Mesoscale model parameterisation of fog in arid environments

Abstract

Fog is a natural hazard as it causes visibility to drop below 1 km due to condensation processes in the atmosphere. Yet, fog is not a standard output from numerical weather prediction models which are used for operational weather forecasts. Instead, proxy indicators are calculated in the model post processing to indicate if fog is forecast. There are traditionally three approaches to achieve this, the first is to use the liquid water content at the lowest model level, the second is to define visibility based on model output variables and the third is to use a multi rule based approach, also based on model variables. All of these approaches are affected by model bias and model parameterisation which will ultimately affect the forecast skill. While these methods have been reported on in the literature for localities in non-arid regions there has been little application to fog in arid environments. In this thesis we aim to explore and optimise these methods for forecasting fog in arid environments using the Weather Research and Forecasting (WRF) numerical weather prediction (NWP) model. The main study site was Abu Dhabi International Airport in the United Arab Emirates (UAE). The first fog climatology for the region is presented in Chapter 2 based on Meteorological Aerodrome Reports (METAR) and radiosonde data, an important long term data set. In addition, the first spatial dynamics of fog is presented in Chapter 3 utilising the Spinning Enhanced Visible and InfraRed Imager (SEVIRI) on board the Meteosat satellite. An important result from the fog climatology analysis helped guide the methodology in Chapter 5, “A rule-based method for diagnosing radiation fog in an arid region from NWP forecasts”. This was related to the surface inversion depth, which was deeper on fog mornings. A rule was applied to the model where inversion depth needed to be greater than 250 m to diagnose fog. This ruled helped separate fog days from non-fog days in the model simulations and improved the fog forecast. The satellite data was used extensively for a qualitative assessment of the modelled fog patch. Model physics is introduced in Chapter 4, more specifically, the land surface model in WRF. A nocturnal cold bias in the model simulations was identified at the Abu Dhabi study site. This resulted in overactive saturation and therefore false alarms in the forecast. To address this cold bias, the model code was adapted to indirectly adjust the thermal roughness length per land cover type. For the arid desert this meant a decrease in the roughness length of heat, Z0t, which will decrease the coefficient of heat exchange, CH, and therefore increase of land surface temperature and by association increase the 2m air temperature. The thermal roughness length was adjusted through the Zilitinkevich parameter (Czil), an input variable in the model. The default value for all land cover types is 0.1, but we found that a value of 0.5 for desert improved the mean gross error in late afternoon/early evening temperature up to 1.48 °C. This chapter showed that the bias could be solved in part through adjusting the Czil parameter.