Assessing uncertainties in crop model simulations using daily bias-corrected Regional Circulation Model outputs

Outputs from the Florida State University/Center for Ocean-Atmospheric Prediction Studies (FSU/COAPS) regional spectral model were linked to the CERES-Maize dynamic crop model, and the sources of uncertainty in yield prediction at 3 sites in the southeastern USA were examined. Daily incoming solar radiation, Tmax and Tmin, and rainfall output data were obtained from 1987 to 2004 of retrospective forecasts (hindcasts) that contained 20 ensemble members. These raw hindcasts were bias-corrected on their cumulative probability functions by using the historical daily weather records prior to the 18 yr hindcasted period. Six combinations of the 4 meteorological variables from raw and bias-corrected hindcasts and climatological values were used as sets of weather inputs into the CERES-Maize crop model. Uncertainties related to these combinations of sets of weather inputs were analyzed. The bias-correction method improved values of monthly statistics of the ensemble compared to the raw hindcasts in relation to the observed data. The number and length of dry spells were also made more accurate with this correction. The main source of uncer- tainty in linking the FSU/COAPS climate model to the CERES-Maize crop model was the specific timing of the occurrence of dry spells during the cropping seasons. Plant growth stress caused by soil water deficit during crucial phenological states largely affects simulated yields. Operationally, the inability of FSU/COAPS to accurately predict the timing of the occurrence of dry spells makes its climate forecasts less useful for farmers wishing to optimize planting dates and crop varieties for crops with short crucial phenological phases, such as maize.