ORIGINAL PAPER

An analysis of fog events at Belgrade International Airport

Katarina Veljović & Dragana Vujović & Lazar Lazić &

Vladan Vučković

Received: 18 July 2013 /Accepted: 3 January 2014# Springer-Verlag Wien 2014

Abstract A preliminary study of the occurrence of fog atBelgrade “Nikola Tesla” Airport was carried out using astatistical approach. The highest frequency of fog has occurredin the winter months of December and January and farexceeded the number of fog days in the spring and the begin-ning of autumn. The exceptionally foggy months, those hav-ing an extreme number of foggy days, occurred in January1989 (18 days), December 1998 (18 days), February 2005(17 days) and October 2001 (15 days). During the wintermonths (December, January and February) from 1990 to2005 (16 years), fog occurred most frequently between 0600and 1000 hours, and in the autumn, between 0500 and 0800hours. In summer, fog occurred most frequently between 0300and 0600 hours. During the 11-year period from 1995 to 2005,it was found that there was a 13 % chance for fog to occur ontwo consecutive days and a 5 % chance that it would occur3 days in a row. In October 2001, the fog was observed overnine consecutive days. During the winter half year, 52.3 % offog events observed at 0700 hours were in the presence ofstratus clouds and 41.4 % were without the presence of lowclouds. The 6-h cooling observed at the surface preceding theoccurrence of fog between 0000 and 0700 hours rangedmainly from 1 to 4 °C. A new method was applied to assessthe probability of fog occurrence based on complex fogcriteria. It was found that the highest probability of fog occur-rence (51.2 %) takes place in the cases in which the relativehumidity is above 97 %, the dew-point depression is 0 °C, thecloud base is lower than 50 m and the wind is calm or weak1 h before the onset of fog.

1 Introduction

Fog commences in a wide variety of conditions, reduces visi-bility and is a common barrier to all types of traffic (Willett1929; Croft 2003), particularly at airports. Fog that descends toground level over much of the land and lasts for a long periodcan lead to chaos at main international airports. Owing toweather conditions with poor visibility, aircraft have to divertto another airport, resulting in a considerable cost increase forairlines. To find a solution to this problem, some attempts havebeen made to develop methods to aid fog dispersal—seeding orheating (Schaefer 1946; Silverman and Weinstein 1974). Thedisadvantage of these methods is that they have some success attemperatures below freezing only. Thereby, timely prediction offog at airports increases public safety and has a high economicvalue. In addition, improved knowledge of the atmosphericconditions under which fog occurs can lead to a more accurateforecast of onset and dissipation of fog.

For these reasons, fog forecasting is a big challenge, and fogis the subject of a number of studies. Pagowski et al. (2004)stated that surface observations of fog date back nearly200 years. Taylor (1917) conducted experiments aimed towardsa better understanding of the physics of fog. To investigate therelative importance of the processes responsible for fog forma-tion and evolution, different numerical models have been de-veloped and used. There is no doubt that the processes in theboundary layer and the surface soil are of the greatest impor-tance for the formation and duration of fog. Several one-dimensional (1D) models have been developed to forecast fogat the local scale, but they are not much useful at localitycharacterised by heterogeneous terrain (Müller et al. 2007).The first 1D model, developed by Zdunkowski and Nielsen(1969), did not contain parameterization for the liquid watersedimentation nor turbulence exchange coefficients. Dew de-position at the surface is seen as a key factor in the formation offog (Lala et al. 1975, Pilié et al. 1975, Brown and Roach 1976,

K. Veljović (*) :D. Vujović : L. Lazić :V. VučkovićFaculty of Physics, Institute of Meteorology, University of Belgrade,Dobracina 16, Belgrade, Serbiae-mail: katarina@ff.bg.ac.rs

Theor Appl ClimatolDOI 10.1007/s00704-014-1090-6

Turton and Brown 1987). As a result of dew deposition at thesurface, downward transport of moisture occurs, which leads tothe formation of nocturnal dew-point inversions that can extendto 200 m above the surface (Cotton and Anthes 1989). Bergotet al. (2005) demonstrated how the use of surfacemeasurementsthrough the coupling of an assimilation scheme and a 1Dmodel, can significantly improve the prediction of fog.

As for fog forecasting, it seems that no confident method iscurrently available for the nowcasting. It is generally knownthat a deficit in the density of horizontal visibility observationsis considerable, and NWP models do not provide the time offog formations and fog clearances with sufficient accuracy.Likewise, success in fog forecasting is smaller because of highspatial and temporal fog variability due to changeable fogcharacteristics in different conditions and stages of its evolution(Bergot 2013). To alleviate the fog variability problem, Mülleret al. (2010) recommend a second-moment cloud-water schemewith a parameterization of the Köhler theory, which is com-bined with a mixed-phase Ferrier microphysics scheme.

In terms of fog definition, different visibility thresholds areused in fog studies (Tardif and Rasmussen 2007). According tothe international definition, fog forms when water vapour in theair, at the surface, condenses into minute water droplets andhorizontal visibility reduces to less than 1,000 m (UKMeteorological Office UKMO 1991). This can be achievedby either adding moisture to the air or by dropping the ambientair temperature to the dew point. In normal circumstances, fogoften (but not always) begins to form at a relative humidityslightly below 100%, but the fog formation is possible at lowerhumidity as well. For example, “urban” fog, related to theunusually large number of fine particles present in the loweratmosphere near the surface, commonly forms and exists at arelative humidity well below 100 % and is often particularlydense and, on some occasions, dry. These characteristics help toprolong the life of such fog type. Horizontal and vertical meanvelocities in fogs are typically small. However, fog formationshould not be expected only in windless conditions. There werecases of long lasting fogs occurring with strong winds andaccompanied by heavy rain (Gultepe 2007).

Two main physical processes responsible for fog formationare radiation cooling and mixing. Depending on the formationmechanism that dominates, fog can be roughly classified asradiation fog and as advection fog. According to the fogclassification, established byWillett (1929) and later modifiedby George (1951) and Byers (1959), additional fog types areupslope, frontal and steaming (evaporation) fogs. The last fogtype is formed by the evaporation of rain or water surfacewarmer than the air. Petterssen (1956) suggested fog classifi-cation based on temperature conditions under which fog oc-curs: liquid fog (T>−10 °C), mixed-phase fog (−10 °C>T>−30 °C) and ice fog (T<−30 °C). However, ice fog can form atT=−20 °C with steady conditions in the absence of mixing(Gultepe 2007).

The most studied fogs are those caused by radiationcooling. Radiation fog occurs at night, within the stableboundary layer, and usually lasts only a short time aftersunrise because the sun’s radiation gradually warms the sur-face. It can occur at any time of year but is most commonduring late autumn and early winter. Conditions favouring theformation of radiation fog are a clear sky or existence oflimited to thin high clouds, ample moisture in the surface layerand very light wind (Taylor 1917) accompanied with a tem-perature inversion (Willett 1929). The temperature inversionis caused by radiative cooling of the ground, and air in thelayers near the ground, and the heating of upper layers byadiabatic compression during development of anticyclones. Aclear sky facilitates more rapid cooling by providing a freeescape of long-wave radiation from the surface, and thus,increases low-level inversion strength. As Savoie and McKee(1995) showed, in the winter, inversions can be very persis-tent. The net solar energy during the day is not sufficient todestroy a stable layer alone. Along with snow cover, andsteady synoptic conditions, daily energy balance would tendto create stable layers or increase the strength of stable layersalready existing. Thermally stable air weakens and may stopturbulence above ground. A light wind limits vertical mixingand enhances cooling of the air near the ground (Roach 1995).In case of land radiation fog, it has been found that initial fogformation occurs when 10 m wind drops to 2 m s−1 or less(Findlater 1985). Following Gultepe (2007) in addition to thefog appearing by cooling under clear sky conditions, radiativefog has onemore aspect, fog induced by lowering stratus base.This cloud base lowering is due to radiative cooling at thecloud top which causes downward turbulent transport of coolair by eddies. Then, the sub-cloud layer is cooled by the latentheat of evaporation and that results in cloud base lowering(Pilié et al. 1979; Dupont et al. 2012).

The objective of this study is to achieve a better under-standing of climatological aspects of fog at Belgrade “NikolaTesla” Airport. The aim is to provide the climatology of thefog event occurrences at the airport, by using statistical find-ings about local dynamic and physical processes that takeplace during and proximate the fog occurrences, as well asto characterise the event itself.

Here, fog occurrence at Belgrade “Nikola Tesla” Airport isanalysed. The paper is organised as follows. Section 2 de-scribes geomorphological features of Belgrade “Nikola Tesla”Airport, data and adopted methodology. Section 3 presents theresults. Finally, Section 4 presents concluding remarks aboutmain findings and accordance with other studies.

2 Data and method

In Belgrade, the occurrence of fog is characteristicallyassociated with anticyclonic conditions in the winter

K. Veljović et al.

half year when there is overnight radiative cooling ofthe moist, and usually shallow, layer of air close to theground.

Location of the Belgrade “Nikola Tesla” Airport(44°49′10″ N, 20°18′25″ E) is depicted in Fig. 1. Theairport has an elevation of 101 m above sea level and issituated in the southern part of Pannonian valley, ap-proximately 17 km west from the centre of Belgrade.The closest high mountain to the airport is Avala,23 km to the southeast with a peak of 511 m. Towardthe north and northwest, the airport is occupied byPannonian valley and is under the influence of theweather predominant above the Polish and Ukrainianplateau. Hence, no mountain exists that has a directimpact on weather and climate in the area in whichthe airport lies. The airport is under two prevailing

winds throughout the year: a westerly wind and asoutheasterly local wind named Košava. The westerlyflow is the most evident, but with low intensity, asidefrom strong gusts from the Atlantic Ocean that causethunderstorms to form in late spring or early summer.

Fig. 1 A map showing the position of Belgrade “Nikola Tesla” Airport with respect to the Danube and Sava rivers, Avala Mountain, Pannonian valleyand the city of Belgrade. Elevations above sea level (in metres) are shown on the upper right corner of the left part of the figure

Fig. 2 Total annual number of days with fog at Belgrade “Nikola Tesla”Airport for the period 1973 to 2005 with linear trend. The straight line isobtained by linear regression

Fig. 3 Top, total number of days with fog per month; bottom, the dottedline—absolute monthly maximum number of days with fog per month;the straight line—mean number of days with fog per month. Analysis wasmade for Belgrade “Nikola Tesla”Airport and for the period 1973 to 2005

An analysis of fog events at Belgrade International Airport

Košava is strong, the southeastern wind in Serbia andsome nearby countries, and blows from September/October to March/April. It begins in the CarpathianMountains and travels northwest along the Danubethrough the Iron Gate region, where it gains character-istics of strong wind, and then continues to Belgrade.Some gusts of this wind can exceed 18–28 m s−1. Thepresence of the Danube and Sava rivers has an impacton fog (advection-radiation type) occurrence and synop-tic processes responsible for weather in this region.

Fog occurrences are defined here by using the visibilitythreshold corresponding to the foregoing internationaldefinition.

To identify fog events, half-hourly surface observa-tions from the National Climatic Data Centre serviceare used. The data source comprises half-hourly data ofvisibility, wind speed and direction, temperature anddew point, as well as cloud base height and cloudtype. The data are summarised based on UTC time.Belgrade local time is UTC+1. In a 35-year period,

from January 1973 to December 2005, 1,529 fogevents were recorded. It should be noted that the dataset is missing some measurements for April and May1999. In addition, measurements of the minimum tem-perature as well as radiosonde measurements (RAOB(Belgrade, station Košutnjak, elevation 203 m)) at0000 UTC are considered.

A novel method to assess the probability of fog occurrencebased on complex criteria described in Radinovic andBanjevic (1996) is applied. This method was originally devel-oped to provide assessment of hail risk in Serbia using param-eters that could be good indicators of cumulonimbus clouddevelopment. The idea of applying this methodology to thefog is due to the fact that cumulonimbus clouds also formunder a wide variety of conditions and are predictable with ahigh uncertainty (Ćurić et al. 2003). Here, parameters that arepotential indicators of fog occurrence are chosen. Four para-meters are chosen: 2-m relative humidity (RH), 2-m dew-point depression (DD), cloud base (CB) and wind velocity(v). All parameters are analysed at 1-h increments before fog

0 20 40 60 80 100 120

Mist, number of days per month

0

20

40

60

80

100

Fog,

num

ber

of d

ays

per

mon

thy = 0.75473 * x + 13.897; R=0.8998

Mist vs. fog, per month, for the period 1995-2005Fig. 4 The monthly correlationbetween number of days with fogand mist at Belgrade “NikolaTesla”Airport for the period 1995to 2005. The straight line isobtained by linear regression. R iscorrelation coefficient

Fig. 5 Diurnal frequency of fogonset at Belgrade “Nikola Tesla”Airport for the period 1990 to2005, for four seasons (seelegend), dashed line; the 2-monthmoving average, solid line

K. Veljović et al.

occurrence. The parameters take values from the boundedintervals (thresholds) that are found to be the most consistentwith fog occurrences. These intervals are divided into threesubintervals for each parameter.

The complex criterion (Cp) is calculated as:

Cp ¼ 1

N

X

i¼1

N nfinfi þ ni

where nfidepicts the number of fog events for a given value ofthe ith parameter, ni is the number of non-fog events for thesame value of the ith parameter. N is the total number ofparameters.

Probability of fog occurrence for the ith parameter is de-fined as:

pi ¼nfi

nfi þ ni:

3 Results and discussion

Over the period from 1973 to 2005, the total number ofdays with fog ranges from 30 to 72/year. As can be

seen in Fig. 2, in this period, there is a linear increaseof the number of fog days (0.273 days year−1), themaximum number of fog days being in the year of2003, which was 72 days.

The temporal distribution of fog-event numbers indi-cates that events occur mostly in winter months (De-cember and January) and far exceed the number of fogdays in the spring and the beginning of autumn(Fig. 3). The number of fog days in winter is almostsix times greater than in summer and 1.5 times greaterthan in autumn. Namely, in winter (December, Januaryand February (DJF)), there were 751 days with fog,whilst in summer (June, July and August (JJA)), andautumn (September, October and November (SON)),there were 128 and 503 days with fog, respectively.The exceptionally foggy months, containing an extremenumber of days with fog, were January 1989 (18 days),December 1998 (18 days), February 2005 (17 days)and October 2001 (15 days), when there is a minimumof solar radiation. The minimum frequency is found insummer.

Inspecting the relationship between monthly occur-rences of mist (1,000 m≤visibility≤2,000 m) and fogfor the period from 1995 to 2005, a very high correla-tion (R=0.89) between the two was found (Fig. 4). The

Tim

e of

day

(U

TC

)

Month

sunset

sunrise

Fig. 6 Top, UTC times of sunrise(black line) and sunset (grey line).Bottom, monthly frequency (inpercent) of the onset of advective(grey bars) and radiation (blackbars) fog, for Belgrade “NikolaTesla”Airport. Type classificationwas made following Tardif andRasmussen (2007). The datacover the period 1995 to 2005

An analysis of fog events at Belgrade International Airport

relationship between fog and mist is strong, linear, andpositive. It can be seen that two points (not shown, butthey refer to the months of April and May) deviate fromthe rest, possibly because the data set for April andMay 1999 was incomplete. In accordance to Quanet al. (2011), the occurrence of mist/haze and fog wasinfluenced by weather conditions, such as static stableweather condition and abundant vapours (for fog). Inaddition, the aerosol concentration might also be a fac-tor that influences the number of fog days.

As can be expected, for long nights in winter, fogoccurred most frequently at night and more frequentlythan in summer (Fig. 5). Generally, fog events tend tocommence during the last half of the night and early inthe morning. During the winter months (DJF) fromJanuary 1990 to December 2005 (16 years), fog oc-curred between 0000 and 1000 UTC and most

commonly between 0500 and 0900 UTC. In spring,fog occurred most frequently between 0400 and 0600 UTC;in summer, between 0300 and 0600 UTC; and in au-tumn, between 0400 and 0700 UTC. Winter fog fre-quency is almost four times greater than summer andspring fog frequency. Daytime fog occurrence is remark-ably low in spring, summer, and autumn.

We examined an existence of statistical dependenceamong successive occurrences of fog events over the11-year period from 1995 to 2005. It was found (notshown) that there was a 13 % chance that fog wouldoccur on two consecutive days and a 5 % chance that itwould occur 3 days in a row. One extreme event hadnine consecutive days of fog in October 2001. Thisresult indicates that successive occurrences of a fogevent cannot be used as a proper means to predictnext-day fog occurrences.

5 27

10871

12-30 -20 -10 0 10 20

Minimum temperature [°C]

0

200

400

600

800

1000

1200

Vis

ibili

ty [

m]

visibilitynumber of fog events

-6

-4

-2

0

2

4

6

8

C]

Tair - Tsoil [ C] during fog events, Belgrade Airport, 1995-2005

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Fig. 7 Top, visibility thresholds(in metres) and number of fogevents as a function of minimumtemperature, T5cm (in degreesCelsius). Bottom, air–soiltemperature differences, Tair–T5cm(in degrees Celsius); all for fogevents observed at 0700 hours atBelgrade “Nikola Tesla” Airportduring the period 1995–2005

K. Veljović et al.

Classification of fog events presented in this article is basedon the two main mechanisms of fog forming. Therefore,events are grouped into two categories: advection and radia-tion fog. Following the classification criterion for these twofog types that are proposed by Tardif and Rasmussen (2007),we examined data for an 11-year period (1995–2005). Thedistinction between advection and radiation fog was defined(e.g., whilst with radiation fog, the observed wind speed at thesurface is below 2.5 m s−1 (calm or weak wind condi-tions), the advection fog is associated with moderate

winds—greater than 2.5 m s−1). Classification criterionis applied to data for a period of 4 to 6 h before fogonset. During this period, 544 fog events were analysedand abovementioned fog types were selected. Thoseevents that did not meet any of the criteria were notclassified. Monthly distribution of advection and radia-tion fog onset in the form of frequency, along with thetimes of sunset–sunrise (which are related to the radia-tive cooling potential), is presented in Fig. 6. Resultsshow that radiation and advection fog are the mostprevailing types in autumn, winter, and early spring-time (monthly sum of the frequencies of these two typesfrom September to May is approximately 50 % orgreater, ∼76 % in April). Throughout almost the entireyear, radiation fog is more common than advection fog.Frequency of advection fog onset is the highest for thefirst 4-month period of the year. That is in agreementwith a high frequency of cyclones in the Adriatic Sea,during the same period. Also, frequency of advectionfog onset is higher than that of radiation fog onset onlyin April (1.28 times). Monthly frequency of advectionfog onset is increasing from January to April, July toSeptember and October to December. The greatest fre-quencies of radiation fog onset are in November (48 %)and December (46.2 %). This is not surprising bearingin mind that, in that period, there is a maximum surfaceradiation cooling along with the permanent inflow ofwater vapour from precipitation—mean monthly relativehumidity during this period is over 80 %, based onclimatological data. In addition, monthly trends in radi-ation fog onset are positive for the periods of Septemberto November, February to March and May to June.

-24 -20 -16 -12 -8 -4 0 4 8 12 16 20 24Temperature (°C)

200

400

600

800

1000

1200

1400

Hei

ght (

m)

NovemberOctoberDecemberJanuaryFebruaryMarch

Mean monthly T and Td profiles during fog events at 0000 UTC, Belgrade

Fig. 8 Monthly summarised average vertical profiles of temperature(solid lines) and dew-point temperature (dashed lines) in the surface layerfor Belgrade (Košutnjak station, 203 m) at 0000 UTC during the fogsoccurred in the winter half year (October–March) for the period 1995 to2005

0 50 100 150 200Fog events

-10

-8

-6

-4

-2

0

2

ΔΤ/Δ

τ [°

C/6

h]

Temperature changes 6 hours before fog onsets during the period 1995-2005Fig. 9 Six-hour temperaturechanges (in degrees Celsius)before fog onsets observed for0000–0070 hours at BelgradeInternational Airport for theperiod 1995 to 2005

An analysis of fog events at Belgrade International Airport

Analysing fog events observed at 0700 hours for an11-year period, it was found that 52 % of fogs were inthe stratus cloud presence and 41.4 % of fog eventswithout low clouds. Moreover, the average minimumtemperature in the presence of fog was −1.6 °C; thehighest number of fogs occurred in the conditions withminimum temperatures ranging from −10 to 0 °C(Fig. 7, top). Following Petterssen’s classification,85.6 % were liquid fogs, and the remainder (14.4 %)consisted of mixed-phase fogs. Being that only 2.3 % offogs occurred at a temperature below −20 °C and thelowest value of minimum temperature observed duringfog was −24 °C, ice fogs are rare or almost improbableat Belgrade “Nikola Tesla” Airport. In addition, 92.6 %of fog events were in the presence of a temperatureinversion. The temperature difference between air andsoil (T2m–T5cm) was 2.8±2.16 °C (Fig. 7, bottom).

Inversion layers have an impact on microclimates,and among other things, may be helpful in understand-ing climatological aspects of fog. To calculate thesurface-based inversion height and strength, RAOB datarelated to the cases with fogs at 0000 UTC during thewinter half year (from October to March) for the 11-year period are considered. Data showed that almost allfog events (∼96 %) occurring at 0000 UTC were ac-companied by surface-temperature inversions. Tempera-ture profiles are summarised on a monthly basis. Cli-matological monthly vertical profiles of temperature anddew-point in the surface layer (up to 850 mb level) areshown in Fig. 8. November and December have thehighest inversion depths that, when averaged over anumber of fog events, are about 350–550 m thick with2–5 °C temperature differences from the base to the topof the inversion. In terms of inversion strength in thepresence of fogs, along October and March show thehighest temperature increase (about 2 °C/100 m), andOctober has the most shallow inversion layer (∼150 mdeep). The temperature inversions in December, Januaryand February are similar to each other in strength(∼1 °C/100 m). Additionally, October has the warmestfogs, 7° warmer than November, until March has thecoldest fogs. Fog events during the most strong inver-sion in March could be a consequence of a largeamount of available moisture from precipitation fromprevious months and relatively cooled surface layer ofthe atmosphere. Dew-point depressions in cases of fogevents at midnight are about 2 °C in October, Novem-ber, and January, and about 3 °C in December, Februaryand March. Note that Belgrade “Nikola Tesla” Airporthas approximately 100 m lower elevation than the sta-tion from which RAOB data are considered.

The 6-h temperature cooling prior to fog occurrencesbetween 0000 and 0700 hours for the 11-year period is

Fig. 10 Four 12-h successive surface analyses of surface temperature (inKelvin; shading, see key) and mean sea-level pressure (contours) overEurope (ECMWF), from 0000 UTC 17 October 2001 (top panel) to1200 UTC 18 October 2001 (bottom). The isobar contour interval is500 Pa. The isotherm shading interval is 5 K. Location of Belgrade“Nikola Tesla” Airport is cross-marked

K. Veljović et al.

shown in Fig. 9. It was found that the average 6-hdecrease of air temperature before fogs was 2.45±1.65 °C. It is similar to the findings of Menut et al.(2013) who found maximum cooling of 3 °C/3 h beforefog events in the Paris area.

Fog occurrence is also analysed considering the syn-optic characteristics in relation to main meteorologicalforms development. The selected case is 17–19 October2001; overnight, fog began forming on 17 October 2001within a region of higher pressure and lasted for 44 h.The surface pressure and temperature weather maps overEurope before and during the fog formation (from0000 UTC 17 October 2001 to 1200 UTC 18 October2001) are shown in Fig. 10.

The synoptic situation over Europe on 18 October2001 at 0000 UTC is characterised by a deep steady-state cyclone located to the west of Ireland, over theeastern Atlantic, and a powerful blocking stationaryanticyclone over Scandinavia, Central, and Eastern Eu-rope which extends by a ridge even to the south of theMediterranean Sea. A high-pressure system that bringssettled weather and clear sky allows the fog to form.

Fog is most likely to occur when the dew-pointdepression is 2° or lower and the wind is 2.5 m s−1

or less (Fig. 11). The fog occurred after a couple ofsouth-south-easterly wind days. The winds were week,below 3 m s−1, at first but by the time the mist andthen fog occurred the winds became light and veeredto the west and southwest at the surface whilst ananticyclone over eastern Europe slowly became weaker.During the foggy days, the mean temperature wasmainly constant and in agreement with the Belgrademonthly average for October, 12 °C. The influence ofsolar heating is obvious, and this fog lasted for only afew hours after sunrise, so it can be classified as aradiation fog.

Focusing on assessing of the probability of fog oc-currence, after a careful inspection of the dataset (forthe period of 1995 to 2005) to analyse the conditionsprior to fog development, significant intervals for pa-rameters are chosen as proxies of fog occurrence

probability. The thresholds are chosen subjectively,inspecting plots shown in Fig. 12 and taking intoaccount the values of the parameters that were mostfrequently appearing. It can be seen that during the pre-fog, 1-h parameter values are distributed in small inter-vals, and the most favourable fog-preceding conditionsare approximately RH higher than 90 %, CB below200 m, and v below 3 m s−1. Significant intervals forparameters are chosen and divided in three subintervalsas follows: for RH (91, 94], (94, 97] and (97, 100]%,for dew-point depression [0.5, 1), [0.1, 0.5) and [0, 0.5)°C, for CB [100, 150), [50, 100) and [0, 50) m, and forwind speed [2.5, 3.5), [1.5, 2.5) and [0, 1.5) m s−1.Given that fog forms quickly, the number of fog eventsis calculated for the given value of the parameter 1 hbefore fog onset. The probability of fog occurrence forspecified values of parameters (Pi) and the probabilityfor Cp are shown in Table 1. It was found that:

& Conditions in which RH is above 91 % but below94 %, 1 h before fog development gave a 30.9 %chance of fog occurrence. The dew point depressionbetween 0.5 and 1 °C favours fog formation in25.8 % of cases, as CB between 100 and 150 mand v between 2.5 and 3.5 m s−1 gave a 10 and12.8 % probability of fog occurrence, respectively.Under all these conditions, there is a 20.1 % chanceof fog forming after 1 h.

& If RH is between 94 and 97 %, 1 h before fogdevelopment, the probability of fog occurrence is34.9 %. If the dew-point depression is between 0.1and 0.5 °C, fog will form in 44.1 % of all cases.CB between 50 and 10 m is suitable for fog devel-opment after 1 h in 45 % of cases, whereas, asexpected, v between 1.5 and 2.5 m s−1 gives a smallprobability of fog occurrence, approximately 16 %.The complex criterion calculated for these conditionsis 34 %.

& Best conditions for fog forming in 1 h are cases inwhich RH is above 97 % (pRH is 43.8 %), the dew-point depression is lower than 0.1 °C (pDD is

Fig. 11 Mean temperature (T (indegrees Celsius)), dew-point (Td(in degrees Celsius)), windintensity FF (in metres persecond; left vertical axis) andvisibility (in metres; right verticalaxis) as a function of time (UTC)on 17 and 18 October 2001.Dotted line denotes isoline of1,000-m visibility

An analysis of fog events at Belgrade International Airport

41.7 %), CB is lower than 50 m (pCB is 97.9 %)and the wind is weaker than 1.5 m s−1 (pv is21.3 %). The complex criterion gives a 51.2 %chance of fog forming in the next hour. It is inter-esting that, in conditions where the cloud base islower than 50 m, the chance for fog forming in thenext hour is very high considering the very commonradiation fog type at Belgrade location.

In addition, the correlations among factors that couldbe important for fog evolution are calculated. Fogevents are characterised by the correlation coefficientsof 0.64 and 0.58 between RH and visibility and RH andair temperature, respectively. The correlation betweenvisibility and low cloud presence is positive but lessstrong (0.46) and similar to the correlation betweenvisibility and wind speed (0.41). The height of thecloud base and the wind speed are in negativecorrelation.

4 Conclusions and recommendations

This study represents the first attempt towards a comprehen-sive analysis of fog climatology for Belgrade “Nikola Tesla”Airport. To achieve this, the objective statistical analysis ofmeteorological variables was carried out. The purpose of thisstudy is twofold.

Firstly, the study links fog development to conditionsin the lower troposphere and is in general agreementwith a number of previous research conclusions aboutfog-forming conditions but for different locations (e.g.Tardif and Rasmussen 2007, Gultepe 2007, Westcottand Kristovich 2009, Stolaki et al. 2009, Menut et al.2013).

Secondly, the study suggests a new method to assess theprobability of fog occurrence based on applying the above-explained complex fog criterion. Phenomena compatible withfog occurrences at the airport are identified, and probabilities offog occurrences under different atmospheric conditions arequantified.

The above results demonstrate that the occurrence offog at the airport is very high according to the interna-tionally accepted definition of fog. The 35-year fogclimatology (1973 to 2005) showed the highest fogoccurrence to be during the wintertime when insolationwas weak. The number of winter fog days was almostsix times greater than in summer and 1.5 times greaterthan in autumn. There was an increasing trend in thenumber of days with fog over the foregoing time period(0.273 days year−1). The result indicates that there is alinear relationship between occurrences of fog and mistat Belgrade “Nikola Tesla” Airport. Furthermore, fogevents were categorised into two types, using criteriathat were based on primary physical mechanisms re-sponsible for fog formation. Low visibility observationsassociated with either advection or calm conditions werefrequent. Fog occurrences were most likely when thedew-point depression is 2° or lower, and the wind is2.5 m s−1 or less. Looking at diurnal and seasonaltrends, lower visibilities were highly concentrated dur-ing late night and early morning hours throughout the

0 50 100 150 200 250 30080

82

84

86

88

90

92

94

96

98

100

RH

[%

]Relative humidities 1 hour before fog onsets during the period 1995-2005

0 50 100 150 2000

500

1000

1500

2000

2500

3000

Clo

ud b

ase

heig

ht [

m]

Cloud base heights 1 hour before fog onsets during the period 1995-2005

0 50 100 150 200Fog events

0

2

4

6

8

10

Win

d sp

eed

[m/s

]

Wind speeds 1 hour before fog onsets during the period 1995-2005

Fig. 12 Top, relative humidity (in percent); middle, cloud base height (inmetres); and bottom, wind intensity (in metres per second) 1 h before fogonsets at Belgrade “Nikola Tesla” Airport for the period 1995 to 2005

K. Veljović et al.

year, and peaks of probability of fog developmentthroughout the seasons follow dawn. That suggests thatradiation fog at Belgrade “Nikola Tesla” Airport is verycommon, and the fog occurrence is mainly due toradiative cooling, particularly in autumn morning hours.In addition, as daytime fog occurrences at Belgrade“Nikola Tesla” Airport are remarkably low in spring,summer and autumn, we can conclude that radiation fogis a prevailing fog type here. Development of advectionfog is more predominant in early spring and quiteuncommon in late spring and early summer. If we lookat the lowest frequencies of advection and radiation fogin late spring and early summer (May–June–July), wecan be sceptical about the number of fog occurrencesexplained by classifying them as radiation or advectionfog being that fog related to the lowering cloud baseand fog in precipitation were not taken into account.Conversely, the higher frequency of radiation than ad-vection fog in this period can be explained by the factthat this fog type is driven by the radiative coolingpotential along with moisture content (available mainlyfrom precipitation). It is known that this is the periodwith most of the precipitation in Belgrade: mean month-ly precipi ta t ion is in May (∼70.7 mm), June(∼90.4 mm) and July (∼66.5 mm). In one rare event,fog lasted nine consecutive days. In an attempt to fore-cast fog occurrence for the next day on the basis of fogevents on the current day, persistence of fog events wasanalysed and it was concluded that there are no possi-bilities of forecasting of fog persistence.

The increase in fog frequency observed at the airportis clear. Could this also be an effect of the rise in thelevel of air pollution in the nearby urban area? Toexplore this possibility, a more detailed study wouldbe required than the one attempted in this paper, asthe effect should be observable over a much wider area.Similar analyses will be of interest for studying thenature of fog dissipation at the airport as well.

It could be said that the most favourable conditionsfor the occurrence of fog were those in which theminimum temperature is ranging between −10 and

10 °C (liquid fogs). The average 6-h temperaturecooling prior to fog onsets observed between 0000 and0700 UTC was 2.45±1.65 °C.

Lowering of the cloud base at 50 m or below, along withrelative humidity higher than 97 % and calm or light winds(weaker than 1.5 m s−1) 1 h before fog events, gave thecomplex criterion probability of fog occurrence of 51.2 %.This probability is in agreement with the probability for fogoccurrence in the presence of stratus cloud, calculated fromthe observations (52 %).

Finally, a better understanding of the conditions underwhich fog occurs at Belgrade “Nikola Tesla” Airport shouldbe useful to arrive at the next step—an attempt atimplementing a 1D COBEL model (Bergot and Guedalia1994) as a new integrated 1D fog and low cloud forecastmethod that uses local observations.

Acknowledgements The Ministry of Science of Serbia under Grant176013 supported this research. The authors would also like to thankMr.Dragomir Bulatovic for technical assistance.

References

Bergot T (2013) Small-scale structure of radiation fog: a large-eddysimulation study. Quart J Roy Meteor Soc 139:1009–1112

Bergot T, Guedalia D (1994) Numerical forecasting of radiationfog. Part I: numerical model and sensitivity tests. Mon Weather Rev122:1218–1230

Bergot T, Carrer D, Noilhan J, Bougeault P (2005) Improved site-specificnumerical prediction of fog and low clouds: a feasibility study.Weather Forecast 20:627–646

Brown R, Roach WT (1976) The physics of radiation fog. II: a numericalstudy. Quart J Roy Meteor Soc 102:335–354

Byers HR (1959) General meteorology. McGraw-Hill, New YorkCotton WR, Anthes RA (1989) Storm and cloud dynamics. Academic,

San DiegoCroft PJ (2003) Fog. In: Holton JR, Curry JA, Pyle JA (eds) Encyclopedia

of atmospheric sciences. Academic, San Diego, pp 777–792Ćurić M, Janc D, Vujović D, Vučković V (2003) The 3-D model char-

acteristics of a Cb cloud which moves along a valley. MeteorolAtmos Phys 84:171–184

Dupont JC, Haeffelin M, Protat A, Bounil D, Boyouk N, Morille Y(2012) Stratus-fog formation and dissipation: a 6-day case study.Bound-Layer Meteorol 143:207–225

Table 1 Values of complex criterion for probability of fog occurrence (Cp(in percent)) at Belgrade “Nikola Tesla”Airport, were calculated based onfour parameters: relative humidity (RH), dew-point depression (DD),

cloud base (CB) and wind speed (v) and probabilities of the occurrenceof fog related to the individual values of parameters (p (in percent))

RH (%) pRH±sl (%) DD=T–Td (°C) pDD±sl (%) CB (m) pCB±sl (%) v (m s−1) pv±sl (%) Cp±sl (%)

91–94 30.9±0.9 0.5–1.0 25.8±0.8 100–150 10.0±1.4 2.5–3.5 12.8±0.5 20.1±0.4

94–97 34.9±2.1 0.1–0.5 44.1±0.8 50–100 45.0±2.2 1.5–2.5 16.1±0.6 34.0±0.5

97–100 43.8±0.9 <0.1 41.7±4.9 <50 97.9±2.1 <1.5 21.3±0.6 51.2±0.5

See text, for more details

sl significance level

An analysis of fog events at Belgrade International Airport

Findlater J (1985) Field investigations of radiation fog formations atoutstations. Meteorol Mag 114:187–201

George JJ (1951) Fog. In: Malon TF (ed) Compendium of meteorology.Amer Meteor Soc. pp. 1183–1187

Gultepe I (2007) Fog and boundary layer clouds: fog visibility andforecasting. Birkhäuser Verlag AG, Basel

Lala GG, Mandel E, Jiusto JE (1975) Numerical evaluation of radiationfog variables. J Atmos Sci 32:720–728

Menut L, Mailer S, Dupont JC, HaeffelinM, Elias T (2013) Predictabilityof the meteorological conditions favourable to radiative fog forma-tion during the 2011 ParisFog Campaign. Bound-Layer Meteorol.doi:10.1007/s10546-013-9875-1

Müller MD, Schmutz C, Parlow E (2007) A one-dimensional ensembleforecast and assimilation system for fog prediction. Pure ApplGeophys 164:1241–1264

Müller MD, Masbou M, Bott A (2010) Three-dimensional fog forecast-ing in complex terrain. Quart J Roy Meteor Soc 136:2189–2202

Pagowski M, Gultepe I, King P (2004) Analysis and modeling of anextremely dense fog event in Southern Ontario. J Appl MeteorClimatol 43:3–16

Petterssen S (1956) Weather analysis and forecasting. Volume II, weatherand weather systems. McGraw-Hill, New York, p 266

Pilié RJ, Mack EJ, KocmondWC, Rogers CW, Eadie WJ (1975) The lifecycle of valley fog. Part I: micrometeorological characteristics. JAppl Meteor 14:347–363

Pilié RJ, Mack EJ, Rogers CW, Katz U, Kocmond WC (1979) Theformation of marine fog and development of fog-stratus systemsalong the California Coast. J Appl Meteorol 18:1275–1286

Quan J, Zhang Q, He H, Liu J, Huang M, Jin H (2011) Analysis of theformation of fog and haze inNorth China Plain (NCP). AtmosChemPhys 11:8205–8214

Radinovic D, Banjevic D (1996) Development of improved criteria forhail suppression seeding in Serbia. J Weather Modif 28:35–38

Roach W (1995) Back to basics: Fog: part 2—the formation and dissi-pation of land fog. Weather 50:80–84

Savoie MH, McKee TB (1995) The role of wintertime radiation inmaintaining and destroying stable layers. Theor Appl Climatol 52:43–54

Schaefer VJ (1946) The production of ice crystals in a cloud ofsupercooled water droplets. Science 104:457–459

Silverman BA, Weinstein AI (1974) Fog. In: Hess WN (ed) Weather andclimate modification. Wiley, New York, pp 355–383

Stolaki SN, Kazadzis AS, Foris DV, Karacostas TS (2009) Fog charac-teristics at the airport of Thessaloniki, Greece. Nat Hazards EarthSyst Sci 9:1541–1549

Tardif R, Rasmussen RM (2007) Event-based climatology and typologyof fog in the New York City Region. J App Meteorol Climatol 46:1141–1168

Taylor GI (1917) The formation of fog and mist. Quart J RoyMeteor Soc43:241–268

Turton JD, Brown R (1987) A comparison of a numerical model ofradiation fog with detailed observations. Quart J Roy Meteor Soc113:37–54

UK Meteorological Office (UKMO) (1991) In: Lewis RPW (ed)Meteorological glossary, 6th edn. HMSO, London, p 335

Westcott NE, Kristovich DAR (2009) A climatology and case study ofcontinental cold season dense fog associated with low clouds. J ApplMeteor Climatol 48:2201–2214

Willett HC (1929) Fog and haze, their causes, distribution, and forecast-ing. Mon Weather Rev 56(11):435–459

Zdunkowski WB, Nielsen BC (1969) A preliminary prediction analysisof radiation fog. Pure Appl Geophys 75:278–299

K. Veljović et al.