Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2014 Analysis of Prospective Fog Warning Systems Using AWOS/ASOS Station Data Throughout the State of Florida Justin Rivard

Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected] FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

ANALYSIS OF PROSPECTIVE FOG WARNING SYSTEMS USING

AWOS/ASOS STATION DATA THROUGHOUT THE STATE OF FLORIDA

By

JUSTIN RIVARD

A Thesis submitted to the Department of Earth Ocean and Atmospheric Sciences in partial fulfillment of the requirements for the degree of Master of Science

Degree Awarded: Summer Semester, 2014 Justin Rivard defended this thesis on July 17, 2014. The members of the supervisory committee were:

Peter Ray Professor Directing Thesis

Jeffrey Chagnon Committee Member

Robert Hart Committee Member

The Graduate School has verified and approved the above-named committee members, and certifies that the thesis has been approved in accordance with university requirements.

ii This thesis is dedicated to my parents, Denis and Darlene Rivard and my brother Shawn, who have always guided and supported my career ambitions. Most of all, this thesis is dedicated to my life long friend, Katie Carlton. Who, for seven years of my life, was my motivation to do well and make a better life for myself. Without her support and dedication, I would not have been able to achieve my goals. I am forever thankful I was able to spend a major part of my life with her and learn so much.

iii ACKNOWLEDGMENTS

Many thanks are due to many people. Especially to my major professor, Peter Ray for his supervision, knowledge, and patience. Dr. Jeffrey Chagnon and Dr. Hart for their input and guidance, and to all my friends and family.

iv TABLE OF CONTENTS

ListofTables...... vii ListofFigures ...... viii ListofAbbreviations...... xi Abstract...... xii

1 Introduction 1 1.1 FogTypes...... 6 1.1.1 AdvectionFog...... 7 1.1.2 UpslopeFog...... 8 1.1.3 FrontalFog ...... 9 1.1.4 RadiationFog...... 9 1.2 TheFogForecastingProblem ...... 12 1.2.1 Numerical Weather Prediction Limitations ...... 14 1.3 FogForecastingTechniques ...... 15 1.3.1 The Croft et al. Conceptual Model for the Southern U.S...... 16 1.3.2 UPSAirlinesConceptualModel ...... 17 1.3.3 Forecasting Using Model Output Statistics ...... 18 1.3.4 OtherFogForecastingTechniques ...... 20

2 Methodology 22 2.1 ThesisGoals...... 22 2.2 TheData ...... 23 2.3 Methodology for Detecting Fog at an AWOS/ASOS Station ...... 27 2.4 Methodologies for Forecasting Fog at an AWOS/ASOS Station...... 28

3 Results 31 3.1 Fog/FireClimatologyofFlorida ...... 31 3.1.1 Topography ...... 34 3.1.2 Fire ...... 34 3.2 Forecasts...... 36 3.2.1 MesonetDataClimatology...... 36 3.2.2 ConditionsFavorableforFogFormation ...... 42 3.2.3 CrossoverTemperature...... 45 3.2.4 SkillScores ...... 47 3.2.5 Detecting/DiagnosingFog ...... 48 3.2.6 Forecast ...... 49 3.2.7 Characteristics of the Fog Forecast Model ...... 52 3.2.8 ModelResultsandComparison ...... 54

v 4 Conclusions 66

References...... 68 BiographicalSketch ...... 72

vi LIST OF TABLES

3.1 Climatological reference forecast example. Percentages represent what a fore- cast chance of fog would be dependent on month of year and station...... 47

3.2 Climatological probability of fog for a given lead time and dewpoint depression 50

3.3 Skill scores for MOS 12-UTC forecast for fog formation ...... 54

3.4 Skill scores for 3-hour model forecast for 12-UTC ...... 55

3.5 BrierandHeidkeskillscore ...... 55

vii LIST OF FIGURES

1.1 Observed soundings taken on 09/29/2012 at 12z for Jacksonville, FL (left) and Tampa Bay, FL (Right). (Courtesy of The University of Wyoming) ...... 3

1.2 Meteogram of Jacksonville, FL for 24-hr period starting January 29, 2012 at 00Z.(PlymouthStateUniversity) ...... 4

1.3 Fog and Smoke Related Crash Density for 2006–2010. Units are in crashes per square mile. Data from Department of Transportation (DOT) accident reports. 5

1.4 Percentage of total fog occurrence given by time of day. Floridaonly...... 11

1.5 Percentage of total fog occurrence given by month of year. Floridaonly. . . . 12

2.1 All mesonet stations in Florida. Primary stations are defined as AWOS, ASOS, FAWN, and FWMD sites. Secondary sites are defined as individual and pri- vately owned weather stations. Latitiude and longitude data from NOAA. . . 23

2.2 PrimaryMesonetstationsinFlorida...... 24

2.3 ContingencyTable ...... 29

3.1 Climatological location and frequency of fog in Florida, averaged per year from 2006 to 2010. GIS kriging technique used to draw contours ...... 32

3.2 Estimated amount of low visibility occurrences using Low Visibility Occurrence Risk Index (LVORI) (Lavdas and Achtemeier, 1995)...... 33

3.3 Topography of Florida in meters using shapefile data from the United States GeologicSurvey(USGS)...... 35

3.4 Smoke dispersion modeling results at Paynes Prairie site. Concentrations in mi- crograms per meter-cubed. Top left emission rate = 0.0003gs−1m−2, Top right emission rate = 0.0006gs−1m−2, Bottom left emission rate = 0.0008gs−1m−2, Bottom right emission rate = 0.0016gs−1m−2. Outermost contour represents 100 micrograms per meter-cubed. Second outermost contour represents 500 microgramspermeter-cubed...... 37

3.5 Number of prescribed burns and wild fires in Florida from 2006 to 2010. Data plottedinmonthlyintervals ...... 38

3.6 Total amount of wild fires from 2006 to 2010 for first six-months of year. Values represent wildfire events per 1000 square kilometers per county. Data from DOF wildfirecounts ...... 39

viii 3.7 Total amount of wild fires from 2006 to 2010 for last six-months of year. Values represent wildfire events per 1000 square kilometers per county. Data from DOFwildfirecounts ...... 40

3.8 Distance in miles to the closest fire at the time of a fog event ...... 41

3.9 Distribution of distances, in miles, from crash to the closest AWOS/ASOS station 41

3.10 Dewpoint depression, temperature and wind speed conditions present when AWOS and ASOS reports fog (left) and no fog (right). Amounts are tallied based on all METAR reports and are divided by total amount of eachevent . 57

3.11 Precipitation and cloud cover conditions present when AWOS and ASOS re- ports fog (left) and no fog (right). Amounts are tallied based on all METAR reports and are then divided by total amount of each event ...... 58

3.12 Average cooling rates examined for fog and no fog events for all seasons and all stations. These events are defined as to whether fog did, or did not occur, between 11 and 12 UTC. Error bars represent a 99% confidence interval. . . . 59

3.13 Average cooling rates examined for fog and no fog events during the cool season (i.e. November to March) only. These events are defined as to whether fog did, or did not occur, between 11 and 12 UTC. Error bars represent a 99% confidence interval...... 59

3.14 Conditions pre-existing fog and no-fog events occurring between 11 and 12- UTC. Top- Nocturnal dewpoint depression changes, Middle- Nocturnal wind speed changes, Bottom- Nocturnal cloud cover percentage changes. Error bars represent99%confidenceintervals ...... 60

3.15 Example atmospheric soundings iillustrating the theory of the cross over tem- perature. Top: Dewpoint (green) and temperature (black) profile at noon. Middle: Dewpoint (green) and temperature (black) profile at warmest part of day. Bottom: Dewpoint (green) and temperature (black) profile at the time fogoccurs...... 61

3.16 Average variations in temperature and dewpoint when fog did occur (right), and did not occur (left), for all stations, only for cool season. Error bars represent a99%confidenceinterval...... 62

3.17 Skill scores for detecting fog at the time a fog event is reported using only wind speed and dewpoint depression. Percent Correct (PC), Probability of Detection (POD), False Alarm Rate (FAR), and Critical Success Index (CSI) are plotted. PC, POD, and FAR are plotted on primary Y-axis. CSI is plotted on secondary Y-axis ...... 63

ix 3.18 Nocturnal hourly cooling rates for temperature and dewpoint, averaged over all stations. Equations displayed are least-squares linear regression. Error bars representa99%confidenceinterval...... 63

3.19 Forecast temperature error for 12-UTC using a 10, 6, and 3-hour forecast. Results represent the percentage of time the forecasting method had a specific temperatureerror...... 64

3.20 Skill scores for forecasting fog, 3-hours in advance, when always forecasting fog belowaforecasteddewpointdepression ...... 64

3.21 Equation used in model to determine fog forecast. Top equation represents the conditions that must be met in order for fog to not be forecasted. Bottom equation represents conditions that must be met for fog to be forecasted . . . 65

x LIST OF ABBREVIATIONS

AERMOD American Meteorological Society Environmental Protection Agency Regulatory Model AMS American Meteorological Society ASOS Automated Surface Observing System AWOS Automated Weather Observing System CCN Cloud Condensation Nuclei CSI CriticalSucessIndex DOT Department of Transportation FAR FalseAlarmRate FAWN Florida Automated Weather Network FDOT Florida Department of Transportation FOG Fog Operational Guidance FSU Florida State University GOES Geostationary Orbiting Satellite kt Knots LCL Lifted Condensation Level MADIS Meteorological Assimilation Data Ingest System MOS Model Output Statistics NAS Naval Air Station NWS National Weather Service PBL Planetary Boundary Layer PC PercentCorrect POD Probability of Detection RH RelativeHumidity SFWMD South Florida Water Management District UCF University of Central Florida UPS UnitedPostalService

xi ABSTRACT

Fog and smoke can combine to form dangerous zero visibility conditions along roadways throughout the state of Florida. The ability to forecast when and where fog will occur is problematic. Fog can occur over large and small scales, and is dependent on many me- teorological and geographic variables. This study used Automated Weather Observation

Stations (AWOS) and Automated Surface Observing Systems (ASOS) throughout the state of Florida to develop a climatology to ascertain what conditions are necessary for radiation fog development. Forecasted dewpoint depression, wind speed, cooling rates, the derived vertical hydrolapse, and other variables were shown to all affect fog formation. Using this information, a fog forecasting model was developed. The model was used to determine a three-hour binary forecast for the early morning hours, every day, at the location of the mesonet stations used. The model would predict fog if meteorological conditions preceding the forecasting time met a series of threshold levels. The goal was to make the model easy to deploy so that law enforcement can make a fast decision of whether to warn the public about potentially dangerous road conditions. The model was compared to other forecasting techniques such as the Model Output Statistics (MOS) fog product and climatology. After comparing the model to reference forecasts, it was found that the model outperformed clima- tology by a significant margin and was able to detect more fog events than MOS. However, the model had a higher false alarm rate and lower percent forecasts correct compared to

MOS .

xii CHAPTER 1

INTRODUCTION

The implications of dense fog to personal safety has been documented and acknowledged by many scientific authors (e.g Croft et al. 1997, Westcott 2006, Tardiff and Rasmussen 2007,

Johnson 1992, and Lavdas and Achtemeier 1995). The impact of fog events is evident on many aspects of transportation, including aviation, marine, and ground (Croft, 1995). In the case of aviation, planes can be stranded on tarmacs for hours waiting for fog to clear, or passengers from across the world can be stranded if their destination is covered by fog. At the

San Francisco International Airport (SFO), fog delays have become so common that NASA has become involved to help the fog problem with a new weather-coordinating program currently under development at Ames Research Center. In addition, according to the US

Bureau of Transportation Statistics, in July 2011, 3,086 flights were delayed coming into San

Francisco Airport due to hazardous fog. This has resulted in more than 3,800 hours of lost travel time (The Examiner, 2012).

Due to advanced weather observation networks in airports, specifically AWOS and ASOS stations, the impacts from fog on aviation are generally economical, and very rarely result in fatalities. The effect that fog has on ground transportation, however, has been much more severe. According to the US Department of Transportation, of the 6,301,000 accidents that occurred between 1995 and 2008, 38,000 of those crashes were fog related. During that same time period, fog events accounted for eight percent of the total weather-related accident

1 fatalities in the United States. According to Pisano (2008), it was estimated that 23 percent of delays on highways across the US are due to snow, ice, and fog. This amounts to an estimated 544 million vehicle-hours of delays per year.

The effects fog has on transportation are seen across the continental US and many other countries. Florida specifically has had a poor record when it comes to fog/smoke related crashes. The most recent example of a fog/smoke related crash in Florida occurred on

January 29, 2012. The multi vehicle crash occurred on I-75 south of Gainesville, Florida, and was responsible for 11 fatalities and more than 17 injuries. The combination of a low-lying stretch of highway, and a forest fire at nearby Paynes Prairie, created visibilities close to zero. Figure 1.1 displays the observed thermodynamic soundings for surrounding

Jacksonville and Tampa Bay, FL on January 29, 2012 at 12z. Both soundings show a low- level inversion at the surface and drier air aloft. Figure 1.2 shows the weather conditions present in Jacksonville, FL for the for 24-hr period starting January 29, 2012 at 00Z. The hours between 07Z and 13Z show relative humidity levels fluctuating around the 100% level, and as seen in Figure 1.2 goes above 100

Four years prior, in January of 2008; a 70-car pileup took place in Polk County, Florida.

Smoke from a prescribed burn, heavy fog, and lack of surveillance created the dangerous conditions leading to the accident which killed five people and injured many others (Paxton et al., 2009). Sheriff Grady Judd of Polk County described the fog as a wall of smoke and fog which had visibilities close to zero. Since 1996, eight other fog/smoke related accidents have occurred on Florida highways, involving at least 109 vehicles and 18 fatalities (Paxton et al., 2009). Currently, the National Weather Service (NWS) will issue a dense fog advisory

2 72206 JAX Jacksonville Intl 72210 TBW Tampa Bay Area 100 16310 m 100 16380 m SLAT 30.50 SLAT 27.70 SLON −81.70 SLON −82.40 SELV 9.00 SELV 13.00 SHOW 10.58 SHOW 14.86 LIFT 16.76 LIFT 10.07 LFTV 16.69 LFTV 10.28 SWET 91.01 SWET 56.99 KINX 0.70 KINX −2.70 13850 m CTOT 11.70 13940 m CTOT −13.9 VTOT 22.70 VTOT 26.10 TOTL 34.40 TOTL 12.20 CAPE 0.00 CAPE 0.00 CAPV 0.00 CAPV 0.00 CINS 0.00 CINS 0.00 200 12070 m CINV 0.00 200 12150 m CINV 0.00 EQLV −9999 EQLV −9999 EQTV −9999 EQTV −9999 LFCT −9999 LFCT −9999 LFCV −9999 LFCV −9999 10650 m BRCH 0.00 10720 m BRCH 0.00 BRCV 0.00 BRCV 0.00 LCLT 274.0 LCLT 280.0 LCLP 889.3 LCLP 898.4 300 9430 m MLTH 283.3 300 9510 m MLTH 288.7 MLMR 4.61 MLMR 7.03 THCK 5543. THCK 5614. PWAT 11.87 PWAT 15.36

400 7410 m 400 7470 m

500 5750 m 500 5810 m

600 600

700 3141 m 700 3166 m

800 800 1557 m 1566 m

900 857 m 900 855 m 1000 207 m 0.4 1 2 4 7 10 16 24 32 40g/kg 1000 196 m 0.4 1 2 4 7 10 16 24 32 40g/kg

−40 −30 −20 −10 0 10 20 30 40 −40 −30 −20 −10 0 10 20 30 40 12Z 29 Jan 2012 University of Wyoming 12Z 29 Jan 2012 University of Wyoming

Figure 1.1: Observed soundings taken on 09/29/2012 at 12z for Jacksonville, FL (left) and Tampa Bay, FL (Right). (Courtesy of The University of Wyoming) when two conditions are expected: (1) visibilities are projected to be lower than 1/4 mile anywhere over the forecast domain, and (2) the reduction in visibility is expected to last greater than two-hours (NWS, 2009). The NWS also issues a fire weather wind forecast for the 6.1 m level. This forecast helps predict how wind conditions affect smoke and burn directions, however, the drawback of this method, as pointed out by Paxton et al. (2009), is that drainage flows can occur below this level. The tools available to the NWS for fog forecasting could be improved, and when a warning is issued, it is typically too late and/or not accurate enough for localized areas of fog. Additionally, in remote areas, the places

3 Figure 1.2: Meteogram of Jacksonville, FL for 24-hr period starting January 29, 2012 at 00Z. (Plymouth State University) under warning lack the infrastructure to warn drivers efficiently.

Between 2003 and 2007, the state of Florida had the third highest number of fog and smoke related fatal crashes in the US (Abdel-Aty et al., 2012). According to Florida accident reports, a total of 994 crashes have been reported as being fog related from 2003 to 2007, with areas of high crash densities located in cities, coastal regions, and low-lying areas (Abdel-

Aty et al., 2012). The physiography of these accident prone locations are conducive for fog formation, and when combined with smoke, can become extremely opaque. From 2006 to

4 ! ! !! ! ! !! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! !! !! !! !! ! ! !! !!! ! ! !! ! ! !! ! ! ! !! !!! !!! ! !!!!! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! !!! ! !! §10 !! ! ! !! !! ! ! ! ! ¨¦ ! ! ! !! !! ! !!!!!! ! ! ! ! !! !!! ! ! ! !!!!!! ! ! ! !!!!! ! !! ! ! ! !! ! ! ! ! ! ! !!(!!! ! ! ! ! ! ! ! !!!!! !!! ! ! ! ! ! ! ! ! !!!!! !! !!! ! ! ! !! ! ! !! !! ! ! !! ! ! §10 ! ! ! !!!!!!!! ¨¦ ! ! ! ! ! ! ! !! ! ! ! ! !!! ! ! !!! ! !!!!!!!!(!! !! ! ! ! ! ! ! !!!!!!! !!!! ! ! ! ! !! ! ! ! !!! ! ! !!! !! ! ! !!!! ! ! ! ! ! !! !!!!!!! Tallahassee! ! ! ! ! !! !!!! !!! ! ! !! ! !! ! ! ! ! ! ! ! !!! ! ! ! !!! ! Jacksonville! !!! ! ! !! ! ! ! ! ! ! !! ! ! !! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! !! !! ! ! ! ! ! ! ! !!! ! !! ! ! !!!! ! !! ! !! !!! !! !!!! !!! ! ! ! !! ! ! !! ! !! ! !! ! !! ! ! !! !!! ! ! ! ! ! ! ! ! !! !! ! ! !! !! ! ! ! ! ! !! !! ! ! !!!! ! !! ! ! !! !! !! ! ! !!!!! !!!! !! !! ! !! !! ! ! ! !!!! !! ! !! ¨¦§75!! ! !!!!! ! ! !! ! ! ! ! ! ! ! !!!!!! ! !! ! ! ! !! ! ! ! ! !! ! ! ! !! !! !! !!! ! !! ! ! ! ! !! ! ! !! !!!!!! ! !! !! ! ! !! ! ! ! ! ! ! !! !! ! ! ! ! ! !! !!! ! ! !! ! ! ! !!!! !!! ! !! ! ! !!! !! ! ! !! ! !! ! ! ! !!!!!! !!!!!! !! !! ! !! !! !!!!! ! ! ! ! ! ! !!(!! !! !!! !! !! !! ! ! !! ! ! !!!!!!!! ! ! ! ! !! ! ! !! ! ! !! !! ! ! ! ! !!!!!! ! ! Orlando! !!! ! ! ! !!¨¦§!4 ! !! ! ! !! ! ! !! ! !!!!! ! !!! ! !!! Tampa! !!! ! ! ! ! ! !!!!! ! ! ! ! ! !!! ! !! !! ! ! !!!!!!!!! ! !!! ! §95! ! ¨¦ ! !!! ! ! !!! ! !!!! 2006 - 2010 Florida Fog and Smoke !! !! !!! !! ! ! !!! ! !!! !! ! !! !!! !!!!!!!!! !! ! ! ! ! ! !!!!! !!!! !!!!!!!! !! !!! !! !!!!!! !! ! !!! ! !!!!!!!!! !! !! !!!! ! !!! ! ! ! (!!!! !!!!! !!!! ! !! ! !! !!!!!!! ! ! ! !!!!! ! !!!!! ! Florida's! !!!!!!! !!!!! ! ! ! !!! ! !!! ! Related Crash Density !!!!!! !! !!!! ! ! !! ! ! !! !!! ! !! ! ! ! ! !!!! ! !! ! !! !! ! !! ! ! !!! !!! !! ! ! ! ! !! TNPK !!! !! ! ! ! !! ! !!!! ! ! ! !! ! !!!! ! !!! !! (Unit: Count of Crashes/Square Miles) !!! ! ! ! ! !!! ! ! ! !! !! ! ! !! ! !! ! !! !! ! ! !!! ! ! ! §75! ! ! !! ! !! ¨¦ ! ! ! ! ! ! ! !! ! ! !!! ! ! ! ! ! !! ! ! ! ! ! ! ! !! ! ! 0 - 0.02 ! ! !! !! !! ! !!! ! ! !!! ! !! ! !!! ! !!!! ! ! ! ! !! !! !! ! ! ! ! ! ! ! !! ! ! ! ! 0.03 - 0.04 ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! !!! ! ! !! ! !!! ! ! ! !!! ! ! !! ! ! 0.05 - 0.06 ! ! !! ! !! ! ! ! ! ! ! !!!!!!!!!!! ! ! ! ! !! ! ! ! ! !!!!!!!!! ! ! ! ! ! !! ! ! ! !!!!! ! !!!! ! !! ! !! ! ! ! ! !!!!!! ! ! !! ! !!!! ! !! ! ! !! 0.07 - 0.08 ! ! ! ! !! ! ! ! !!!! ! !! ! ! ! !! ! !! ! ! !! ! ! ! 0.09 - 0.1 ! !! ! ! !!!!!! ! !!!! ! ! !!! !!!!!!! ! ! ! ! ! ¨¦§75 ! ! !!! !! ! ! 0.11 - 0.12 ! !!!! ! !!!!!! !!! ! !!! ! ! !!! !!! ! !!!! ! 0.13 - 0.14 !!!! ! ! !!!!! (!! ! ! !! !! !!!!! Miami! ! !! 0.15 - 0.16 !! ! ! ! ¯ ! 0.17 or above 0 100 200 Miles

!!

Figure 1.3: Fog and Smoke Related Crash Density for 2006–2010. Units are in crashes per square mile. Data from Department of Transportation (DOT) accident reports.

2010 the locations and quantity of fog/smoke related crashes were very similar to that found by Abdel-Aty et al., *(2012). Figure 1.3 shows fog and smoke related crash densities between

2006 and 2010. This figure is a function of both fog occurrence and population, it does not show where the most fog events occur, but rather the quantity of accidents resulting from an unknown number of fog events. Major cities and low-lying areas are most problematic.

Also, it is important to note that crash densities along Interstate-4, between Tampa and

Orlando, are much higher than the other highways in the state of Florida.

The term superfog, coined by Achtemeier (2003), has been used to describe extreme fog events. It occurs when smoke from forest fires, which contain many hydroscopic particles, are released into the atmosphere and interact with significant atmospheric moisture. When

5 the two air masses combine it creates an extremely dense and opaque fog-smoke mixture.

Additionally, as the smoke is released, excess moisture is released from smoldering organic materials, such as plants, logs, and stumps. The combination of smoke and moisture creates a very humid smoke, and when cooled via mixing or from radiative processes, creates superfog.

This effect, along with other atmospheric conditions at the time of accidents plays a major role in determining fog intensity, the time it takes to form, and the duration of an event.

1.1 Fog Types

The definition of fog is an observed visibility below one kilometer, resulting from the pres- ence of suspended water droplets and/or ice crystals (National Oceanic and Atmospheric Ad- ministration, 1995). According to Houghton (1985), fog generally occurs when water droplets are suspended in air that is within ten percent of saturation. Typically, there are three pri- mary physical processes that can make unsaturated air become saturated: these include cooling the air temperature, adding moisture, and vertically mixing air parcels with different humidities and temperatures (Duynkerke., 1990). There are many other atmospheric and local factors that can exacerbate these mechanisms; including vegetation, horizontal and vertical winds, radiation fluxes, soil moisture, and topographic effects. Once fog has formed the primary mechanisms influencing further fog development and intensity are radiational cooling, gravitational droplet settling, fog microphysics, and cloud cover (Duynkerke., 1990).

Synoptic, dynamic, and microphysical conditions will normally control what type of fog will generally form. Willet (1928) created an all-inclusive fog classification system, later

6 revised by Byers (1959), which is comprised of 11 different types of fog, each of which were categorized by the physical processes undergone and the atmospheric scenario under which fog formed. Most fog types classified by Byers (1959), however, are merely derivatives of the four distinct types of fog as described by Stull (1999). They are: advection fog, upslope fog, frontal fog, and radiation fog. The four fog types are defined below.

1.1.1 Advection Fog

Advection fog occurs when a warm moist air mass moves over a cool surface (American

Meteorological Society, 2000). The warmer air mass loses heat through conduction to the cooler surface, thus lowering the temperature to its dewpoint (Stull, 1999). The surfaces in which advection fog can form include: cold water or ground, and ground covered with snow or ice. Advection fog is found in marine environments, around coastal areas, as water sources provide the moisture and heat necessary to facilitate this fog type. However, the natural land- breeze thermal circulation that occurs in coastal regions during the early morning hours can limit the progress of advection fog. Therefore, overlying synoptic wind speeds and directions are critical in determining whether advection fog will form. According to the US Department of Transportation (1975) advection fog deepens as wind speed increases up to approximately

15 kt. Wind speeds stronger than 15 kt will induce turbulence and mixing, leading to fog dissipation.

Sea-fog is another form of advection fog where warm moist air advects over cooler ocean air. Through conduction, the warm air cools to its dewpoint (Baars, 2009). Sea fog typically occurs in regions of cold ocean currents to the west of continents, such as over the northeast

7 Pacific Ocean off the coast of California (Baars, 2009). Sea fog is most problematic for marine transportation, and only affects ground transportation when bridges or other roadways are over a sufficient amount of water. Both sea and land advection fogs are often more opaque and longer lasting than radiation fog. This is because advection fog, once formed, can experience radiational cooling on top of the fog layer Stull (1999), this exacerbates the rate of cooling in the warm-moist air mass, creating a more dense fog. The dissipation of advection fog is similar to that of most fogs. If the relatively cooler surface becomes warmer, saturation levels would not be sufficient for fog. In addition, synoptic patterns, such as fronts, pressure systems, and wind direction can act to remove advection fog (Stull, 1999).

1.1.2 Upslope Fog

Upslope fog forms as a result of adiabatic expansion and cooling of the air as it is orographically lifted up the side of a hilly surface (Kolb and Goodmanson, 1945). Similarly with advection fog, upslope fog can form with moderate to strong winds under cloudy skies

(NWS, 2010). Under stable conditions this ground-level cloud will form when the air parcel reaches its lifted-condensation-level (LCL). If additional condensation nuclei are added to the air mass by sources such as smoke or other particles, the fog will be more dense and longer lasting. The most important factors affecting the formation of upslope fog are: lapse rate of the parcel, moisture levels at the surface and on top of the hill, wind speed, and hill shape (Kolb and Goodmanson, 1945). Upslope fog will typically persist on the upslope side of the hill until the forcing at lower levels subsides, and/or there is a change in temperature or humidity levels.

8 1.1.3 Frontal Fog

Frontal fog, also known as precipitation fog, is usually divided into three types: warm- front prefrontal fog, cold-front post-frontal fog, and frontal-passage fog (Byers, 1944). Pre- frontal fog occurs in the cool stable air mass ahead of a warm-front when warm stratiform precipitation falls on the cool side of the front. As the rain falls, it evaporates and raises the dewpoint of the surrounding air, making the cloud lower towards the surface (Gultepe et al., 2008). Due to its location and climate, the Northeastern US is most at risk for this type of fog, as mid-latitude cyclones occurring during the fall and winter bring the conditions nec- essary for prefrontal fog development. The mechanisms that form cold front post frontal fog are very similar to the aforementioned prefrontal fog. Evaporation from falling precipitation humidifies the air behind a cold front and acts to lower the cloud to the surface (Gultepe et al., 2008). This type of fog is unlikely to be widespread due to the limited amount of pre- cipitation that falls behind a cold front. Stationary fronts, however, could provide the ideal environment. Lastly, frontal passage fog can also occur when near saturated air parcels from warm and cold air masses mix together in calm wind environments (Gultepe et al., 2008).

1.1.4 Radiation Fog

Radiation fog forms when radiative fluxes off the surface are sufficient to reduce the air temperature to its dewpoint (American Meteorological Society, 2000). This fog type forms at night and typically requires clear skies and abundant low-level moisture. Clear skies are necessary in order for long-wave radiation to radiate off the earths surface, allowing surface

9 temperatures to decrease rapidly. If dewpoint temperatures are sufficiently high, humidity

levels can reach a critical point, and fog will form. In addition, light winds, typically below

2.5 ms−1 , are also necessary for radiation fog to occur (Taylor, 1917). If wind speeds are too strong, turbulence within the boundary layer will result, and low-level moist air will mix with drier air aloft. However, if winds are too calm, gravity will force the suspended water droplets to settle on the ground, creating dew/frost. Other favorable conditions for radiation fog formation include: a small dew point depression at sunset, low-lying areas or valleys, and large amounts of condensation nuclei.

Radiation fog forms upward from the ground as the night progresses and is usually deepest and most opaque around sunrise. Initially, the fog density decreases with height as temperatures at low-levels increase with height. However, as the fog continues to thicken at lower levels, it restricts the surface/ground from emitting long-wave radiation. When conditions reach this point, the maximum radiative cooling level moves upward toward the top of the fog layer. This results in denser fog at the top of the layer and initiates sinking cold thermals which act to turbulently mix the fog (Stull, 1999). Radiative cooling at the top of the fog can act to maintain and strengthen the fog intensity (Stull, 1999).

Radiation fog generally begins dissipating when the sun rises in the early morning hours, initiating mixing in the boundary layer. Through this method, the surface warms quickly as it absorbs short-wave radiation and then warms the surrounding air. The water vapor droplets evaporate into the warmer air, resulting in dissipation of the fog. Radiation fog can also dissipate if there is a change in the overlying synoptic conditions, such as fronts or winds, or the dynamic forcing is altered.

10 Figure 1.4: Percentage of total fog occurrence given by time of day. Florida only.

In the southern United States, radiation fog is most problematic during the winter as the length of night is sufficiently long enough to drop the air temperature to the dewpoint.

Using AWOS and ASOS data from 1947-2012 the observed frequency of fog formation is shown in Figure 1.5 and Figure 1.4. Fog is most frequent in the cold months and relatively infrequent during the warm months, fog is also a nocturnal event, peaking just before dawn and rare during day-time hours.

11 Figure 1.5: Percentage of total fog occurrence given by month of year. Florida only.

1.2 The Fog Forecasting Problem

Advances in the understanding of fog formation have been made through many field and numerical experiments (Tardiff and Rasmussen, 2007). However, due to the variable and sudden nature of fog events, as well as the small scale over which it occurrs, the ability to forecast fog remains challenging. Herzegh et al. (2004) noted that fog forecasting is often done through observations (nowcasting); however, this has proven difficult, as remote sens- ing techniques involving satellite and radar tend to be ineffective. For example, fog often occurs at too low of an elevation for conventional radar to identify it, and satellite imagery has difficulty in distinguishing fog from low level stratus (Westcott ,2006), although some

12 wavelength techniques may allow for identification of widespread fog events (e.g. Ellrod

2002, and Bendix et al. 2005). Not only is fog difficult to identify with conventional sensing

techniques, but fog is also highly variable in time and space, it may occur many times in

many different areas, or not at all (Ratzer, 1988). In addition, fog is dependent on many

physical variables on different length scales within the boundary layer, including microphys-

ical, dynamical, mesoscale conditions, and the overlying synoptic flow (Gultepe et al. 2006,

Westcott 2006). Croft et al. (1997) attributes the fog forecasting problem to the fact that

fog is a boundary layer phenomenon, and since the boundary layer is initially set-up by syn-

optic scale factors, and then later is also affected by dynamic mesoscale factors. Prediction

of the interactions between these scales is not accomplished through current models. Also,

microphysical processes further act to complicate the interactions.

As a result, fog is very difficult to forecast and model, particularly in regions of Florida, where radiation fluxes and advection can both determine whether fog will form. The fog- forecasting problem becomes exacerbated when the effects of forest fires (smoke) are inte- grated into the forecast. The addition of smoke not only results in microphysical changes in condensation-nuclei, but according to Achtemeier (2003), the burning of organic material leads to excess moisture in the atmosphere, which could result in fog forming faster and having a longer duration.

Over the past half-century there have been numerous developments and methods to

solve the fog-forecasting problem (e.g Leipper 1995, Gultepe et al. 2006, Croft et al. 1997,

Duynkerke 1990); however most of the modeling and forecasting studies were only designed

to be used in a specific location or region. Therefore, a fog model or forecasting method

13 for one location may not apply to different locations, where topography, climate, and other

environmental variables increase in complexity (Tardiff and Rasmussen, 2007). For example,

fog has been studied extensively on the west coast of the United States by many authors

(e.g. Leipper 1995, and Lewis 2003). Tardiff and Rasmussen (2007) examined fog in the New

York City area, and Croft et al. (1997) conceived a conceptual model for fog forecasting in

the Southern region of the United States. The need, however, for a fog forecasting method

remains for much of the United States, including Florida, which, as mentioned before, has

had many fatal car crashes from sudden, dense fog events.

The objectives of this project include: researching what conditions generally lead to fog formation in the state of Florida, researching how the current weather observation network across the state can be used to forecast fog on a short time scale, investigate the effects that smoke from control burns has on fog formation, develop a forecasting tool to improve fog forecasting, and test the skill of the developed model to other forecasting techniques.

1.2.1 Numerical Weather Prediction Limitations

The NWS forecasts fog using a variety of tools, including model output statistics (MOS), conceptual models, and many other methods as described by Baker et al. (2002) and Cox (2007).

However, Ballard et al. (1991) writes that forecasting fog tendencies, such as formation, intensity, and dissipation remains one the most difficult problems for a forecaster. Leip- per (1995) regards fog forecasting as a difficult task which involves predicting the formation and dissipation of a cloud at a certain time and specific location in space.

14 Despite the fact that numerical weather prediction models have continued to advance with time, the progress made within these models, in terms of forecasting fog, has been slow (Zhou et al., 2007). NWP model resolution is too coarse to model local scale fog, and models have difficulty because the cloud parameterization schemes only function well for clouds at high levels (Miller et al., 2005). Also, to predict fog, NWP uses local solutions, aided by mesoscale models. Since the domain for these forecasts encompass the entire continental US, the computer resources available are often insufficient (Zhou et al., 2007). Additionally, Gultepe et al. (2006) notes that the NWP models’ coarse vertical resolution and oversimplified cloud physics result in an inability to resolve mesoscale processes affecting fog formation. Another problem with NWP, as stated by Gultepe et al. (2006), is the model calculates visibility with relationships between liquid water content and visibility (Gultepe et al, 2006). In other words, NWP models use liquid water content (LWC) in order to predict the possibility of fog. However, since surface observations are spatially insufficient in determining this variable, initiating the model with the LWC could give a bad representation of future atmospheric conditions.

1.3 Fog Forecasting Techniques

Due to the inadequacies of NWP models, other methods have been developed to forecast fog. Croft et al. (1997) created a simple conceptual model for the southern region of the

United States by employing well known physical fog formation principals to new forecasting techniques, and Baker et al. (2002) developed the United Parcel Service (UPS) Airlines

15 conceptual model. These models are described below.

1.3.1 The Croft et al. Conceptual Model for the Southern U.S.

The conceptual model developed by Croft et al. (1997) for the southern United States employs boundary layer characteristics such as air mass type, cloud condensation nuclei available, moisture availability, and dynamic forcing. The different variables that must be accounted for in the model represents the scale lengths and processes that affects fog formation; which include synoptic, mesoscale, microphysical, and dynamic effects. The model itself is comparable to a forecast decision tree, where the forecaster answers certain questions about atmospheric conditions, which then leads him/her to a guidance forecast.

To use the model effectively the forecaster would first judge the overall synoptic pattern to determine what type of air mass would be affecting the area of concern (Croft et al., 1997).

The model has a wide range of synoptic conditions it can account for, including everything from maritime air masses to continental air masses, and allow the forecaster to identify the concentrations of CCNs and relevant drop sizes. This step helps provide information to the forecaster of how heavy/opaque the fog will be as well as its duration (Croft et al., 1997).

When this step is completed the forecaster would look at moisture availability in the area as well as dynamic forcing, specifically looking for the development of a surface layer inversion. Dynamic forcing mechanisms are assessed according to base-state flow, local circulations, and thermodynamic lifting (Croft et al., 1997). From the dynamic forcing, the forecaster can determine the duration of the fog event as well as the extent of the occurrence.

Moisture availability is determined according to how much moisture can be realized from

16 condensation through cooling. This variable gives the forecaster an idea of the spatial extent

of fog, and when combined with CCN and drop sizes observations, can give a good idea of

the fog intensity. The assessment of all these variables allow the forecaster to move within

the bases of the model to determine how likely fog is to form, its intensity, and duration

(Croft et al. 1997).

1.3.2 UPS Airlines Conceptual Model

As discussed previously, fog can have severe economic consequences on the airline indus- try. The United Postal Service (UPS) Airlines Company ships packages across the world, and is often affected by fog delays due to the high number of arrivals and shipments at sunrise (Baker et al. ,2002). The fog forecasting problem motivated UPS airlines to develop a conceptual model using practical quantitative forecasting tools.

The main idea behind this model was to account for vertical tendencies that affect the fog formation process. Baker et al. (2002) noted that surface-based approaches to fog forecasting often fail to account for the very important changes that occur above and below the forecasted altitude. The UPS airlines fog conceptual model includes many of the variables which are typically ignored in the surface-based models, including; the vertical distribution of humidity in the potential fog layer, the turbulent mixing potential of the lower boundary layer and the surface temperature below the fog layer. The UPS method attempts to quantify these variables in order to better assess vertical atmospheric changes, and how they affect fog formation (Baker et al., 2002). This method is discussed in more detail in chapter 3 due to its use in this project.

17 1.3.3 Forecasting Using Model Output Statistics

Model Output Statistics is commonly used by the National Weather Service to objec- tively interpret numerical model output and produce site-specific guidance for a 6-84 hour period, at three hour increments (NWS, 2008). MOS relates observed weather variables to different predictors with a statistical approach. The predictors that MOS uses include; NWP model forecasts, past observations, geoclimatic data, and linear regressions between the pre- dictors and predictands. MOS specializes in objectively interpreting NWP models based on historical data and predicting events forced by synoptic-scale systems. MOS is also able to quantify uncertainty in NWP models’ forecasts and can adjust for certain biases within the

NWP model (NWS, 2003). MOS is able to account for some local effects, however, it lacks the computing power necessary to consider every local effect. Also it is also unable to predict based on mesoscale forces.

When forecasting fog, MOS uses probabilistic and categorical guidance for obstructions

to visibility. Typically, MOS will produce one of the five forecasts for the obstruction to

visibility category. They are listed below:

1. No non-precipitating obstructions

2. Haze, smoke, or dust

3. Light fog or mist (fog with visibility of 5/8 mi or greater)

4. Dense fog or ground fog (fog with visibility of <5/8 mi)

5. Blowing snow, dust or sand

18 To determine which obstruction to visibility forecast to issue, MOS will correlate ob- servations, such as ceiling and visibility to NWP model forecasts, such as predicted values of relative humidity, precipitable water, temperature, wind speed, etc. (Croft et al., 1997).

According to the WMO (1991), the predictors based on observations are the most important input in the model for giving an accurate short-range forecast.

Croft et al. (1997) examined the statistical prediction of dense fog for the 24 and 6 hour

MOS forecast for the cities of Jackson, Mississippi; Mobile, Alabama; and New Orleans,

Louisiana. Croft et al. (1997) found that the best statistical predictor for dense fog for the

24-hour forecast was the gridded relative humidity, while the 6 hour MOS forecasts leading indicator was the latest observed visibility at the station during model initiation. The predictors used for these models, identified as statistically significant, were shown to have nearly identical correlation coefficients to that of persistence nowcasting (Croft et al., 1997).

This suggests that the MOS fog product is very similar to a persistence forecast.

Although the resolution and accuracy of MOS has improved from Croft et al’s. (1997) study, most of the inadequacies remain. For example, MOS is unable to account for extreme climatic conditions. This is a problem when smoke from forest fires occurs and alters the microphysical conditions affecting fog formation, or vertical profiles differ immensely from climatological means. Additionally, since MOS only adjusts forecasts based on statistical relationships, it is unable to correct for NWP model physics, analysis schemes, or param- eterizations. Additionally, if the weather observation network is spatially insufficient for recording observations, MOS will predict with less accuracy (Croft et al., 1997).

19 1.3.4 Other Fog Forecasting Techniques

There have been many other attempts to develop fog forecasting techniques. Some are based on climatology (e.g. Johnson and Grashel 1992, and Jarvis et al. 2001), where past atmospheric conditions are evaluated during fog events, and then are used to forecast fog when alike conditions are forecasted to be present. Gurka (1978), Ellrod (2007), Gultepe et al. (2008) employed remote sensing techniques, primarily using infrared imagery from the

Geostationary Orbiting Satellite System (GOES) to detect fog location and depths. Others have focused on improving the microphysical parameters in numerical models, Gultepe et al. (2006) suggested a new parameterization for fog visibility in numerical NWP models, where he said that both droplet number concentration and liquid water content (LWC) are a better gauge in determining visibility within NWP models compared to the current scheme which only uses relationships between visibility and LWC (Gultepe et al., 2006). Other methods including statistical relationships, numerical modeling, operational modeling, and conceptual models, have all been used to forecast fog (Croft et al., 1997).

Unfortunately, many of these methods are incapable of forecasting or detecting fog in local areas. Many numerical models are incapable of forecasting fog due to excessive assumptions and poor resolution (Croft et al., 1997). Also, NWP depends on accurate initiation in order to forecast fog on a small scale. Unfortunately, the current weather observation network is incapable of providing enough detail to support the NWP model initiation. Climatological techniques are incapable of handling any extreme conditions that may be present at any time. An example would include an influx of CCN from control burns or forest fires in a

20 specific area. Lastly, conceptual models act to give forecasters a good idea of whether there will be fog or not; however, they do not do a good job in determining localized fog as these models depend on the data available and a forecasters skill.

Since the objective of this study is to create a forecasting method which is easy to imple- ment and use, the skill will not be much higher than the previously mentioned forecasting methods. However, the model developed takes much less time and needs much less data compared to numerical and statistical models, and as shown later, only sacrifices a relatively small amount of skill.

21 CHAPTER 2

METHODOLOGY

2.1 Thesis Goals

The aim of this investigation is to determine when and where fog formed and then look at the surrounding meteorological data to see if there is any way to forecast fog formation using climatology. We needed to identify locations where fog was reported and when it occurred. This data would form the backbone of the developed climatology. We first sought locations through inquiries to law enforcement agencies and other entities, however to the low amount of responses recieved, it was decided that human observations would not be used. Next we sought information from ASOS and AWOS stations, particularly at airports.

The expectation is that they would report fog when it occurred and differentiate that from the days when it did not occur. The project would contain the following steps:

• We would identify locations where we could reliably know each day if fog formed or did not form.

• Data from each site would be used to develop a comprehensive climatology of fog events in Florida.

• A forecast model would be developed incorporating meteorological conditions shown to affect fog formation.

• The data set used to develop the model would also be used to determine statistics to see how accurately the model predicts both the occurrence and non-occurrence of fog.

• The forecast model would be compared to other forecasting methods such as climatol- ogy and MOS.

22 2.2 The Data

There are four institutional mesonet networks as well as a large number of private sites, all of which measure many meteorological variables. Some of the private mesonet sites are well maintained, and others have unknown maintenance status. Much of the data are archived at

MADIS where all the incoming data are subjected to some quality control. Data generally arrive in 30-minute or 1-hour intervals, but some as often as 15-minute. All stations are shown in Fig. 2.1.

!!! !!! !!!!!!!!!! !!!!!!!!!!!!!!! !! !! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!! !!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!! !!!!!!!!!! ¨¦§10 !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!! !!!!!!!! !!!!!!!!!! ! !!!!!!!! !!! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!! ! !!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!(!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! ! !!!!!!!!!!!!!! !!!!! ! Jacksonville !!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!! 10 ! ! !!!!!!!!!!!!!!!!!!!!!!!!!(!!!!!!!!!!!!!!!! ! ¨¦§ !!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!! !! Tallahassee!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!! !!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!! !!!!!!!!!!!!!!!!!!! !!! !!!!! !!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!! !! !!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!! !!!!!!!!!! !!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!! ! ! !!!!!!!! !!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!! !!!!!!!!!! !!!!!!!!!! !!!!! !!!!!!!!!!! !!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! ! !! ! ! !!! !!!!!! ! !!!!!!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! 75 !!!!!!!!!! !!! !!!!!!!!!!!!!!!!!!!!!!!! !!!!! ¨¦§ !!!!!!!!!! !!!!! !!!!!!!!!!! !!!!!!! !! ! !!! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! ! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! ! !!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! (! !!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !! !!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!! !!!!!!!!!!!!!!!! 4 Orlando !!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!¨¦§! !!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!! !!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!! ¨¦§95!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!! ! !!!!!!!!!!!!!!!! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!(! !!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!! !!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Florida's !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!! !!!! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!! Tampa! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!! TNPK ! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! ! !!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!! ! !!!!!!!!!! !!!!!!!!!!!!!!! ! !!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!! !!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !! !!!!!!!!!!!! !!!!! !!!!!!!!!! !!!!!!!! ! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !! ! ! ! ¨¦§75!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!! ! ! !!!!!!!!!!!!!!!!!!!!! Weather Stations in Florida !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!! !!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!! !!!!!!!!!!!!!!!!!!!!!!!! !! ! !!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! !!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ! Florida MESO Stations (Primary) !! ! ! !!!!!!!!!!!!!!! !!! !!! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!! ! ¨¦§75 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!! ! ! Other Stations (Secondary) !!!!!!!!!!!! !! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!! !!!!!!!!!!!!!!!! !!!!! !!!! !!!!! ! !!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Miami!!!!!!(! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! (! Florida Major Cities !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! ¯ !! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!! ! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 0 100 200 !!! !!!!! Miles ! ! !!!!!!!!!! !!!!!!!!!!!!!!!

!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!

Figure 2.1: All mesonet stations in Florida. Primary stations are defined as AWOS, ASOS, FAWN, and FWMD sites. Secondary sites are defined as individual and privately owned weather stations. Latitiude and longitude data from NOAA.

23 The most extensive sites with well-documented commitment to maintenance are the Au- tomated Surface Observing System (ASOS), Automated Weather Observing System (AWOS),

Florida Automated Weather Network (FAWN), and South Florida Water Management Dis- trict (SFWMD) networks. Although SFWMD is one of five water management districts, it is the only district that maintains an extensive mesonet. These four networks are the primary mesonet stations that were evaluated to be used, as they are assured to be well-maintained.

These are illustrated in Fig. 2.2

!! *# *# ! 10 ! ¨¦§ *# *# ! (! ! ! ! ! (! ! Tallahassee ¨¦§10 *# *# ! ! ! ! ! Jacksonville ! *# *# ! ! ! *# *#

*# ! *# ! *# §!75 ! ¨¦ *# ! *# ! *# *# *# ! Orlando! ! *# (! ! ? 4 ! ! ! ¨¦§ ? ¨¦§95 Tampa !*# ? ! ! ! *#!*# Florida's ! (!! ! *# ! *# *# TNPK ! *# ? ! ! ! *# *# *# ? ! ? *# ? ! *# ? ? ! ? ! *# ? ? ? *#! ? ? ¨¦§75 ?*# !? Primary Weather Stations in Florida ! ? ! ? ? *#! ? ! ASOS ? ? ? ! ? ? ! ! AWOS ¨¦§75 *#! ! ! *# FAWN ? ! Miami!(! ? FWMD ?! *# (! ¯ Florida Major Cities ? ? 0 100 200 Miles

! !

Figure 2.2: Primary Mesonet stations in Florida.

Only the highest quality data were used to develop the climatology in this project. As

noted in the ASOS and AWOS user guidelines, AWOS and ASOS stations have rigorous

24 maintenance schedules and their data is heavily QA’d. Since other weather observation

networks do not undergo comprehensive QA, it was determined that data from AWOS and

ASOS stations are of the best quality, and therefore were the only data used in this study.

The time period selected from the data-set was between January 1st, 2006 and December

31st, 2010. Since the model developed is statistical, data prior to 2006 was not used due to

the potential changes in climate and data source locations.

There are a total of 93 AWOS and ASOS stations throughout the state of Florida, 77

of which are located at airports. Upon review of the data quality for these stations, it was

determined that 16 of these stations did not create data that was reliable or seemed correct.

For example, some stations would never report fog throughout the time of the study, or would

report a very small amount of fog events. If these stations were found to be in a geographically

favored region for fog formation, then that station’s data was removed from the study. Also,

if at any observation time, a station was missing temperature, dewpoint, wind speed, or

visibility readings, that observation was removed from the data set. Additionally, the data

was filtered to remove other bad data such as extreme outliers, duplicates, and times when

the station was down for maintenance or repair.

AWOS and ASOS stations typically take observations in one-hour intervals. However, if there is significant weather occurring, AWOS and ASOS stations will/can take observations more frequently. In the data set used, these stations typically took observations 10-minutes before the end of every hour. Any extra observations taken during that hour due to a special weather event were removed. In order to create a temporally homogenous data set the following steps were performed: extra observations recorded due to extreme weather events

25 were removed, time stamps were rounded to the nearest hour, and only overlapping times

from all stations were kept.

As mentioned previously, there are many factors that affect fog formation. The AWOS

and ASOS stations report many variables, some of which are believed to be important to

fog formation and others that are not. Meteorological variables that fog formation has been

known to be dependent on, which are reported by these stations, include: temperature,

dewpoint, wind speed, cloud cover, precipitation, and visibility. When the visibility falls

below the threshold value of 5/8 of a mile, the station will report fog. If the visibility is

greater than 5/8 of a mile, but obscure conditions still exist, the METAR will report mist.

Occasionally however, automated stations will still report fog even when the visibility is

greater than 5/8 of a mile. Most of the time this is due to fog forming just above or below

the visibility sensor, or, this can also occur when the visibility sensor registers highly volatile

readings which fluctuate above and below the fog-threshold value. When this occurs, the

METAR will report ”patchy fog”.

According to the ASOS and AWOS user guidelines, both systems have known resolutions

and error ranges when reporting. ASOS stations have a temperature resolution of 1.0◦F and

a maximum error of 1.8◦F, where-as AWOS stations have a temperature resolution of 1◦F and a maximum error of 2◦F. The difference in resolution and error between the two systems

will distort the temperature climatology somewhat, however, the effects are expected to be

small.

Fire data for both prescribed burns and wildfires are taken daily by the Department

of Forestry (DOF). The DOF reports size, longevity, acres burned, and the location of all

26 fires throughout the state. In this study, fire data was taken from 2006-2010, and was used to develop a climatology of fire events. This climatology includes; the normalized average number of wildfires per county, the proximity of wildfire events to fog events, and the trend of wildfires and prescribed burns from 2006 to 2010. Additionally, this study utilized an air dispersion model called the American Meteorological Society/Environmental Protection

Agency Regulatory Model (AERMOD) to look at the maximum radius smoke from a wild

fire would have on visibility. Four different emission rates were utilized to determine the maximum impacts wildfires could have.

2.3 Methodology for Detecting Fog at an

AWOS/ASOS Station

To get a better understanding of when and how fog forms, the data-set was separated into two files: the first file only contains times in which fog was reported, and the second file only includes data when no fog occurred. After completing this, analysis was run on both

files in order to examine what conditions existed during fog and no-fog events. Specifically, computer programs written in Perl were used to categorize the conditions when fog formed and did not form. For example, when fog did form, the dewpoint depressions were calculated and stored in a separate file, these numbers were then plotted showing the percentage of time that fog formed within a specific dewpoint depression range. This was also completed with wind speed, cloud cover, temperature, and precipitation.

27 After finding out which variables were most important to fog formation, the next step was to figure out how well fog can be detected using an algorithm based on these variables.

Using climatology, threshold values of dewpoint depression and wind speed were used to detect fog. A program was created to optimize the highest success rate when detecting fog below threshold values of wind speed and dewpoint depression. The highest success rate was determined using a contingency table as shown in Fig. 2.3. The contingency table was used to calculate four different skill scores; the percent correct (PC), false alarm rate

(FAR), probability of detection (POD), and critical success index (CSI). The PC gives the percentage of forecasts correct divided by the total number of forecasts made, the FAR gives the percentage of time a ”yes” forecast was issued but was not observed, the POD gives the percentage of time that an event was forecasted and that event was observed. The CSI is the total number of correct events forecasted divided by the total number of ”yes” forecasts plus the number of misses (hits + false alarms + misses). The CSI is not affected by the number of non-event forecasts that verify.

2.4 Methodologies for Forecasting Fog at an

AWOS/ASOS Station

The ability to forecast radiation fog is mainly derived from the ability to forecast the cooling rate throughout the night. It is well known that fog formation is dependent on more than just dewpoint depression; mixing, upper level moisture, and radiation fluxes all

28 Observed a: an event is forecast and the event occurs Yes No b: an event is forecast and the event does not occur Yes a b c: an event is not forecast and the event occurs Forecast d: an event is not forecast and the event does not occur No c d

a + d Percent Correct (PC) = a + b + c + d a Probability of Detection (POD) = a + c b False Alarm Rate (FAR) = a + b a Critical Success Index (CSI) = a + b + c

Figure 2.3: Contingency Table play a role, however, the objective of this project is to create an easy, efficient way for law enforcement to know when fog may occur. This project employs two different techniques to forecast fog. They are listed below:

• The first technique would be binary, and would forecast fog solely on the forecasted dewpoint depression.

• The second method will use a model/algorithm that uses all the variables fog formation is dependent on to make a binary forecast.

To forecast fog, a night time cooling rate must be established. Using climatology, an average night-time cooling rate was calculated per hour from 00-UTC to 12-UTC. Using the estimated cooling rate, and assuming the dewpoint temperature remains relatively constant throughout the night, a forecast for dewpoint depression is made. This information, as well as other factors, such as precipitation, cooling rates at 00-UTC, daytime minimum

29 dewpoint, and wind speed were used as inputs in the model to forecast fog. The model was

run on the 2006 to 2010 data set, and was used to make a 3-hour forecast. To evaluate the

performance of the model, the same contingency table used for detecting fog, would also be

used to evaluate the success of the fog forecast. In addition to the model, a program was

built to optimize the best threshold values for predicting fog when the forecasted dewpoint

depression is at a certain level.

To determine if the model has skill, it was first evaluated against climatology. A cli-

matological forecast predicts fog, based on the average percentage of time that a fog event

occurred at an individual station for each month of the year. For example, if fog occurred

10% of the time in December for the Tallahassee station, then the climatology forecast would

say there is a 10% chance of fog every day for the month of December in Tallahassee. Using

climatology as the zero-skill forecast, and running this forecast on the data, we compared

the success rate of the climatology forecast to the model forecast.

The forecast model was then compared to the MOS fog forecast product. To do this,

MOS forecasts between 2006 and 2010 were downloaded. Currently, MOS does not have a

3-hour fog forecast for 12-UTC. Due to this, we could not compare the skill of the three hour

forecast of the model developed to a MOS three hour forecast. However, since both methods

forecast for 12-UTC, a comparison was made using the latest MOS output for a 12-UTC

forecast and the three-hour model forecast for 12-UTC.

In order to objectively compare the forecast methods developed in this study to clima- tology and MOS, two well-known skill scores were used. The first being the Brier Skill Score

(BSS) and the second being the Heidke Skill score (HSS).

30 CHAPTER 3

RESULTS

3.1 Fog / Fire Climatology of Florida

Most of the fog in Florida is radiation fog. Unlike advection fog, radiation fog is most

often found in “patches”, and may extend over several miles. Just as often, radiation fog

may be well less than a mile in extent and can be found intermittently along the ground.

Elevation is often an important factor, where differences in a few feet can be the determining

factor to where fog forms. Often there are “favorite” depressions where fog most frequently

forms. The average number of fog days in Florida per year is shown in Fig. 3.1.

Note that fog is generally more prevalent in the panhandle than in peninsular Florida,

yet most crashes are in the central peninsula. The visibility-related crashes are a product of

the propensity of fog and the amount of vehicular traffic.

The maximum at Tallahassee is probably due to two factors. The first is the placement of the airport in a locally low area. Secondly, there is a synoptic condition in the summer that favors fog formation in the lower Southeastern U.S.

Lavdas and Achtemeier (1995) contoured the amount of low visibility events using the

Low Visibility Occurrence Risk Index (LVORI). The LVORI index is a tool which uses dispersion indexes and relative humidity to determine the risk of a dense fog/smoke occurring.

Their results, illustrated in Fig. 3.2, shows the general trend and location of highest fog events is similar to the climatology in this study.

31 ! ! ! ! ! ! ! §10 ! ! ! ¨¦ Tallahassee ! ! ! !! ! Jacksonville! ! !

! Gainesville _! !

¨¦§75 ! ! ! ! ! ! Orlando! ! ! ¨¦§4 Average Fog Days Per Year (2006-2010) ! ! ! Tampa! ! ! ¨¦§95 ! !! ! ! ! _ Paynes Prarie ! ! AWOS/ASOS Stations ! ! Fog Days per Year !

0 - 5 ! 5 - 10 ! 10 - 15 ! ! 15 - 20 ! ! ! 20 - 25 75¨¦§ ! ! 25 - 30 Miami! 30 - 35 ! ! 35 - 40 ¯ ! 40 - 45 45 - 55 020 40 80 120 160 Miles

Figure 3.1: Climatological location and frequency of fog in Florida, averaged per year from 2006 to 2010. GIS kriging technique used to draw contours

Fog usually forms when the temperature is at or near the dew point and the mixing ratio is relatively large; that is, there is sufficient moisture in the air. It is not only a thermodynamic problem, the physical terrain, soil moisture, and vegetation all make a difference.

Compounding the problem of determining where fog has formed is the lack of data of where and when fog has formed. Fog is reported by ASOS and AWOS stations that are often located at airports. Thus, in the Panhandle of Florida, for example, the ASOS site is at the Tallahassee airport (TLH), which happens to be located in a low lying area where fog forms frequently. This location is not representative of the weather conditions over most of

32 Figure 3.2: Estimated amount of low visibility occurrences using Low Visibility Occurrence Risk Index (LVORI) (Lavdas and Achtemeier, 1995). the area surrounding it, or even Tallahassee. As seen in Fig. 3.1 there is little data in the

Everglades and thus almost no reported fog events.

There are two very important considerations in the climatology of fog and the impacts it has on vehicular traffic. The first is simply what is considered fog, or fog that is dense enough to affect safe driving. The second is the local “spotty” nature of fog. Instruments are likely to be placed neither where fog most frequently forms, nor where fog is likely to form. Thus, data from any station only approximately represents the conditions of its surroundings. These are both extremely important considerations when interpreting these

33 results. In addition, fog-related accidents do not necessarily occur where fog is most frequent,

but where the product of the occurrence of fog and the density of traffic is the greatest.

3.1.1 Topography

Florida is relatively flat with a ridge in the central peninsula. However, fog is much more likely to occur in lower elevations, often characterized by changes of only a few feet. In general, these small differences are not resolved in much of the topographical data, and are best known from the climatology of fog occurrence. Unfortunately, the data on fog occurrence is only very coarsely known. An illustration of the topographic features of Florida is given in Fig. 3.3

3.1.2 Fire

Fire contributes both moisture, and more importantly, cloud condensation nuclei (CCN)

to the formation of fog. Although CCN are naturally ubiquitous, the more CCN, the more

droplets will form. However, the same total water content distributed among many more

smaller drops will significantly decrease the visibility. A large majority of the emissions

from a forest fire is PM2.5, defined as microscopic particulates with a diameter less than

2.5 micrometers. Smoke, in and of itself, can form a “smoke fog” made up of PM2.5 that can restrict visibility and produce an irritating acrid smell. For this to be a significant driving hazard, a large wildfire very close to the roadway is required. According to Wang et

3 al., 2006 , under dry conditions, concentrations of PM2.5 would have to exceed 500µm to lower visibility below 500 meters. However, under saturated conditions, PM2.5 concentrations

34 ¦¨§10 ¦¨§295

¦¨§4

¦¨§95

Florida Elevation Map ¦¨§275 Elevation (meters) 0-10 10-20 20-30 30-40 595 40-50 ¦¨§75 ¦¨§ 50+ ¯ 030 60 120 180 240 Miles

Figure 3.3: Topography of Florida in meters using shapefile data from the United States Geologic Survey (USGS)

of 100µm3 can lower visibility to less than 500 meters. To determine the spatial extent of high PM2.5 concentrations from wild fires, the air dispersion model AERMOD was used.

A fictional 500 acre wildfire was simulated in Paynes Prairie off of US-441, and the fire was configured to emit four different emission rates. Averaging the dispersion properties with the meteorologic conditions every hour for the entire year yielded the results seen in Fig. 3.4.

As the emission rate increases, concentration levels greater than 100µm3 are observed further

from the site. Under saturated conditions, and assuming a typical wildfire emission rate,

the visibility could significantly decrease within two kilometers. Using wild fires intensity,

35 therefore, could show high risk areas of superfog around the wildfire site.

The number of prescribed fires far exceeds the number of wild fires. It can be seen from Fig. 3.5 that the total number of wild fires increase as the number of prescribed burns decreases and vice versa. Since wild fires are unplanned, they pose a greater risk to public safety. Because of this, dense fog and smoke is more likely to form when a wildfire is present rather than a prescribed burn. It is useful to look at the time of year and the locations where wild fires are more abundant. The distribution of the total number of wild fires for the first half of the year, occurring from 2006 to 2010 are given in Fig. 3.6, and for the last six months of the year in Fig. 3.7. From these figures it can be seen that the first six months of the year are more prone to wild fires than the latter half. Therefore, the chance of a dense fog resulting from event is increased during this time period.

To examine the possible impact of fire-generated CCN and their effects on the production of fog, we look at the distance of a fog event to the closest fire that occurred within the past

24 hours. Fig. 3.8 shows that the median distance was 3 miles, but many events ranged from two to ten miles. What effect that actually had is not clear, except that fires are widespread enough and frequent enough to contribute additional CCN.

3.2 Forecasts

3.2.1 Mesonet Data Climatology

There are several approaches that can be used to improve fog forecasts. Of prime im- portance is knowing where fog typically forms. As explained in the fog climatology section,

36 Figure 3.4: Smoke dispersion modeling results at Paynes Prairie site. Concentra- tions in micrograms per meter-cubed. Top left emission rate = 0.0003gs−1m−2, Top right emission rate = 0.0006gs−1m−2, Bottom left emission rate = 0.0008gs−1m−2, Bottom right emission rate = 0.0016gs−1m−2. Outermost contour represents 100 micrograms per meter-cubed. Second outermost contour represents 500 micrograms per meter-cubed.

37 Figure 3.5: Number of prescribed burns and wild fires in Florida from 2006 to 2010. Data plotted in monthly intervals this is only well-known at selected locations, typically around airports. Out of necessity, we have chosen to forecast fog using two methods. The first uses threshold dewpoint depression values, and the second uses a fog forecasting model derived from climatology. The assump- tions we employ include (1) we have available climatological data, and (2) the only other data we have is that which comes from sensors that report standard meteorological data at one-hour intervals. For the algorithm development, we have restricted ourselves to just the

ASOS and AWOS stations because (1) they are reliable, and (2) they alone report visibility.

ASOS and AWOS sites are never located at the point of a crash. It is useful to know how far to expect the data to be removed from the point of interest. The distance from known fog related crashes to the nearest ASOS/AWOS site is shown in Fig. 3.9. Fifty percent of the crashes were within 10 miles of an ASOS/AWOS site, and 70% were within 15 miles.

38 Figure 3.6: Total amount of wild fires from 2006 to 2010 for first six-months of year. Values represent wildfire events per 1000 square kilometers per county. Data from DOF wildfire counts

39 Figure 3.7: Total amount of wild fires from 2006 to 2010 for last six-months of year. Values represent wildfire events per 1000 square kilometers per county. Data from DOF wildfire counts

40 1200

1000

800

600

400 Number ofNumber Fog Events

200

0 1 2 3 4 5 6 7 8 9 10 Distance of Fog to Closest Fire (miles)

Figure 3.8: Distance in miles to the closest fire at the time of a fog event

Figure 3.9: Distribution of distances, in miles, from crash to the closest AWOS/ASOS station

41 3.2.2 Conditions Favorable for Fog Formation

It is useful to examine what atmospheric conditions are typically present when fog oc- curs and does not occur. It is important to note that the following findings were used as criterion in the model to forecast fog or no fog, conditional on the value of the meteorolog- ical parameters. Fig. 3.10 shows the distributions of dewpoint depression, wind speed, and temperature that are present during fog events and non-fog events. It should be noted that the data used in these graphs are taken at hourly intervals, and the quantity of no-fog events is much greater than fog events. It can be seen that fog typically forms when the dewpoint depression is 2◦F or less. Only 18% of fog events occurred when the dewpoint depression was above 2◦F. It is possible that these events were a result of advection fog, which can be common along coastal areas, or this could be the result of the temperature recording errors as explained in 2.2. Wind speed also shows importance during fog events, 90% of fog events occurred when the wind speed is less than 4-knots. In the bottom panel of Fig. 3.10, the relevance of temperature to fog formation is shown. The majority of fog events occurred between 50◦F and 70◦F, and never occurred when temperatures exceeded 80◦F.

Fig. 3.11 shows that fog events do not occur with any particular cloud cover amount.

24-hour rainfall amounts, however, shows that when fog occurred precipitation amounts were greater than 0.80 inches 17% of the time. Because precipitation is considered a relatively rare event on an hourly basis, these graphs are skewed toward zero rainfall regardless of fog or no-fog events. Therefore, the 17% is significant, as fog events only encompass 3% of the events in the data.

42 Cloud cover, at the time of a fog event did not show a discernable pattern when compared to no-fog events. This may be due to several factors; first, AWOS/ASOS stations measure cloud cover by sending a single detection beam straight upward. This can create many inconsistencies and poor readings. Second, when fog does occur, the sensor can register a false reading of low stratus, giving the impression that it is cloudy during fog events. Because of these inconsistencies, it is impossible to use measurements of cloud cover to diagnose a fog event at an AWOS/ASOS site.

To gain a better understanding of conditions preceding a fog event, the conditions 12- hours prior to a fog event, occurring between 11 and 12-UTC, were examined. Anytime fog occurred at any station between 11 and 12 UTC the temperature profile for the previous

12 hours was studied. The cooling rates for each hour was calculated for every station that reported fog. These cooling rates were then averaged. The same steps were then completed when fog did not occur between these times. The average cooling rates are plotted in

Fig. 3.12. It should be noted that the error bars seen in this figure, and the figures going forward, are overestimates of the confidence interval. This is because some of the events examined are not independent events, as meteorological conditions from one hour may affect other hours in the study. As seen in Fig. 3.12, there is a difference in cooling rates from 0-1

UTC between the two events. When fog occurs between 11 and 12-UTC, faster cooling rates between 0 and 1 UTC are observed. Fog is typically a cool season phenomena, and because the atmosphere during this period is much drier, cooling rates vary differently. Therefore, in order to accommodate for this, the cooling rates between the two events were looked at during the cool season only, which is defined as months between November and March.

43 This graphic is depicted in Fig. 3.13. The cooling rates between fog and no-fog events are

different, with fog events having faster cooling rates in the early night-time hours. However,

the cooling rates, instead of being 0.5 degrees apart from 0-1 UTC when all seasons are

included, is only 0.3 degrees apart during the cool season.

In addition to cooling rates, we must evaluate how other meteorological variables behave

during nights before a fog event occur. Fig. 3.14 shows the nocturnal behavior of dewpoint

depression, wind speed, and cloud cover when there is fog or no fog present between 11 and

12-UTC. From the figure, it is clear that in order for fog to form, the dewpoint depression

at 0-UTC usually must be less than 7◦F. Anything above this level greatly decreases the chance for fog to form, as the typical cooling rate would not allow for the atmosphere to reach saturation. In addition, wind speed showed a pattern between the two events. It is generally known that calm winds are necessary for fog formation. However, Fig. 3.14 shows that wind speeds are typically less than 5-kts, starting at 0-UTC, and then gradually decrease throughout the night. This suggests that wind speeds must remain light, not only at the time when fog is occurring but throughout the night as well. The bottom of Fig. 3.14, shows how cloud cover changes before fog occurs. As seen in the figure, there is no clear pattern distinguishing between fog and no-fog events. As mentioned before, this is most likely due to the fact that AWOS/ASOS stations are limited in their ability to accurately assess cloud coverage.

44 3.2.3 Crossover Temperature

The crossover temperature, first described by Baker et al. (2002), is a concept that more

accurately assesses the vertical measure of boundary layer humidity using surface based

observations. The vertical profile of relative humidity (hydrolapse) is of large importance

to fog formation. However it is largely ignored since the hydrolapse is not a standard

meteorological variable that is reported (Baker et al., 2002). In order to accommodate for this

atmospheric variable, the UPS forecasters developed a method, using surface observations to

infer what the vertical humidity profile looks like. This method uses the assumption that the

hydrolapse can be determined by looking at the dewpoint temperature trend throughout the

warmest part of the day. This is because, throughout the warmest part of the day, mixing

is typically maximized as the boundary layer reaches its peak height. If humidity levels

drop during this time, this implies that water vapor is being vertically transported into drier

air aloft, thus lowering the dewpoint. If however, during this time dewpoint levels do not

decrease at the surface, it can be assumed that moisture does not decrease with height (Baker

et al., 2002). Using this information, the UPS forecasters defined crossover temperature as

the minimum dewpoint observed during the warmest daytime hours. If the temperature

is forecasted to fall below this level, fog is forecasted, if the forecasted temperature is 3◦F below the crossover temperature, dense fog is forecasted. Fig. 3.15 illustrates this process.

The three example soundings show what the atmospheric profile would look like at noon the day before a fog event (top), at the warmest part of the day (middle), and at the time the fog occurs (bottom). It can be seen that at the warmest part of the day (middle

45 graph) the dewpoint profile is uniform throughout the planetary boundary layer (PBL) and

decreases at the surface. This is because mixing is maximized during this time. Using

this information, a forecaster would forecast fog only if the forecasted temperature is below

the dewpoint temperature at this time. The bottom graph shows the surface temperature

decreases below the dewpoint depression at the top of the PBL which is assumed to remain

constant. Therefore, assuming the crossover temperature theory is correct, fog should be

forming at the surface.

The humidity profile of the PBL is important to the formation of fog. It is important to

have saturation at the surface and throughout the PBL. If only the surface reaches saturation,

moisture could be moved and dispersed upward into the drier air aloft. Therefore, this

method essentially requires the forecast of saturation throughout the PBL, in which no

moisture could be displaced upward. This method assumes that air is not being mixed

vertically between the PBL and upper levels.

Fig. 3.16 shows the average temperature and dewpoint trends for fog events and no fog

events only during the cool season. To support Baker et al’s. (2002) theory, the average

nighttime minimum temperature must fall below the average minimal dewpoint experienced

during the warmest part of the day. The graph on the right of Fig. 3.16 shows the average

minimal temperature (58◦F) occurring at 12z is below the average minimal dewpoint (59◦F) experienced during the warmest part of climatological average cycle. The figure on the left shows the expected results when no fog occurred. It can be seen that the nighttime temper- ature does not go below the daytime minimum dewpoint. This preliminary analysis shows the validity of the UPS forecasters technique. This technique will be used in conjunction

46 Table 3.1: Climatological reference forecast example. Percentages represent what a forecast chance of fog would be dependent on month of year and station.

Station JAN JUL SEP DEC AAF 21.43% 4.61% 3.36% 15.89% MIA 1.94% 0.66% 1.34% 0.00% NPA 8.20% 0.70% 0.72% 9.93% PGD 9.15% 0.00% 2.74% 10.60% OBE 13.11% 0.82% 4.88% 11.48%

with the aforementioned variables to forecast fog.

3.2.4 Skill Scores

It is important both in this section and the following section to have a means of quan- tifiably evaluating a diagnosis or forecast, and also have an objective function to measure skill. Skill is defined as the improvement of a forecast over a reference forecast. Often the reference forecast can either be persistence, climatology, some combination of the two, or any other forecasting method. As mentioned previously, this study will use both the MOS fog product and climatology as the reference forecast. Table. 3.1 shows the fog forecast based on climatological averages for some of the stations and months used in this study. The percentages shown in the table give the chance of fog occurring for any day within a given month at a station specified. For example, during the month of January, at station AAF, the climatological forecasts would predict a 21.43% chance for fog every day within that month.

There are several ways of scoring a forecast and the skill involved. As mentioned earlier, the Brier and Heidke Skill Scores are the primary measure of skill in this study.

47 The Brier score (Brier, 1950; Brier and Allen, 1952; and Brier, 1956) can be expressed

as:

1 2 BS = (yk − ok) (3.1) N X

where yk is the forecast (or the reference forecast) and ok is the observation. A skill score can be developed which examines how much better the forecast is over some reference forecast, such as climatology.

The Briers skill score is represented as:

BS BSS =1 − (3.2) BS ref

A skill score attributed to Heidke (1926) examines the improvement over the reference measure, defined as the proportion correct that would be expected by random forecasts that are statistically independent of the observations.

2(ad − bc) HSS = (3.3) [(a + c)(c + d)+(a + b)(b + d)

Where variables a, b, c, and d are represented in the contingency table as seen in Fig. 2.3.

3.2.5 Detecting/Diagnosing Fog

First we will look at how feasible it is to detect the presence of fog, possessing only the standard meteorological information available (with the exception of visibility) at selected mesonet sites in Florida. As shown in Fig. 3.10, dewpoint depression and wind speed have a

48 large impact in differentiating between fog and no-fog events. Therefore, these two variables were used in order to detect fog. A program was created that would optimize a dewpoint depression and wind speed level, in which to say there is, or is no fog present. This level would be determined where the critical success index (CSI) would be highest. Fig. 3.17 shows the calculated skill scores defined in Fig. 2.3. CSI, PC, POD, and FAR were calculated for a given dewpoint depression and wind speed between the hours of 11 and 12-UTC. Fig. 3.17 shows the optimum level in which to say there is fog present. This occurs when the dewpoint depression is less than 1.8◦F and the wind speed is less than 4-kts. Using these levels we are able to detect approximately 55% of fog events. Since the number of no-fog events significantly outweighs the number of fog events, the FAR is always extremely high no matter what levels are chosen for dewpoint depression and wind speed.

3.2.6 Forecast

In the previous section, we described how reliable a zero-hour forecast would be. The forecast problem is then how well we can predict conditions that create fog 3, 6, or 10 hours in advance. We examine two types of forecast methods—one which uses only dewpoint depression, and the other which uses all meteorological variables shown in the climatology to affect fog formation (or fog/no fog forecasts).

Table 3.2 shows the probability that a fog event will occur based on the dew-point depression ten, six and three hours before. The probabilities were calculated based on the climatological averages using the five year data set. It can be seen that as the forecast

49 Table 3.2: Climatological probability of fog for a given lead time and dewpoint depression

Dewpoint Depression 10-hour 6 hour 3 hour 0 10.1% 28.4% 33.8% 0–1 8.1% 13.5% 14.3% 1–2 4.8% 8.1% 5.0% 2–3 4.0% 1.9% 2.9% 3–4 2.7% 1.6% 0.5% 4–5 3.8% 1.6% 0.3% > 5 1.4% 0.2% 0.0%

length decreases there is a higher certainty that a fog event will occur when the dew-point depression is less than 2◦F.

To make a forecast for fog, a nocturnal cooling rate is determined. In this case we have only looked at the winter months when most fog events occur. There could be seasonal and geographic adjustments which would increase the skill somewhat. For all stations we look at the cooling rates and how the temperature might be forecast at the time of fog formation.

This is shown for all stations during the winter months in Fig. 3.18. From this figure it can be seen that after 21:00 eastern-standard-time (EST), the cooling rate decreases linearly until 07:00 EST. In the same figure, we also examined how the dewpoint temperature rate changes throughout the night. As expected, the dewpoint remains relatively constant. It is obvious that if the overlying synoptic conditions change, this would have a large impact on the dewpoint and temperature. However, including these synoptic factors is not within the scope of this project. Future studies may wish to include this as part of a forecast model.

Eq. 3.4 shows the linear equation that represents the cooling rates throughout the night.

50 This equation was used to model the temperature at any time after 20:00 local time. The variable x represents the amount of minutes after 20:00 local time. In order to forecast temperature for 10, 6, and 3 hours in advance, this equation was used to predict the cooling rate at any time and was then subtracted from the existing temperature at the beginning of the forecast time. In order to verify how well the equation performed, it was run on the existing five year data set. A measure of its accuracy is shown in Fig. 3.19. This graph represents how accurate the equation was for a 10, 6, and 3-hour forecast. From this figure, it can be seen that for a 3-hour forecast, Eq. 3.4 yielded a temperature accuracy of less than 2◦F more than 70% of the time. The 6-hour forecast accurately predicted within 2◦F approximately 50% of the time, and the 10-hour forecast showed a wide range of errors, mostly failing to predict the temperature within 3◦F. This implies that accurately predicting fog 10 and 6-hours in advance is not possible using a model which employs a linear algorithm to forecast temperature. Synoptic as well as local conditions would have to be considered in order for the forecast to be more accurate.

y =0.00131 × x +0.8366 (3.4)

It should be noted that this method of forecasting dewpoint depression has inherent

flaws. One being that it does not consider any localized weather events or trends at any particular location. Additionally, since the equation incorporated both fog and no-fog events when developed, the dewpoint depression forecasted will be larger in magnitude than the actual on a fog night and vice versa when no fog occurs. Future studies may want to create

51 a temperature forecasting algorithm that is site and season specific.

Since the method above performs poorly for ten and six-hour forecasts, we limited the forecasting time to three-hours. Using the linear equation to forecast temperature, and assuming the dewpoint would remain constant throughout the night, a dewpoint depression can now be forecasted for 12-UTC. Using these forecasted dewpoint depressions, we now test the optimal forecasted dewpoint depression in which we will have the highest success when forecasting fog. Fig. 3.20 shows the skill scores for forecasting fog when the dewpoint depression is below a certain level. It can be seen from the Heidke skill score and critical success index, that the optimal dewpoint depression to always forecast fog below is 1.5◦F.

This level yields a CSI of 13.2% and a HSS of 20.3%. However, with a probability of detection of only 38%, these results are not ideal or reliable when forecasting fog across the state.

From these results, it is obvious that using a one variable forecasting system is not efficient or ideal. Therefore, we need to incorporate more meteorological variables into an algorithm/model that will give a binary forecast for fog formation. This model is discussed in the next section.

3.2.7 Characteristics of the Fog Forecast Model

Using the conditions that we know must exist for fog to form, an algorithm was built to forecast fog three-hours in the future. The model predicts fog or no fog based upon the meteorological conditions and patterns shown to impact fog formation. The variables included in the model are wind speed, temperature, dewpoint depression, cooling rate, cross- over temperature, and precipitation in the past 24-hours. The model incorporates different

52 threshold values that must be met in order for the model to predict fog for 12-UTC. In order for the three hour forecast to predict fog, different criterion must be met. The criterion are a combination of forecasted values and past atmospheric observations. They are listed below:

• If the visibility at the time of model initiation is less than 5/8ths mile always forecast fog no matter what the other conditions are below. If the visibility is greater than 5/8ths of a mile then the following conditions must be met to forecast fog:

1. The dewpoint depression at the time of model initiation must be equal to or less than 3◦F. 2. The cooling rate between 01-UTC and 02-UTC must be greater than 0.5◦F, unless the dewpoint depression is less than 3◦F at model initiation. 3. The dewpoint cooling rate between 01-UTC and 02-UTC must not exceed 1.0◦F 4. The wind speed at the time of model initiation must be less than 5-kts 5. The forecasted temperature at 11-UTC must be below the cross-over temperature. 6. The forecasted dewpoint depression at 11-UTC must be less than 2◦F

The criterion above were developed given the climatological trends found in section 3.2.2.

For example, as seen in Fig. 3.10, it was found that wind speed typically was always below

5-knots prior to a fog event. Therefore, in order for the model to predict fog, the recorded wind speeds must always be below 5-knots 12 hours prior to the forecasting time. The dewpoint depression and cooling rate thresholds followed the same application as the wind speed criterion. Subsequent adjustments (e.g. changing the dewpoint depression threshold, adjusting the cooling rate at 01-UTC, etc..) were then made to increase skill after combining the conditions. Fig. 3.21 below illustrates the criteria above in equation form. The user would have to subsitute past conditions as well as forecasted quantities into the equations to get a result. If the quantities satisfy either of the equations then that event is forecasted.

53 In order to determine if the fog forecasting methods developed in this study are useful,

they are compared to methods currently used. In this study the techniques developed will

be compared against the MOS fog product and climatology for the five year period. Since

MOS gives a binary forecast, it will be compared against the binary model in this study. The

skill scores for the MOS forecasts for 12-UTC are shown in Table 3.3 below: Additionally,

we will compare the climatology reference forecast and the model forecast directly using the

Brier and Heidke Skill Scores.

It is important to note that the model was not run on an independent data set when

calculating skill. The skill scores calculated are based upon model runs on the data set which

was used to create the model. Because of this, the skill scores calculated in the next section

are overestimates of what the overall skill of the model should be. Future studies may want

to evaluate the model using a different time period.

Table 3.3: Skill scores for MOS 12-UTC forecast for fog formation

Type Skill Score PC 97.03% POD 32.35% FAR 61.79% CSI 21.19%

3.2.8 Model Results and Comparison

Upon running the model on the five-year data set, the skill scores achieved are listed in

Table 3.4

54 Table 3.4: Skill scores for 3-hour model forecast for 12-UTC

Type Skill Score PC 71.24% POD 69.10% FAR 83.18% CSI 15.64% BIAS 4.11%

Table 3.5: Brier and Heidke skill score

Type Skill Score Brier 0.68 Heidke 0.24

Despite the model’s various inputs and conditions, derived from climatology, it is unable to outperform the MOS fog product. The only exception to this, however, was the probability of detection skill score, where the model far outperformed MOS in these cases. As expected, this resulted in a higher false alarm rate, resulting in a lower CSI. In order to see if the model shows skill against other forecasting methods, it is compared against climatology and the proportion of forecasts correct by chance using the Brier and Heidke Skill scores, respectively.

The Brier and Heidke skill scores, as discussed in 3.2.4 are able to quantify the improvement over a reference forecast. Both skill scores range from infinity to one, a skill score of 1 indicates a perfect score, 0 indicates no skill, and anything less than 0 means the reference forecast is better than the model. The Heidke and Brier skill scores are shown below.

55 Both the Heidke and Brier skill scores are greater than zero, meaning the model is an improvement over a climatological forecast or a forecast which is correct by pure chance.

The model developed is unable to take into account synoptic patterns or abrupt weather changes. In the model’s current form, it uses a single point as the source for the all the inputs. Therefore a neighboring front or weather pattern moving into the forecast area would increase the error of the model. Additionally, the model uses climatological cooling rates to forecast temperature during the early morning hours. If the temperature variation during these hours differs greatly from climatology, the fog forecast would have a greater chance of being incorrect. Since frontal systems are a cool season phenomena for Florida, forecast errors are expected to be greater during this time compared to summer months, which has less drastic temperature and humidity changes. The model is also unable to forecast for advection fog. It could be assumed that many of the fog events which occur under higher wind speeds and dewpoint depression levels are advection fog events. Due to the different meteorological conditions this type of fog occurs under, a different model would need to be developed to forecast this event.

56 Figure 3.10: Dewpoint depression, temperature and wind speed conditions present when AWOS and ASOS reports fog (left) and no fog (right). Amounts are tallied based on all METAR reports and are divided by total amount of each event

57 Figure 3.11: Precipitation and cloud cover conditions present when AWOS and ASOS reports fog (left) and no fog (right). Amounts are tallied based on all METAR reports and are then divided by total amount of each event

58 Figure 3.12: Average cooling rates examined for fog and no fog events for all seasons and all stations. These events are defined as to whether fog did, or did not occur, between 11 and 12 UTC. Error bars represent a 99% confidence interval.

Figure 3.13: Average cooling rates examined for fog and no fog events during the cool season (i.e. November to March) only. These events are defined as to whether fog did, or did not occur, between 11 and 12 UTC. Error bars represent a 99% confidence interval.

59 Figure 3.14: Conditions pre-existing fog and no-fog events occurring between 11 and 12-UTC. Top- Nocturnal dewpoint depression changes, Middle- Nocturnal wind speed changes, Bottom- Nocturnal cloud cover percentage changes. Error bars represent 99% confidence intervals

60 Figure 3.15: Example atmospheric soundings iillustrating the theory of the cross over temperature. Top: Dewpoint (green) and temperature (black) profile at noon. Middle: Dewpoint (green) and temperature (black) profile at warmest part of day. Bottom: Dewpoint (green) and temperature (black) profile at the time fog occurs.

61 Figure 3.16: Average variations in temperature and dewpoint when fog did occur (right), and did not occur (left), for all stations, only for cool season. Error bars represent a 99% confidence interval.

62 Figure 3.17: Skill scores for detecting fog at the time a fog event is reported using only wind speed and dewpoint depression. Percent Correct (PC), Probability of Detection (POD), False Alarm Rate (FAR), and Critical Success Index (CSI) are plotted. PC, POD, and FAR are plotted on primary Y-axis. CSI is plotted on secondary Y-axis

Figure 3.18: Nocturnal hourly cooling rates for temperature and dewpoint, averaged over all stations. Equations displayed are least-squares linear regression. Error bars represent a 99% confidence interval.

63 Figure 3.19: Forecast temperature error for 12-UTC using a 10, 6, and 3-hour forecast. Results represent the percentage of time the forecasting method had a specific temperature error

Figure 3.20: Skill scores for forecasting fog, 3-hours in advance, when always fore- casting fog below a forecasted dewpoint depression

64 Figure 3.21: Equation used in model to determine fog forecast. Top equation represents the conditions that must be met in order for fog to not be forecasted. Bottom equation represents conditions that must be met for fog to be forecasted

65 CHAPTER 4

CONCLUSIONS

Forecasting fog accurately in time and space using any method other than numerical or statistical modeling is challenging. Fog occurs under many different meteorological conditions and variables, however, there is not a high correlation between the occurrence of fog and any of the meteorological variables it is known to be dependent on. Furthermore, the data available via AWOS/ASOS stations are not spatially sufficient or accurate enough to give a clear picture of what conditions fog forms under. Therefore, creating a model or algorithm based on climatology is not ideal, and a forecaster should use MOS whenever possible.

Despite MOS far outperforming the developed model, the model was able to outperform climatology easily. It should be noted that the success of the model over climatology was due in large part to how poor a forecasting tool climatology is. Regardless, the model shows value in this comparison. Additionally, due to the lack of computing power needed to run the model, it can be used to forecast for any time in the future, and only needs a relatively small amount of input data. This is valuable because MOS can take up to four hours to run, and only makes forecasts in three hour intervals. During the morning hours, when fog occurs most often, this can be problematic as MOS only gives forecasts for fog at 09z and

12z. As seen in Figure 1.4, fog most often occurs between these times. The model developed could provide a means of forecasting fog at the times when MOS does not.

In conclusion, the model developed is an easy to implement method to forecast fog.

66 Although it is not as accurate as MOS, the model proves to be much more conservative

(forecasts fog more frequently) than MOS, as seen by the POD figure. Thus, in terms of public safety, the model serves the purpose of warning the public that there could be fog in the forecast period.

67 REFERENCES

Abdel-Aty, M.A., M. Ahmed, J. L. Lee, Q. Shi, M. Abuzwidah, 2012: Synthesis of visibility detection systems. Final Report Contract from Florida DOT to UCF Orlando, Florida, Contract BDK78 — Task Work order 87711. 134 pp.

Achtemeier, G. L., 2003: On the origins of “super fog” - a combination of smoke and wa- ter vapor that produces zero visibility over roadways. Second International Wildfire Fire Ecology and Fire Management Congress and Fifth Symposium on Fire and Forest Meteorology, Athens, GA J8.9.

American Meteorological Society., 2000: Glossary of Meteorology, 2 Allen Press. 638pp.

Baars, J. A., M. Witiw, and A. Al-Habash, 2009: Determining fog type in the Los Angeles basin using Historic surface observation data. Poroc. 16th Conf. on Probability and Statistics in the Atoms. Sci., Long Beach, CA.

Baker, R. J. J. Cramer, and J. Peters, 2002: Radiation fog: UPS Airlines conceptual models and forecast methods. 10th Conf. on Aviation, Range and Aerospace Meteorology Amer. Meteor. Soc. Portland, OR 154-159.

Ballard, S., B. Golding, and R Smith, 1991: Mesoscale model experimental forecasts of the hair of northeast Scotland. Mon. Wea. Rev., 119, 2107-2123.

Bendix, J., B. Thies, J. Cermak, and Nauss, 2005: Ground fog detection form space based on Modis daytime data - a feasibility study. Wea. and Forecasting, 20 989-1005.

Brier, G. W., 1950: Verification of forecasts expressed in terms of probability. Mon. Wea. Rev. 78(1). 1-3.

Brier, G. W., 1957: Effect of errors in estimating probabilities on the usefulness of probability forecasts. Bull. Amer. Meteor. Soc., 38, 76-78.

Brier, G. W, and R. A. Allen, 1952: Verification of weather forecasts. Compendium of Meteorology. T.F. Malone, Ed. Amer. Meteor. Soc. 841-848.

Byers, H. B., 1944: General Meteorology 1st ed., McGraw-Hill, 645 pp.

Byers, H. B., 1959: General Meteorology 3rd ed., McGraw-Hill, 481 pp.

Cox, R. E., 2007: Applying fog forecasting techniques using AWIPS and the internet.

68 Croft, P. J., 1995: Forecasting significant fog in southern Alabama. National Weather Digest 19. 10-16.

Croft, P. J., R. L. Pfost, J. M. Medlin, and G. A Johnson, 1997: Fog forecasting for the Southern Region: A conceptual model approach. Wea. Forecasting 12 545-556.

Duynkerke, P.G., 1990: A comparison of model simulations with detailed observations. Mon. Wea. Rev. 119, 119-324.

Ellrod, G. P., 2002: Estimation of low cloud base heights at night from satellite infrared and surface temperature data. National Weather Digest 26, 39-44.

Ellrod, G.P., 2007: Event bases climatology and typology of fog in the New York City region. Wea. Forecasting.,10, 606-619.

Gultepe, I., M. Muller, and Z. Boybeyi, 2006a: A new visibility parameter for warm-fog applications in numerical weather prediction models. j. Appl. Meteor. Climatol., textbf45, 1469-1480.

Gultepe, I., M. Pagowski, and J. Reid, 2006b: A satellite-based fog detection scheme using screen air temperature. Wea. Forecasting, 22, 444-456.

Gultepe, I., G. Pearson, J. a. Milbrandt, B. Hansen, S. Platnick, P. Taylor, M. Gordon, J.P. Oakley, and S. G. Cober, 2008: The fog remote sensing and modeling field project. Bulletin American Meteor. Soc., 90, 341-359.

Gurka, J. J., 1978: The use of enhanced vidible imagery for prediction the time of fog dissipation. Preprints, Conference on Weather Forecasting and Analysis and Aviation Meteorology, St.Louis, MO Amer. Meteor. Soc. 54-57.

Heidke, P., 1926: Berechnung des Erfolges und der G¨ute der Windst¨arkevorhersagen im Sturmwarnungsdienst. Geor. Ann., Stockh. 8, 310-349.

Herzegh, P. H., R. L. Bankert, B. K. Hansen, M. Tryhane, and G. Wiener, 2004: Recent progress in the development of automated analysis and forecast products for ceiling and visibility conditions. 20th Conf. on Interactive Information Processing Systems for Meteorology, Oceanography, and Hydrology, Seattle, WA.

Houghton, D. D., 1985: Handbook of Meteorology, 1st ed., John Wiley and Sons, 1461 pp.

Jarvis, M. R., H. E. Fuelberg, P. H. Ruscher and A. I . Watson, 2001: Improving forecasts of dense fog over north Florida. 18th Conf. Wea. Analysis and Flrecasting Fort Lauderdale, FL. 218-222.

69 Johnson, G. a. and J. Grashel, 1992: Sea fog and stratus: A major aviation and marine hazard in the northern Gulf of Mexico Preprints, Symp. on Weather Forecasting Atlanta, GA, 1992.

Kolb, L. L., and M. M. Goodmanson, 1945: Upslope weather along the inland Alaska air route. J. of Meteor. 1, 98-108.

Lavdas, L.; Achtemeier, G.L. 1995. A fog and smoke risk index for estimating roadway visibility hazard. National Weather Digest. 20. 26-33.

Leipper, D. L.,1995: Fog forecasting objectively in the California Coastal area using Libs. Wea. and Forecasting,10, 741-462

Lewis, J.m 2003: Sea fog off the California Coast: Viewed in the context of transient weather systems. J. of Geo. Res. 108. D15.

Miller, M. D., M. Masbou, A Bott, and Z Janjic, 2005: Fog prediction in a 3d model with parameterized microphysics Preprints, WWRP INt. Symp. on Nowcasting and very short range forecasting Toulouse, France. Meteo-France page 6.26

Paxton, C. H., J. M. Collins, R. J. Davis, N. Petro, and A. Williams, 2009: Central Florida smoke and fog–Carnage on Interstate-4 Abstract for the Submission to The Symposium on Urban High Impact Weather Phoenix, AZ. National Weather Service P1.11

Pisano P. A., L. C. Goodwin, and M. A. Rossetti 2008: U.S. highway crashes in adverse road weather conditions 88th American Meteorological Society Annual Meeting; FHWA, Noblis and RITA Washington, D.C. U.S. DOT Federal Highway Administration P1.15

Ratzer, M. A., 1988: Middle Atmosphere Dynamics, 2, of Noaa Technical Service Publica- tions NWS CR. Scientific Services Division. Central Region. U. S. Dept. of Commerce, National Oceanic and Atmospheric Administration. National Weather Service.

Stull, R. B., 1999: Meteorology for Scientists and Engineers 2nd ed., Brooks/Cole, 502 pp.

Stone, R. G., 1936: Fog in the United States and Adjacent Regions American Geographical Society, 26. 111-134

Tardiff, R. and R. M. Rasmussen, 2007: Event based climatology and typology of fog in the New York City region. Appl. Meteor. and Climo, 46, 1141-1168.

Taylor, G., 1917: The formation of fog and mist. Royal Meteorological Society. 43, 241-268.

70 Wang, J.L, Y.H. Zhang, M. Shao, X.L. Liu, L.M. Zeng, C.L. Cheng, X.F. Xu, 2006: Quanti- tative relationship between visibility and mass concentrations of PM2.5 in Beijing. J. of Env. Sciences. 18. 475-481

Westcott, N. E., 2006: Some aspects of dense fog in the midwestern United States. Wea. and Forecasting, 22. 457-465.

Willett, H. C., 1928: Fog and haze, their causes, distribution and forecasting. Mon. Wea. Rev., 56, 457-468.

Zhou B., J. Du, B. Ferrier, J. McQueen, and G. Dimego, 2007: Numerical forecasts of fog - central solutions. 18th Conference on Numerical Weather Prediction, Park City UT. Amer. Meteor. Soc.

71 BIOGRAPHICAL SKETCH

Justin Rivard was born in Providence, RI on December 14th 1989 to his dear parents Denis and Darlene. In 2007, he attended Plymouth State University to study meteorolog,y and in May, 2011 graduated Summa Cum Laude with a Bachelor of Science degree. In the fall of 2011, Justin moved to Tallahassee, Florida to pursue his Masters degree at Florida State

University.

While an undergraduate, Justin worked as an intern at the National Oceanic and Atmo- spheric Administration and received the Hollings scholarship in his junior year. In his second year of graduate school, Justin began working as a Meteorologist for the Florida Department of Environmental Protection, specializing in air quality modeling and analysis.

72