ENSO climate risk: predicting crop yield variability and coherence using cluster-based PCA

The El Nino-Southern Oscillation (ENSO) has, in recent years, contributed to increases in the yields of major agricultural (annual) crops like wheat and barley in Canada. How such forcing alters the pattern of yield variation across different geographic scales and across large agricultural landscapes like the Canadian Prairies is less understood. Yet, such questions are of major importance in forecasting future cereal crop production. We explore the potential impact of ENSO on wheat and barley across the Canadian Prairies/Western Canada using a multi-scale, cluster-based principal component analysis (PCA) model that integrates machine-learning (K-means clustering) to predict areas of high climate risk. These risk areas are separable clusters of subregions that show similar ENSO-yield correlation response (spatial coherency). Benchmarking this spatial model to non-spatial models indicates that spatial coherency leads to gains in prediction skill. Incorporating spatial coherency increased the skill in predicting crop yield; reducing RMSE error by up to 26-34% (spring wheat) and 2-4% (barley). We infer that accounting for spatial coherency improves the accuracy and reliability of crop yield forecasts.