SEPTEMBER 1997

CROFT ET AL.

545

Fog Forecasting for the Southern Region: A Conceptual Model Approach

PAUL J. CROFT
Jackson State University, Jackson, Mississippi

RUSSELL L. PFOST
National Weather Service, Jackson, Mississippi

JEFFREY M. MEDLIN
National Weather Service, Mobile, Alabama

G. ALAN JOHNSON
National Weather Service, Slidell, Louisiana
(Manuscript received 1 February 1996, in final form 21 April 1997)
ABSTRACT
The prediction of fog occurrence, extent, duration, and intensity remains difficult despite improvements in
numerical guidance and modeling of the fog phenomenon. This is because of the dependency of fog on microphysical and mesoscale processes that act within the boundary layer and that, in turn, are forced by the prevailing
synoptic regime. Given existing and new technologies and techniques already available to the operational
forecaster, fog prediction may be improved by the development and application of a simple conceptual model.
A preliminary attempt at such a model is presented for the southern region of the United States (gulf coastal
states) and requires information regarding cloud condensation nuclei, moisture availability (or saturation), and
dynamic forcing. Each of these factors are assessed with regard to their extent and evolution with time. An
illustration, and potential application, of how the model could be used is detailed as no extensive operational
testing has yet been completed. Instead, the model is applied in hindcast to verify its application. Successful
use of the model will require an operational forecaster to assimilate all available tools including climatology,
numerical guidance, sounding analysis, model diagnostic software, and satellite imagery. These must be used
to characterize and quantify the nature of the local and regional boundary layer in the forecast region according
to macroscale forcing and moisture availability, the initial local settings and boundary layer, qualitative assessment
of cloud condensation nuclei, and the interaction of these in time and space. Once identified, the evolution of
the boundary layer may be forecast with regard to the overall environment for fog occurrence, its likely extent,
intensity, and duration.

1. Introduction
The impact and significance of fog to personal safety
and local economies has been documented by many
authors (e.g., Croft et al. 1995; Johnson and Graschel
1992; Martin and Suckling 1987; George 1960). These
impacts range from delays in aviation, marine, and surface transportation and deliveries to serious accidents
caused in part by poor visibility. Regional airport and
marine port operations are often slowed or delayed at
a cost of several thousands of dollars a day (Garmon et
al. 1996). There have also been deadly consequences

Corresponding author address: Dr. Paul J. Croft, Department of

Physics and Atmospheric Sciences, Jackson State University, 1400
J. R. Lynch Street, P.O. Box 17660, Jackson, MS 39217-0460.
E-mail: pcroft@stallion.jsums.edu

q 1997 American Meteorological Society

from fog, such as the Amtrak disaster just north of Mobile Bay, Alabama, in 1993 and the Mobile Bayway
chain-reaction highway crash of 1995.
In the day to day forecast operations of the National
Weather Service the decision of whether or not to forecast significant fog (i.e., fog that leads to economic and
or life-threatening impacts) is a complex process. The
forecast decision is based on the assimilation of data
and the use of numerical and mesoscale tools by forecasters and the forecasters own experience. As several
different approaches may be taken when forecasting fog,
results can vary greatly for any given situation. For
example, one forecaster may rely on persistence and
climatology while another may rely solely on the analysis of raw data in real time. In either case a forecaster
is considering the processes involved in fog development.

546

WEATHER AND FORECASTING

a. Fog
Fog is often defined as simply a cloud on the ground
[e.g., the textbook by Lutgens and Tarbuck (1995)] that
has formed through a cooling or humidification process.
In some cases it has been imprecisely defined as a mist
or as suspended water droplets that have formed as air
reached or approached saturation (Huschke 1959). A
more precise definition of fog indicates that it occurs
when water droplets form and are suspended in air that
is within 10% of saturation (Houghton 1985). However,
these definitions do not consider the observed droplet
concentration of fog that determines a fogs thickness
(intensity or opacity) and its microphysical and dynamic
character.
Cloud drops typically range from 10 to 40 mm in
diameter (Young 1993) and mist (or drizzle) from 100
to 500 mm (e.g., Neuberger 1957; Wayne 1993). The
drop size distribution of fog may range from 2 to 65
mm, with the distribution in some marine fogs reported
to be as low as 1 to 10 mm (Houghton 1985; Jiusto
1981). These differences are a function of the cloud
condensation nuclei available and their characteristics.
More recent study indicates that marine environments
produce low droplet concentrations with large droplets,
whereas continental environments produce high droplet
concentrations with smaller droplets (Twohy et al.
1995). Although high concentrations of cloud condensation nuclei limit droplet sizes, it is the droplet concentration that determines fog opacity.
The prediction of fog demands that a forecaster mentally integrate various scale interactions that can lead to
fog. This natural atmospheric integration is, however,
an inverse problem. The forecaster, despite knowledge
of the conditions associated with fog, must anticipate
and predict fog occurrence a priori. Therefore forecasters often use fog types as a guide [e.g., radiative,
advective, and combinatorial; George (1960)]. Fog
types are further classified according to the synoptic
regime that produces them, the geographic region in
which they form, or the evolutionary processes that affect their formation and spread. For example, Stull
(1988) cites two kinds of radiation fog based upon their
character. One is most dense near the ground and more
diffuse with height. The other has a sharp top, similar
to stratocumulus, and forms within a well-mixed stable
layer. These differences clearly result from variations
between stable and thermal (convective) internal boundary layer evolutions.
b. The fog forecast problem
The forecasts of fog occurrence, its extent, duration,
and intensity are difficult operationally as fog is a
boundary layer phenomenon that often displays great
variability in time and space. Further, as the boundary
layer is driven and initially set up by the synopticscale circulation, the forecast of fog is, to a first ap-

VOLUME 12

proximation, determined by the macroscale. However,

fog occurrence often displays mesoscale features as determined by regional characteristics of, and contributions from, the boundary layer. The diagnosis and prediction of these interactions is not readily accomplished
with current operational models. Further, the interactions may be complicated by microphysical processes
within and outside of fog masses or when decoupling
of the boundary layer occurs.
Numerous local and regional studies and modeling
attempts have been made concerning fog and fog forecasting. Many of these have led to the development of
fog forecasting techniques based on climatology and
checklists (e.g., Garmon et al. 1996; M. Sutton 1994,
unpublished manuscript; Johnson and Graschel 1992)
and empirical and statistical relationships (e.g., Bacinschi and Filip 1976; Gimmestad 1993). Others have
focused on the use of satellite imagery (e.g., Gurka
1978; Ellrod et al. 1989), numerical modeling (e.g., Barker 1977; Forkel and Zdunkowski 1986; Bergot and
Guedalia 1994), operational models (e.g., Burroughs
and Alpert 1993), and the development (or proposed
development) of operational conceptual models (e.g.,
Leipper 1994; Croft et al. 1995). A few have considered
the microphysical processes involved in fog occurrence
(e.g., Meyer and Lala 1990; Ackerman et al. 1995) in
an operational setting. Unfortunately, many numerical
models are either incapable of recreating the fog process
or suffer from poor parameterizations, gross assumptions, or the lack of certain physical data. As such, specific and highly specialized conceptual model approaches are more likely to provide the operational forecaster
with the tools necessary to predict fog [e.g., the LIBS
approach; Leipper (1995)] according to an overall understanding of the fog process.
2. Traditional fog forecasting
As fog occurs within the boundary layer, a forecaster
must focus on the evolution of weather across all scales
that may lead to saturation of all or some portion of the
boundary layer. However, the application of the commonly used forecast funnel approach (UCAR 1991)
will be impractical unless an assessment is made of the
boundary layers ability to support fog dynamically and
thermodynamically. One typical approach is to study
persistence and climatology to determine synoptic regimes that lead to boundary layer saturation on a large
scale. The same approach could be used to detail the
thermodynamic environment necessary for fog. Thus,
to a first approximation, a forecaster may predict fog
based on regional and site-specific climatologies.
a. Climatology
For portions of Mississippi, Louisiana, and Alabama,
the greatest average number of dense fog days (visibility
less than mi) occurs near the coastline (Fig. 1) and

SEPTEMBER 1997

CROFT ET AL.

FIG. 1. Isopleths of the average number of dense fog days (visibility

less than mi) yr21 for portions of Alabama, Florida, Louisiana, and
Mississippi.

during the cool season (Fig. 2). The frequency is highest

during the early morning hours (Fig. 3). Thus on a regional basis (and subsynoptic scale), fog occurrence is
strongly dependent upon radiative flux and other cooling
mechanisms that lead to saturation of the boundary layer. This flux may also be accomplished, and/or enhanced, through advective and boundary layer modification processes.
However, this approach fails to fully assess the dynamic of the fog occurrence and reveals little about
mesoscale variations of the boundary layer over time.
For example, a coastal zone with steep topographic variation can lead to nocturnal drainage flows that interact
with ambient synoptic flows (e.g., Golding 1993). In a
second case, during the fall when river water temperatures are relatively high (compared to the air above)
and strong radiational cooling occurs through a deep
layer of dry air, fog forms immediately above the water
surface. The fogs extent is then a function of the dendritic pattern of the river and local saturation of the
boundary layer. Unfortunately, the operational assessment and quantification of such local climatic characteristics is difficult and often relies heavily on forecaster
experience.

547

FIG. 2. Same as in Fig. 1 except for the cool season (October

March).

cording to partial correlation coefficients) is the grid

binary 1000-mb relative humidity, an indicator of sufficient boundary layer saturation. This predictor indicates that when a critical value of relative humidity is
reached or exceeded, fog is expected. Other leading predictors include stability and mixing ratios, and monthly
relative frequencies of visibilities less than 3 mi depending on which site is examined. The effective correlation coefficients (one indicating perfect correlation)
for these predictors, when combined in separate regression equations for JAN, MOB, and MSY, are 0.158,
0.195, and 0.108, respectively.
An examination of 6-h MOS fog forecast equations
was also made. In this case, for each location the lead
predictor for dense fog is the latest observed obstruction
to visibility at the station at model initial time. Other
leading predictors include dewpoint depression, ceiling
height, and visibility, which are often site specific. Correlation coefficients for each 6-h equation ranged from
0.267 at JAN to 0.427 at MOB. While performance is

b. Numerical guidance
One means by which forecasters may attempt to estimate or quantify mesoscale variations operationally is
through the use of model output statistics, or MOS guidance. Through statistical regression, MOS equation solutions offer insight to mesoscale variations (and microscale climates) and thus potentially fog formation
and persistence. The 24-h MOS cool season fog forecast
equations for Jackson (JAN), Mississippi; Mobile
(MOB), Alabama; and New Orleans Moisant Airport
(MSY), Louisiana; were examined for their statistical
prediction of dense fog (visibility less than mi). For
each location the best statistical predictor (defined ac-

FIG. 3. Relative frequency of fog restriction (i.e., fog reported at

time of observation) by time of day for JAN, MOB, and MSY. Based
on data for the period 194890.

548

WEATHER AND FORECASTING

VOLUME 12

TABLE 1. Selected dense fog days (visibility less than mi) at JAN for the period 199395. Summary information includes date of
occurrence, the principle fog type (i.e., radiative or advective), temperature and dewpoint at time of dense fog, season of occurrence, whether
precipitation was observed during the 24 h preceding the observation of dense fog, the depth of the boundary layer, the lowest visibility
observed, and the vertical wind shear (i.e., backing, veering, or neutral).

Date
24
14
26
16
24
17
2
12
6
12
22
25
30
6
4
23
31
6
9
9
17
18
26
31
20
21
2

Mar 93
May 93
Oct 93
Nov 93
Nov 93
Dec 93
Jan 94
Jan 94
Feb 94
Feb 94
Apr 94
Jun 94
Jun 94
Aug 94
Oct 94
Oct 94
Oct 94
Nov 94
Nov 94
Dec 94
Dec 94
Dec 94
Dec 94
Dec 94
Feb 95
Mar 95
Jun 95

Type

Sfc T/Td
(8F)

Radiation
Radiation
Radiation
Advection
Advection
Radiation
Advection
Radiation
Radiation
Advection
Radiation
Radiation
Radiation
Radiation
Radiation
Radiation
Advection
Radiation
Advection
Advection
Radiation
Radiation
Radiation
Advection
Radiation
Radiation
Radiation

44/44
54/54
56/56
61/61
44/44
38/38
46/46
43/42
53/53
35/35
60/60
69/69
72/71
71/70
59/58
61/60
60/59
48/48
62/62
61/61
55/55
36/35
27/26
53/53
44/44
56/56
63/63

Season

Pcpn last
24 h?

Depth of moist
layer (m)

Lowest
visibility
(mi)

Winds

Spring
Spring
Fall
Fall
Fall
Winter
Winter
Winter
Winter
Winter
Spring
Summer
Summer
Summer
Fall
Fall
Fall
Fall
Fall
Fall
Fall
Fall
Winter
Winter
Winter
Spring
Spring

No
Yes
No
Yes
No
No
Yes
Yes
Yes
Yes
No
Yes
Yes
Yes
No
No
No
Yes
No
No
Yes
No
No
Yes
Yes
No
Yes

60
40
475
100
55
500
895
525
455
505
450
10
480
985
44
411
450
50
1920
960
1422
80
104
1883
55
50
465

,
,

,
,

,
,

Back
Neut.
Back
Veer
Neut.
Back
Veer
Back
Back
Veer
Neut.
Veer
Neut.
Neut.
Back
Back
Veer
Back
Veer
Veer
Back
Back
Back
Veer
Neut.
Back
Neut.

improved, the predictors identified as statistically significant fall within the realm of persistence nowcasting
and are traditionally used by operational forecasters.
Clearly the correlation coefficients illustrate a basic inability of current operational numerical model statistical
guidance to provide adequate information for reliable
prediction of fog 624 h in advance.
c. Sounding analysis
Therefore fog forecasting requires a direct examination of the boundary layer environment in which
dense fog occurs. To illustrate, rawinsonde data for JAN
were obtained and analyzed for the period 199395 for
days on which dense fog (visibility less than mi, 0.8
km) was observed. A total of 53 days were identified.
In order to isolate and remove inconsistencies in the
data, those days in which precipitation was occurring
at the time of the dense fog were removed. Also, if
surface observational data showed that the fog was transient (i.e., lasting less than an hour) or very localized
(e.g., ground fog with vertical extent less than a few
meters), these occurrences were also removed. Data for
27 days remained and each event was categorized as an
advective (8) or radiative (19) case. The associated temperature, dewpoint, season of occurrence, whether or

not precipitation occurred within the previous 24 h, the

depth of the moist layer [inferred from sounding information using SHARP: the Skew-T Hodograph and Research Program, by Hart and Korotky (1991)], the lowest visibility observed, and surface to 700-mb wind
characteristics (veering or backing) were determined
(Table 1).
Temperatures during dense fog events varied from
238 to 228C (278728F) and most occurred during the
cool season. The fall of precipitation within the previous
24 h did not appear to be necessary for dense fog as
this condition was met in only half of the cases examined. The average depth of the moist layer (as defined
by the saturated layer of the sounding) for the advective
cases was 846 m, while the average depth for radiative
cases was 350 m [and this is consistent with Jiusto
(1981)]. Dense radiation fog occurred under weak backing (cold air advection) or neutral wind pattern (Fig.
4a), whereas dense advection fog occurred under a veering (warm air advection) wind pattern (Fig. 4b). Very
dry air aloft, often with more than 208C of dewpoint
depression, was evident in all cases and typically produced a goalpost when temperature and dewpoint
were plotted on a skew-T diagram [and is similar to the
threshold criteria determined by Leipper (1995)]. This
goalpost represents an open atmospheric window for

SEPTEMBER 1997

CROFT ET AL.

549

FIG. 5. Conceptual model for fog forecasting. Each base surface

represents the character of cloud condensation nuclei present in the
boundary layer and a variety of atmospheric combinations of moisture
availability (or saturation) and dynamic forcing supportive of fog
occurrence, extent, and intensity. Movement on and between base
surfaces represents transition processes.

a. Description

FIG. 4. Typical soundings associated with dense fog (visibility less

than mi) at JAN for (a) radiative (backing) and (b) advective
(veering) cases.

longwave radiative cooling, which can strengthen a

near-surface inversion layer.
3. Conceptual model
As climatology and numerical guidance provide both
synoptic-scale and site-specific information, and as
sounding analysis is strictly site specific, none of these
techniques alone are appropriate for mesoscale prediction of fog. In combination, and with a sufficient degree
of specificity in time and space, they can allow mesoscale fog predictions to be made. To achieve this, a framework for a simple operational conceptual model approach to local and regional fog forecasting, similar to
that by Leipper (1995), is presented for the southern
region of the United States (gulf coast states). The conceptual model is based on macroscale forcing and moisture availability (or saturation), the character of the
boundary layer, a qualitative assessment of the significance of fog microphysics (according to cloud condensation nuclei), and the interaction and change of
these in time and space. The model (Fig. 5) is intended
to ultimately serve as a collection of surfaces in space
and time that represent varying atmospheric conditions
that result in varying extents, intensities, and durations
of fog.

The model surfaces (Fig. 5) represent the character

of boundary layer air in a forecast region in terms of
the unique distribution of cloud condensation nuclei.
These base surfaces range from maritime to continental
and allow a forecaster to assess the nature, and initial
concentrations of, local (or imported) cloud condensation nuclei and their associated drop sizes. This provides
information as to the intensity (opacity or thickness) of
fog and its duration. If there is a local variation, or a
forced synoptic change, the forecaster simply moves
from one base surface to another. In the process, a forecaster assesses the impact of changes in cloud condensation nuclei and drop size as the change takes place
and at any boundaries between different air masses.
Once an initial forecast time base surface is established, a forecaster may move laterally on the face
of that surface to quantify moisture availability (or saturation) and dynamic forcing. Moisture availability is
assessed according to moisture present and moisture that
can be realized through condensational cooling. Dynamic forcing (primarily with regard to lifting and cooling mechanisms and moisture pooling and transport) is
assessed according to the predominance of boundary
layer effects (e.g., base-state flow, local circulations, and
surface parcel thermodynamics) or synoptic flows peculiar to the forecast region. This must be determined
through the application of basic meteorological principles for any location to separate boundary layer
growth from boundary layersynoptic interaction and
from synoptically forced boundary layer growth.
For both moisture and dynamics, a mixed zone or
transition is also possible. In the case of moisture availability the forecaster can determine the extent of fog
(i.e., patchy versus widespread) as well as intensity (in
combination with cloud condensation nuclei information). From dynamic forcing the forecaster can assess

550

WEATHER AND FORECASTING

the duration (e.g., all-night radiative event) and extent

of fog occurrence (given saturation). The approach is
similar to Leipper (1995) in that synoptically produced
initial conditions are quantified according to the development of a surface boundary layer inversion. The evolution of the boundary layer is then considered with
regard to airmass modification (and associated continental and marine interactions) and fog formation.
b. Illustration
To illustrate the models use, consider fog occurrence
across the gulf coastal states following the arrival of a
cooler and drier air mass, its modification, and the approach of a frontal system. Initially the forecaster determines that a continental-base surface is present and
that moisture availability across the region is low with
synoptic-scale forcing. This places the forecaster in the
lower-left portion of the bottom surface (Fig. 5) and
makes fog unlikely (or very localized and very limited
in extent and duration). However, if synoptic forcing
weakens and boundary layer processes become dominant, the forecaster moves to the upper-left portion of
the surface. In this case, radiational cooling is maximized overnight and other areas of fog form in the vicinity of local moisture/saturation maxima (e.g., patchy
fog). A forecaster might therefore anticipate fog of varying intensity, and of longer duration, in a number of
locations in a forecast region.
Within 2448 h, the air mass in place modifies according to local factors through energy exchange processes (e.g., evaporative flux) and as related to physiographic features and location (i.e., over land, water,
or coastal interface). Differential heating and cooling,
local cloud condensation nuclei, and moisture sources
contribute to this modification. Therefore the forecaster
is now on a mixed-base surface (i.e., the middle surface
in Fig. 5) with greater moisture availability. If another
night of radiative cooling takes place, fog formation may
be widespread, more intense, and of longer duration. At
the same time, local differences in the base surface conditions may exist and create sharp boundaries (e.g., as
related to landsea-breeze circulations or valley winds).
For example, moisture availability will be greatest near
the coast, and cloud condensation nuclei will more likely
be dominated by maritime sources rather than continental and local sources. Such local variations within
the local boundary layer can be inferred from isentropic
surfaces and the resulting sharp boundaries (and variations in intensity) between fog and no-fog areas well
known in satellite imagery.
Following this airmass modification period, a cold
frontal system approaches the gulf coastal states. The
increased pressure gradient often leads to synoptic-scale
forcing and results in the dominance of advective processes that act to homogenize the air across the forecast
region (in many cases). In response, the initial base
surface evolves from one that is mixed to one that is

VOLUME 12

more maritime with greater moisture availability and

maritime cloud condensation nuclei. This puts the forecaster closer to, or on, the maritime base surface on the
lower-right corner. In this situation, if fog occurs, it is
more likely to behave as a synoptic-scale mass with
embedded mesoscale features. Fog would be expected
to be widespread and of long duration with much variation in intensity due to local mixing. Any cooling
through advective lift (e.g., upslope flow), or by passage
over a cold ground surface, would enhance fog formation.
4. Operational application
The preceding illustration of the conceptual model
(section 3) was intended to provide the forecaster guidance on how to apply the model. The model is purposely
simplistic to allow for its ultimate application to, and
modification for, any operational forecast scenario or
location. The model does not provide specific threshold
values for any specific operational situation as thresholds will vary significantly from location to location
(e.g., see Leipper 1995), are not readily known, or have
not as yet been thoroughly nor completely investigated.
When known (e.g., Leipper 1995), forecasts of fog occurrence, extent, intensity, and duration may be quantified and verified. The conceptual model here presents
a framework in which forecasters develop an integrated
assessment of how microphysical and thermodynamic
conditions and dynamic processes, in association with
local and mesoscale effects and interactions, result in
variations in fog occurrence, extent, intensity, and duration.
To further illustrate the potential utility of this simple
operational conceptual model, as no operational testing
has been completed, two specific fog situations are examined in hindcast. The first is a radiational event that
occurred over northern Mobile Bay in Alabama. This
event was notable because it resulted in the worst multiple-vehicle accident in U.S. history despite the intense
fogs rather limited extent and duration. The second situation considers advective sea fog over the Gulf of Mexico. In each case the operational tools used in fog forecasting are highlighted and the initial conceptual model
applied.
a. Mobile bayway disaster
The chain reaction collision that occurred during sunrise on 20 March 1995 on the Mobile I-10 Bayway was
the largest ever in U. S. history. Nearly 200 vehicles
were involved and over 100 persons were injured. Insurance losses have been estimated in the hundreds of
millions of dollars. Miraculously, only one person was
killed. Despite the collisions notoriety, the fog occurrence itself was not particularly noteworthy or unusual.
Indeed, based on available GOES-8 channel-differenced
satellite imagery (Fig. 6), the areal extent of the fog and

SEPTEMBER 1997

CROFT ET AL.

551

FIG. 6. Satellite imagery sequence (channel differenced) for Mobile Bay and vicinity. Hourly images for 0145 through 0445 EST (upper
left, upper right, lower left, lower right) show the initial local development of dense fog. The last image shows the increased intensity and
extent of fog prior to the multiple-vehicle accident described in the text.

stratus was initially rather limited and represented a fairly typical radiational event for the given time of year.
Eyewitness reports indicated that the top of the fog layer
was ragged. Several days prior a synoptic-scale event
provided a 2-in. basin average rainfall associated with
extensive runoff. This was followed by cooler and drier
boundary layer air that quickly modified.
An examination of the initial hour (0000 UTC 20
March 1995) 80-km early eta gridded model output and
a composite of multipressure level synoptic-scale features was completed (not shown). Immediately prior to
the fog event strong anticyclonic flow existed through
a deep tropospheric layer and contributed to regional
subsidence and the drying out of midlevel air parcels
as they descended dry adiabatically. This would place
the forecaster between the continental-and mixed-base
surfaces (Fig. 5) following initial airmass modification.
The surface pressure gradient was weak and supported
nearly calm boundary layer winds over the region with
a 4-mb pressure difference between west Florida and
east Texas. Soundings upstream of Mobile from JAN
and SIL (Figs. 7a, b) indicated very dry air extending
from the surface to approximately 400 mb with a goal-

post appearance just prior to the fog event. At some

middle-tropospheric levels dewpoint depressions were
nearly 208C. Surface isodrosotherms revealed dewpoints of 168C (608F) over far southwestern Alabama,
indicating limited moisture availability and thus extent
of fog formation.
The depth of the boundary layer was assumed to be
no more than 25 mb deep (or less than 50 m) based on
eyewitness accounts and modified sounding analyses
(not shown). This information was used to select the
appropriate isentropic surface (295 K) to determine
whether any acceleration or mixing of air in the boundary layer was possible. Figure 8 depicts the pressure
distribution and wind flow pattern on the 295-K isentropic surface about 3 h prior to the development of the
strong low-level thermal inversion. Figure 9 depicts
wind vectors, mixing ratio (dashed), and hand-analyzed
streamlines found on the 295-K isentropic surface for
the same time. The col in the streamline pattern near
Mobile Bay contributed to the light wind conditions
throughout the entire lower troposphere. Any airmass
modification (moisture and temperature) over Mobile
Bay was therefore related to the bay water temperature

552

WEATHER AND FORECASTING

VOLUME 12

bulent boundary layer mixing and the resultant vertical

exchange of high-momentum air into the boundary layer.
Applying and summarizing the conceptual model approach, a maritime base surface was located over Mobile
Bay and experienced maximum boundary layer forcing
(i.e., radiative cooling). This culminated in the preferential formation of fog in those regions with radiative
cooling rates sufficient to achieve saturation where high
moisture was present. The final fog plume (Fig. 6)
was found along and to the east of the Mobile River
valley, and thus fog extent was determined in part by
local physiography. In addition, the close proximity of
paper mills (located in the vicinity of the accident site)
and saltwater source (Mobile Bay) likely provided for
localized modification and pooling of the aerosol distribution. A resulting wide drop size distribution of
cloud condensation nuclei (related to local industry
emissions) may have contributed to the fogs intensity
(or optical thickness) by increasing droplet concentrations. Accident victims indicated that the fog appeared
as a solid wall with near-zero visibility when they drove
into it. Radiative effects, related to sunrise, as well as
chemical effects, may have also played a role in the
fogs intensity.
FIG. 7. SHARP soundings for (a) JAN and (b) SIL for 0000 UTC
20 March 1995.

at the time (which was unavailable). There was also a

neutral pressure advection pattern coincident with the
relative maximum of low-level moisture over the bay,
which persisted through 1200 UTC. This prevented tur-

b. Sea fog assessment

Sea fog and stratus can affect extensive areas of the
northern Gulf of Mexico as well as the immediate coastal plain. Although there is an abundance of moisture
available over the gulf, sea fog development, if it occurs,
and its onshore motion is a function of coastal zone

FIG. 8. Initial (0000 UTC) 80-km early Eta Model pressure distribution (mb) and wind (m
s21) on the 295-K isentropic surface on 20 March 1995. Each half wind barb equals 2.5 m
21
s .

SEPTEMBER 1997

CROFT ET AL.

553

FIG. 9. Initial (0000 UTC) 80-km early Eta Model wind vectors (m s21), mixing ratio (g
kg21), and hand-analyzed streamlines.

processes (e.g., convergence and friction) and gradients

in sea surface temperatures. This is particularly true
during the winter and early spring months (December
March) when cold dry air pushes south to the gulf coast
and overlies the warm gulf waters. In addition, cold
waters develop near the coastline due to drainage of
cold river basin waters to the coast and the presence of
cold air masses over the shallow coastal waters. These
conditions often lead to an unstable layer of near-surface
evaporative, or cold advection, fog near the coast. When
the flow is reversed (i.e., a return flow event), warm
moist air is forced inland over colder air and ground
surfaces resulting in a cooling of the warm air. The
cooling can lead to stratus and warm advection fog over
time as a deeper and more stable boundary layer develops.
Despite the existence of an operational advective sea
fog forecast model for the northeast coast (Burroughs

FIG. 10. Locations of supplementary aviation weather reporting

stations over the northern Gulf of Mexico.

and Alpert 1993), none currently exists for any portion

of the Gulf of Mexico. As a result, other methods must
be used to assess the possibility of fog occurrence in
this region, especially given the lack of adequate and
accurate model initialization. Therefore efforts have focused on developing graphs, based on empirical evidence, to predict sea fog occurrence and intensity. These
have focused on the relationship between atmospheric
temperature, dewpoint, and wind with sea surface temperatures. For example, climatological data for 10 supplementary aviation weather reporting stations across
the northern Gulf of Mexico (Fig. 10) were collected
for the winter and early spring months (Johnson and
Graschel 1992). Due to limited data availability, these
data were obtained for the years 198586 and 198890
(and have since been discarded).
Upon examination of this dataset, distinctions were
made between those conditions associated with the occurrence or nonoccurrence of fog. Fog intensity (based
on visibility thresholds) for both warm advection (Figs.
11ac) and cold advection (Fig. 11d) cases were also
considered. Each graph suggests parameter relationships
for observing advective sea fog based on empirical evidence that have been applied successfully in operations
by the Slidell National Weather Service Office for the
last several years. They allow a forecaster to physically
assess the importance of the magnitude of the air and
sea temperature difference, the dewpoint depression,
and wind. The air and sea temperature difference, and
dewpoint depressions, offer the forecaster a surrogate
measure of moisture availability and air mass, boundary
layer, modification. The wind provides information regarding synoptic versus boundary layer forcing. The
above information is obtained operationally by the Sli-

554

WEATHER AND FORECASTING

VOLUME 12

FIG. 11. Advective sea fog occurrence, and restriction to visibility, as a function of (a) wind
speed vs dewpoint depression, (b) dewpoint depression vs airsea temperature difference, (c)
dewpoint depression vs dewpointsea temperature difference for warm advection sea fogs, and
as a function of (d) relative humidity vs airsea temperature difference for cold advection sea
fogs.

dell National Weather Service Office using observations

and model gridded data.
Applying and summarizing the conceptual model approach, a maritime-base surface is present when air and
sea temperature differences are minimal and winds light.
Moisture availability (or saturation) is maximized, as are
boundary layer processes under clear skies (i.e., radiational
cooling) and sea fog results. The presence of these conditions over a large area can lead to widespread, thick,
and long-duration fog. In the event advection is added,
sea fog may begin to move toward the coast. As the fog
mass encounters the coastal zone, it undergoes a fundamental transformation due to turbulent mixing. In addition,
the fog mass passes over a changing surface, which will
force changes in the nuclei and drop size distributions.
These are often manifest in the observed weather (e.g.,
fog and drizzle at the coast versus denser widespread fog
inland) as a fog mass moves (and/or redevelops) onshore.
Thus the base nuclei surface will become mixed, moisture
availability (or saturation) will decrease (unless cooling
by the land, or through lift over the land occurs), and
advective flow diminish.
c. Forecasting tools
The use of the conceptual model presented requires
forecasters to acquire basic information about cloud

condensation nuclei, moisture availability, and dynamic

forcing. An assessment of the local aerosol field, and
its evolution in time and space, can be based on a local
knowledge of nuclei sources and sinks. The makeup and
character of these could be determined by direct measurement (e.g., using satellite data, although presently
not possible in the operational environment) or qualitatively inferred. The resulting intensity of fog may then
be considered (pragmatically) to be a function of the
concentration of water droplets suspended in the air.
This concentration is dependent upon the drop size distribution. However, as the distribution of aerosols
(amount and size distribution) is to a large extent determined by advection of these into a region, aside from
the local origination of aerosols, variations in intensity
over time must be considered. The use of MOS guidance
and gridded model data (including trajectories) can assist the forecaster in identifying relevant flow pattern
variations that will affect the local aerosol composition.
Once the aerosol field is evaluated, an assessment of
stability and moisture according to the synoptic regime
(e.g., airmass type) and boundary layer contributions
require diagnosis of cooling and/or lifting mechanisms
that could result in fog formation. This may be accomplished by application of model diagnostic software programs (e.g., isentropic analysis with personal computer

SEPTEMBER 1997

555

CROFT ET AL.

based Interactive Display and Diagnostic System,

PCGRIDDS, and mesoeta analysis fields) and by consideration of local physiographic effects. A forecaster
may use the SHARP program in conjunction with gridded data from the rapid update cycle (or RUC) model
(or the mesoeta) to diagnose the boundary layer. Indeed,
model predictions of temperature and skin temperature
have been used to forecast fog for space shuttle landing
operations (Garner and Batson 1995).
A quantification of boundary layer characteristics regionally may be aided by remote sensing techniques,
including satellite and radar (or profilers). Satellite data,
particularly from the GOES-8 platform, offer high-resolution imagery in both visible and infrared channels.
Channel differencing techniques (e.g., Ellrod et al.
1989; Ellrod 1995) provide forecasters with definitive
regional temperature and moisture distributions relevant
to fog extent. Although these techniques are presently
most useful in identifying areas of fog after their formation, their operational promise is great. For example,
precipitable water imagery can be used to diagnose
return flow events (Johnson and Rabin 1993), and
statistical clustering techniques applied to satellite information allow for the determination of low-level humidity gradients (Fuelberg et al. 1995).
Radar information may also be considered for its ability to detect density gradients and thus inversions. The
clear-air mode capability of the WSR-88D provides information regarding gradients of moisture and temperature that often appear within the ground clutter pattern.
A careful study of typical ground clutter patterns in time
lapse during super refractive conditions with regard to
its average time of appearance, its reflective intensity,
its horizontal size, and its rate of change can help to
characterize the local rate of thermal change in the
boundary layer. Radar-derived products, such as storm
total precipitation, may also be useful in identifying
areas that may have higher levels of moisture in the
boundary layer.
5. Summary
To forecast fog more precisely, in terms of fog occurrence, extent, intensity, and duration, operational meteorologists must understand and diagnose the thermodynamics, kinematics, and microphysics of the fog
process. Presently the thermodynamic and kinematic
components are readily quantified, or inferred, whereas
the microphysics and boundary layer processes are not.
Therefore, as numerical models have so far been inadequate in this regard, it is important that an operational forecaster develop and use a conceptual fog forecast model. The successful use of such a model will
depend on the forecasters ability to assimilate the appropriate tools and techniques available in real time.
Prior to this however, a forecaster must ingest and
properly conceptualize the various elements of the model in order to provide quantification of each for the given

forecast zone. This can be achieved through the application of the latest mesoscale tools and observations in
hindcast as well as climatic study. In the process, quantification of fog forecasting may be accomplished and
critical thresholds determined for the base surfaces
of the conceptual model presented here. Until improvements in operational models (and in particular operational cloud physics) can be made, a forecaster must
mentally solve the fog inverse problem to truly forecast
fog rather than simply respond to its formation. Only
in this way will the regional prediction of fog in space
and time improve.
Acknowledgments. The authors wish to thank Fred
Settelmaier of the Techniques Development Laboratory
for printouts of MOS equation predictors and James
Purdom for satellite imagery of the central gulf coastal
states. Thanks also go to Jeff Garmon for his assistance
with the technical aspects of the study and to Robert E.
(Gene) Merritt for his personal comments regarding the
Mobile Bayway disaster. Thanks also to Southern Region, as well as Monesa Watts, for help in the drafting
of figures. We are very grateful for the assistance provided by the reviewers and editor in completing this
paper.
REFERENCES
Ackerman, A. S., O. B. Toon, and P. V. Hobbs, 1995: A model for
particle microphysics, turbulent mixing, and radiative transfer
in the stratocumulus-topped marine boundary layer and comparisons with measurements. J. Atmos. Sci., 52, 12041236.
Bacinschi, D., and L. M. Filip, 1976: Forecasting nomogram for
meteorological visibility. Meteor. Hydrol., 1, 3338.
Barker, E. H., 1977: Maritime boundary-layer model for the prediction of fog. Bound.-Layer. Meteor., 11, 267294.
Bergot, T., and D. Guedalia, 1994: Numerical forecasting of radiation
fog. Part I: Numerical model and sensitivity tests. Mon. Wea.
Rev., 122, 12181230.
Burroughs, L. D., and J. C. Alpert, 1993: Numerical fog and visibility
guidance in coastal regions. Tech. Procedures Bull. 398, 8 pp.
[Available from Environmental Science Information Center,
NOAA, U.S. Dept. of Commerce, 3300 Whitehaven St. N.W.,
Washington, DC 20235.]
Croft, P. J., D. Darbe, and J. Garmon, 1995: Forecasting significant
fog in southern Alabama. Natl. Wea. Dig., 19, 1016.
Dornan, C. E., E. Maturi, and J. Steger, 1989: Detection of fog at
night using dual channel GOES-VAS imagery. Preprints, 12th
Conf. on Weather Analysis and Forecasting, Monterey, CA,
Amer. Meteor. Soc., 515520.
Ellrod, G. P., 1995: Advances in the detection and analysis of fog at
night using GOES multispectral infrared imagery. Wea. Forecasting, 10, 606619.
Forkel, R., and W. Zdunkowski, 1986: Two-dimensional numerical
model for fog forecasting. Ann. Meteor., 23, 4647.
Fuelberg, H. E., P. A. Rao, and D. W. Hillger, 1995: Clustering of
satellite sounding radiances to investigate low-level humidity
gradients. J. Appl. Meteor., 34, 15251535.
Garmon, J. F., D. L. Darbe, and P. J. Croft, 1996: Forecasting significant fog on the Alabama coast: Impact climatology and forecast checklist development. NWS Tech. Memo. NWS SR-176,
16 pp. [Available from Scientific Services Division, Southern
Region NWS, 819 Taylor Street, Room 10A23, Fort Worth, TX
76102.]

556

WEATHER AND FORECASTING

Garner, T., and B. Batson, 1995: Use of NGM regional profile for
fog forecasting for space shuttle landing operations at KSC. Preprints, Sixth Conf. on Aviation Weather Systems, Dallas, TX,
Amer. Meteor. Soc., 189191.
George, J. J., 1960: Weather Forecasting for Aeronautics. Academic
Press, 673 pp.
Gimmestad, C., 1993: A statistical forecasting method for fog at
Olympia, Washington. Western Region Technical Attachment
93-15, 6 pp. [Available from Environmental Science Information
Center, NOAA, U.S. Dept. of Commerce, 3300 Whitehaven St.
N.W., Washington, DC 20235.]
Golding, B. W., 1993: A study of the influence of terrain on fog
development. Mon. Wea. Rev., 121, 25292541.
Gurka, J. J., 1978: Use of enhanced VIS imagery for fog detection
and prediction. National Environmental Satellite Services. Satellite Applications Information Note 78/4, 6 pp. [Available from
Satellite Applications Laboratory, U.S. Dept. of Commerce,
NOAA/NESDIS, Washington, DC 20233.]
Hart, J. A., and W. Korotky, 1991: The SHARP workstation, version
1.50A skew-t hodograph analysis and research program for
the IBM compatible PC. NOAA/NWS, Charleston, WV, 30 pp.
[Available from Scientific Services Division, Southern Region
NWS, 819 Taylor Street, Room 10A23, Fort Worth, TX 76102.]
Houghton, D. D., Ed., 1985: Handbook of Meteorology. John Wiley
and Sons, 1461 pp.
Huschke, R. E., Ed., 1959: Glossary of Meteorology. Amer. Meteor.
Soc., 638 pp.
Jiusto, J. E., 1981: Fog structure. Clouds, Their Formation, Optical
Properties and Effect, P. V. Hobbs and A. Deepak, Eds., Academic Press, 495 pp.
Johnson, G. A., and J. Graschel, 1992: Sea fog and stratus: A major
aviation and marine hazard in the northern Gulf of Mexico.
Preprints, Symp. on Weather Forecasting, Atlanta, GA, Amer.
Meteor. Soc., 5560.

VOLUME 12

, and R. Rabin, 1993: Highlights of return flow events from the

Gulf of Mexico during 19891992 wintersA forecasting perspective. Preprints, 13th Conf. on Weather Analysis and Forecasting, Vienna, VA, Amer. Meteor. Soc., 135138.
Leipper, D. F., 1994: Fog on the U.S. west coast: A review. Bull.
Amer. Meteor. Soc., 75, 229240.
, 1995: Fog forecasting objectively in the California coastal area
using LIBS. Wea. Forecasting, 10, 741762.
Lutgens, F. K., and E. J. Tarbuck, 1995: The Atmosphere. PrenticeHall, 462 pp.
Martin, M. D., and P. W. Suckling, 1987: Winter fog and air transportation in Sacramento, California. Climatol. Bull., 21, 1622.
Meyer, M. B., and G. G. Lala, 1990: Climatological aspects of radiation fog occurrence at Albany, New York. J. Climate, 3, 577
586.
Neuberger, H., 1957: Introduction to Physical Meteorology. The
Pennsylvania State University, 271 pp.
Stephens, P., 1993: Report to the National Science Foundation of a
workshop on the role of clouds in climate. Reno, NV, 17.
[Available from National Science Foundation, Office of NOAA
Programs, Arlington, VA 22230.]
Stull, R. B., 1988: An Introduction to Boundary Layer Meteorology.
Kluwer Academic, 666 pp.
Twohy, C. H., P. A. Durkee, B. J. Huebert, and R. J. Charlson, 1995:
Effects of aerosol particles in the microphysics of coastal stratiform clouds. J. Climate, 8, 773783.
UCAR, 1991: Forecast process. Forecasters Multimedia Library, Cooperative Program for Operational Meteorology, Education, and
Training, University Corporation for Atmospheric Research,
Boulder, CO.
Wayne, R. P., 1993: Chemistry of Atmospheres. Clarendon Press, 447
pp.
Young, K. C., 1993: Microphysical Processes in Clouds. Oxford University Press, 427 pp.