Estimation of canopy bulk density for modeling fire behavior in boreal forest of Northeastern China using forest inventory and multisource satellite remote sensing data

Canopy bulk density (CBD) data is essential for modeling the behavior of crown fires. Few studies generated spatially continuous CBD maps at landscape scale in China. This study developed a framework to estimate CBD using Chinese forest inventory data and multisource remote sensing imagery. We devised an indirect method for calculating plot level CBD based on the load-over-depth concept, integrating Chinese National Forest Inventory plots data and species-specific allometric equations. Pearson correlation and recursive feature elimination were used to identify optimal predictors for ensemble learning algorithms, including Random Forest (RF), Gradient Boosting Decision Tree, and eXtreme Gradient Boosting. RF demonstrated the highest performance among these algorithms. The R 2 for the model were 0.67 and 0.57, respectively, and the RMSE values were 0.0219 kg/m 3 and 0.0269 kg/m 3 . The final 10-m resolution CBD map generated by the RF model for the study area, which revealed values ranging from 0.0180 to 0.1476 kg/m 3 , with a mean of 0.0808 kg/m 3 . Pixel-level uncertainty in spatial extrapolation was also quantified based on the RMSE and rRMSE. Additionally, explainable analysis identified the contribution of each selected feature. The accuracy of the resulting maps can serve as a baseline for the Chinese boreal forest.