A crop model cross calibration for use in regional climate impacts studies

Crop simulation models are widely used to assess the impacts of and adaptation to climate change in relation to agricultural production. However, a substantial mismatch often exists between the spatial and temporal scale of available data and the requirements of crop simulation models. Conventional model calibration methods which concentrate on a model's performance at plot scale cannot be used for large scale regional simulation (especially for climate change impacts assessments), given the limited observed data and the iterative calibration needed. One primary purpose of regional simulation is to predict the spatial yield variation and temporal yield fluctuation. This purpose could be fulfilled through model input calibration in which the objective of the calibration focuses on spatial or temporal agreement between simulated and observed values. This study examines the performance of CERES-Rice at the regional scale across China using a cross calibration process based on limited experiment data, agroecological zones (AEZ) and 50km×50km grid scale geographical database. Model performance is evaluated using rice yields from experimental sites at the plot scale, and/or observed yield data at the county scale. Results suggest: the CERES-Rice model was able to simulate the site-specific rice production with good performance in most of China, with a root mean square error (RMSE)=991kgha−1 and a relative RMSE=14.9% for yield across China. The cross calibration process, in which AEZ-scale parameter values were derived, gave a relative bigger bias to yield estimation, with a RMSE=1485kgha−1 and a relative RMSE=22.5%, but achieved a reasonable agreement with observed maturity day and yield at spatial scale. The bias rose further if this cross calibrated model was used to simulate the real farmer rice yields at a regional scale, with a RMSE=2191kgha−1 and relative RMSE=34% across China. The pattern of yield variation was captured spatially by the model in most of the rice planting areas, but not temporally. The sources of uncertainties were analyzed for both plot scale and regional scale simulation. This calibration process could be incorporated into climate change integrated assessment and adaptation assessment, especially for those developing counties with limited observed data.