Assimilation of Sentinel-2 Estimated LAI into a Crop Model: Influence of Timing and Frequency of Acquisitions on Simulation of Water Stress and Biomass Production of Winter Wheat

Abstract

The Sentinel-2 (S2) Toolbox permits for the automated retrieval of leaf area index (LAI). LAI assimilation into crop simulation models could aid to improve the prediction accuracy for biomass at field level. We investigated if the combined effects of assimilation date and corresponding growth stage plus observational frequency have an impact on the crop model-based simulation of water stress and biomass production. We simulated winter wheat growth in nine fields in Germany over two years. S2 LAI estimations for each field were categorized into three phases, depending on the development stage of the crop at acquisition date (tillering, stem elongation, booting to flowering). LAI was assimilated in every possible combinational setup using the ensemble Kalman filter (EnKF). We evaluated the performance of the simulations based on the comparison of measured and simulated aboveground biomass at harvest. The results showed that the effects on water stress remained largely limited, because it mostly occurred after we stopped LAI assimilation. With regard to aboveground biomass, we found that the assimilation of only one LAI estimate from either the tillering or the booting to flowering stage resulted in simulated biomass values similar or closer to measured values than in those where more than one LAI estimate from the stem elongation phase were assimilated. LAI assimilation after the tillering phase might therefore be not necessarily required, as it may not lead to the desired improvement effect.

Keywords

• data assimilation
• leaf area index
• ensemble Kalman filter
• dynamic crop simulation model
• Sentinel-2
• remote sensing