Parameters optimization of WOFOST model by integration of global sensitivity analysis and Bayesian calibration method
- Publication Type:
- Journal Article
- Nongye Gongcheng Xuebao/Transactions of the Chinese Society of Agricultural Engineering, 2016, 32 (2), pp. 169 - 179
- Issue Date:
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
© 2016, Editorial Department of the Transactions of the Chinese Society of Agricultural Engineering. All right reserved. Crop model calibration and validation are essential for model evaluation and application. It is important for model application to accurately estimate the values of crop model parameters and further improve the capacity of model prediction. In the previous researches, trial-and-error method was widely used in model calibration and validation. The deficiency of this method was subjective selection of parameter values and time-consuming processes. To overcome these issues, the optimization methods such as general likelihood uncertainty estimation (GLUE), genetic algorithm (GA) and shuffled complex evolution (SCE-UA) algorithm were alternative method for model calibration and validation. However, it is a problem to decide which parameters for optimization. It is essential to select the most sensitive parameters among hundreds of parameters in the crop model for optimization. To avoid subjective selection of parameters for calibration and validation, we used the global sensitivity analysis method of model parameters and the Markov Chain Monte Carlo (MCMC) method based on Bayesian theory to optimize the crop genetic parameters in the WOFOST (world food studies), and the data of three-year winter wheat field experiment in Luancheng in the North China Plain were adopted. The main objectives were: 1) to analyze the sensitivity and uncertainty of WOFOST brought by 55 crop genetic parameters using the extended Fourier amplitude sensitivity test; 2) to calibrate and validate the WOFOST using the MCMC method after sensitivity analysis. We found that: 1) The most sensitive parameters for maximum leaf area index (MAXLAI) in the crop growth period were successively: specific leaf area at development stage of 0, 0.5, 0.6, and 0.75, maximum CO2 assimilation rate at development stage of 1.5, and maximum relative increase in LAI (RGTLAI); 2) The most sensitive parameters for total above ground production (TAGP) in the crop growth period were successively: maximum CO2 assimilation rate at development stage of 1.5 (AMAXTB150), specific leaf area at development stage of 0 (SLATB00), life span of leaves growing at 35℃, extinction coefficient for diffuse visible light at development stage of 0 (KDIFFTB00), maximum CO2 assimilation rate at development stage of 1.8 (AMAXTB180), efficiency of conversion into storage organs (CVO); 3) The parameter sensitivity for MAXLAI and TAGP in potential and rain-fed production level was almost coincident, which indicated that yield level didn't influence the parameter sensitivity results; 4) Eleven sensitive parameters were selected for optimization by using the MCMC method. The first calibration and validation strategy (i.e. the data in 1998-1999 for calibration and those in 1999-2000 and 2000-2001 for validation), was better than other 2 strategies. 5) WOFOST simulation was much improved if the optimized parameters by the MCMC method were adopted. The index of agreement was higher than 0.9 and the relative root mean square error was less than 20%. However, WOFOST performed worse in rain-fed case because water stress factor was added to limit crop growth. The results indicate that more sensitive parameters should have priority in adjusting values for model calibration and validation. In addition, the MCMC method is a feasible optimization method for WOFOST calibration and validation.
Please use this identifier to cite or link to this item: