Accurate estimation of forest tree species diversity at regional scales is essential for biodiversity monitoring, forest resource management, and ecological conservation. Because tree species differ in canopy spectral responses and phenological dynamics, multi-season remote sensing observations can provide critical information for characterizing species composition and diversity patterns. However, the potential contribution of seasonal image features to improving remote-sensing-based tree species diversity estimation has often been insufficiently considered. In this study, the Yichun forest region in Heilongjiang Province, northeastern China, was selected as the study area. Sentinel-1, Sentinel-2, and topographic data were integrated to extract multi-seasonal spectral, vegetation index, texture, radar, and topographic features. The Boruta algorithm was used for feature selection, and random forest (RF), extreme gradient boosting (XGBoost), k-nearest neighbor (KNN), support vector regression (SVR), Bayesian regularized neural network (BRNN), and Stacking ensemble learning were developed to estimate and map Richness, Shannon, and Gini–Simpson indices. The results showed that: (1) Sentinel-2 optical features were the primary information source for tree species diversity estimation, topographic factors further improved model performance, and Sentinel-1 radar features mainly provided complementary structural information; (2) seasonal remote sensing features differed in their predictive ability, with Richness performing better in spring, while Shannon and Gini–Simpson achieved higher accuracy in winter. The four-season fusion scenario produced the highest accuracy for all three indices, with optimal R² values of 0.51, 0.63, and 0.57, respectively; (3) the Stacking ensemble generally improved estimation accuracy and model stability, although the optimal model differed among diversity indices, with Stacking, SVR, and RF performing best for Richness, Shannon, and Gini–Simpson, respectively; and (4) summer Sentinel-2 NDVI, GNDVI, and NDWI contributed strongly to all three indices, elevation was particularly important for Richness, and winter vegetation indices and autumn red-edge bands and texture features were also informative for Shannon and Gini–Simpson. These findings indicate that integrating multi-seasonal remote sensing features and multi-source data using machine learning models can effectively improve forest tree species diversity estimation, providing technical support for regional forest biodiversity monitoring and precision forest management.

Remote sensing provides an effective means of estimating forest aboveground biomass (AGB) over large areas. However, empirical models calibrated in a specific region may lose predictive accuracy when transferred elsewhere because predictor–AGB relationships can vary with forest structure and environmental conditions. In addition, the complex nonlinear relationships learned by many machine learning models are not explicitly linked to ecological processes, limiting their ecological interpretability. To address these challenges, we developed a height-based allometric framework that incorporates the kernel normalized difference vegetation index (kNDVI) derived from Sentinel-2 imagery and leaf area index (LAI) indirectly estimated from UAV-LiDAR point clouds into the scaling relationship between mean stand height and AGB. By allowing the scaling exponent to vary with canopy spectral and structural conditions, the proposed framework was designed to account for spatial variation in the height–biomass relationship while retaining an explicit allometric structure. Its performance was compared with four commonly used machine learning methods: k-nearest neighbors (kNN), random forest (RF), extreme gradient boosting (XGBoost), and Bayesian Regularized Neural Network (BRNN). Within the calibration region, the proposed allometric model achieved an R² of 0.67, an RMSE of 17.74 Mg ha⁻¹, and an rRMSE of 15.17%, closely approaching the performance of RF, which was the most accurate machine learning model (R² = 0.69, RMSE = 17.20 Mg ha⁻¹, and rRMSE = 14.71%). More pronounced differences emerged when the models were evaluated using independent samples located outside the calibration area. Under this spatial transfer test, the allometric model retained an R² of 0.48, with an RMSE of 22.31 Mg ha⁻¹ and an rRMSE of 20.83%. In comparison, BRNN was the best-performing machine learning method in the same external validation but achieved an R² of 0.32, an RMSE of 25.44 Mg ha⁻¹, and an rRMSE of 23.75%. Introducing either LAI or kNDVI into the scaling exponent provided greater explanatory value than using a fixed exponent, while the kNDVI-based formulation (R² = 0.41, RMSE = 23.60 Mg ha⁻¹, and rRMSE = 22.03%) exhibited stronger spatial transferability than its LAI-based counterpart (R² = 0.37, RMSE = 24.42 Mg ha⁻¹, and rRMSE = 22.80%). Although all evaluated models experienced accuracy losses outside the calibration region, the reduction was substantially greater for the machine learning approaches. The proposed framework therefore provides predictive accuracy comparable to that of machine learning models within the calibration area while offering greater ecological interpretability and stronger performance under spatial transfer. This approach provides a basis for large-area forest AGB estimation by integrating optical observations with canopy structure information and could potentially be extended to applications combining optical imagery with structural information derived from satellite LiDAR. Future research could incorporate spatial autocorrelation and spatial heterogeneity into the proposed framework to further reduce uncertainty in forest AGB estimation.