Urban building height mapping in Beijing-Tianjin-Hebei megalopolis by synergy of spaceborne LiDAR and multisource geospatial datasets

Building height is a critical parameter for characterizing urban three-dimensional (3D) structure, providing valuable insights into rapid urbanization and its impacts on resource consumption, urban climate, carbon emissions, and population estimation. Although spaceborne photon-counting LiDAR can deliver high-precision building height samples over large areas, its discrete sampling principle necessitates the integration of multi-source data to generate continuous wall-to-wall building height information. In this study, we propose a land use type-based modeling framework that couples an XGBoost regressor with residual kriging (XGBoostK) to map building heights. Applied to the Beijing-Tianjin-Hebei megalopolis, this approach successfully produced a 30 m resolution building height map for 2020. Main findings are fourfold: (1) The importance of building height-related explanatory variables varies across different land use types (residential, commercial, industrial, transportation, and public management and service), emphasizing the scientific necessity of land use type-based analysis and modeling. (2) Among mainstream machine learning models (GAM, SVR, RF, and XGBoost), XGBoost achieved the optimal performance for building height prediction, with a lower RMSE of 5.35 m and a shorter runtime of 7.06 s. (3) Incorporating prediction residuals from XGBoost into the XGBoostK framework effectively reduced the mapping RMSE from 5.22 m to 4.32 m. (4) Cross-comparison with very-high-resolution imagery, 3D models, and two existing height layers indicates that our product accurately characterizes building height distribution and exhibits superior accuracy. The proposed workflow provides methodological support for large-scale building height mapping and promotes the in-depth application of building height in urbanization research and associated management practices.