The Correspondence Ratio in Forecast Evaluation

Abstract

The correspondence ratio is developed to evaluate output from an ensemble of numerical weather prediction models. This measure is a simple extension of the threat score, or critical success index, to more than two fields and is used to measure the divergence of the forecast fields. The ratio is compared with two commonly used measures: the anomaly correlation, and the mean square error. Results indicate that the correspondence ratio is sensitive to the bias and, when calculated for several threshold values, can provide information beyond that supplied by the mean-square error and anomaly correlation measures. The correspondence ratio is particularly useful in evaluating discontinuous fields, such as precipitation. While no one measure can provide a complete assessment of forecast success, this ratio provides useful information that can increase our understanding of model forecast quality.