6-011 (P) The Joint International Conference on “Sustainable Energy and Environment (SEE)”

1-3 December 2004, Hua Hin, Thailand

Forecasting Fog at Donmuang Airport in Winter

Sathaporn Ruangjun* and Robert H. B. Exell
The Joint Graduate School of Energy and Environment , King Mongkut’s University of Technology Thonburi, Bangkok, Thailand

Abstract: A method has been developed for predicting visibility over Donmuang airport at 0700 a.m. in winter from surface
meteorological observations at 0100 a.m. and 0500 a.m. using multiple linear regression. Data from the Thai Meteorological
Departments weather station at Bangkok International Airport in December, January and February were used. For each month two
models were found: one containing all the available surface observations, and one omitting the insignificant observations. The model
forecast consists of the probabilities of fog, poor visibility, and good visibility, respectively.

Keywords: Fog, Visibility, Forecast, Donmuang Airport, Bangkok, Winter.

1. INTRODUCTION made during the preceding early morning hours.

Data from the meteorological stations of the Royal Thai Air
Widespread fog causes problems of safety for air, sea and Force (RTAF) and the Thai Meteorological Department (TMD)
land traffic due to low visibility. In particular, when fog at Donmuang Airport were used in a linear regression model
occurs over airfield runways and taxiways, pilots and air from which probability values were calculated for fog(visibility
traffic control personals are concerned about the risk of less than 1 km), poor visibility (1 km to 4.8 km) and good
accidents. In Thailand, fog occurs over airfields in winter, visibility (over 4.8 km).
producing low visibility and a low ceiling over the ground. Another set of data uesd was from the Royal Thai Air Force
Therefore, there is a need for research on fog forecasting so Base at Donmuang. Results from this data set are available
that reliable warnings of low visibility due to fog can be issued elsewhere [3].
to pilots.
The World Meteorological Organization (WMO) reports 2. MATERIALS AND METHOD
“fog” (with symbol FG) when visibility is less than 1,000
meters, and “mist” (with symbol BR) when visibility is from The models tested in this research for predicting fog at
1,000 to 9,000 meters [1]. Weather records show that fog Donmuang airport were multiple linear regression using
occurs as the surface temperature approaches the dew point surface meteorological observations at by the International
temperature and the relative humidity is more than 75 percent. airport. Donmuang airport is 20 km north of Bangkok city. The
Under these conditions visibility may be reduced to less than 1 data sets used for finding and testing the regression models are
km. The creation of these conditions at a specific location listed in Table 1.
depends on many factors, including the synoptic situation, the The Independent Variables were values at 0100 and 0500
wind speed and direction, the stability of the atmosphere, the Local Standard Time of the following observations
air temperature and the dew point temperature. At Donmuang
Air Force Base, latitude 13° 55’ N , longitude 100° 36’ E, (i) T, Screen air temperature, (0 C)
about 20 km north of the city of Bangkok, and 45 km north of (ii) C, Cloud cover (oktas)
the coast of the Gulf of Thailand, fog sometimes occurs near (iii) T-Td, Dew point depression (0 C)
the surface and produces poor visibility in the morning in (iv) WSIN X, E-W component of surface wind (knots)
winter (December to February). (v) WCOS X, N-S component of surface wind (knots)
Report by Vinai Thongphasuk(2002)[2] concerns the (vi) LOG10 VIS, The logarithm to base 10 of surface
formation of fog over Donmuang Airport in the northeast visibility (km)
monsoon season during the periods 9-11 February 1999 (vii) P, Pressure (inches of mercury)
and13-16 February 2000. The conditions favorable for fog at (viii) P, Pressure change in the last 24 hours (inches of
sunrise were found to be as follows: mercury)

• Pressure falling in a high pressure over Thailand. The logarithm of the visibility is chosen instead of the
• An easterly wind above the surface with speed 5 visibility itself for two reasons. The first, the extinction and
knots scattering of light passing though the atmosphere are
• A clear sky during the night multiplicative with respect to distance, and second, small
• Upper air soundings at 0700 local standard changes in visibility at a short distance are more important than
time(LST) at Bang Na about 30 km south of the same changes in visibility at long distances. The logarithm
Donmuang showed a temperature inversion near the therefore gives a natural scale for the processes influencing
surface and stable air above visibility which is expanded at short distances.
Use of log10 vis instead of the actual visibility which in all
• Dew point at the surface almost equal to the air
reports falls between 100 m and 10 km, transforms the
temperature giving a relative humidity almost 100 %.
observations into the interval –1 to +1. Since log10 (1)= 0 and
The minimum visibility in thick fog was 50 meters at log10 (4.8270) = 0.6837, we have log10 vis ≤ 0 for fog, 0 <
06:00 LST and 150 meters at 06:30 LST. log10 vis ≤ 0.6837 for poor visibility, and log10 vis ≥ 0.6837
The objective of the research reported in this paper was to for good visibility.
find a model for forecasting fog at dawn over Donmuang Air The SPSS program version 9.0 was used with the variables
Port during the winter months of December, January and defined above to find multiple linear regression equations
February using routine surface meteorological observations relating the dependent variable to the independent variables.
The full regression equations used for the forecasting of log10
vis from the observations had the form
Corresponding author: ruangjun31@yahoo.com

704
6-011 (P) The Joint International Conference on “Sustainable Energy and Environment (SEE)”
1-3 December 2004, Hua Hin, Thailand

log10 vis = a0+a1T+a2C+a3(T-Td)+a4WSIN X+a5WCOS X Log10 vis(0700) = -27.455+0.014Td-0.042SIN X-0.008RH

+ a6 log10vis+a7P+a8 ∆ P (1) +1.009 Log10 vis(0500)+0.927P+0.01W (6)

The coefficients a0,…,a8 were obtained by the “ENTER” Table 2 Model properties
method of the SPSS program. Another set of coefficients for a __________________________________________________
simplified regression equation omitting non-significant terms
was obtained by the “STEPWISE” method of the program. Properties Model
Values of the multiple correlation coefficient R, the multiple
coefficient of determination R2 , and the standard error of TMD-111 TMD-511 TMD-121 TMD-521
estimate (S) were also given in the program outputs. __________________________________________________
The Standard Error of Estimate(S) is a measure of the
scattering of the observed values of the dependent variable yi = Input data time 0100 0500 0100 0500
log10 visi about the corresponding values yest,i = Method of ENTER ENTER STEPWISE STEPWISE
Analysis
a0+a1Ti+…+a8 ∆ Pi estimated from the regression equation. We Correlation 0.575 0.763 0.572 0.763
assume that the scattered values have a normal distribution with Coefficient ( R )
mean yest,i which is different for each data point, and a standard Standard error 0.2574 0.2038 0.2578 0.2035
deviation, which is a constant for each regression equation. of Estimate (S)
This normal distribution has been used to calculate for each
set of observations specifying the values of the independent
variables the probabilities of fog, poor visibility, and good
visibility as follow.
Let yest,i be a predicted value of log10 vis using one of the
regression models, and let S be the standard error of estimate of 1.0 M M
0.8 H
the model. Let Z = (log10 vis – yest,i)/S be the corresponding 0.6
standard values of log10 vis for this distribution. Also let Z0 and
FORECAST 0.4
0.2 M H F
Z1 be the standard values corresponding to vis = 1 km and vis = 0.0
4.8 km respectively. Then the probabilities of fog, poor -0.2
visibility and good visibility are given respectively by
-0.4 H F F
-0.6
-0.8
P{fog} = P{Z< Z0} = erf (Z0),
-1.0
P{poor visibility} = P{ Z0 ≤ Z ≤ Z1} -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
= erf(Z1) – erf(Z0),
P{good visibility} = P{Z > Z1}
= 1 – erf(Z1) (2) ACTUAL

Table 1 Data sets used

Fig. 1 TMD-111 model.
December January, February
To find regression models 1997 - 2001 1999 - 2003 To
test regression models 2003 2004
1.0 M M H
3. RESULTS 0.8
0.6
FORECAST

0.4 M H F
Here we present the regression models found from combined 0.2
0.0
data covering the winter period December through February. -0.2
The model names, specification, and statistics are listed in table -0.4
-0.6 H F F
2. The actual regression equations found from the input data -0.8
were as follows: -1.0
TMD-111: -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Log10 vis(0700) = -19.534-0.004T+0.106C+0.019
(T-Td)+0.01WSIN X+0.007WCOS X ACTUAL
+1.219 Log10 vis(0100)+0.641P+ 0.340 ∆ P (3)
TMD-511:
Fig. 2 TMD-511 model.
Log10vis(0700)=-28.0374+0.008T+0.207C+0.025
(T-Td)-0.011WSIN X-0.001WCOS X
+1.008 Log10 vis(0500)+0.925P+ 0.334 ∆ P (4)
TMD-121:

Log10 vis(0700) = -26.082+0.109C+0.022(T-Td)+0.010

WSIN X +1.215 Log10 vis(0100)+0.856P (5)
TMD-521:

705
6-011 (P) The Joint International Conference on “Sustainable Energy and Environment (SEE)”
1-3 December 2004, Hua Hin, Thailand

1.0 M M H
0.8
0.6 M
0.4
FORECAST

0.2 H F
0.0
-0.2
-0.4
-0.6 H F F
-0.8
-1.0
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

ACTUAL

Fig. 5 Predicted probability classes (high visibility) on 8

Fig. 3 TMD-121 model. December 2003.

In Fig. 6 the value of log10 vis(0700) predicted from the

meteorological data at 0100 LST was 0.1323, which makes the
1.0 M M H
predicted visibility 1.4 Km(poor). The probabilities of fog, poor
0.8 visibility and good visibility were 30.36%, 68.04% and 1.61%,
0.6
0.4 respectively. The actual visibility on this day(26 December
FORECAST

0.2 M H F
0.0 2003) at 0700 LST was 1 Km.
-0.2
-0.4 H F F
-0.6
-0.8
-1.0
-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

ACTUAL

Fig. 4 TMD-521 model.

Figures 1 to 4 show the results of the model tests as plots of Fig. 6 Predicted probability classes (low visibility) on
the predicted values of log10 vis versus observed values of log10 26 December 2003.
vis at 0700 LST.
Boxes labelled H represent hits (correct forecasts). Boxes 4. CONCLUSIONS
labelled M represent misses (failures to predict fog or poor
visibility). Boxes labelled F represent false alarms (actual Fog and poor visibility are difficult to forecast. The models
visibility better than predicted). developed in this research are not perfect, but they have been
Examples of the predicted probabilities of fog, poor proved to give results that will be of value to forecasters with
visibility and good visibility are illustrated in Figs 5 and 6. In their predictions. At the present time computer programs are
Fig. 5 the value of log10 vis(0700) predicted from the being developed in the Royal Thai Air Force Base, Weather
meteorological data at 0100 LST was 0.8476, which makes the division by means of which forecasters can obtain probabilities
predicted visibility 7.0 Km(good). The probabilities of fog, of fog, poor visibility and good visibility at 0700 LST from the
poor visibility and good visibility were 0.05%, 26.18% and weather observations at 0100 LST and 0500 LST.
73.12%, respectively. The actual visibility on this day (8
December 2003) at 0700 LST was 9 km. ACKNOWLEDGEMENTS
Details of the calculations for the results presented above
may be found in reference [3]. This reference also gives details The authors gratefully acknowledge the contribution of
of models calculated from observations made by the Royal Thai [RHBE, RTAF, TMD].
Air Force at Donmuang during the same period.
REFERENCES
[1] Suwanpong, S.(1974). Weather Instrument Handbook,
Thai Meteorological Department, pp. 2-3.
[2] Thongphasuk, V. (2001). Forecasting Dense Fog Covering
Donmuang Aerodrome, Thai Meteorological Department,
pp. 41- 42.
[3] Ruangjun, S. (2004) Fog Formation over Donmuang Air
Force Base During Winter, Msater degree thesis, JGSEE,
King Mongkut’s University of Technology Thonburi.

706