A parsimonious Bayesian crop growth model for water-limited winter wheat

Dynamic crop models are widely used to simulate crop production, but are often complex and thus face parameter non-identifiability issues. In this study, we demonstrate a simple dynamic model for crop growth within a Bayesian hierarchical framework and quantify improvements in predicted crop growth patterns by inclusion of a dynamic water balance. Seven crop parameters and four water balance parameters were estimated by the model from data on leaf area index, biomass, yield, and plant available water over the growing season across multiple environments. Posterior median values for Willmott agreement index (d) and Nash-Sutcliffe efficiency (NSE) showed that the model predicted leaf area index (LAI; d =0.89; NSE =0.62) and biomass well (d =0.98; NSE =0.92) with less success for plant available water (PAW; d =0.75; NSE =-0.03) and grain yield (d =0.90; NSE =0.40). Inclusion of a water balance component raised median NSE from 0.57 to 0.62 for LAI, 0.74 to 0.92 for biomass, and -0.58 to 0.40 for grain yield. The median and highest density interval (HDI) were biologically plausible for the parameters canopy light extinction coefficient (median=0.46; HDI=[0.38, 0.55]), maximum leaf area index (median=6.66; HDI=[6.06, 7.24]), and crop evapotransipration coefficients for the initial (median=0.03; HDI=[0.000004, 0.08]) and mid -season (median=0.32; HDI=[0.23, 0.42]) growth stages. Posterior values for the potential radiation use efficiency (RUEp; g MJ(-1) (C-d)(-1)) parameter (median=2.17; HDI=[1.99, 2.34]) were not biologically plausible indicating a need for model improvement.