Use of climate indices to predict corn yields in southeast USA

The impact of large-scale oceanic and atmospheric climate patterns on county corn yields in Alabama, Florida, and Georgia were evaluated for the period 1970-2005. Associations between detrended corn yield residuals, precipitation. surface temperature, and climate indices were explored by correlation analysis. Significant correlations were found with indices of the Pacific-North American pattern, tropical North Atlantic and eastern tropical Pacific sea surface temperatures, and two indices of the Bermuda high. The summer index of the Bermuda high (BHI) was the only index to be significantly correlated with yield residuals during the critical summer tasselling period; a time when corn is most Susceptible to water stress. Due to the hi-h degree of multi-collinearity found between the five indices, leave-n-out cross-validated principal component regression was conducted using all possible combinations of indices to predict corn yield residuals. Three indices produced models with the greatest skill using both lagged (known prior to planting) and concurrent indices (cross-validated Pearson's r: 0.679) and using lagged indices only (cross-validated Pearson's r: 0.569). Using the cross-validated models 99.2 and 96.9% of predicted county yields showed predictive skill (based on tercile hit scores) using both concurrent and lagged indices, and lagged-only indices, respectively. The cross-validated model using lagged-only indices indicated the results that can be achieved using known climate index values as early as the winter before the spring planting. Copyright (C) 2008 Royal Meteorological Society