The global availability of high spatial resolution images makes mapping tree species distribution possible for better management of forest resources. Previous research mainly focused on mapping single tree species, but information about the spatial distribution of all kinds of trees, especially plantations, is often required. This research aims to identify suitable variables and algorithms for classifying land cover, forest, and tree species. Bi-temporal ZiYuan-3 multispectral and stereo images were used. Spectral responses and textures from multispectral imagery, canopy height features from bi-temporal stereo imagery, and slope and elevation from the stereo-derived digital surface model data were examined through comparative analysis of six classification algorithms including maximum likelihood classifier (MLC), k-nearest neighbor (kNN), decision tree (DT), random forest (RF), artificial neural network (ANN), and support vector machine (SVM). The results showed that use of multiple source data—spectral bands, vegetation indices, textures, and topographic factors—considerably improved land-cover and forest classification accuracies compared to spectral bands alone, which the highest overall accuracy of 84.5% for land cover classes was from the SVM, and, of 89.2% for forest classes, was from the MLC. The combination of leaf-on and leaf-off seasonal images further improved classification accuracies by 7.8% to 15.0% for land cover classes and by 6.0% to 11.8% for forest classes compared to single season spectral image. The combination of multiple source data also improved land cover classification by 3.7% to 15.5% and forest classification by 1.0% to 12.7% compared to the spectral image alone. MLC provided better land-cover and forest classification accuracies than machine learning algorithms when spectral data alone were used. However, some machine learning approaches such as RF and SVM provided better performance than MLC when multiple data sources were used. Further addition of canopy height features into multiple source data had no or limited effects in improving land-cover or forest classification, but improved classification accuracies of some tree species such as birch and Mongolia scotch pine. Considering tree species classification, Chinese pine, Mongolia scotch pine, red pine, aspen and elm, and other broadleaf trees as having classification accuracies of over 92%, and larch and birch have relatively low accuracies of 87.3% and 84.5%. However, these high classification accuracies are from different data sources and classification algorithms, and no one classification algorithm provided the best accuracy for all tree species classes. This research implies the same data source and the classification algorithm cannot provide the best classification results for different land cover classes. It is necessary to develop a comprehensive classification procedure using an expert-based approach or hierarchical-based classification approach that can employ specific data variables and algorithm for each tree species class.

Tree species information is an important factor in forest resource surveys, and light detection and ranging (LiDAR), as a new technical tool for forest resource surveys, can quickly obtain the 3D structural information of trees. In particular, the rapid and accurate classification and identification of tree species information from individual tree point clouds using deep learning methods is a new development direction for LiDAR technology in forest applications. In this study, mobile laser scanning (MLS) data collected in the field are first pre-processed to extract individual tree point clouds. Two downsampling methods, non-uniform grid and farthest point sampling, are combined to process the point cloud data, and the obtained sample data are more conducive to the deep learning model for extracting classification features. Finally, four different types of point cloud deep learning models, including pointwise multi-layer perceptron (MLP) (PointNet, PointNet++, PointMLP), convolution-based (PointConv), graph-based (DGCNN), and attention-based (PCT) models, are used to classify and identify the individual tree point clouds of eight tree species. The results show that the classification accuracy of all models (except for PointNet) exceeded 0.90, where the PointConv model achieved the highest classification accuracy for tree species classification. The streamlined PointMLP model can still achieve high classification accuracy, while the PCT model did not achieve good accuracy in the tree species classification experiment, likely due to the small sample size. We compare the training process and final classification accuracy of the different types of point cloud deep learning models in tree species classification experiments, further demonstrating the advantages of deep learning techniques in tree species recognition and providing experimental reference for related research and technological development.

Species-rich subtropical forests have high carbon sequestration capacity and play important roles in regional and global carbon regulation and climate changes. A timely investigation of the spatial distribution characteristics of subtropical forest aboveground biomass (AGB) is essential to assess forest carbon stocks. Lidar (light detection and ranging) is regarded as the most reliable data source for accurate estimation of forest AGB. However, previous studies that have used lidar data have often beenbased on a single model developed from the relationships between lidar-derived variables and AGB, ignoring the variability of this relationship in different forest types. Although stratification of forest types has been proven to be effective for improving AGB estimation, how to stratify forest types and how many strata to use are still unclear. This research aims to improve forest AGB estimation through exploring suitable stratification approaches based on lidar and field survey data. Different stratification schemes including non-stratification and stratifications based on forest types and forest stand structures were examined. The AGB estimation models were developed using linear regression (LR) and random forest (RF) approaches. The results indicate the following: (1) Proper stratifications improved AGB estimation and reduced the effect of under- and overestimation problems; (2) the finer forest type strata generated higher accuracy of AGB estimation but required many more sample plots, which were often unavailable; (3) AGB estimation based on stratification of forest stand structures was similar to that based on five forest types, implying that proper stratification reduces the number of sample plots needed; (4) the optimal AGB estimation model and stratification scheme varied, depending on forest types; and (5) the RF algorithm provided better AGB estimation for non-stratification than the LR algorithm, but the LR approach provided better estimation with stratification. Results from this research provide new insights on how to properly conduct forest stratification for AGB estimation modeling, which is especially valuable in tropical and subtropical regions with complex forest types.

Many studies have investigated the effects of spectral and spatial features of remotely sensed data and topographic characteristics on land-cover and forest classification results, but they are mainly based on individual sensor data. How these features from different kinds of remotely sensed data with various spatial resolutions influence classification results is unclear. We conducted a comprehensively comparative analysis of spectral and spatial features from ZiYuan-3 (ZY-3), Sentinel-2, and Landsat and their fused datasets with spatial resolution ranges from 2 m, 6 m, 10 m, 15 m, and to 30 m, and topographic factors in influencing land-cover classification results in a subtropical forest ecosystem using random forest approach. The results indicated that the combined spectral (fused data based on ZY-3 and Sentinel-2), spatial, and topographical data with 2-m spatial resolution provided the highest overall classification accuracy of 83.5% for 11 land-cover classes, as well as the highest accuracies for almost all individual classes. The improvement of spectral bands from 4 to 10 through fusion of ZY-3 and Sentinel-2 data increased overall accuracy by 14.2% at 2-m spatial resolution, and by 11.1% at 6-m spatial resolution. Textures from high spatial resolution imagery play more important roles than textures from medium spatial resolution images. The incorporation of textural images into spectral data in the 2-m spatial resolution imagery improved overall accuracy by 6.0–7.7% compared to 1.1–1.7% in the 10-m to 30-m spatial resolution images. Incorporation of topographic factors into spectral and textural imagery further improved overall accuracy by 1.2–5.5%. The classification accuracies for coniferous forest, eucalyptus, other broadleaf forests, and bamboo forest can be 85.3–91.1%. This research provides new insights for using proper combinations of spectral bands and textures corresponding to specifically spatial resolution images in improving land-cover and forest classifications in subtropical regions.

Estimating individual tree diameters at breast height (DBH) from delineated crowns and tree heights on the basis of airborne light detection and ranging (LiDAR) data provides a good opportunity for large-scale forest inventory. Generally, ground-based measurements are more accurate, but LiDAR data and derived DBH values can be obtained over larger areas for a relatively smaller cost if a right procedure is developed. A nonlinear least squares (NLS) regression is not an appropriate approach to predict the aboveground biomass (AGB) of individual trees from the estimated DBH because both the response variable and the regressor are subject to measurement errors. In this study, a system of compatible individual tree DBH and AGB error-in-variable models was developed using error-in-variable regression techniques based on both airborne LiDAR and field-measured datasets of individual Picea crassifolia Kom. trees, collected in northwestern China. Two parameter estimation algorithms, i.e., the two-stage error-in-variable model (TSEM) and the nonlinear seemingly unrelated regression (NSUR), were proposed for estimating the parameters in the developed system of compatible individual tree DBH and AGB error-in-variable models. Moreover, two model structures were applied to estimate AGB for comparison purposes: NLS with AGB estimation depending on DBH (NLS&DD) and NLS with AGB estimation not depending on DBH (NLS&NDD). The results showed that both TSEM and NSUR led to almost the same parameter estimates for the developed system. Moreover, the developed system effectively accounted for the inherent correlation between DBH and AGB as well as for the effects of measurement errors in the DBH on the predictions of AGB, whereas NLS&DD and NLS&NDD did not. A leave-one-out cross-validation indicated that the prediction accuracy of the developed system of compatible individual tree DBH and AGB error-in-variable models with NSUR was the highest among those estimated by the four methods evaluated, but, statistically, the accuracy improvement was not significantly different from zero. The main reason might be that, except for the measurement errors, other source errors were ignored in the modeling. This study implies that, overall, the proposed method provides the potential to expand the estimations of both DBH and AGB from individual trees to stands by combining the error-in-variable modeling and LiDAR data and improve their estimation accuracies, but its application needs to be further validated by conducting a systematical uncertainty analysis of various source errors in a future study.

High-precision automatic identification and mapping of forest tree species composition is an important content of forest resource survey and monitoring. The airborne hyperspectral image contains rich spectral and spatial information, which provides the possibility of high-precision classification and mapping of forest tree species. Few-shot learning, as an application of deep learning, has become an effective method of image classification. Prototypical networks (P-Net) is a simple and practical deep learning network, which has significant advantages in solving few-shot classification problems. Considering the high band correlation and large data volume associated with airborne hyperspectral images, how to fully extract effective features, filter or reduce redundant features is the key to improving the classification accuracy of P-Net, in order to extract effective features in hyperspectral images and obtain a high-precision forest tree species classification model with limited samples. In this research, we embedded the convolutional block attention module (CBAM) between the convolution blocks of P-Net, the CBAM-P-Net was constructed, and a method to improve the feature extraction efficiency of the P-Net was proposed, although this method makes the network more complex and increases the computational cost to a certain extent. The results show that the combination strategy using Channel First for CBAM greatly improves the feature extraction efficiency of the model. In different sample windows, CBAM-P-Net has an average increase of 1.17% and 0.0129 in testing overall accuracy (OA) and kappa coefficient (Kappa). The optimal classification window is 17 × 17, the OA reaches 97.28%, and Kappa reaches 0.97, which is an increase of 1.95% and 0.0214 along with just 49 s of training time expended, respectively, compared with P-Net. Therefore, using a suitable sample window and applying the proposed CBAM-P-Net to classify airborne hyperspectral images can achieve high-precision classification and mapping of forest tree species.