Assimilation of coupled microwave/thermal infrared soil moisture profiles into a crop model for robust maize yield estimates over Southeast United States

Abstract

Abstract Global food security is one of the most pressing issues of the current century, particularly for developing nations. Agricultural simulation models can be a key component in testing new technologies, seeds and cultivars etc., however, inaccurate input information in addition to model related errors adds to model uncertainties. Satellite observations of soil moisture (SM), vegetation index etc. can be assimilated into crop models to reduce input and model related uncertainties. The goal of this study was to determine if assimilation of satellite-driven SM profiles can improve crop model yield estimations in a commonly used crop model while reducing the reliance on field management information at regional scales. Toward reaching this goal, this study utilized satellite derived microwave and thermal-infrared coupled SM profiles assimilated into a crop model via the Ensemble Kalman Filter over parts of the Southeastern United States from 2006 to 2010. The gridded Decision Support System for AgroTechnology Transfer (GriDSSAT) model was used to test and verify the SM profile assimilation methodology in two specific modes: (a) first using the best state-of-the art climate inputs available without data assimilation (“open-loop”); (b) then enhancing the open-loop model by assimilating into the model Remote sensing Driven SM Profile (RDSMP) data where available. The National Agricultural Statistical Services (NASS) reported yield data at county levels were used for comparison and validation purposes. For non-irrigated regions, the GriDSSAT model simulation (rain-fed) showed an overall absolute error of nearly 36 % in comparison with the reported NASS yields, whereas the error in crop yields after data assimilation was only 29 %. Over irrigated counties, the GriDSSAT model simulation (irrigated) showed an error of nearly 24 % while using the data assimilation model simulation without irrigation information also showed overall error of ∼23 %. Compared to the open-loop model simulation and improvement in simulated crop yields of nearly 2.5 times was found over irrigated regions (23 % vs 58 %).