What question did this study set out to answer?

This research aims to evaluate various data assimilation schemes using E-WOFOST and PROSAIL models for accurately predicting winter wheat yield.

May 25, 2026Open Access

Comparison between different data assimilation schemes by integrating the E-WOFOST and PROSAIL models on winter wheat yield prediction

Puntos clave

This research aims to evaluate various data assimilation schemes using E-WOFOST and PROSAIL models for accurately predicting winter wheat yield.
Integrated E-WOFOST and PROSAIL models across 23 field sites in Henan Province
Developed and tested 14 data assimilation schemes combining remote-sensing data, soil moisture, and assimilation algorithms
Analyzed the performance of assimilation schemes using metrics such as RMSE and R²
CI red-edge index showed best performance with R² of 0.70 and RMSE of 732.71 kg ha-1
PSO optimization reduced RMSE by 13.25 kg ha-1 compared to GA and by 201.40 kg ha-1 compared to SCE-UA
The best assimilation scheme yielded R² of 0.71 with RMSE of 719.46 kg ha-1, demonstrating robust performance across multiple years and regions

Resumen

Accurate estimation of the winter wheat yield is critical for agricultural management and food security. Crop models are essential tools for simulating crop growth and predicting yield. However, uncertainties owing to variability in crop model parameters cannot be ignored. We integrated the enhanced WOrld FOod STudies (E-WOFOST) and PROSAIL models to compare different data assimilation schemes for yield prediction across 23 field sites in Henan Province. A comparison between data assimilation schemes was conducted based on three aspects: (i) remote-sensing crop data, (ii) soil moisture (SM), and (iii) assimilation algorithms. These aspects were crosscombined to form 14 assimilation schemes. The assimilation results of different remote-sensing crop data showed that vegetation indices (VIs) outperformed Sentinel-2 (S2) reflectance, while the latter generally surpassed the use of satellite products and measured values. Reanalysis of products incorporating deep-layer SM achieved better results than using only surface-layer measured SM. Among the five VIs, the chlorophyll index red edge (CI red-edge ) demonstrated the best performance ( R 2 =0.70; RMSE=732.71 kg ha -1 ; nRMSE=15.46%). Relative to the Genetic Algorithm and Shuffled Complex Evolution-University of Arizona (SCE-UA), Particle Swarm Optimization (PSO) reduced the RMSE/nRMSE by 13.25 ha -1 /0.28% and 201.40 ha -1 /4.25%, respectively. Among the 14 assimilation schemes, the highest accuracy was obtained by assimilating ERA5-Land SM, constraining E-WOFOST with CI red-edge , and optimizing via PSO ( R 2 =0.71, RMSE=719.46 kg ha -1 , and nRMSE=15.18%). Finally, application of the best scheme demonstrated strong interannual performance in Henan (2018–2020 and 2024: R 2 =0.50–0.63; RMSE=719.76–771.67 kg ha -1 ) and transferred effectively to Shandong (2025: R 2 =0.63; RMSE=951.11 kg ha -1 ), demonstrating its robust applicability across different years and regions.

Leer artículo completoexternamente

Preguntar a la IA

Me gusta

Guardar

Ver artículo completo