Abstract: Populus euphratica, the only native tree species along the Tarim River, is ecologically crucial for stabilizing landscapes and conserving biodiversity in arid region. However, its irregular canopy structure and sparse distribution complicate large-scale forest assessments. To overcome this, we developed a two-stage random forest model integrating backpack and unmanned aerial vehicle (UAV) Light Detection and Ranging (LiDAR) data covering 10,143 plots of (30 m × 30 m), satellite imagery, and environmental variables. First, we derived structural attributes: leaf area index (R² = 0.83, RMSE = 0.21), canopy cover (R² = 0.80, RMSE = 0.07), foliage height diversity (R² = 0.76, RMSE = 0.28), and canopy height (R² = 0.67, RMSE = 2.04 m). Spatially, these attributes exhibited a general decreasing trend from the upper to the lower reaches, with reductions in their mean values ranging from 17.8% (canopy height) to 30.7% (foliage height diversity). These were then used to estimate aboveground biomass (R² = 0.78, RMSE = 32.91 Mg ha^-1). Our approach generated the first 30m-resolution continuous maps of Populus euphratica structure and biomass across the 1,321 km Tarim River basin. Total aboveground carbon stock was 73.50 Tg C, with a mean density of 65.93 ± 21.80 Mg C ha^-1. Validation against forest inventory data confirmed superior performance over global biomass products (rRMSE = 36.26%). This study can provide fundamental data and theoretical support for the monitoring, management, and conservation of riparian forests in arid regions.

Key words: Forest aboveground biomass, Structure attributes, Sparse vegetation, LiDAR, Multi-source remote sensing data