Parameter Estimation and Verification of DSSAT-CERES-Wheat Model for Simulation of Growth and Development of Winter Wheat under Water Stresses at Different Growth Stages
Yao Ning,Zhou Yuangang,Song Libing,Liu Jian,Li Yi,Wu Shufang,Feng Hao,He Jianqiang
DOI: https://doi.org/10.11975/j.issn.1002-6819.2015.12.019
2015-01-01
Abstract:Crop growth simulation models are useful tools to help us understand and regulate the agro-ecological systems in arid areas. In this study, the CERES-Wheat, a wheat growth simulation model in the DSSAT (decision support system for agrotechnology transfer) software, was investigated for its ability to simulate the growth and yield of winter wheat (Triticum aestivum L.) in arid areas and to find the optimal plan for the estimation of genetic parameters and the model verification. Field experiments were conducted under a rainout shelter for winter wheat growing under water stresses at different growth stages in 2 growth seasons (from October 2012 to June 2013 and from October 2013 to June 2014). The whole growth season of wheat was divided into 5 growing stages (wintering, greening, jointing, heading and grain filling). Water stress occurred every 2 continuous stages while irrigations were applied at other stages, which resulted in 4 different levels of water stress period (D1-D4). Two irrigation levels of 40 mm (I1) and 80 mm (I2) were applied. There were a total of 8 treatments, with 3 replicates for each, and the split-plot experiment was designed. An extra control treatment with irrigation in all 5 stages was arranged near. The experimental data were used to run the model. A total of 5 different plans for model calibration and verification were designed and the DSSAT-GLUE, a program package for parameter estimation in DSSAT, was used to estimate the relevant genetic coefficients. Then the 5 plans were compared for the discrepancies between corresponding simulated and observed values of phenological phase, single grain weight, biomass, yield and soil moisture so as to determine the accuracy of CERES-Wheat model to simulate the agro-ecological processes of winter wheat farming system in arid areas. The results showed that 2 genetic coefficients P1V (days required to complete vernalization at optimum vernalizing temperature) and G3 (standard dry weight of a single tiller without stress at maturity) varied remarkably under different scenarios of water stress. The coefficients of variation were 19.07% and 16.34%, respectively. It suggested that the values of these 2 parameters were influenced heavily by genotype-environment interactions. The rest of parameters were relatively independent of water stress scenarios since the coefficients of variation were all less than 10%. The DSSAT-GLUE package was proved to have good convergence since the estimated values of most genetic coefficients converged into narrow ranges. For output variables, the different plans of model calibration and verification showed great discrepancy. Plan 1 (model calibration used the data from the CK treatments with sufficient irrigation and model verification used the data from the rest of treatments in the 2 growth seasons) was proved to be the optimal one since its absolute relative error (ARE) and relative root mean squared error (RRMSE) for model calibration were the lowest, only 4.89% and 5.18%, respectively. When water stresses occurred during the heading and grain-filling stages, CERES-Wheat model was able to correctly simulate the dynamic changes in growth and development of wheat as well as the soil moisture content. However, when water stresses occurred during the wintering and greening stages, there were relatively large simulation errors. When water stress occurred earlier and severer, the simulation accuracy was lower. In addition, CERES-Wheat model could not correctly simulate the phenological discrepancies caused by different water stress scenarios because current algorithm for phenology estimation was only based on temperature and photoperiod but neglecting the secondary effects by water stress. Thus an improvement on current phenology algorithm of winter wheat was needed. The results of leave-one-out cross validation showed that the overall error was about 15%-18% for CERES-Wheat model to simulate winter wheat growth and yield under different scenarios of water stress designed in this study. In general, there were some limitations for CERES-Wheat model to simulate winter wheat growth under arid conditions. It was necessary to research into the mechanism and simulation method of winter growth responding to water stresses in early vegetative stage, if CERES-Wheat was expected to be applied more widely in the management and research of winter wheat production in arid and semi-arid areas in China.