Comparisons among four different upscaling strategies for cultivar genetic parameters in rainfed spring wheat phenology simulations with the DSSAT-CERES-Wheat model

Abstract Cropping system models are widely used to assess the impacts of and adaptation practices to climate change on agricultural production. However, crop growth simulations at large scales have often lacked consideration of variation in crop cultivars, which were represented by different sets of genetic coefficients in crops models. In this study, taking the phenology of spring wheat (Triticum aestivum L.) as an example, we compared four different strategies for upscaling genetic parameters in phenology simulations at large scales with two experimental datasets. The first dataset was from field experiments comprising 40 different spring wheat cultivars at Altay (2014) and Yangling (2015–2017) station; the second dataset was historical (2010–2014) observed phenology records from 57 national agro-meteorological observation stations in China. The four strategies were the representative cultivar estimated at a single site (SSPs), the representative cultivar estimated at the 57 sites (NRPs), the various representative cultivars estimated at different agro-ecological zones (RRPs), and the virtual cultivars generated from the posterior distributions (VCPs). The posterior distributions aforesaid were established based on the calibrated parameter values of the 40 different spring wheat cultivars planted in Yangling. Then, 1000 sets of VCPs were randomly sampled from the posterior distributions. The results indicated that both the SSPs and NRPs strategy obtained large errors and uncertainties in spring wheat phenology simulations in China since only one representative cultivar was used. The RRPs strategy achieved the second high and the highest accuracy in anthesis and maturity data simulations. The VCPs strategy obtained the highest accuracy in anthesis simulation but relative larger errors in maturity simulation. The VCPs strategy can be directly used in large-scale crop growth simulations without tedious process of calibration. Hence, this strategy is recommended in areas where observations are scarce and for model users who not good at model parameter estimation.