In this paper, the random forests method and the support vector machine in machine learning are explored and compared to the traditional statistical-based maximum likelihood method with 126 features from Sentinel-2A images. The spectral reflectance of 12 bands, 96 texture parameters, 7 vegetation indices, and 11 phenological parameters are successfully extracted from Sentinel-2A images in 2017. The classification result shows that the optimal combination of 13 features yields overall accuracies of traditional classification and machine learning classification of 88.96% and 98%, respectively. Short-wave infrared information shows a significant effect on distinguishing rice, corn, and soybean. The water vapor band plays a significant role in distinguishing between corn and rice. In the multiclassification problem, the machine learning methods have robustness with the identification accuracy of greater than 95% for each crop type, whereas the traditional classification result shows imbalanced accuracies for different crops.

Vegetation phenology plays a key role in terrestrial ecosystem nutrient and carbon cycles and is sensitive to global climate change. Compared with spring phenology, which has been well studied, autumn phenology is still poorly understood. In this study, we estimated the date of the end of the growing season (EOS) across the Greater Khingan Mountains, China, from 1982 to 2015 based on the Global Inventory Modeling and Mapping Studies (GIMMS) normalized difference vegetation index third-generation (NDVI3g) dataset. The temporal correlations between EOS and climatic factors (e.g., preseason temperature, preseason precipitation), as well as the correlation between autumn and spring phenology, were investigated using partial correlation analysis. Results showed that more than 94% of the pixels in the Greater Khingan Mountains exhibited a delayed EOS trend, with an average rate of 0.23 days/y. Increased preseason temperature resulted in earlier EOS in most of our study area, except for the semi-arid grassland region in the south, where preseason warming generally delayed EOS. Similarly, EOS in most of the mountain deciduous coniferous forest, forest grassland, and mountain grassland forest regions was earlier associated with increased preseason precipitation, but for the semi-arid grassland region, increased precipitation during the preseason mainly led to delayed EOS. However, the effect of preseason precipitation on EOS in most of the Greater Khingan Mountains was stronger than that of preseason temperature. In addition to the climatic effects on EOS, we also found an influence of spring phenology on EOS. An earlier SOS led to a delayed EOS in most of the study area, while in the southern of mountain deciduous coniferous forest region and northern of semi-arid grassland region, an earlier SOS was often followed by an earlier EOS. These findings suggest that both climatic factors and spring phenology should be incorporated into autumn phenology models in order to improve prediction accuracy under present and future climate change scenarios.

Net primary productivity (NPP) is the total amount of organic matter fixed by plants from the atmosphere through photosynthesis and is susceptible to the influences of climate change and human activities. In this study, we employed actual NPP (ANPP), potential NPP (PNPP), and human activity-induced NPP (HNPP) based on the Hurst exponent and statistical analysis to analyze the characteristics of vegetation productivity dynamics and to evaluate the effects of climate and human factors on vegetation productivity in Northeast China (NEC). The increasing trends in ANPP, PNPP, and HNPP accounted for 81.62%, 94.90%, and 89.63% of the total area, respectively, and ANPP in 68.64% of the total area will continue to increase in the future. Climate change played a leading role in vegetation productivity dynamics, which promoted an increase in ANPP in 71.55% of the area, and precipitation was the key climate factor affecting ANPP. The aggravation of human activities, such as increased livestock numbers and intensified agricultural activities, resulted in a decrease in ANPP in the western grasslands, northern Greater Khingan Mountains, and eastern Songnen Plain. In particular, human activities led to a decrease in ANPP in 53.84% of deciduous needleleaf forests. The impact of climate change and human activities varied significantly under different topography, and the percentage of the ANPP increase due to climate change decreased from 71.13% to 53.9% from plains to urgent slopes; however, the percentage of ANPP increase due to human activities increased from 3.44% to 21.74%, and the effect of human activities on the increase of ANPP was more obvious with increasing slope. At different altitudes, the difference in the effect of these two factors was not significant. The results are significant for understanding the factors influencing the vegetation productivity dynamics in NEC and can provide a reference for governments to implement projects to improve the ecosystem.

Northeast China is located at high northern latitudes and is a typical region of relatively high sensitivity to global climate change. Studies of the land surface phenology in Northeast China and its response to climate change are important for understanding global climate change. In this study, the land surface phenology parameters were calculated using the third generation dataset from the Global Inventory Modeling and Mapping Studies (GIMMS 3g) that was collected from 1982 to 2013 were estimated to analyze the variations of the land surface phenology in Northeast China at different scales and to discuss the internal relationships between phenology and climate change. We examined the phonological changes of all ecoregions. The average start of the growing season (SOS) did not exhibit a significant trend throughout the study area; however, the end of the growing season (EOS) was significantly delayed by 4.1 days or 0.13 days/year (p < 0.05) over the past 32 years. The SOS for the Hulunbuir Plain, Greater Khingan Mountains and Lesser Khingan Mountains was earlier, and the SOS for the Sanjing, Songnen and Liaohe Plains was later. In addition, the EOS of the Greater Khingan Mountains, Lesser Khingan Mountains and Changbai Mountains was later than the EOS of the Liaohe Plain. The spring temperature had the greatest impact on the SOS. Precipitation had an insignificant impact on forest SOS and a relatively large impact on grassland SOS. The EOS was affected by both temperature and precipitation. Furthermore, although temperature had a lag effect on the EOS, no significant lag effect was observed for the SOS.

Quantifying crop residue cover (CRC) on field surfaces is important for monitoring the tillage intensity and promoting sustainable management. Remote-sensing-based techniques have proven practical for determining CRC, however, the methods used are primarily limited to empirical regression based on crop residue indices (CRIs). This study provides a systematic evaluation of empirical regressions and machine learning (ML) algorithms based on their ability to estimate CRC using Sentinel-2 Multispectral Instrument (MSI) data. Unmanned aerial vehicle orthomosaics were used to extracted ground CRC for training Sentinel-2 data-based CRC models. For empirical regression, nine MSI bands, 10 published CRIs, three proposed CRIs, and four mean textural features were evaluated using univariate linear regression. The best performance was obtained by a three-band index calculated using (B2 − B4)/(B2 − B12), with an R²_cv of 0.63 and RMSE_cv of 6.509%, using a 10-fold cross-validation. The methodologies of partial least squares regression (PLSR), artificial neural network (ANN), Gaussian process regression (GPR), support vector regression (SVR), and random forest (RF) were compared with four groups of predictors, including nine MSI bands, 13 CRIs, a combination of MSI bands and mean textural features, and a combination of CRIs and textural features. In general, ML approaches achieved high accuracy. A PLSR model with 13 CRIs and textural features resulted in an accuracy of R²_cv = 0.66 and RMSE_cv = 6.427%. An RF model with predictors of MSI bands and textural features estimated CRC with an R²_cv = 0.61 and RMSE_cv = 6.415%. The estimation was improved by an SVR model with the same input predictors (R²_cv = 0.67, RMSE_cv = 6.343%), followed by a GPR model based on CRIs and textural features. The performance of GPR models was further improved by optimal input variables. A GPR model with six input variables, three MSI bands and three textural features, performed the best, with R²_cv = 0.69 and RMSE_cv = 6.149%. This study provides a reference for estimating CRC from Sentinel-2 imagery using ML approaches. The GPR approach is recommended. A combination of spectral information and textural features leads to an improvement in the retrieval of CRC.

Time series of Normalized Difference Vegetation Index (NDVI) derived from multiple satellite sensors are crucial data to study vegetation dynamics. The Land Long Term Data Record Version 4 (LTDR V4) NDVI dataset was recently released at a 0.05 × 0.05° spatial resolution and daily temporal resolution. In this study, annual NDVI time series that are composited by the LTDR V4 and Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI datasets (MOD13C1) are compared and evaluated for the period from 2001 to 2014 in China. The spatial patterns of the NDVI generally match between the LTDR V4 and MOD13C1 datasets. The transitional zone between high and low NDVI values generally matches the boundary of semi-arid and sub-humid regions. A significant and high coefficient of determination is found between the two datasets according to a pixel-based correlation analysis. The spatially averaged NDVI of LTDR V4 is characterized by a much weaker positive regression slope relative to that of the spatially averaged NDVI of the MOD13C1 dataset because of changes in NOAA AVHRR sensors between 2005 and 2006. The measured NDVI values of LTDR V4 were always higher than that of MOD13C1 in western China due to the relatively lower atmospheric water vapor content in western China, and opposite observation appeared in eastern China. In total, 18.54% of the LTDR V4 NDVI pixels exhibit significant trends, whereas 35.79% of the MOD13C1 NDVI pixels show significant trends. Good agreement is observed between the significant trends of the two datasets in the Northeast Plain, Bohai Economic Rim, Loess Plateau, and Yangtze River Delta. By contrast, the datasets contrasted in northwestern desert regions and southern China. A trend analysis of the regression slope values according to the vegetation type shows good agreement between the LTDR V4 and MOD13C1 datasets. This study demonstrates the spatial and temporal consistencies and discrepancies between the AVHRR LTDR and MODIS MOD13C1 NDVI products in China, which could provide useful information for the choice of NDVI products in subsequent studies of vegetation dynamics.