Quantifying Understory Vegetation Cover of Pinus massoniana Forest in Hilly Region of South China by Combined Near-Ground Active and Passive Remote Sensing

Abstract

Understory vegetation cover is an important indicator of forest health, and it can also be used as a proxy in the exploration of soil erosion dynamics. Therefore, quantifying the understory vegetation cover in hilly areas in southern China is crucial for facilitating the development of strategies to address local soil erosion. Nevertheless, a multi-source data synergy has not been fully revealed in the remote sensing data quantifying understory vegetation in this region; this issue can be attributed to an insufficient match between the point cloud 3D data obtained from active and passive remote sensing systems and the UAV orthophotos, culminating in an abundance of understory vegetation information not being represented in two dimensions. In this study, we proposed a method that combines the UAV orthophoto and airborne LiDAR data to detect the understory vegetation. Firstly, to enhance the characterization of understory vegetation, the point CNN model was used to decompose the three-dimensional structure of the pinus massoniana forest. Secondly, the point cloud was projected onto the UAV image using the point cloud back-projection algorithm. Finally, understory vegetation cover was estimated using a synthetic dataset. Canopy closure was divided into two categories: low and high canopy cover. Slopes were divided into three categories: gentle slopes, inclined slopes, and steep slopes. To clearly elucidate the influence of canopy closure and slope on the remote sensing estimation of understory vegetation coverage, the accuracy for each category was compared. The results show that the overall accuracy of the point CNN model to separate the three-dimensional structure of the pinus massoniana forest was 74%, which met the accuracy requirement of enhancing the understory vegetation. This method was able to obtain the understory vegetation cover more accurately at a low canopy closure level (Rlow2 = 0.778, RMSElow = 0.068) than at a high canopy closure level (RHigh2 = 0.682, RMSEHigh = 0.172). The method could also obtain high accuracy in version results with R2 values of 0.875, 0.807, and 0.704, as well as RMSE of 0.065, 0.106, and 0.149 for gentle slopes, inclined slopes, and steep slopes, respectively. The methods proposed in this study could provide technical support for UAV remote sensing surveys of understory vegetation in the southern hilly areas of China.