Welcome to visit Scientia Silvae Sinicae，Today is

Scientia Silvae Sinicae ›› 2026, Vol. 62 ›› Issue (2): 173-185.doi: 10.11707/j.1001-7488.LYKX20240765

• Research papers • Previous Articles     Next Articles

Accurate Monitoring of Sparse Forest Canopy Closure Based on Sentinel-2 and GBRT Model: a Case Study on the Returning Farmland to Forest Project in the Inner Mongolia

Tiancan Wang1,2,3,Gegentana 4,Xiaosong Li1,2,*(),Yuelianggaoke 5,Tong Shen1,2,Chaochao Chen1,2,3,Yubo Zhi1,2,3,Licheng Zhao1,2,Cuicui Ji5   

  1. 1. Aerospace Information Research Institute, Chinese Academy of Sciences　Beijing 100094
    2. International Research Center of Big Data for Sustainable Development Goals　Beijing 100094
    3. University of Chinese Academy of Sciences　Beijing 100101
    4. Forestry and Grassland Work Station of Inner Mongolia　Hohhot 010010
    5. Smart City Academy, Chongqing Jiaotong University　Chongqing 400074
  • Received:2024-12-16 Revised:2025-02-17 Online:2026-02-25 Published:2026-03-04
  • Contact: Xiaosong Li E-mail:lixs@aircas.ac.cn

PDF (PC)

12

Abstract:

Objective: The purpose of this study is to integrate high-resolution UAV data with Sentinel-2 satellite imagery and utilize the Gradient Boosting Regression Tree algorithm to achieve accurate monitoring of sparse forest canopy closure in the Returning Farmland to Forest Project areas, thereby providing technical support for effectiveness assessment of the new round of the project. Method: UAV LiDAR and visible light image data were collected in typical areas of returning farmland to forest. Combined with Sentinel-2 remote sensing images in the growing and non-growing seasons of 2024 and terrain data, a gradient boosting regression tree model was established to estimate the canopy closure of the sparse forest, and its accuracy and discrimination ability were evaluated. Result: With UAV-based LiDAR point cloud and visible light images of 90 open forest plots of land for returning farmland to forests, the canopy height model (CHM) combined with the threshold segmentation method were used to construct 5 764 sparse forest canopy closure sample points. Based on multi-temporal Sentinel-2 remote sensing image features and topographic information, a gradient lifting regression tree model was established to realize the accurate monitoring of sparse forest canopy closure, and the model coefficient of determination (R2) was 0.731, the root mean squared error (RMSE) was 0.028, and the mean absolute error (MAE) was 0.021. The vegetation indexes and reflectance of non-growing seasons and the elevation were the key factors for estimating sparse forest canopy closure. Conclusion: The gradient boosting tree regression model constructed by combining high-precision UAV LiDAR data and Sentinel-2 remote sensing images can better predict the sparse forest canopy closure, and has good stability under the influence of different geographic environments and vegetation types, which is of great significance for the effectiveness assessment of the new round of returning farmland to forest projects in Inner Mongolia.

Key words: Sentinel-2, UAV, returning farmland to forest, Inner Mongolia, sparse vegetation, canopy cover, gradient boosting regression tree

CLC Number: