elevation (DEM) data) demonstrated best prediction performance. This produced map at 4 m spatial resolution, root mean square error (RMSE) was reduced 6.47 initial 3.99 after calibration, relative RMSE (rRMSE) 36.52% 22.53%. demonstrates feasibility using regional confirms superior performance algorithm enhancing accuracy. findings provide valuable insights into multidimensional impacts key environmental extraction, supporting precise monitoring carbon stock assessment terrains regions.
Language: Английский