We examined observation-based fog occurrence at three Czech monitoring sites: Praha 4 - Libuš, Košetice and Churáňov, representing different environments - urban, rural and mountain - over a time span of 27 years (1989-2015). We searched for a simple model describing fog occurrence fitting the observed air pollution and meteorological data. For our analysis we used a generalized additive model, GAM, with (penalized) spline components to capture possible nonlinear and a priori unknown functional relationships. In order to cope with the binary nature of the data (indicators of fog presence on individual days), we employed a logistic regression GAM model fitted by a maximizing penalized likelihood (where the penalty coefficients were estimated via cross-validation). After testing several physically motivated models, being guided by AIC and physical interpretation of the components, we arrived at a model which uses the following explanatory variables: relative humidity, ambient SO2 concentrations, ambient NOx concentrations, air temperature and seasonality. All associations between the response and the analysed explanatory variables were highly significant. According to our results, the most important explanatory variables modelling the fog probability were relative humidity and air pollutants. Interestingly, we observed an increasing trend in fog occurrence at all three sites under review starting around the mid 2000s.
Capsule: The most important explanatory variables modelling the fog probability at three Central European sites were humidity, SO2 and NOx. An increasing trend in fog occurrence has been observed since the mid 2000s.
Keywords: Air pollution; Fog; Generalized additive model; Meteorology.
Copyright © 2018 Elsevier B.V. All rights reserved.