Multi-model estimation of wetland vegetation biomass combining UAV LiDAR, hyperspectral, and ZY-1 02E spaceborne 2.5m-fused multispectral data: A case study of Qilihai Wetland, China

Estimating plant biomass in wetland ecosystems using remote sensing and mathematical models is crucial for assessing carbon sequestration potential and for wetland conservation, management, and research. This study proposes a two-step framework for aboveground biomass (AGB) estimation in the Qilihai Wetland by integrating UAV LiDAR, hyperspectral images, and 2.5 m-resolution ZY-1 02E satellite imagery. First, biomass was estimated for seven regions using UAV data by comparing support vector machine (SVM), extreme gradient boosting (XGBoost), gradient boosting machine (GBM), and random forest (RF) models to identify the optimal estimation model. The results showed that RF achieved the highest prediction accuracy (R2 = 0.922), with canopy height, vegetation type, and several narrowband indices identified as key predictors, while SVM performed the poorest (R2 = 0.616). Second, UAV-derived biomass was applied to the satellite image to compare five models (SVM, XGBoost, GBM, RF, and the convolutional neural network (CNN)). CNN achieved the best performance (R2 = 0.806), outperforming RF (R2 = 0.721) and significantly surpassing the accuracy of direct field–satellite modeling (R2 = 0.434). The high-resolution AGB map revealed clear spatial heterogeneity, with higher biomass in mixed regions of Phragmites australis and Typha orientalis and in communities closer to water sources. By integrating the predictive capabilities of CNN with the interpretability of RF, this two-step framework significantly enhanced the robustness and ecological relevance of biomass estimation. The findings underscore the pivotal role of multi-source data fusion in improving the accuracy of wetland AGB estimates. Moreover, the high-resolution biomass distribution map provides essential guidance for the conservation, management, and ecological restoration of the Qilihai Wetland.