Assimilating remote sensing data into a crop model improves winter wheat yield estimation based on regional irrigation data

Abstract

Irrigation plays an important role in crop yield production in arid and semi-arid regions. However, irrigation effects have not been well addressed in the application of crop models at a regional scale due to limited data availability, which constrains the reliability and accuracy of simulation results. Assimilating remote sensing information into crop models can provide a viable approach to reduce associated uncertainties. In this study, regional irrigation data for winter wheat (Triticum aestivum L.) grown on the Loess Plateau was used to calibrate and validate the ChinaAgrosys (China Agricultural System) crop model at site and regional scales. Remote sensing data was then assimilated into the ChinaAgrosys crop model under four assimilation schemes. Two remotely sensed assimilation state variables (i.e., LAI and NDVI) and two assimilation algorithms (i.e., PSO (Partical swarm optimization) and SCE-UA (Shuffled complex evolution)) were considered. During the winter wheat growing season on the Loess Plateau, 30.6% of the wheat production area was irrigated once, 6.7% was irrigated two times, 3.7% was irrigated three times, and the remaining wheat area was rainfed. The R2 values between maturity date, LAI, and yield simulated by the ChinaAgrosys crop model and observations at 21 agrometeorological stations on the Loess Plateau were greater than 0.73, 0.44, and 0.60, respectively, during 2010–2015. The accuracy and spatial heterogeneity of winter wheat yield estimation were effectively improved by assimilating remote sensing data into the ChinaAgrosys crop model based on regional irrigation data. Under the four assimilation schemes, the combination of PSO+NDVI produced the highest accuracy for yield estimation in Hongtong county (92.8%), followed by SCE-UA+NDVI (92.0%). Our results demonstrated the importance of accounting for the spatial heterogeneity of water availability when applying a crop model in arid and semi-arid regions. Additionally, our analysis regarding different assimilation state variables and algorithms indicated that both simulation accuracy and calculation efficiency should be considered when assimilating remote sensing data into a crop model for simulating crop growth at regional scales.