Wheat is one of the main crops in China, and crop yield prediction is important for regional trade and national food security. There are increasing concerns with respect to how to integrate multi-source data and employ machine learning techniques to establish a simple, timely, and accurate crop yield prediction model at an administrative unit. Many previous studies were mainly focused on the whole crop growth period through expensive manual surveys, remote sensing, or climate data. However, the effect of selecting different time window on yield prediction was still unknown. Thus, we separated the whole growth period into four time windows and assessed their corresponding predictive ability by taking the major winter wheat production regions of China as an example in the study. Firstly we developed a modeling framework to integrate climate data, remote sensing data and soil data to predict winter wheat yield based on the Google Earth Engine (GEE) platform. The results show that the models can accurately predict yield 1~2 months before the harvesting dates at the county level in China with an R² > 0.75 and yield error less than 10%. Support vector machine (SVM), Gaussian process regression (GPR), and random forest (RF) represent the top three best methods for predicting yields among the eight typical machine learning models tested in this study. In addition, we also found that different agricultural zones and temporal training settings affect prediction accuracy. The three models perform better as more winter wheat growing season information becomes available. Our findings highlight a potentially powerful tool to predict yield using multiple-source data and machine learning in other regions and for crops.

Wheat is a leading cereal grain throughout the world. Timely and reliable wheat yield prediction at a large scale is essential for the agricultural supply chain and global food security, especially in China as an important wheat producing and consuming country. The conventional approach using either climate or satellite data or both to build empirical and crop models has prevailed for decades. However, to what extent climate and satellite data can improve yield prediction is still unknown. In addition, socio-economic (SC) factors may also improve crop yield prediction, but their contributions need in-depth investigation, especially in regions with good irrigation conditions, sufficient fertilization, and pesticide application. Here, we performed the first attempt to predict wheat yield across China from 2001 to 2015 at the county-level by integrating multi-source data, including monthly climate data, satellite data (i.e., Vegetation indices (VIs)), and SC factors. The results show that incorporating all the datasets by using three machine learning methods (Ridge Regression (RR), Random Forest (RF), and Light Gradient Boosting (LightGBM)) can achieve the best performance in yield prediction (R²: 0.68~0.75), with the most individual contributions from climate (~0.53), followed by VIs (~0.45), and SC factors (~0.30). In addition, the combinations of VIs and climate data can capture inter-annual yield variability more effectively than other combinations (e.g., combinations of climate and SC, and combinations of VIs and SC), while combining SC with climate data can better capture spatial yield variability than others. Climate data can provide extra and unique information across the entire growing season, while the peak stage of VIs (Mar.~Apr.) do so. Furthermore, incorporating spatial information and soil proprieties into the benchmark models can improve wheat yield prediction by 0.06 and 0.12, respectively. The optimal wheat prediction can be achieved with approximately a two-month leading time before maturity. Our study develops timely and robust methods for winter wheat yield prediction at a large scale in China, which can be applied to other crops and regions.

Maize is an extremely important grain crop, and the demand has increased sharply throughout the world. China contributes nearly one-fifth of the total production alone with its decreasing arable land. Timely and accurate prediction of maize yield in China is critical for ensuring global food security. Previous studies primarily used either visible or near-infrared (NIR) based vegetation indices (VIs), or climate data, or both to predict crop yield. However, other satellite data from different spectral bands have been underutilized, which contain unique information on crop growth and yield. In addition, although a joint application of multi-source data significantly improves crop yield prediction, the combinations of input variables that could achieve the best results have not been well investigated. Here we integrated optical, fluorescence, thermal satellite, and environmental data to predict county-level maize yield across four agro-ecological zones (AEZs) in China using a regression-based method (LASSO), two machine learning (ML) methods (RF and XGBoost), and deep learning (DL) network (LSTM). The results showed that combining multi-source data explained more than 75% of yield variation. Satellite data at the silking stage contributed more information than other variables, and solar-induced chlorophyll fluorescence (SIF) had an almost equivalent performance with the enhanced vegetation index (EVI) largely due to the low signal to noise ratio and coarse spatial resolution. The extremely high temperature and vapor pressure deficit during the reproductive period were the most important climate variables affecting maize production in China. Soil properties and management factors contained extra information on crop growth conditions that cannot be fully captured by satellite and climate data. We found that ML and DL approaches definitely outperformed regression-based methods, and ML had more computational efficiency and easier generalizations relative to DL. Our study is an important effort to combine multi-source remote sensed and environmental data for large-scale yield prediction. The proposed methodology provides a paradigm for other crop yield predictions and in other regions.

Precise and timely information on crop spatial distribution over large areas is paramount to agricultural monitoring, food security, and policy development. Currently, automatically classifying crop types at a large scale is challenging due to the scarcity of ground data. Although previous studies have indicated that transductive transfer learning (TTL) is a promising method to address this problem, it performs poorly within regions where crop compositions and phenology differ largely. Here we transferred random forest classifiers trained in limited regions with diversified growing conditions and land covers to the rest of the study area where ground data are scarce, with more than 130,000 Sentinel-2 images processed using the Google Earth Engine (GEE) platform. We established the 10 m crop maps for four major crops (i.e., maize, rapeseed, winter, and spring Triticeae crops) across 10 European Union (EU) countries from 2018 to 2019. The final crop maps had a high accuracy with overall accuracy generally greater than 0.89, with user’s accuracy and producer’s accuracy ranging from 0.72 to 0.98. Moreover, the resulting maps were consistent with the NUTS-2 level official statistics, with R2 consistently greater than 0.9. We further analyzed the crop rotation patterns and found that the rotation intervals across these EU countries were generally at least one year. Maize was dominantly rotated with winter Triticeae crops or converted to other land covers in the following year. Rapeseed was generally grown in rotation with winter Triticeae crops, whereas the rotation patterns of winter and spring Triticeae crops were more diversified. Red Edge Position (REP) and Normalized Difference Yellow Index (NDYI) played significant roles in crop classification across the EU. This study highlights the potential of the developed TTL method for crop classification over large spatial extents where labeled data are limited and the differences in crop compositions and phenology are relatively large.

Identifying the rapeseed (Brassica napus L.) flowering dates are important for planting area estimation, growth monitoring, and yield estimation. However, there is currently a lack of data on rapeseed flowering dates at the parcel scale. In this study, a new spectral index (Normalized Rapeseed Flowering Index, NRFI) is proposed to detect rapeseed flowering dates from time series data generated from Landsat 8 OLI and Sentinel-2 sensors. This study also analyzed the feasibility of using the backscattering coefficients (VV, VH, and VV/VH) of Sentinel-1 to detect the flowering dates of rapeseed at the parcel scale. Based on the spectral and polarization characteristics of 718 rapeseed parcels collected in 2018, we developed a method to automatically identify peak flowering dates by the local maximum of NRFI series and the local minimum of VH and VV, along with the maximum of VV/VH. The results show that most of the peak flowering dates derived from Sentinel-1 and Sentinel-2 can be confirmed by the in-situ phenological observations at the Deutscher Wetterdienst (DWD) stations in Germany. The NRFI outperforms the Normalized Difference Yellow Index (NDYI) in identifying the peak flowering dates from Landsat 8. The derived medians of peak flowering dates by NRFI, NDYI (Sentinel-2), and VH are similar, while a systematic delay is observed by NDYI (Landsat 8). The method with the spectrum and backscattering coefficients will be a potential tool to identify crop flowering dynamics and map crop planting area.

Drought forecasting for basin water management requires accounting for precipitation deficits, climatic water balance, and soil-water storage across appropriate timescales. This study evaluates index timescales, lead times, and five machine learning models for the Yellow River Basin using 21 ERA5-Land variables from 1990 to 2025. We constructed the standardized precipitation index (SPI), Penman–Monteith standardized precipitation evapotranspiration index (SPEI-PM), and root-zone standardized soil moisture index (SSI-rootzone). A staged comparison used training-window standardization and rolling-origin cross-validation, followed by evaluation during 2021–2025. Scale screening retained 90-day indices. At a seven-day lead, histogram gradient boosting achieved R² values of 0.807, 0.852, and 0.995 for SPI, SPEI-PM, and SSI-rootzone, respectively. R² remained positive at 30 days but became negative at 90 days. Across 8548 grid cells, forecasts reproduced drought-area dynamics but underestimated severe and extreme drought extent, particularly for SPEI-PM. GPM and SMAP, excluded from model development, provided external comparisons; independently derived indices correlated with SPI and SSI-rootzone forecasts at 0.669 and 0.517. These findings show that evaluating forecasts for drought preparedness requires consideration of temporal aggregation, lead time, spatial extremes, and product dependence rather than basin-average accuracy alone.