When projecting future crop production, the skill of regional scale (>100 km resolution) crop models is limited by the spatial and temporal accuracy of the calibration and weather data used. The skill of climate models in reproducing surface properties such as mean temperature and rainfall patterns is of critical importance for the simulation of crop yield. However, the impact of input data errors on the skill of regional scale crop models has not been systematically quantified. We evaluate the impact of specific data error scenarios on the skill of regional scale hindcasts of groundnut yield in the Gujarat region of India, using observed input data with the GLAM crop model. Two methods were employed to introduce error into rainfall, temperature and crop yield inputs at seasonal and climatological timescales: (1) random temporal resequencing, and (2) biasing values. We find that, because the study region is rainfall limited, errors in rainfall data have the most significant impact on model skill overall. More generally, we find that errors in inter-annual variability of seasonal temperature and precipitation cause the greatest crop model error. Errors in the crop yield data used for calibration increased root mean square error by up to 143%. Given that cropping systems are subject both to a changing climate and to ongoing efforts to reduce the yield gap, both potential and actual crop productivity at the regional scale need to be measured. We identify three key endeavours that can improve the ability to assess future crop productivity at the regional scale: (i) increasingly accurate representation of inter-annual climate variability in climate models; (ii) similar studies with other crop models to identify their relative strengths in dealing with different types of climate model error; (iii) the development of techniques to assess potential and actual yields, with associated confidence ranges, at the regional scale.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)