Forests, commonly viewed as the Earth’s lungs, play a crucial role in mitigating greenhouse gas emissions, regulating the globe, and maintaining ecological equilibrium. The assessment of aboveground biomass (AGB) serves as a pivotal indicator for evaluating forest quality. By integrating remote sensing images with a small number of ground-measured samples to map, forest AGBs can significantly reduce time and labor costs. Current research mainly focuses on improving the accuracy of mapping forest AGBs, such as integrating multiple-sensors remote sensing data and models. However, due to uncertainties associated with remote sensing images and complexities inherent in forest structures, the accuracy of mapping forest AGBs is constrained by both the quantity and distribution of ground samples available. The development of transfer learning methods can fully utilize ground-based measurement data and enable the application of samples across regions and time. To evaluate the potential of transfer learning methods in mapping forest AGBs, this study conducted a spatial–temporal transfer of spectral variables (SVs) and prediction models (PMs) using a direct-push transfer method, and a new evaluation metric, relative change of R-squared (RCRS), was proposed to assess the transferability of SVs and PMs. The results showed that the transferability of SVs and PMs in the spatial target domain is obviously greater than that in the temporal target domain. Compared to the temporal target domain, the RCRS for transfer SVs in the spatial target domain was lower by 20.89 (oak) and 20.88 (Chinese fir) and for transfer PMs by 24.16 (oak) and 24.79 (Chinese fir). Tree species is also one of the main factors affecting the spatial and temporal transfer of SVs, and it is challenging to transfer SVs between different tree species. The results also show that nonparametric models have better generalization performance, and their transferability is much greater than that of parametric models.

As one of the important types of forest resources, mapping forest stock volume (FSV) in larch (Larix decidua) forests holds significant importance for forest resource management, carbon cycle research, and climate change monitoring. However, the accuracy of FSV mapping using common spectral and texture features is often limited due to their failure in fully capturing seasonal changes and growth cycle characteristics of vegetation. Phenological features can effectively provide essential information regarding the growth status of forests. In this study, multi-temporal Sentinel-2 satellite imagery were initially acquired in the Wangyedian Forest Farm in Chifeng City, Inner Mongolia. Subsequently, various phenological features were extracted from time series variables constructed by Gaussian Process Regression (GPR) using Savitzky–Golay filters, stepwise differentiation, and Fourier transform techniques. The alternative features were further refined through Pearson’s correlation coefficient analysis and the forward selection algorithm, resulting in six groups of optimal subsets. Finally, four models including the Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Multiple Linear Regression (MLR) algorithms were developed to estimate FSV. The results demonstrated that incorporating phenological features significantly enhanced model performance, with the SVM model exhibiting the best performance—achieving an R² value of 0.77 along with an RMSE value of 46.36 m³/hm² and rRMSE value of 22.78%. Compared to models without phenological features, inclusion of these features led to a 0.25 increase in R² value while reducing RMSE by 10.40 m³/hm² and rRMSE by 5%. Overall, integration of phenological feature variables not only improves the accuracy of larch forest FSV mapping but also has potential implications for delaying saturation phenomena.

Spectral features (SFs) and texture features (TFs) extracted from optical remote sensing images can capture the structural composition and growth information of forests, and combining remote sensing variables with a few ground measurement samples is a common method for mapping forest stock volume (FSV). However, the accuracy of mapping FSV using optical images with a high spatial resolution (one meter or sub-meters) is often lower than medium resolutions (larger than 10 m) using the same types of features and approaches. To overcome the limitations of high spatial resolution images in mapping FSV, down-scaled images with spatial resolution ranging from 1 to 30 m were obtained by GF-2 image to interpret the relationships between spatial resolutions of features and the accuracy of mapping FSV, and combination strategies of variables with various spatial resolutions were proposed to improve the accuracy of mapping FSV. The results show that the spatial resolution of features significantly affects the performance of employed models in estimating FSV, the sensitivity between SFs and FSV gradually increases with the decreasing of spatial resolution, and the optimal spatial resolutions of two types of features (SFs and TFs) are not synchronized in mapping forest FSV. After using combination strategies of variables with various spatial resolutions, the accuracy of mapping FSV is significantly higher than those derived from variable sets with the same spatial resolutions. It is proved that TFs derived from GF-2 images have great potential to improve the accuracy of mapping FSV, and the contribution of features depends on the approaches of extracting and combination strategies.

Forest aboveground biomass (AGB) serves as a crucial quantitative indicator that reflects the carbon sequestration capacity of forests, and accurately mapping AGB is pivotal for assessing forest ecosystem stability. However, mapping AGB in subtropical evergreen broadleaf forests in southern China presents challenges due to their complex canopy structure, stand heterogeneity, and spectral signal saturation. The phenological features reflecting seasonal vegetation dynamics are conducive to over-coming these challenges. By analyzing differential spectral reflectance patterns during the non-growing (Jan–Mar, Nov–Dec) versus growing (Apr–Oct) seasons, this study established a phenological feature-based methodology for improving AGB estimation in subtropical evergreen broadleaf forests. Subsequently, four time series vegetation indices (VI), namely NDVI, EVI2, NDPI, and IRECI were employed to extract phenological features (PFs) for mapping forest AGB using a multiple linear regression model (MLR), K-nearest neighbor model (KNN), support vector machine model (SVM), and random forest model (RF). The results demonstrated significant differences in Sentinel-2 spectral reflectance (740–1610 nm bands) between the growing and non-growing seasons. The PFs demonstrated the highest distance correlation coefficient (0.57), significantly outperforming other baseline feature types (0.44). Furthermore, seasonal changes in NDVI and NDPI were found to better reflect AGB accumulation in evergreen broadleaf forests compared to EVI2 and IRECI. Incorporating diverse PFs derived from all four VI significantly enhanced the accuracy of AGB mapping by yielding rRMSE values ranging from 21.01% to 25.06% and R² values ranging from 0.40 to 0.58. The results inferred that PFs can be considered a key factor for alleviating spectral signal saturation problems while effectively improving the accuracy of AGB estimation.

Polarimetric Synthetic Aperture Radar (PolSAR) images with dual polarization modes have great potential to map forest stock volume (FSV) by excellent penetration capabilities and distinct microwave scattering processes. However, the response of these SAR data to FSV is still uncertain in the non-growing season. To further interpret the response of FSV to different dual polarization SAR images, three types of dual polarization SAR images (GF-3, Sentinel-1, and ALOS-2) were initially acquired in coniferous planted forest in the non-growing season. Then, sensitivity between FSV and all alternative features extracted from each type of SAR image was analyzed to express the response of FSV to dual polarization SAR images with bands and polarization modes in the non-growing season in deciduous (Larch) and evergreen (Chinese pine) forests. Finally, mapped FSV using single and combined dual polarization images were derived by optimal feature sets and four machine learning models, respectively. The combined effects were also analyzed to clarify the difference of bands and polarization modes in deciduous and evergreen forests in the non-growing season. The results demonstrated that the backscattering energy from different sensors is significantly different in Chinese pine, and the difference is gradually reduced in Larch forests. It is also implied that the polarization mode is more important than penetration capability in mapping forest FSV in deciduous forest in the non-growing season. By comparing the accuracy of mapped FSV using single and combined images, combined images have more capability to improve the accuracy and reliability of mapped FSV. Meanwhile, it is confirmed that compensation effects with bands and polarization modes not only have great potential to delay the saturation phenomenon, but also have the capability to reduce errors caused by overestimation.

Forest growing stem volume (GSV) is regarded as one of the most important parameters for the quality evaluation and dynamic monitoring of forest resources. The accuracy of mapping forest GSV is highly related to the employed models and involved remote sensing features, and the criteria of feature evaluation severely affect the performance of the employed models. However, due to the linear or nonlinear relationships between remote sensing features and GSV, widely used evaluation criteria inadequately express the complex sensitivity between forest GSV and spectral features, especially the saturation levels of features in a planted forest. In this study, novel feature evaluation criteria were constructed based on the Pearson correlations and optical saturation levels of the alternative remote sensing features extracted from two common optical remote sensing image sets (GF-1 and Sentinel-2). Initially, the spectral saturation level of each feature was quantified using the kriging spherical model and the quadratic model. Then, optimal feature sets were obtained with the proposed criteria and the linear stepwise regression model. Finally, four widely used machine learning models—support vector machine (SVM), multiple linear stepwise regression (MLR), random forest (RF) and K-neighborhood (KNN)—were employed to map forest GSV in a planted Chinese fir forest. The results showed that the proposed feature evaluation criteria could effectively improve the accuracy of estimating forest GSV and that the systematic distribution of errors between the predicted and ground measurements in the range of forest GSV was less than 300 m³/hm². After using the proposed feature evaluation criteria, the highest accuracy of mapping GSV was obtained with the RF model for GF-1 images (R² = 0.49, rRMSE = 28.67%) and the SVM model for Sentinel-2 images (R² = 0.52, rRMSE = 26.65%), and the decreased rRMSE values ranged from 1.1 to 6.2 for GF-1 images (28.67% to 33.08%) and from 2.3 to 6.8 for Sentinel-2 images (26.85% to 33.28%). It was concluded that the sensitivity of the optimal feature set and the accuracy of the estimated GSV could be improved using the proposed evaluation criteria (less than 300 m³/hm²). However, these criteria were barely able to improve mapping accuracy for a forest with a high GSV (larger than 300 m³/hm²).