Concurrent data assimilation and model-based optimization of irrigation scheduling

Abstract

Crop models can be combined with optimization algorithms in order to develop management tools. However, such model-based tools are inherently affected by the imperfectness of the model on which they are based. In this paper we describe a procedure in which data assimilation and partial re-parametrization of the model embedded in the optimization procedure used to determine irrigation scheduling are performed before each optimization run. Furthermore, sensitivity analysis is performed before performing data assimilation, which ensures that only influential parameters are adjusted. The procedure was tested via simulation with DSSAT-CROPGRO for a hypothetical processing tomato crop in Davis, CA, in 2010–2019. Several scenarios that differed in terms of the measurements assumed to be available for assimilation (leaf area index, biomass and/or soil water content) were simulated. The results were compared to a benchmark scenario involving a perfect model, as well as a scenario in which data assimilation was not performed. The analysis focused on the overall performance of the irrigation schedule (yield vs. irrigation amount) derived using the model rather on the accuracy of the estimated model parameters. Assimilating weekly measurements of leaf area index led to overall performance that was within 3% of the benchmark performance. Adding weekly measurements of biomass or daily measurements of soil water content did not improve the performance. On the other hand, assimilating only daily soil water content measurements led to poorer results (5% decrease compared to benchmark) and also affected the repeatability of the results. Defining dynamically the subset of parameters for calibration via sensitivity analysis rather than calibrating a fixed subset of parameters or all parameters was beneficial, both in terms of overall performance and repeatability of the results. Overall, concurrent data assimilation and model-based optimization has potential to enhance irrigation scheduling decision making, particularly in water limited environments.