Forest Aboveground Biomass (AGB) is a key parameter for assessing forest productivity and global carbon content. In previous studies, AGB has been estimated using various prediction methods and types of remote sensing data. Increasingly, there is a trend towards integrating various data sources such as Light Detection and Ranging (LiDAR) and optical data. In this study, we constructed and compared the accuracies of five models for estimating AGB of forests in the upper Heihe River Basin in Northwest China. The five models were constructed using field and remotely-sensed data (optical and LiDAR) and algorithms including Random Forest (RF), Support Vector Machines (SVM), Back Propagation Neural Networks (BPNN), K-Nearest Neighbor (KNN) and the Generalized Linear Mixed Model (GLMM). Models based on the RF algorithm emerged as being the best among the five algorithms irrespective of the datasets used. The Random Forest AGB model, using only LiDAR data (R² = 0.899, RMSE = 14.0 t/ha) as the input data, was more effective than the one using optical data (R² = 0.835, RMSE = 22.724 t/ha). Compared to LiDAR or optical data alone, the AGB model (R² = 0.913, RMSE = 13.352 t/ha) that used the RF algorithm and integrated LiDAR and optical data was found to be optimal. Incorporation of terrain variables with optical data resulted in only slight improvements in accuracy. The models developed in this study could be useful for using integrated airborne LiDAR and passive optical data to accurately estimate forest biomass.

Winter wheat is one of the major cereal crops in China. The spatial distribution of winter wheat planting areas is closely related to food security; however, mapping winter wheat with time-series finer spatial resolution satellite images across large areas is challenging. This paper explores the potential of combining temporally aggregated Landsat-8 OLI and Sentinel-2 MSI data available via the Google Earth Engine (GEE) platform for mapping winter wheat in Shandong Province, China. First, six phenological median composites of Landsat-8 OLI and Sentinel-2 MSI reflectance measures were generated by a temporal aggregation technique according to the winter wheat phenological calendar, which covered seedling, tillering, over-wintering, reviving, jointing-heading and maturing phases, respectively. Then, Random Forest (RF) classifier was used to classify multi-temporal composites but also mono-temporal winter wheat development phases and mono-sensor data. The results showed that winter wheat could be classified with an overall accuracy of 93.4% and F1 measure (the harmonic mean of producer’s and user’s accuracy) of 0.97 with temporally aggregated Landsat-8 and Sentinel-2 data were combined. As our results also revealed, it was always good to classify multi-temporal images compared to mono-temporal imagery (the overall accuracy dropped from 93.4% to as low as 76.4%). It was also good to classify Landsat-8 OLI and Sentinel-2 MSI imagery combined instead of classifying them individually. The analysis showed among the mono-temporal winter wheat development phases that the maturing phase’s and reviving phase’s data were more important than the data for other mono-temporal winter wheat development phases. In sum, this study confirmed the importance of using temporally aggregated Landsat-8 OLI and Sentinel-2 MSI data combined and identified key winter wheat development phases for accurate winter wheat classification. These results can be useful to benefit on freely available optical satellite data (Landsat-8 OLI and Sentinel-2 MSI) and prioritize key winter wheat development phases for accurate mapping winter wheat planting areas across China and elsewhere.

In the last two decades, machine learning (ML) methods have been widely used in digital soil mapping (DSM), but the regression kriging (RK) model which combines the advantages of the ML and kriging methods has rarely been used in DSM. In addition, due to the limitation of a single-model structure, many ML methods have poor prediction accuracy in undulating terrain areas. In this study, we collected the SOC content of 115 soil samples in a hilly farming area with continuous undulating terrain. According to the theory of soil-forming factors in pedogenesis, we selected 10 topographic indices, 7 vegetation indices, and 2 soil indices as environmental covariates, and according to the law of geographical similarity, we used ML and RK methods to mine the relationship between SOC and environmental covariates to predict the SOC content. Four ensemble models—random forest (RF), Cubist, stochastic gradient boosting (SGB), and Bayesian regularized neural networks (BRNNs)—were used to fit the trend of SOC content, and the simple kriging (SK) method was used to interpolate the residuals of the ensemble models, and then the SOC and residual were superimposed to obtain the RK prediction result. Moreover, the 115 samples were divided into calibration and validation sets at a ratio of 80%, and the tenfold cross-validation method was used to fit the optimal parameters of the model. From the results of four ensemble models: RF performed best in the calibration set (R²_c = 0.834) but poorly in the validation set (R²_v = 0.362); Cubist had good accuracy and stability in both the calibration and validation sets (R²_c = 0.693 and R²_v = 0.445); SGB performed poorly (R²_c = 0.430 and R²_v = 0.336); and BRNN had the lowest accuracy (R²_c = 0.323 and R²_v = 0.282). The results showed that the R² of the four RK models in the validation set were 0.718, 0.674, 0.724, and 0.625, respectively. Compared with the ensemble models without superimposed residuals, the prediction accuracy was improved by 0.356, 0.229, 0.388, and 0.343, respectively. In conclusion, Cubist has high prediction accuracy and generalization ability in areas with complex topography, and the RK model can make full use of trends and spatial structural factors that are not easy to mine by ML models, which can effectively improve the prediction accuracy. This provides a reference for soil survey and digital mapping in complex terrain areas.