Multi-environment evaluation of peanut lines by model simulation with the cultivar coefficients derived from a reduced set of observed field data

A major limitation of the application of a crop simulation model is the determination of cultivar coefficients, as the recommended procedure requires extensive data sampling throughout the growing season which is very impractical when a large number of lines are involved or when critical resources are limited. Our previous study has shown that a reduced set of experimental data can be used to accurately estimate the cultivar coefficients of peanut lines as used by the CSM-CROPGRO-Peanut model. The objectives of this study were to verify our previous finding and to evaluate the derived cultivar coefficients in assisting multi-environment evaluation of peanut lines with the CSM-CROPGRO-Peanut model. Nine peanut lines in a regional yield trial (Set I) and ten peanut lines in a standard yield trial (Set II) were grown during the dry and rainy seasons of 2005. Data were collected on plant growth and development following the optimum protocol from our previous study. These data were used for model calibration to derive the cultivar coefficients of the individual peanut lines. Model calibration showed simulated values of phenology and growth characteristics of the peanut lines that were close to the corresponding observed values, with the coefficient of determination (r2) and the index of agreement (d) close to optimal values of 1, and a normalized root mean square error (RMSEn) smaller than 35%. Genetic variation among lines in cultivar coefficients was also observed. The initial model evaluation with data collected in the 2004 rainy season confirmed that model prediction was good for independent data, i.e., giving high values of r2 and d; and small RMSEn. The derived cultivar coefficients were shown to enable the CSM-CROPGRO-Peanut model to satisfactorily mimic yield ranking and stability of peanut lines in the Set I and Set II yield trials with 10 and 8 environments, respectively. Among the top five highest yielding lines based on mean observed pod yield (upper 56% for the Set I yield trial and upper 50% for the Set II yield trial), four lines were identified by model simulation in both sets. Also, the same top yielding lines in the two sets were identified by both simulation and experimentation. The model predicted similar GGE biplot patterns as present in observed trials, and also identified the same stable lines as the observed data. It is concluded that a reduced set of field data can be used for model application in assisting the multi-environment evaluation of peanut lines.