Evaluation of ceres-wheat simulation of wheat production in China
Crop models have been used extensively to assess the impacts of environmental change, but few studies have evaluated their performance at the regional scale. Here, we evaluate the performance of CERES-Wheat (Triticum aestivum L.) in simulating regional spatial and temporal characteristics of wheat production in China. The model uses genetic coefficients of representative cultivars within relatively homogeneous Agro Ecological Zones on a 50- by 50-km grid resolution with supported soil and daily weather inputs. Simulated maturity dates and yields are compared with observations (1998–2000) at 141 experimental agricultural stations, and with county-scale census yields (1980–2000). The model is evaluated with respect to its ability to capture the regional and temporal patterns of yield, measured by statistics including relative root mean square error (RMSE) and the index of agreement (d). Results show that the model captures the spatial patterns of yield variability reasonably well when compared to observed field data (RMSE% = 22.8% and d = 0.85) and census data (RMSE% = 27.0% and d = 0.76). Simulation of interannual variability in national production is very good (RMSE% = 8.2%, and d = 0.85). In spatial terms, simulation of variability varies considerably; good results are achieved in some of the main wheat planting areas, but not in others. The paper ends with a discussion of likely reasons for these differences in performance.