Field-level crop yield mapping with Landsat using a hierarchical data assimilation approach

Spatially explicit information on crop yields is essential for farmers and agricultural agencies to make well-informed decisions. One approach to estimate crop yield with remote sensing is data assimilation, which integrates sequential observations of canopy development from remote sensing into model simulations of crop growth processes. However, it remains challenging to efficiently scale this approach to large areas while maintaining reliable prediction at field scales. In this paper, we explored the factors limiting the generalization of the data assimilation approach and found that the accuracy of crop model prediction and the systematic model errors can significantly affect the performance of data assimilation and the yield estimation. To address these issues, we propose a hierarchical data assimilation framework, which enables maize yield estimation at field levels across large areas for the Midwestern US with no a priori knowledge about the management of individual fields. This approach applies data assimilation algorithms at two spatial scales. At the county scale, we adopted a Markov Chain Monte Carlo algorithm to recalibrate uncertain and sensitive model parameters based on aggregated Leaf Area Index (LAI) time series derived from Landsat images and county-level yield statistics. Using the county-specific models, we assimilated LAI time series into crop model simulations using Ensemble Kalman Filter for individual fields or pixels. This method was validated by multiple field-level maize yield datasets across major production states in the US Midwest. The Root Mean Squared Error ranges from 1.4 to 2.3 ton/ha, and the percentage error is between 9% and 21%. The hierarchical data assimilation framework provides a novel solution that downscales county-level yield statistics to 30-meter resolution yield maps, which can inform between and within field maize yield variability. This study contributes valuable insights towards practical large-scale crop yield mapping at high resolutions.