Forest growing stem volume (GSV) reflects the richness of forest resources as well as the quality of forest ecosystems. Remote sensing technology enables robust and efficient GSV estimation as it greatly reduces the survey time and cost while facilitating periodic monitoring. Given its red edge bands and a short revisit time period, Sentinel-2 images were selected for the GSV estimation in Wangyedian forest farm, Inner Mongolia, China. The variable combination was shown to significantly affect the accuracy of the estimation model. After extracting spectral variables, texture features, and topographic factors, a stepwise random forest (SRF) method was proposed to select variable combinations and establish random forest regressions (RFR) for GSV estimation. The linear stepwise regression (LSR), Boruta, Variable Selection Using Random Forests (VSURF), and random forest (RF) methods were then used as references for comparison with the proposed SRF for selection of predictors and GSV estimation. Combined with the observed GSV data and the Sentinel-2 images, the distributions of GSV were generated by the RFR models with the variable combinations determined by the LSR, RF, Boruta, VSURF, and SRF. The results show that the texture features of Sentinel-2’s red edge bands can significantly improve the accuracy of GSV estimation. The SRF method can effectively select the optimal variable combination, and the SRF-based model results in the highest estimation accuracy with the decreases of relative root mean square error by 16.4%, 14.4%, 16.3%, and 10.6% compared with those from the LSR-, RF-, Boruta-, and VSURF-based models, respectively. The GSV distribution generated by the SRF-based model matched that of the field observations well. The results of this study are expected to provide a reference for GSV estimation of coniferous plantations.

Satellite imagery of 25–30 m spatial resolution has been recognized as an effective tool for monitoring the spatial and temporal dynamics of forest cover at different scales. However, the precise mapping of forest cover over fragmented landscapes is complicated and requires special consideration. We have evaluated the performance of four global forest products of 25–30 m spatial resolution within three flatland subregions of Ukraine that have different forest cover patterns. We have explored the relationship between tree cover extracted from the global forest change (GFC) and relative stocking density of forest stands and justified the use of a 40% tree cover threshold for mapping forest in flatland Ukraine. In contrast, the canopy cover threshold for the analogous product Landsat tree cover continuous fields (LTCCF) is found to be 25%. Analysis of the global forest products, including discrete forest masks Global PALSAR-2/PALSAR Forest/Non-Forest Map (JAXA FNF) and GlobeLand30, has revealed a major misclassification of forested areas under severe fragmentation patterns of landscapes. The study also examined the effectiveness of forest mapping over fragmented landscapes using dense time series of Landsat images. We collected 1548 scenes of Landsat 8 Operational Land Imager (OLI) for the period 2014–2016 and composited them into cloudless mosaics for the following four seasons: yearly, summer, autumn, and April–October. The classification of images was performed in Google Earth Engine (GEE) Application Programming Interface (API) using random forest (RF) classifier. As a result, 30 m spatial resolution forest mask for flatland of Ukraine was created. The user’s and producer’s accuracy were estimated to be 0.910 ± 0.015 and 0.880 ± 0.018, respectively. The total forest area for the flatland Ukraine is 9440.5 ± 239.4 thousand hectares, which is 3% higher than official data. In general, we conclude that the Landsat-derived forest mask performs well over fragmented landscapes if forest cover of the territory is higher than 10–15%.

As an important vegetation canopy parameter, the leaf area index (LAI) plays a critical role in forest growth modeling and vegetation health assessment. Estimating LAI is helpful for understanding vegetation growth and global ecological processes. Machine learning methods such as k-nearest neighbors (kNN) and random forest (RF) with remote sensing images have been widely used for mapping LAI. However, the accuracy of mapping LAI in arid and semi-arid areas using these methods is limited due to remote and large areas, the high cost of collecting field data, and the great spatial variability of the vegetation canopy. Here, a novel and modified kNN method was presented for mapping LAI in arid and semi-arid areas of China using Sentinel-2 and Landsat 8 images with field data collected in Ganzhou and Kangbao of China. The modified kNN was developed by integrating the traditional kNN estimation and RF classification. The results were compared with those from kNN and RF regression alone using three sets of input predictors: (i) spectral reflectance bands (input 1); (ii) vegetation indices (input 2); and (iii) a combination of spectral reflectance bands and vegetation indices (input 3). Our analysis showed that in Ganzhou, the red-edge bands of the Sentinel-2 image had a high correlation with LAI. Using the red-edge band-derived vegetation indices increased the accuracy of mapping LAI compared with using other spectral variables. Among the three sets of input predictors, input 3 resulted in the highest prediction accuracy. Based on the combination, the values of RMSE obtained by the traditional kNN, RF, and modified kNN were 0.526, 0.523, and 0.372, respectively, and the modified kNN significantly improved the accuracy of LAI prediction by 29.3% and 28.9% compared with the kNN and RF alone, respectively. A similar improvement was achieved for input 1 and input 2. In Kangbao, the improvement of the prediction accuracy obtained by the modified kNN was 31.4% compared with both the kNN and RF. Therefore, this study implied that the modified kNN provided the potential to improve the accuracy of mapping LAI in arid and semi-arid regions using the images.

The co-evolution of multi-cities has emerged as the primary form of urbanization in China in recent years. However, the processes, patterns, and coordination are not well characterized and understood, which hinders the understanding of the driving forces, consequences, and management of polycentric urban development. We used the Continuous Change Detection and Classification (CCDC) algorithm to integrate all available Landsat 5, 7, and 8 images and map annual land use and land cover (LULC) from 2001 to 2017 in the Chang–Zhu–Tan urban agglomeration (CZTUA), a typical urban agglomeration in China. Results showed that the impervious surface in the study area expanded by 371 km² with an annual growth rate of 2.25%, primarily at the cost of cropland (169 km²) and forest (206 km²) during the study period. Urban growth has evolved from infilling being the dominant type in the earlier period to mainly edge-expansion and leapfrogging in the core cities, and from no dominant type to mainly leapfrogging in the satellite cities. The unfolding of the “cool center and hot edge” urban growth pattern in CZTUA, characterized by higher expansion rates in the peripheral than in the core cities, may signify a new form of the co-evolution of multi-cities in the process of urbanization. Detailed urban management and planning policies in CZTUA were analyzed. The co-evolution of multi-cities principles need to be studied in more extensive regions, which could help policymakers to promote sustainable and livable development in the future.

Gross primary production (GPP) determines the amounts of carbon and energy that enter terrestrial ecosystems. However, the tremendous uncertainty of the GPP still hinders the reliability of GPP estimates and therefore understanding of the global carbon cycle. In this study, using observations from global eddy covariance (EC) flux towers, we appraised the performance of 24 widely used GPP models and the quality of major spatial data layers that drive the models. Results show that global GPP products generated by the 24 models varied greatly in means (from 92.7 to 178.9 Pg C yr⁻¹) and trends (from −0.25 to 0.84 Pg C yr⁻¹). Model structure differences (i.e., light use efficiency models, machine learning models, and process-based biophysical models) are an important aspect contributing to the large uncertainty. In addition, various biases in currently available spatial datasets have found (e.g., only 57% of the observed variation in photosynthetically active radiation at the flux tower locations was explained by the spatial dataset), which not only affect GPP simulation but more importantly hinder the simulation and understanding of the earth system. Moving forward, research into the efficacy of model structures and precision of input data may be more important for global GPP estimation.