Crop yield estimation is crucial for improving agricultural management and ensuring food security. Accurate remote sensing-based yield estimation typically relies on the optimal selection of the most relevant variables as predictors from those variables, such as vegetation indexes and physiological indicators. However, although crop growth indexes from traditional crop growth analysis have been proven to correlate with crop yields, their potential for remote sensing-based yield estimation has not been explored. In this study, we developed a remote sensing-based formula to derive the Relative Growth Rate (RGR), a common crop growth index, and introduced it to the rice yield estimation models. Its estimation performance was evaluated by comparing with the Ratio Vegetation Index (RVI), Difference Vegetation Index (DVI), Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), Enhanced Vegetation Index (EVI), Normalized Difference Water Index (NDWI), Gross Primary Productivity (GPP), and Leaf Area Index (LAI). The results from 5-fold cross-validation showed that the RGR-based models outperformed the models based on other predictors both individually (R² = 0.71, RMSE = 832.50 kg/ha) and incorporated with environmental indicators (R² = 0.78, RMSE = 719.63 kg/ha). The RGR-based model exhibited superior spatial robustness and temporal scalability to other predictors and achieved an R² value above 0.75 up to 16 days before maturity for within-season estimation. Model interpretability analysis indicated that the RGR from the vegetative growth phase to the early reproductive phase was more influential than in the late reproductive phase, and yield increased as the growth rate rose in most periods. These findings suggest the potential of RGR in remote sensing-based yield estimation.

Satellite and UAV-based remote sensing have been widely used for agricultural systems monitoring jointly. How to quantitatively optimize the efficiency of integrating these two techniques remains largely understudied. To address this gap, we, for the first time, formulate the configuration of satellite–UAV integrated system as a spatial sampling optimization problem and propose an SSO (spatial sampling optimization) model that jointly optimizes the spatial locations and flight paths of UAV sampling within the satellite monitoring area. The SSO model enables maximizing the accuracy of monitoring under a given cost constraint. We obtained comprehensive data in rapeseed fields and conducted experiments based on the SSO model. We compared the sampling effectiveness of the SSO model with that of simple random sampling, systematic sampling, equal stratified sampling and Neyman stratified sampling. The results showed that the SSO-optimized plan had the highest sampling efficiency, which was at least 38.7% higher than that of the best-performing conventional method (Neyman stratified sampling). Under the same cost constraint, the SSO-optimized sampling scheme can have 11.1% more sampling points than the conventional sampling scheme. The Elite Genetic Algorithm (EGA) performed well in solving the SSO model. The error of the SSO-optimized scheme was reduced by 27.3% and the sampling distance was reduced by 7000 to 8000 m on average. In conclusion, the proposed SSO model helps to optimize the configuration of satellite–UAV integrated remote sensing, thereby improving the cost-effectiveness of agricultural monitoring systems. We call for considering cost constraints and increasing efficiency in agricultural system monitoring and government censuses in the future.

Accurate rice yield estimation is essential for food security. Two key factors affecting estimation accuracy are the long-term upward trend in yield over time and regional heterogeneity across space. Current studies predominantly employ statistical detrending methods (e.g., moving averages, linear regression) to isolate temporal trends. However, such methods rely on prior assumptions about the time–yield relationship and may introduce systematic bias when these assumptions break down. Meanwhile, the individual contributions of temporal and spatial information, and their interactive effects, have not been systematically evaluated within a unified framework. We selected 112 rice-growing counties across six U.S. states (2000–2021), using vegetation index (Normalized Difference Vegetation Index), meteorological indicators (growing degree days, killing degree days, and cumulative precipitation), and spatiotemporal variables (year, longitude, and latitude). We designed six input configurations to compare conventional detrending against direct temporal variable inclusion, testing across four model architectures (Long Short-Term Memory, Random Forest, XGBoost, and Transformer). Results showed that: (1) directly inputting year significantly outperformed detrending across all models, with the combined spatiotemporal configuration achieving the best performance (LSTM R² = 0.61 vs. 0.54 for detrending); (2) year was the most important predictor in SHAP analysis, with spatiotemporal variables ranking higher than most meteorological and remote sensing variables; (3) spatial information consistently improved accuracy and mitigated systematic bias for extreme yield regions; (4) the combined configuration performed best across different states, years (including extreme climate events), and yield levels, achieving near-end-of-season accuracy at the grain-filling stage (1.5–2 months before harvest). This study demonstrates that integrating raw spatiotemporal data directly into deep learning models is more effective than statistical detrending, offering a simpler and more robust approach for large-scale crop yield estimation.