Global sensitivity and uncertainty analysis of a dynamic agroecosystem model under different irrigation treatments

Abstract

Savings in consumptive use through limited or deficit irrigation in agriculture has become an increasingly viable source of additional water for places with high population growth such as the Colorado Front Range, USA. Crop models provide a mechanism to evaluate various management methods without performing costly and time-consuming experiments, e.g., field studies investigating irrigation scheduling and timing effects on crop growth. Few studies have focused on CERES-Maize crop model parameterization with respect to water-stressed conditions, and the model has previously been shown to overestimate evapotranspiration (ET) for limited irrigation treatments (stress during vegetative stage). It is therefore desirable to quantify the effects of CERES-Maize input parameters on model output responses typically used for calibration and/or important in limited irrigation management, including vegetative growth, crop yield, and ET. A sensitivity analysis (SA) utilizing the Morris one-at-a-time screening and Sobol’ variance-based methods was performed on CERES-Maize v4.5 input parameters affecting water balance and crop growth including soil hydraulic properties, phenological growth properties, and radiation use efficiency. CERES-Maize output responses of interest for the SA included anthesis date, maturity date, leaf number per stem, maximum leaf area index, yield, and cumulative ET. The SA study utilized five years of multi-replicate field management data (both full and limited irrigation treatments) for each combination of random input parameters. Results comparing the Morris mean and the Sobol’ total sensitivity index showed very high correlation between the two, indicating that in this study the computationally cheaper Morris method could have been used as an effective indicator of input parameter sensitivity. For the full irrigation treatment, CERES-Maize output responses were mostly sensitive to crop cultivar parameters. For the limited irrigation treatment, CERES-Maize leaf area index, yield, and ET output responses were highly influenced by soil lower limit and drained upper limit input parameters, which define water holding capacity. There was also a greater amount of interaction between input parameters for the limited irrigation treatment than for full irrigation. An uncertainty analysis was also conducted using model outputs from the Sobol’ SA method. In some cases, cumulative ET had higher values for limited irrigation than for full irrigation, further indicating the need to evaluate model processes governing ET under water stress. A new methodology for systematic calibration of CERES-Maize, based on the Morris and Sobol’ sensitivity indices for the two irrigation treatments, is proposed for future model evaluation as sensitivity differences between treatments indicates that existing CERES-Maize calibration procedures (typically based on non-stressed crops) may need to be reconsidered in cases of water stress.