Accurate forest above-ground biomass (AGB) is crucial for sustaining forest management and mitigating climate change to support REDD+ (reducing emissions from deforestation and forest degradation, plus the sustainable management of forests, and the conservation and enhancement of forest carbon stocks) processes. Recently launched Sentinel imagery offers a new opportunity for forest AGB mapping and monitoring. In this study, texture characteristics and backscatter coefficients of Sentinel-1, in addition to multispectral bands, vegetation indices, and biophysical variables of Sentinal-2, based on 56 measured AGB samples in the center of the Changbai Mountains, China, were used to develop biomass prediction models through geographically weighted regression (GWR) and machine learning (ML) algorithms, such as the artificial neural network (ANN), support vector machine for regression (SVR), and random forest (RF). The results showed that texture characteristics and vegetation biophysical variables were the most important predictors. SVR was the best method for predicting and mapping the patterns of AGB in the study site with limited samples, whose mean error, mean absolute error, root mean square error, and correlation coefficient were 4 × 10 −3 , 0.07, 0.08 Mg·ha −1 , and 1, respectively. Predicted values of AGB from four models ranged from 11.80 to 324.12 Mg·ha −1 , and those for broadleaved deciduous forests were the most accurate, while those for AGB above 160 Mg·ha −1 were the least accurate. The study demonstrated encouraging results in forest AGB mapping of the normal vegetated area using the freely accessible and high-resolution Sentinel imagery, based on ML techniques.Traditional field-based measurements provide the most accurate AGB values, but they are destructive and spatially limited [10,11]. Uncertainty and bias in field measurements obviously exist, particularly those with large trees and tropical issues [4,5]. Combining remote sensing and sample plot data has become a popular method to generate spatially explicit estimations of forest AGB [12,13]. Various types of remote-sensing data are used for forest biomass estimation such as optical sensor data, radio detection and ranging (radar) data, and light detection and ranging (LiDAR) data, with each one having certain advantages over the others [14,15]. Optical sensors were first applied to retrieve the horizontal forest structure and AGB assessments through field sampling, due to their aggregate spectral signatures (reflectance or vegetation indices) with global coverage, repetitiveness, and cost-effectiveness [16,17]. Optical remote sensing data from a number of platforms, such as IKONOS, Quickbird, Worldview, ZY-3, systeme probatoire d'observation de la terre (SPOT), Sentinel, Landsat, and moderate-resolution imaging spectroradiometer (MODIS), with spatial resolutions varying from less than one meter to hundreds of meters, have been used by numerous researchers for biomass estimation [18][19][20]. However, the widespread usage of optical data is limited...

No abstract

Abstract-The classification of tree species through remote sensing data is of great significance to monitoring forest disturbances, biodiversity assessment, and carbon estimation. The dense time series and a wide swath of Sentinel-2 data provided the opportunity to map tree species accurately and in a timely manner over a large area. Many current studies have applied machine learning (ML) algorithms combined with Sentinel-2 images to classify tree species, but it is still unclear which algorithm is more effective in the automotive extraction of tree species. In this study, five machine learning algorithms were compared to identify the composition of tree species with multi-temporal Sentinel-2 images in the JianShe forest farm, Northeast China. Three major types of deep neural networks (Conv1D, AlexNet and LSTM) were tested to classify Sentinel-2 time series, which represent three disparate but effective strategies to apply sequential data. The other two models are Support Vector Machine (SVM) and Random Forest (RF), which are renowned for extensive adoption and high performance for various remote sensing applications. The results show that the overall accuracy of neural network models is better than that of SVM and RF. The Conv1D model had the highest classification accuracy (84.19%), followed by the LSTM model (81.52%), and the AlexNet model (76.02%). For non-neural network models, RF's classification accuracy (79.04%) is higher than that of SVM (72.79%), but lower than that of Conv1D and LSTM. Therefore, the deep neural networks combined with multitemporal Sentinel-2 images can efficiently improve the accuracy of tree species classification.