Mapping bamboo forest and expansion intensity in China by coupling vegetation phenology and C-band SAR with Sentinel-1 and Sentinel-2 images

Abstract

Bamboo forest has undergone dramatic expansion due to climate changes and human activities, and its direct effects on both carbon storage and biodiversity of the forest ecosystem have occurred in tropical and subtropical regions, especially in China. However, the uncertainty in tracking bamboo forest extent and expansion intensity substantially influenced our assessment of them due to the poor resolution and persistent cloud covers in optical images (e.g., Landsat and Sentinel-2). We developed a straightforward and superior algorithm by coupling vegetation phenology and cloud-free SAR using Sentinel-1 and -2 images to identify bamboo forest at large scales and higher spatial resolution. Specifically, (1) this study analyzed the spectral and phenological characteristics during the bamboo forest growth season; (2) the optimum parameter sets for the SAR backscatter were calibrated against field data by using the genetic algorithm; (3) and then we generated bamboo forest and expansion intensity maps at 10 m spatial resolution for seven study regions in China in 2020. The results showed that the Kappa of the maps was 0.89, and the OA was 94.7% in all study areas, compared with the field data. The accuracy among each study area also performed well in mapping bamboo forest, indicated by Kappa varying from 0.82 to 0.94, and OA ranging from 91.1% to 97.33%. Compared to the national forestry inventory map, our results showed a significant positive linear relationship, with a higher R2 (0.96, p < 0.001). With the high-resolution results, we found that the tree was most severely affected by bamboo forest, followed by grassland, sparse vegetation, and shrubland. This product can be a key input for many carbon cycles, climate, and vegetation models.

Keywords

• Bamboo forest
• Expansion intensity
• Vegetation phenology
• SAR
• Google Earth Engine