RESUMEN
Accurate predictions of wheat yields are essential to farmers'production plans and to the international trade in wheat. However, only poor approximations of the productivity of wheat crops in China can be obtained using traditional linear regression models based on vegetation indices and observations of the yield. In this study, Sentinel-2 (multispectral data) and ZY-1 02D (hyperspectral data) were used together with 15709 gridded yield data (with a resolution of 5 m × 5 m) to predict the winter wheat yield. These estimates were based on four mainstream data-driven approaches: Long Short-Term Memory (LSTM), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and Support Vector Regression (SVR). The method that gave the best estimate of the winter wheat yield was determined, and the accuracy of the estimates based on multispectral and hyperspectral data were compared. The results showed that the LSTM model, for which the RMSE of the estimates was 0.201 t/ha, performed better than the RF (RMSE = 0.260 t/ha), GBDT (RMSE = 0.306 t/ha), and SVR (RMSE = 0.489 t/ha) methods. The estimates based on the ZY-1 02D hyperspectral data were more accurate than those based on the 30-m Sentinel-2 data: RMSE = 0.237 t/ha for the ZY-1 02D data, which is about a 5% improvement on the RSME of 0.307 t/ha for the 30-m Sentinel-2 data. However, the 10-m Sentinel-2 data performed even better, giving an RMSE of 0.219 t/ha. In addition, it was found that the greenness vegetation index SR (simple ratio index) outperformed the traditional vegetation indices. The results highlight the potential of the shortwave infrared bands to replace the visible and near-infrared bands for predicting crop yields Our study demonstrates the advantages of the deep learning method LSTM over machine learning methods in terms of its ability to make accurate estimates of the winter wheat yield.
RESUMEN
The goal of this study was to promptly map the extent of corn and soybeans early in the growing season. A classification experiment was conducted for the US Corn Belt and neighboring states, which is the most important production area of corn and soybeans in the world. To improve the timeliness of the classification algorithm, training was completely based on reference data and images from other years, circumventing the need to finish reference data collection in the current season. To account for interannual variability in crop development in the cross-year classification scenario, several innovative strategies were used. A random forest classifier was used in all tests, and MODIS surface reflectance products from the years 2008-2014 were used for training and cross-year validation. It is concluded that the fuzzy classification approach is necessary to achieve satisfactory results with R-squared ~0.9 (compared with the USDA Cropland Data Layer). The year of training data is an important factor, and it is recommended to select a year with similar crop phenology as the mapping year. With this phenology-based and cross-year-training method, in 2015 we mapped the cropping proportion of corn and soybeans around mid-August, when the two crops just reached peak growth.