By employing the unique phenological feature of winter wheat extracted from peak before winter (PBW) and the advantages of moderate resolution imaging spectroradiometer (MODIS) data with high temporal resolution and intermediate spatial resolution, a remote sensing-based model for mapping winter wheat on the North China Plain was built through integration with Landsat images and land-use data. First, a phenological window, PBW was drawn from time-series MODIS data. Next, feature extraction was performed for the PBW to reduce feature dimension and enhance its information. Finally, a regression model was built to model the relationship of the phenological feature and the sample data. The amount of information of the PBW was evaluated and compared with that of the main peak (MP). The relative precision of the mapping reached up to 92% in comparison to the Landsat sample data, and ranged between 87% and 96% in comparison to the statistical data. These results were sufficient to satisfy the accuracy requirements for winter wheat mapping at a large scale. Moreover, the proposed method has the ability to obtain the distribution information for winter wheat in an earlier period than previous studies. This study could throw light on the monitoring of winter wheat in China by using unique phenological feature of winter wheat.

Accurate and timely estimation of crop yield at a small scale is of great significance to food security and harvest management. Recent studies have proven remote sensing is an efficient method for yield estimation and machine learning, especially deep learning, can infer a good prediction by integrating multi-source datasets like satellite data, climate data, soil data and so on. However, there are some bottleneck challenges to improve accuracy. First, the popular remote sensing data used for yield prediction fall into two major groups, time-series data, and constant data. Surprisingly little attention has been devoted to deep learning networks which can integrate the two kinds of data effectively; second, both temporal and spatial features play a role in affecting the yields. But most of the existing approaches employed either CNN (convolutional neural network) or RNN (recurrent neural network). CNN cannot learn temporal patterns, while RNN barely can learn spatial characteristics. This work proposed a novel multi-level deep learning (MLDL) model coupling RNN and CNN to extract both spatial and temporal features. The inputs include both time-series remote sensing data and soil property data and the model outputs yield. We experimented with the model in U.S. Corn Belt states and used it to predict corn yield from 2013 to 2016 at the county-level. The results approve the effectiveness and advantages of the proposed approach over the other methods. In the future, the model will be used on other crops like soybean and winter wheat to assist agriculture decision making.

Despite the importance of tidal ecosystems in the global carbon budget, the relationships between environmental drivers and carbon dynamics in these wetlands remain poorly understood. This limited understanding results from the challenges associated with in situ flux studies and their correlation with satellite imagery which can be affected by periodic tidal flooding. Carbon dioxide eddy covariance (EC) towers are installed in only a few wetlands worldwide, and the longest eddy-covariance record from Georgia (GA) wetlands contains only two continuous years of observations. The goals of the present study were to evaluate the performance of existing MODIS Gross Primary Production (GPP) products (MOD17A2) against EC derived GPP and develop a tide-robust Normalized Difference Moisture Index (NDMI) based model to predict GPP within a Spartina alterniflora salt marsh on Sapelo Island, GA. These EC tower-based observations represent a basis to associate CO2 fluxes with canopy reflectance and thus provide the means to use satellite-based reflectance data for broader scale investigations. We demonstrate that Light Use Efficiency (LUE)-based MOD17A2 does not accurately reflect tidal wetland GPP compared to a simple empirical vegetation index-based model where tidal influence was accounted for. The NDMI-based GPP model was capable of predicting changes in wetland CO2 fluxes and explained 46% of the variation in flux-estimated GPP within the training data, and a root mean square error of 6.96 g C m−2 in the validation data. Our investigation is the first to create a MODIS-based wetland GPP estimation procedure that demonstrates the importance of filtering tidal observations from satellite surface reflectance data.