What are the dominant features of rainfall leading to realistic large‐scale crop yield simulations in West Africa?

A large‐scale crop model is forced by a range of climate datasets over West Africa to test the sensitivity of simulated yields to errors in input rainfall. The model skill, defined as the correlation between observed and simulated yield anomalies over 1968–1990 at the country scale, is used for assessment. We show that there are two essential rainfall features for the model to skillfully simulate interannual yield variability at the country scale: cumulative annual variability and frequency. At such a scale, providing additional information on intraseasonal variability, such as the chronology of rain events, does not improve the model skill. We suggest that such information is relevant at smaller spatial scales but is not spatially consistent enough to impact large‐scale yield variability.