YIELDSTAT – A spatial yield model for agricultural crops

• YIELDSTAT, a model which combines statistical approaches and expert knowledge data to predict regional yields.
• YIELDSTAT considers combinations of soil, site, climate and management information.
• Model validation was performed at three spatial scales (experimental station, county, state) in Thuringia, Germany.
• YIELDSTAT employs CO2 and technology trends for future yield scenario simulations.

The YIELDSTAT model for crop yields, an advanced hybrid of traditional non-linear regression approaches and expert knowledge databases, was developed to predict the spatial distribution of yields for a range of arable crops (winter wheat, winter barley, winter rye, winter triticale, spring barley, oats, potato, sugar beet, winter oil-seed rape, silage maize, clover, clover/grass mix, lucerne, lucerne/grass mix, fodder grass) and two grassland types (intensive, extensive) in eastern Germany across different scales up to the regional scale. YIELDSTAT accounts for a wide range of yield-influencing factors derived from weather, soil, relief and management data, as well as for the long-term changing atmospheric CO2 concentration and for the trend owing to progress in breeding and agro-technology. YIELDSTAT regression modules were derived from several hundred farm data sets from 1975 to 1990 and tested against recent yield observations from the Federal State of Thuringia, Germany. The model test was performed at three different spatial scales. YIELDSTAT successfully reproduced the observed data at all three scales, with a normalised mean bias error of 3.02% across all crops and scales. Model testing also revealed a number of weaknesses in the model, identifying yield-reducing factors that had not been considered previously. All in all, the model proved fitness-for-purpose for simulating spatial yields, also under assumed future climate conditions.