Effect of temporal and spatial scales of weather data on crop yield forecasts

Abstract

A dynamical crop model is used to understand the statistical relationship between climate and wheat yield in China. Using 16 years of regional yield data and a gridded database of climate data, it is shown that a clear statistical relationship is lacking in many points between seasonal mean climate anomalies and yield. Dynamical simulations of yield using daily climate data to drive the crop model demonstrate that this is partly due to the strongly nonlinear relationship between climate and yield, which can thus hamper the application of linear statistical models for forecasting and planning purposes. A case study using individual station data emphasized how local variations in climate could be substantial, and could have marked effects on simulated yields. Due to the memory effect of the soil moisture, little difference was found when driving the crop model with 5 d (pentad) temporally averaged rainfall data. This result is relevant for crop studies, as reliable pentad satellite-based rainfall retrievals are available for a far longer period than daily data. The highly skewed distribution of rainfall implies 10 d or longer averaging of rainfall strongly impacts crop yield, although temperature, a spatially smoother field, could be averaged for dekads or even monthly. Spatially averaging the driving climate data shows that the crop model skill in yield forecasting improves as the spatial scale of weather data increased from 2° × 2° to 0.5° × 5°, even if point-wise sampling errors are greater. © Inter-Research 2012.