Development of crop models for combined seasonal weather and crop forecasting systems

Tim Wheeler, Andrew Challinor and Peter Craufurd

Departments of Agriculture and Meteorology, The University of Reading, Reading, UK.

Email t.r.wheeler@reading.ac.uk

Many crop models have been developed over the last 30 or so years. Existing crop models range in complexity from those based on empirical relationships between crop yield and weather, to process-based simulation models operating with detailed environmental inputs on hourly time-steps. In general, the predictive skill of crop models tends to vary greatly with location and with crop variety (genotype), because current models are poor at representing the interactions between genotype and environment. The accurate representation of crops within combined seasonal weather and crop forecasting systems presents new challenges for crop simulation models. For example, these new forecasting systems require that crops yields are simulated at much larger spatial and temporal scales (regions or states, as opposed to fields or farms). Also, simulated crop yields need to be those found on farms, rather than the yields found in optimal conditions which are simulated by most simulation models. Our approach has been to identify the spatial scale at which there is a strong correlation over time between climate/ weather and crop yield, and to develop the crop model at this scale. The analysis of Challinor et al. (abstract submitted for this conference) demonstrated a strong relationship between total seasonal rainfall and the yields of groundnut across India at a scale of 50-200km. We then developed a simple model of groundnut yields at this spatial scale. The details of this model, and the scaling issues of the simulations will be presented.