A method for simulating risk profiles of wheat yield in data-sparse conditions

Crop models are informative but require detailed climate data that is not always available. In this work, researchers investigated the consequences of using extrapolated growing-season weather data in a crop model. The method relied on using adjustment factors based on differences in long-term averages between the reference and the test sites. Basing the adjustment factors on climate series of 40 years or more, improved the accuracy of the extrapolated climate. For rainfall, 40 years of data led to errors within 10%. Minimum and maximum temperatures were within 0.5 C at most test sites with records of 40 years or more. Accuracy of solar radiation varied little with record length.

Using these extrapolated climate records to model long-term risk profiles of wheat yields was more accurate when the adjustment factors were based on monthly averages compared to seasonal averages. There was less bias in the risk profiles of wheat yields when using extrapolated climates in areas with winter-dominant rainfall. Sites in subtropical and semiarid tropical zones were commonly overestimated and had the largest biases. The best results were obtained when all variables (rainfall, minimum and maximum temperature, and solar radiation) were adjusted using climate records of 80 years or more. For instance, extrapolating all the variables based on a 100-year climate record resulted in 85% of sites with bias of 10% or less. Extrapolating for fewer variables had greater bias and much less improvement with longer climate records.

This information is useful for crop modellers interested in data-poor environments. It can reduce computational time allowing for inclusion of other factors such as soil types. It also assists farmers and agronomists to judge the reliability of climate adjustments for risk profiles created using modelled yield.