Abstract: [Purposes] Accurate building height data are essential for characterizing urban 3D morphology,analyzing spatial structures,and formulating sustainable development strategies.This study aims to address the challenge of fine-scale building height retrieval at the regional level by constructing a high-precision inversion model for the Yangtze River delta urban agglomeration. [Methods] Building height samples are constructed based on GEDI LiDAR data.Multi-source remote sensing data from Landsat 8 and Sentinel-1/2 are integrated to extract spectral,index-based,texture,radar,and statistical features.A random forest algorithm is employed to invert building heights,and the influence of different feature combinations on inversion accuracy is evaluated. [Findings] The combination of spectral,index,texture,radar,and statistical features achieves the highest accuracy at the regional scale (R²=0.740,MAE=6.134 m,RMSE=8.651 m,MSE=74.835).The model demonstrated strong predictive performance across the Yangtze River delta (R²=0.680,MAE=5.217 m,RMSE=6.583 m,MSE=43.332),indicating good adaptability and robustness.Spatially,building height exhibits a “high core-low periphery” pattern and shows significant correlations with urban development indicators,with the total retail sales of consumer goods being most strongly related,reflecting the intrinsic law of urban spatial intensification. [Conclusions] The study confirms the effectiveness of multi-source remote sensing feature fusion for regional-scale building height inversion and highlights model performance differences across regions with varying development characteristics,providing a valuable technical reference for sustainable urban planning and 3D spatial analysis.

Key words: building height, GEDI, multi-source remote sensing, random forest, Yangtze River delta urban agglomeration