Abstract
A model output statistics (MOS) scheme, using a stepwise-selection multiple linear regression approach, is used to estimate marine fog probability at 24 h intervals from 0 to 48 h, for the North Pacific Ocean (30–60°N) summer season. The predictand is uniquely determined by present and past weather, visibility and low-level cloud information in the primary synoptic report. Available predictors include 158 Fleet Numerical Oceanography Center model output parameters as well as monthly climatological fog frequencies. The regression equations, containing at most seven terms, are derived from 0000 GMT synoptic ship report data in the months of July 1976, 1977 and 1979. Evaporative heat flux and fog frequency are among the first four parameters selected for each equation. Reliability and sharpness diagrams are presented to identify specific biases in estimating fog probability. Categorical (percentage correct, threat and Heidke skill) and probabilistic (Brier) scoring methods are used to establish the credibility level of the MOS scheme in relation to climatology and FNOC's advective-fog probability program. The MOS scheme outperforms its competitors with score ranges for an independent August 1979 data set as follows: Heidke-skill, 0.47 to 0.40; threat; 0.45 to 0.42; Brier, 0.27 to 0.34; percentage correct, 78% to 70%.
