A parsimonious, multiple-regression model of wheat yield response to environment

Abstract

A database of nearly 2000 yield observations from winter wheat crops grown in UK trials between 1976 and 1993 was used to develop a new model of effects of weather on wheat yield. The intention was to build a model which was parsimonious (i.e., has the minimum number of parameters and maximum predictive power), but in which every parameter reflected a known climate effect on the UK crop-environment system to allow mechanistic interpretation. To this end, the model divided the effects of weather into phases which were predicted by a phenology model. A maximum set of possible weather effects in different phenological phases on yield was defined from prior knowledge. Two-thirds of the database was used to select which effects were necessary to include in the model and to estimate parameter values. The final model was tested against the independent data in the remaining third of the data set (246 aggregated yield observations) and showed predictive power ( r=0.41), which was improved when comparing against mean annual yields ( r=0.77). The final model allowed the relative importance of the 17 explanatory variables, and the weather effects they represent (defined before fitting), to be assessed. The most important weather effects were found to be: (1) negative effects of rainfall on agronomy before and during anthesis, during grain-filling and in the spring (2) winter frost damage (3) a positive effect of the temperature-driven duration of grain-filling and (4) a positive effect of radiation around anthesis, probably due to increased photosynthesis. The model developed here cannot be applied outside the UK, but the same approach could be employed for applications elsewhere, using appropriate yield, weather and management data.