Forecasts of “normal”
Simon J. Mason, Christopher A. T. Ferro, Willem A. Landman
Abstract
Abstract The difficulty of forecasting “normal” climate conditions is demonstrated in the context of bivariate normally distributed forecasts and observations. Deterministic and probabilistic skill scores for the normal category are less than for the outer category for all‐but‐perfect models. There are two important mathematical properties of the normal category in a three‐category climatologically equiprobable forecast system that affect the scores for this category. First, the normal category can achieve the highest probability less frequently than the outer categories, and far less frequently in contexts of weak to moderate skill. Second, there are upper limits to the probability the normal category can reach. These mathematical constraints suggest that summary measures of skill may underestimate the predictability and forecast‐skill of extreme events, and that subjective inputs to probabilistic forecasts may need to take greater account of limitations to the predictability of normal conditions.