Coefficients for Debiasing Forecasts

Abstract

Skill-score decompositions can be used to analyze the effects of bias on forecasting skill. However, since bias terms are typically squared, and bias is measured in skill-score units rather than in units of the forecasts, such decompositions only partially describe the nature of the bias. They do not describe the direction of bias or measure the amount of correction required to remove bias. Simple ?debiasing? coefficients can be derived to supplement the skill-score decomposition by indicating the direction and amount of bias, relative to unbiased forecasts. Calculation of these coefficients is described and an example is provided.