Forecasting radiation fog at climatologically contrasting sites: evaluation of statistical methods and WRF

Abstract

A six-year climatology of radiation fog has been compiled at two sites: the Research Centre for the Lower Atmosphere (CIBA, Spain) and the Cabauw Experimental Site for Atmospheric Research (CESAR, The Netherlands). These sites are contrasted in terms of geographical situation, climate zone, altitude, humidity and soil water availability. Therefore, several climatological differences in fog abundance, onset, dissipation and duration have been quantified between the two sites. The more humid site (CESAR) is characterised by relatively short radiation fog events distributed throughout the year. However, radiation fog at the drier site (CIBA) is more persistent and appears during late autumn/winter months. In general, its formation requires more time after sunset (approximate to 2 h more), since further cooling is required to reach saturation. The forecast of these fog events has been evaluated through two different approaches. First, we extend the statistical method presented by Menut et al. (2014) (M14). This method uses statistics to define threshold values on key variables for fog formation (pre-fog) and verifies its predictability using observations and numerical model output. We present some of the most appropriate threshold values for the forecasting of pre-fog periods at both sites, which differ from those presented in M14 and depend on the optimisation of the hit rate or the false-alarm rate. Additionally, we also extend M14 by suggesting other variables as potential predictors for fog formation (friction velocity and visibility tendency). Finally, we focus on fog simulation by the Weather Research and Forecasting (WRF) model in terms of liquid water content. The WRF model was able to simulate radiation fog when configured with sophisticated physical options and high resolution. However it failed in simulating the onset, dissipation and the vertical extent of fog (which was overestimated). The model results were extremely sensitive to the spin-up time.