Forests are one of the most important natural resources. Understanding the impact of various factors on forest biomass is crucial for future forest spatial structure and management. Constructing biomass models for different forest types can provide scientific basis for the restoration and conservation of forest ecosystems.

This study focuses on seven typical forest types in the Daxing'anling region of Heilongjiang Province, using data from 1 157 monitoring plots in 2015. Sentinel-2 satellite images and digital elevation model (DEM) data provided by the European Space Agency were used to calculate vegetation indices, texture features, slope, and other variables. By integrating remote sensing data with field survey data and climate data, we established generalized least squares (GLS) biomass models and generalized additive models (GAM) for biomass. Ten-fold cross-validation was used, and the models were evaluated using root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE). Additionally, 328 plots resurveyed in 2020 were used for model validation.

The Generalized additive models (GAM) performed better than the Generalized Least Squares (GLS) models across the seven typical forest types. Specifically, the mean absolute error (MAE) of the GAM was reduced by 1.99% to 27.48% compared to the GLS models, the root mean square error (RMSE) was reduced by 4.29% to 20.87%, and the mean square error (MSE) was reduced by 6.72% to 35.43%. Secondary validation results showed that the prediction accuracy of the generalized additive model (GAM) for each forest type is above 80%.

Generalized additive models are a non-parametric method for constructing biomass models and are suitable for predicting biomass across different forest types in the Daxing'anling region.