A data-driven model for quantifying how maize yields in North-East China respond to summer climate

We demonstrate a data-driven approach that gives skilful annual predictions of province-scale maize yield across China's Northeast Farming Region (NFR), using the June – August (JJA) mean temperature and total precipitation as predictors. Our work builds on a method used to explore climate-related risks to maize production in the USA and South Africa. The approach uses a two-dimensional Gaussian function to parametrise the relationship between the detrended maize yields and summer climate conditions, which allows for non-linear growth responses to the individual climate variables. To enable skilful annual yield predictions we extend the approach in three ways: i) testing and validating out-of-sample yield and yield shock predictions; ii) introducing partial pooling of data across the provinces to allow for systematic differences between them, resulting in more robust model parameter estimation; iii) iterative correction of biases introduced by yield detrending procedures. Maximal yields occur for mean JJA temperature 21–22 °C and total JJA rainfall 400 mm (corresponding to monthly rainfall totals of ~130 mm), which broadly agree with previous applications of similar approaches in different countries, giving confidence that the approach is robust despite inherent approximations. The model also demonstrates skilful retrospective forecasts of province area-average maize yield, giving out-of-sample Pearson correlations of ~ 0.6 (statistically significant at 99.9% level) between forecasts and observations for 1979-2016. Furthermore, including the August Standardised Precipitation-Evapotranspiration Index (SPEI) as an extra predictor improves out-of-sample predictions for low yield events (e.g. yields at least 10% below average) associated with adverse climate conditions, as occurred in Liaoning in 2000. The extended modelling framework can provide maize yield predictions using either weather observations or seasonal climate forecasts, and offers the potential for climate services that could help manage impacts on the food system due to adverse weather conditions.