Harmonizing remote sensing and ground data for forest aboveground biomass estimation

Accurate aboveground biomass (AGB) estimation is crucial for evaluating management and conservation policy of forests. However, the complexity of forest ecosystems and the diversity of geography bring great challenges to traditional biomass estimation methods. This study aims to develop an optimized AGB estimation framework that integrates heterogeneous data sources (i.e., ground survey data, National Forest Continuous Inventory (NFCI) data, and both active and passive remote sensing data) to enhance estimation accuracy and address the needs of future satellite missions and forest monitoring efforts. Using Longyan City, Fujian Province, China, as a case study, we construct a machine learning-based AGB estimation framework and generate high-resolution AGB spatial distribution maps through stepwise variable selection, hyperparameter optimization, and incremental integration of data sources. The effectiveness of this approach was demonstrated by a 0.67 increase in the correlation coefficient R 2 , a 43.57 % reduction in the root mean square error (RMSE), and a 68.00 % reduction in the mean square error (MSE) achieved through the optimal combination of data sources. The optimization framework not only significantly improves AGB estimation accuracy but also facilitates the identification of key areas for afforestation through the generated spatial distribution map, offering a scientific foundation for targeted forest management and ecological restoration. This study highlights the potential of combining heterogeneous data sources with machine learning techniques, providing a scalable solution for future forest monitoring tasks. • Integrating the data from active remote sensing, passive remote sensing, and ground survey can help to refine the estimation accuracy of aboveground biomass (AGB). • The estimation accuracies for different data source combinations yielded correlation coefficient R 2 ranging from 0.02 to 0.69, root mean square errors (RMSE) from 59.83 to 33.77 t/ha, and mean square errors (MSE) from 3590.71 to 1148.91 t/ha. • An optimized estimation framework that effectively enhances accuracy includes data preprocessing, redundant variable elimination, model hyperparameter adjustment, the gradual integration of diverse data sources, and the generation of spatial distribution maps.