Understanding and Reducing False Alarms in Observational Fog Prediction

The reduction in visibility that accompanies fog events presents a hazard to human safety and navigation. However, accurate fog prediction remains elusive, with numerical methods often unable to capture the conditions of fog formation, and observational methods having high false-alarm rates in order to obtain high hit rates of prediction. In this work, 5 years of observations from the Cabauw Experimental Site for Atmospheric Research are used to further investigate how false alarms may be reduced using the statistical method for diagnosing radiation-fog events from observations developed by Menut et al. (Boundary-Layer Meteorol 150:277-297, 2014). The method is assessed for forecast lead times of 1-6 h and implementing four optimization schemes to tune the prediction for different needs, compromising between confidence and risk. Prediction scores improve significantly with decreased lead time, with the possibility of achieving a hit rate of over 90% and a false-alarm rate of just 13%. In total, a further 31 combinations of predictive variables beyond the original combination are explored (including mostly, e.g., variables related to moisture and static stability of the boundary layer). Little change to the prediction scores indicates any appropriate combination of variables that measure saturation, turbulence, and near-surface cooling can be used. The remaining false-alarm periods are manually assessed, identifying the lack of spatio-temporal information (such as the temporal evolution of the local conditions and the advective history of the airmass) as the ultimate limiting factor in the methodology's predictive capabilities. Future observational studies are recommended that investigate the near-surface evolution of fog and the role of non-local heterogeneity on fog formation.