You are on page 1of 64

Florida State University

Libraries
Electronic Theses, Treatises and Dissertations The Graduate School

2021

Estimating Return Periods of Extreme


Tropical Cyclone Winds Accounting for
Effective Surface Roughness at High
Resolution
Robert Walker Whitield

Follow this and additional works at the DigiNole: FSU's Digital Repository. For more information, please contact lib-ir@fsu.edu
.
FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

ESTIMATING RETURN PERIODS OF

EXTREME TROPICAL CYCLONE WINDS ACCOUNTING FOR

EFFECTIVE SURFACE ROUGHNESS AT HIGH RESOLUTION

By

ROBERT WALKER WHITFIELD

A Thesis submitted to the


Department of Earth, Ocean and Atmospheric Science
in partial fulfillment of the
requirements for the degree of
Master of Science

2021
Robert Walker Whitfield defended this thesis on March 18, 2021.
The members of the supervisory committee were:

Mark A. Bourassa
Professor Directing Thesis

Steven D. Cocke
Committee Member

Robert E. Hart
Committee Member

Zhaohua Wu
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
ACKNOWLEDGMENTS

Many, many thanks to my primary advisor, Dr. Mark Bourassa, for his valuable guidance
throughout this project and time spent with thorough revisions of this manuscript. Another huge
thank you to Dr. Steve Cocke, for truly helping shape the project to its final form today and
providing the data and statistical knowledge from which much of the analysis originates. I’d also
like to thank Dr. Robert Hart and Dr. Zhaohua Wu for their important insight as committee
members.

One I cannot emphasize enough, an incredible thank you to my peers Charley Fite and
Evan Jones for their time and knowledge while guiding me through portions of what turned into
a rather extensive coding undertaking. Lastly, as I can always count on, thank you to my friends,
old and new, and family who have continually supported me and kept me sane throughout my
journey in higher education.

At no point, particularly while working entirely remotely during a global pandemic, did
this project take a linear path towards completion, and so it is not without any of the
aforementioned people that this project would have been a success.

iii
TABLE OF CONTENTS

List of Figures.................................................................................................................................. v
Abstract......................................................................................................................................... vii

1. INTRODUCTION ....................................................................................................................... 1

2. PRELIMINARY ASSESSMENT AND PROCESSING OF HURDAT EXTENDED


BESTTRACK DATA ...................................................................................................................... 4

3. A HIGH-RESOLUTION ASSESSMENT OF RETURN PERIODS BASED ON THE


FLORIDA PUBLIC LOSS HURRICANE MODEL .................................................................... 16

4. RETURN PERIODS NEAR COASTAL LOCATIONS .......................................................... 27

5. LAND AND LAND USAGE IMPACTS ON RETURN PERIOD ............................................ 33

6. FUTURE CONSIDERATIONS AND CONCLUDING REMARKS ...................................... 46

References ..................................................................................................................................... 52

Biographical Sketch....................................................................................................................... 55

iv
LIST OF FIGURES

Figure 1: Representation of quadrant and wind radii-separated wind fields sourced from
HURDAT BestTrack Extended Dataset as plotted by Mouch et al. (2019) of Hurricane Irma ...... 5

Figure 2: Plotted HURDAT Extended BestTrack maximum TC-related winds (mph) at discrete
time steps in the Northern Atlantic basin in 2005 ........................................................................... 6

Figure 3: HURDAT Extended BestTrack maximum TC-related winds (mph) in the Northern
Atlantic basin in 2005, smoothed through linear interpolation ....................................................... 7

Figure 4: Plotted probability distribution functions for the observed wind speeds and four fitted
extreme value distributions (3-Parameter Weibull, 3-Parameter Gamma, Gumbel, and
Generalized Pareto Distributions) for HURDAT BestTrack Extended dataset point with one of
the largest available number of recorded maximum TC windspeeds in the domain (n=20) ........... 9

Figure 5: Plotted TRP (units of years) for 𝑥 = 121𝑘𝑝ℎ (75𝑚𝑝ℎ) within the Northern Atlantic
from observed EBTRK data fitted to a Generalized Pareto distribution ....................................... 10

Figure 6: Gridded plot of maximum recorded wind speeds (mph) from HURDAT BestTrack
Extended Dataset over the southeastern United States in 2005 .................................................... 11

Figure 7: Plotted return periods by Keim et al. (2007) using historical tracks and assumed
‘average’ wind field radii correlated to storm intensities .............................................................. 13

Figure 8: A continuous, gridded analysis of expected return period values given equivalent
Category 1 strength TC winds across multiple ocean basins at 10 km resolution by Bloemendaal
et al. (2020) .................................................................................................................................... 14

Figure 9: Vertical wind profile as seen over the transition from smooth to rough surface features,
leading to surface wind speed reduction as a result of the modified flow within the IBL (dotted
line). The IBL grows in height with time over the rough surface until equilibrium is reached,
while the undisturbed PBL above the IBL remains at similar magnitude to marine EL conditions
(Raupach and Thom, 1981) ........................................................................................................... 19

Figure 10: Effects of varying Holland B parameter values on calculated pressure and wind speed
profiles as a function of radial distance from the TC center (Holland, 1980) ............................... 21

Figure 11: Selected locations of varying land use (stars) for the gridded TRP analysis of the
FPLHM. The gridded analysis is taken at one-kilometer resolution and each discrete location
encompasses one of these cells. These marked locations are overlaid on a land use reference map
of the Tampa/Clearwater area developed by Russell et al. (2013) ................................................ 24

Figure 12: TC tracks through Tampa/Clearwater, Florida from the mid 1850’s-present by
NOAA (2021) ................................................................................................................................ 31

v
Figure 13: Modeled wind swath from FPLHM dataset of a south to north moving TC along the
west coast of Florida over Tampa/Clearwater ............................................................................... 34

Figure 14: Return periods for 121 kph (75 mph) TC winds around Tampa/Clearwater as
calculated by modeled FPLHM at 1 km resolution (left) and observed HURDAT BestTrack
Extended at 27.75 km resolution ................................................................................................... 35

Figure 15: Expected TC wind speeds at a logarithmic TRP scale in Tampa/Clearwater, Florida at
a 80 kph/50 mph threshold ............................................................................................................ 36

Figure 16: GPD fit to PDF of modeled FPLHM winds at select locations around Tampa,
Clearwater...................................................................................................................................... 37

Figure 17: Expected winds for a continuous TRP distribution as fit to GPD for six locations
around Tampa/Clearwater. Empirical rank-calculated return periods of FPLHM data plotted as
blue dots; GPD fit plotted as black line with 95% confidence intervals as calculated by the Delta
method plotted as dotted red lines ................................................................................................. 41

Figure 18: GPD calculated at 𝑥! = 50; 95% confidence intervals (red) within ±1mph (top) and
GPD calculated at 𝑥! = 85; 95% confidence intervals within ±18 mph (bottom) ...................... 42

Figure 19: Land use map around Tampa/Clearwater by Russell et al. (2013) ............................. 51

vi
ABSTRACT

In the United States, tropical cyclones (TCs) were the leading cause of natural disaster-
related fatalities and Consumer Price Index (CPI)-adjusted costs averaged over the years 1980-
2017. Accurately assessing TC-induced extreme winds is crucial to proper risk management
related to mitigation of these losses. Several studies attempt to calculate return periods (TRP), or
the statistically estimated average time between events of a given magnitude, for extreme TC
winds at point locations or along coastlines. This study seeks to further previous work by
addressing concerns related to spatial and temporal resolutions, in addition to the consideration
of TC winds as a function of surface roughness and fetch over land. We begin by analyzing
HURDAT BestTrack Extended (1988-2017) data (EBTRK) at 0.25-degree (~27 km) resolution,
assigning a maximum observed windspeed value per grid cell per year, to which several extreme
value distributions (EVDs) are fit to the probability distribution function (PDF). The Generalized
Pareto Distribution (GPD) is found to have the best fit to the EBTRK data, where fitting
parameters are calculated based on these observations using the method of moments. We find the
EBTRK dataset, while more continuous and higher resolution than previous assessments, lacks
sufficient observations for proper fit to an EVD and it appears to not fully consider the effect of
surface roughness on TC wind speed, likely because of great limitations on spatial information in
the data set. To address these issues, we use 59,000 years of modeled TC winds produced by the
Florida Public Loss Hurricane Model (FPLHM), which considers the effects of effective surface
roughness on upstream winds on a fine scale (~90m), output at locations of interest roughly
every kilometer. The study domain is narrowed down to the Tampa/Clearwater area of Florida
for its various terrain and coastline features, from which six locations of varying terrain/coastline
proximity are chosen. We reimplement the EVD fitting procedure to the FPLHM data, again
finding the GPD to be the best fit, and extrapolating out TRP values for the most extreme TC
winds at each location. Empirically calculated TRP values are also plotted for the entire domain to
gain a sense of those associated patterns. We find TRP and subsequent expected winds can vary
dramatically depending on effective surface roughness over distances at least as short as ~1 km
(the finest resolution of the saved modeled data), while distance inland independent of TC
weakening plays a lesser of a role after the first few kilometers. The results indicate a need for
more accurate future TRP analyses that are both at a fine scale and incorporate effective surface

vii
roughness, which also points to the weakness in using historical data in areas with recent changes
in land use.

viii
CHAPTER 1
INTRODUCTION

Of the natural disasters to impact the United States from 1980 to 2017, tropical cyclones
(TCs) were the second-most frequent type to exceed a cost of one billion dollars, only behind
severe storm events, yet their total cost was over three times that of the more frequent severe
(non TC) storms at $850.5 billion in Consumer Price Index-adjusted losses (NOAA, 2021).
While TCs may not be as numerous as other severe weather events in the U.S., it is clear when
they do occur, they result in a significantly higher economic toll. Moreover, TC-related fatalities
topped the list at over 3,400 out of the ~10,000 total deaths attributed to U.S. natural disasters in
the same time span. TC-induced extreme winds contribute both directly to this loss of life and
property, while also indirectly influencing other destructive factors such as storm surge and wave
heights (Robertson et al., 2007). The speed of surface winds, defined as 10m above the surface,
is influenced by more than the size of the storm, features within the storm such as the eyewall,
and the wind speeds aloft. Surface roughness plays an important role in the efficiency at which
momentum from aloft is mixed down to the surface. Therefore, wind intensity is highly
influenced by variations of terrain (Giammanco et al., 2012) and the distance from terrain
features, known as ‘fetch’, particularly as it is observed on and offshore (Wong & Chan, 2007).
In order to mitigate future costs to life and infrastructure and do so in a manner that considers the
spatial variability of risk, it is important to consider TC-induced extreme winds within a risk
analysis assessment that accurately takes into account boundary layer variations on fine spatial
scales. It is critical to understand the risk of TC-induced extreme winds due to their wide variety
of impacts and range of intensities through interactions with on and offshore surface features.

One diagnostic to quantify the risk of extreme winds is through a return period (TRP),
which can be described as the average time between events of the same magnitude that occur at a
given location. Mathematically, it is the inverse of one minus the cumulative distribution
function (CDF), or probability per unit time, of an event equivalent to or exceeding a particular
value from a certain distribution. A number of previous studies (Emanuel & Jagger, 2010;
Bloemendaal et al., 2020) have shown the estimated TRP for the most extreme TC winds can be
on the order of centuries. Researchers continuously struggle with the accurate calculation of

1
longer return periods because accurate records of TC tracks and wind fields span a time scale of
only a few decades; much less time than has passed to appropriately calculate longer RPs
directly from an observational record. In addition, inherent in the study of extremes is a low
number of extreme observations due to their statistical improbability. Various assessments
attempt to mend this problem through the utilization of modeled storms and/or an assumed fitted
distribution (Keim et al., 2007; Hoque et al., 2017). Of the recent assessments to generate RP
analyses for various geographical regions, some implement a similar technique to Keim et al.
(2007) where they use historical data from the National Hurricane center to track and model TCs
over a select number of coastal points of interest. These are discrete locations and lie tens of
kilometers apart from one another. Another recent assessment addressed both point locations in
addition to a 10-kilometer resolution TRP analysis of a number of ocean basins and their
surrounding land masses (Bloemendaal et al., 2020). They synthetically populate 10,000 years of
TC wind data by sampling environmental averages sourced through IBTrACS and empirically
calculate the return period from the data. What is not considered, however, are the effects of
surface roughness on TC winds near landfall or features of a higher resolution analysis.

This study seeks to calculate values of TRP for the higher intensity TC winds represented
by the tail of a generalized extreme-value (GEV) distribution, and to examine spatial changes in
risk on fine scales. These spatial changes in risk will be examined as a function of distance
inland and in conjunction with surface land usage. While some previous studies have examined
TRP on a point-by-point or generalized regional basis, they do not consider spatial differences in
TRP associated with fetch and surface roughness on fine scales. Analysis of HURDAT data
(1988-2017) with the use of fitted EVDs to attempt extrapolation of TRP values (Chapter 2),
demonstrates that both the resolution, short temporal range, and generalization of wind fields that
do not account for surface friction limit the utility in accurate TRP projections. Analyzing TRP at a
higher resolution (Chapters 3 to 5) while considering on and offshore surface roughness will
accomplish two goals: 1) account for surface roughness, particularly inland from the immediate
coastline, that can cause significant variation in the surface wind speed and subsequent TRP
depending on type of land usage, and 2) ensure the risk analysis is not overgeneralized within too
many areas that may have significant changes in surface roughness over a relatively small
distance; a factor that is particularly relevant directly along the coastline.

2
We will explore the viability of such an analysis through the use of the high-resolution
Florida Public Loss Hurricane Model (FPLHM). The FPLHM is a catastrophe model that
stochastically models 59,000 simulated years of TC wind events within the domain of Florida. It
is used to quantify risk of TC wind damages, which can then be examined through a number of
lenses such as a financial or engineering perspective. The model input sources from historical
data to fit distributions of previous observations while providing sufficient modeled surface wind
data for more robust statistical analysis. The study area is narrowed down to the
Tampa/Clearwater area for its diverse terrain and complex coastline features, then TRP analyses
are performed at six chosen locations with varying terrain/coastline proximity. Results show
much higher variability across short distances than previous assessments for expected wind
speeds at varying TRP values and we explore the causes from a meteorological and geographical
perspective.

3
CHAPTER 2
PRELIMINARY ASSESSMENT AND PROCESSING OF HURDAT
EXTENDED BESTTRACK DATA

To begin this study, observational sustained TC wind data sourced through the
HURDAT BestTrack Extended (Colorado State Regional and Mesoscale Meteorology Branch,
1988-2017) dataset (EBTRK) are used to assess the areas appropriate for TRP analysis. This
dataset contains information on storm name, storm dates, locations (latitude/longitude), pressure,
and various wind measurements as sourced through observational and forecast techniques at 6-
hour intervals. HURDAT is the original data containing estimates of these variables by the
National Hurricane Center (NHC); however, this original dataset does not contain any
information regarding storm size, which is needed for wind field estimates. For operational
forecasting since 1988, the NHC puts out estimates of wind radii for 34, 50, and 64 kt for each
quadrant of the storm. The Extended Best Track (EBTRK) combines the original HURDAT and
operational forecast data into a single dataset, which provides us at the time of this study 30
years of estimated TC surface wind data.

The maximum winds are determined for each grid cell with TC data for each of these 30
years and are plotted (Fig. 2) for the Northern Atlantic basin, including over land masses, chosen
at a grid resolution of 0.25 degrees (27.75 km, 17.25 miles) for areas around 27-degrees north.
This resolution allows for best possibility of capturing maximum winds within eyewall features
that have an average diameter of 32-64 km (20-40 miles) and are located near the center of the
TC (NOAA, 2021), while also retaining enough detail as to not overgeneralize any broad region.
The value of a fine resolution map for analysis is limited by several considerations: the lack of
temporal resolution, the limited spatial resolution of the wind speed field for each time, and
precision of the EBTRK data. For example, the smallest resolution at which EBTRK records
wind radii is at 5-nautical mile intervals (9.25 km, 5.75 mi), which while finer resolution than
previous gridded TRP analyses, is still far too large for meaningful localized TRP analysis as will
be demonstrated in Section 2.3. The EBTRK wind radii data is separated by storm quadrant,
those being the furthest extent of the maximum wind field, 64 kt (118 kph) wind field, 50 kt (93
kph) wind field, and 34 kt (63 kph) wind field for the northwest, southwest, northeast, and
southeast quadrants respectively, as can be seen in Figure 1. These data are recorded at 6-hour

4
time intervals with records of landfalling storms being narrowed down to the hour for TCs
occurring 1991 and onward. While these later storms have more continuous data at times and
locations near landfall, the temporal scale is too large for most storm’s wind fields to remain
continuous when plotted within the gridded analysis due to insufficient temporal sampling for
this application.

Figure 1: Representation of quadrant and wind radii-separated wind fields sourced from
HURDAT BestTrack Extended Dataset as plotted by Mouch et al. (2019) of Hurricane Irma.

In the event of a TC with rapid forward motion, the plot of TC-related winds can have
gaps in along the storm track between recorded time steps (Fig. 2). To mend this, wind fields of
TC wind radii are assumed to be continuous and vary smoothly in space and time. The storm
location and wind radii between EBTRK records are linearly interpolated in 30-min time steps.
5
For each of our quarter degree grid cells, wherever the extent of a given wind radii passes over,
that grid cell retains the measurement of only the maximum wind speed recorded at that
particular location and year (e.g., Fig. 3). This holds true as well for the case of multiple TCs
passing over a single location, so that only the greatest wind speed observed at that grid cell is
retained. This process is done for each of the 30 years of available data. While this approach
greatly reduces the problem of temporal sampling, it does not address the coarse wind speed
scale depending only on the maximum winds and the values for which wind radii are provided
(Fig. 1). These details could be improved by assuming a model of the wind speed dependence,
but this improvement was not considered because of the overwhelming problems that will be
demonstrated below.

Figure 2: Plotted HURDAT Extended BestTrack maximum TC-related winds (mph) at discrete
time steps in the Northern Atlantic basin in 2005.

6
Figure 3: HURDAT Extended BestTrack maximum TC-related winds (mph) in the Northern
Atlantic basin in 2005, smoothed through linear interpolation.

Methodology for Calculating Return Periods of HURDAT Data

Once the maximum TC winds have been assigned to a year, gridded, and smoothed, we
can calculate the TRP for any selected windspeed threshold. Values of TRP can be calculated
through a method of moments approach (Christopeit, 1994) where fitting parameters of a well-
fitted distribution are derived from the data itself. Once these fitting parameters are established,
the probability distribution function (PDF), cumulative distribution function (CDF), and
exceedance probabilities can be calculated, which results in a continuous distribution of values
from which RPs can be calculated. The uncertainty in the fitting parameters can be estimated and
propagated through these equations to provide estimates of uncertainty. This approach gives us
the ability to extrapolate that distribution estimate TRP at the most extreme tail values; however,
the uncertainty can be quite large when there are insufficient data, as is often the case in this
example.

In order to calculate TRP through the method of moments, a fitted distribution must be
found that is well-suited to the data. In the case of extremes, we preliminarily examine four

7
known extreme value distributions (EVDs) that could potentially fit the TC wind data: 3-
Parameter Weibull, 3-Parameter Gamma, Gumbel, and Generalized Pareto distributions. For
example, in order to get the best possible fit for this relatively limited dataset, a point is chosen
for a location at which there is the one of the greatest number of recorded windspeed values, that
being a point in the middle of the Northern Atlantic about 400 miles due north of the Dominican
Republic where 𝑛 = 20 over the 30 years of data. The maximum TC wind speed value for each
of these 30 years at this grid cell are collected data and plotted as a PDF for comparison of fit to
the PDFs of EVDs (Fig. 4). By comparing 𝑅" values between the histogram and the EVDs, we
find (and it is visually apparent) the Generalized Pareto distribution (GPD), while not a perfect
fit given the still limited data, is the best fit to the PDF of observations. Assuming a GPD
distribution is valid at all points, values of TRP are determined for the entire domain, with the
fitting parameters estimated for each grid cell. The equations to calculate the GPD fitting
parameters using the method of moments are as follows:

)
#$%&' (
𝜎 = 𝑠𝑐𝑎𝑙𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 = 0.5=𝑋 − 𝑥! @ A * ) +,
B (1)

)
#$%&' (
𝛾 = 𝑠ℎ𝑎𝑝𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 = 0.5 A * ) %,
B (2)

And the equations to calculate the PDF, CDF, and RP for a given wind speed 𝑤 are:

1
%, ,%-(&%&' ) 231
𝜎 N O , 𝛾 ≠ 0
𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝐷𝑒𝑛𝑠𝑖𝑡𝑦 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 (𝑃𝐷𝐹) = M 0 (3)
%(&%& )
𝜎 %, 𝑒𝑥𝑝 N 0 ' O , 𝛾 = 0

1
,%-(&%&' ) 2
1− N O , 𝛾 ≠ 0
𝐶𝑢𝑚𝑢𝑙𝑎𝑡𝑖𝑣𝑒 𝐷𝑒𝑛𝑠𝑖𝑡𝑦 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 (𝐶𝐷𝐹) = M 0 (4)
%(&%&' )
1 − 𝑒𝑥𝑝 N 0
O , 𝛾 = 0

,
𝑅𝑒𝑡𝑢𝑟𝑛 𝑃𝑒𝑟𝑖𝑜𝑑 (𝑅𝑃) = ,%456 (5)

8
Where:

𝑋 = mean of the full time series of annual maximum TC windspeeds at a given grid cell

𝑥! = location parameter or wind speed threshold at which we begin fitting the GPD

𝑆 " = variance of the full time series of annual maximum TC windspeeds at a given grid cell

𝑥 = windspeed for which the return period is calculated

Figure 4: Plotted probability distribution functions for the observed wind speeds and four fitted
extreme value distributions (3-Parameter Weibull, 3-Parameter Gamma, Gumbel, and
Generalized Pareto Distributions) for HURDAT BestTrack Extended dataset point with one of
the largest available number of recorded maximum TC windspeeds in the domain (n=20).

Initial Results and Discussion of EBTRK Data: A Case for Modeled Data

The nature of extremes, which are the primary focus of this study, are that they do not
occur often. In the case of the most extreme TC winds for a given location, these occurrences can
have return periods on the order of hundreds of years, and consequently cannot be accurately

9
estimated with a 30-year record. Nonetheless, gaining some knowledge of the nature of this
relatively small temporal analysis both validates the subsequent adoption of high spatial and
temporal resolution data, and also provides somewhat of a basis to which the results of that
modeled data can be compared for real-world verification.

A preliminary analysis for TRP of windspeed 𝑥 = 121𝑘𝑝ℎ (75 𝑚𝑝ℎ) is chosen due to its
definition as the lower threshold of windspeeds in a Category 1 hurricane while maintaining a
low enough TRP threshold that the majority of the domain is populated with enough data to
observe patterns and robustness. We also take 𝑥! to be 34 kt (63 kph) because that is the first TC
wind radii threshold and allows us to keep as much data as possible to calculate fitting
parameters for the GPD. These values are used in the above equations and calculated for every
grid cell within the domain and plotted (Fig. 5).

Figure 5: Plotted TRP (units of years) for 𝑥 = 121kph/75mph within the Northern Atlantic from
observed EBTRK data fitted to a Generalized Pareto distribution.

Figure 5 does reveal some viable broad scale patterns of storm tracks over the Northern
Atlantic basin, highlighting some areas of most frequent TC occurrence and general intensities.
As we would expect, the bulk of the lowest TRP for the highest winds occur over the open ocean
and decrease dramatically as storm tracks move over land masses. Although the TRP for TC

10
windspeeds are generally reduced inland due to storm decay, the EBTRK dataset only takes into
account the weakening of the entire system itself as it moves over land rather than explicit
reduction of wind fields due to effects of surface roughness. Instead, EBTRK implements the
generalization of the 34kt, 50kt, 64kt, and maximum wind fields respectively. This problem is
exemplified in Figure 6 with a large wind swath moving over southern Florida from Hurricane
Wilma where the centermost portion of the storm containing the highest wind speeds maintains
what appears to be nearly homogenous wind characteristics as it traverses over land. There is
little to no spatial variation associated with moving onshore or offshore on either coast. We also
know this is not an artifact of the linear interpolation as can be seen in Figure 2 where plotting
the raw data shows a time stamp of records while Hurricane Wilma is directly over land. The
plots show some weakening as would be expected as a TC passes over land, however the
reduction of the wind field, particularly from off to onshore, seems unlikely to well represent
real-world phenomena.

Figure 6: Gridded plot of maximum recorded wind speeds (mph) from HURDAT BestTrack
Extended Dataset over the southeastern United States in 2005

Because the wind field reduction as a direct result of TC wind interaction with surface
roughness is not fully represented, it is likely some of these wind fields, especially over
landmasses, are overestimated. Therefore, locations over which there may happen to be an
extreme TC wind swath during this relatively short study period are biased towards lower TRP

11
values for the upper wind extremes purely due to chance, showcasing the need for a much larger
dataset.
Near land, where we might expect spatial differences in roughness due to terrain and the
type of land usage, there may be subsequently related patterns in the values of TRP. Such features
should not appear over the open ocean, where the surface roughness is a function largely of wind
speed. There are areas where it is clear discrete storm tracks have highly influenced the
calculated TRP of that wind threshold, for example within areas of the Caribbean, leading to an
unnaturally sharp gradient between low and high TRP for the same TC wind value within a small
geographic region. Notably, within those areas affected by strong TC wind swaths, TRP for high
wind speeds are much lower than their surroundings, indicating a strong bias towards higher TC
wind speeds likely due to pure chance within this short time frame. With such little data to
support the analysis, the presence or absence of outliers within the data, such as the extremes we
are focusing on, can have a large impact on the calculated CDF and subsequent return periods.
The appearance of storm tracks over the open ocean (Fig. 5) indicates that sampling is
insufficient. These results clearly point to a need for more data to make accurate assessments of
return period. Despite the low TC wind threshold of 𝑥 = 121 𝑘𝑝ℎ (75 𝑚𝑝ℎ) for the example
calculation of TRP, which provides the best chance for robust results given the limited data from
EBTRK, it is clear the lack of data is hindering the results. This lack of data in addition to some
shortcomings with appropriate calculations for the entire domain, particularly accurate
windspeeds over landmasses as a direct function of surface roughness need to be addressed.
As compared to some of the most similar previous TRP assessments by other researchers,
while overall patterns remain intact, more detailed examination reveals some notable differences.
One assessment by Keim et. al (Fig. 7) takes a different approach where they examine discrete
locations of interest immediately along the eastern and Gulf of Mexico coastlines in the United
States. Their analysis makes use of historical data and assigns an assumed wind field for
respective storm intensities. Notably, this assessment is not continuous along the coastline, nor
does it consider locations inland or offshore. Therefore, while beneficial to the given locations of
interest, this assessment may not be representative of expected conditions for locations
elsewhere. Similar patterns in TC landfalling patterns are found to our assessment, however there
are differences in magnitudes of expected wind speeds where our EBTRK assessment predicts
longer TRP.

12
Figure 7: Plotted return periods by Keim et al. (2007) using historical tracks and assumed
‘average’ wind field radii correlated to storm intensities.

Another assessment by Bloemendaal et al. (2020) takes this concept further by expanding
results out into a continuous, gridded assessment of expected TC winds across multiple ocean
basins and inland locations down to a 10 km resolution (Fig. 8).

13
Figure 8: A continuous, gridded analysis of expected return period values given equivalent
Category 1 strength TC winds across multiple ocean basins at 10 km resolution by Bloemendaal
et al. (2020).

Their methods source through IBTrACS data to synthetically populate 10,000 years of
modeled TCs. For inland locations, they assume both a general inland decay function derived
from United States landfalling events and wind field reduction upon circulation center contact
with land. This wind model also does not explicitly include the influence of land, which they
acknowledge may result in overestimation of wind speeds for some inland locations. Magnitudes
are in the same general range as compared to our preliminary assessment, however there is a
noted lack of spatial variation with the assessment by Bloemendaal et al. Considering the
assumptions made with their wind model, particularly over land, in addition to their much larger
synthetically populated dataset, this may be why their gridded analysis appears much more
homogenous as compared to the EBTRK analysis. The commonality between these assessments
14
by the previous researchers and the current EBTRK analysis is the lack of explicit consideration
of land interactions in addition to a much higher resolution analysis to capture any possible
smaller scale variability.
While the analysis of EBTRK data are good for a general, real-world assessment of TC
winds and their overall patterns, we have found that it does lack the requirements for a robust,
localized analysis of extreme winds. The wind fields are strongly generalized to their extent of
wind radii and overall storm decay rather than observations at the surface level that will be
affected by boundary layer conditions. For the open ocean, this is not a bad proxy, however
winds over land can vary dramatically and a single radius of expected winds for an entire storm
quadrant is unrealistic in practice. Even if this was not a problem with EBTRK, we still lack the
resolution at which a true sense of the boundary layer effects will have on the wind. The smallest
wind radii interval for EBTRK is 5-nautical miles (9.25 km, 5.75 mi), and the average a quarter
of the angular distance around a TC would provide a lot of smoothing. As stated, wind variation,
particularly from off to onshore, can vary dramatically within a spatial extent of tens of meters. It
is clear then a peak resolution of 5-nautical miles will come nowhere near capturing this level of
variability, therefore again over-generalizing a geographic region. At this point, for a more
robust analysis, we can conclude a dataset is required that has a high spatial resolution, a much
larger temporal extent, and takes into account boundary layer features that will affect TC wind
speeds.

15
CHAPTER 3

A HIGH-RESOLUTION ASSESSMENT OF RETURN PERIODS BASED


ON THE FLORIDA PUBLIC LOSS HURRICANE MODEL

Despite the drawbacks in using HURDAT for the TRP analysis, there are clear advantages
to using observational data when extrapolating those results out to the real world. While we may
not have an extensive observational dataset for TC wind fields, The Florida Public Loss
Hurricane Model (FPLHM) attempts to mend this and a number of other concerns. The FPLHM
is a catastrophe model design to quantify risk of TC wind events. It is an interdisciplinary effort
combining the work of leading meteorologists, statisticians, geographic information systems
(GIS) experts, engineers, computer scientists, and actuarial scientists to quantify structural losses
due to TCs. The FPLHM models thousands of years of synthetic events that are statistically
consistent with historical observations in order to create robust statistics from which potential
losses may be calculated. More specifically, it is a high-resolution TC wind model (certified
annually 2006-onward) comprised of 59,000 modeled years of TC wind measurements at
roughly 1-kilometer resolution as a function of effective surface roughness down to a resolution
of 90-meters across the entire state of Florida. The dataset used in this study, however, is taken at
a resolution of roughly 1-km for locations of interest where infrastructure exists. Effective
surface roughness can be defined as the composite or mean of local surface roughness values
within a given grid size of heterogenous roughness conditions (Hansen, 1993). This is considered
when calculating upstream fetch for modeled wind speeds at each of the 1-kilometer spaced
points. The FPLHM was created as a catastrophe model to estimate the risk of TC damage to
residential properties within the state of Florida and is one of only five models certified by The
Florida Commission on Hurricane Loss Projection Methodology to do so.
A number of key meteorological considerations in the FPLHM should be noted prior to
explanation of their implementation within the model. In our new analysis, we are now explicitly
considering the effects of surface roughness at local scales as we expect those to play an
important role in determining expected TC near-surface windspeeds, which then correlate to
subsequent TRP values. Generally, surface roughness is the surface-based, geometric variability
as a result of terrain and/or features such as vegetation and man-made structures. This is
parameterized as a roughness length, which is proportional to a length scale of turbulence
16
interacting with the surface. Effective roughness length will be further examined in Section XX,
where we will examine risk around the Tampa/Clearwater region on the west coast of Florida.
The variability in elevation in this area of Florida is negligible with respect to wind interaction
(citation), therefore surface features related to main-made structures and vegetation are the
primary focus leading to TC wind variability when accounting for those values as a function of
the surface roughness.
At local scales, we consider the main driving forces behind tropical cyclones winds to be
the pressure gradient force leading directly to the strength of the winds in conjunction with the
Coriolis force and opposing frictional forces, which can be parameterize in terms of surface
roughness length and stratification (alternatively, a drag coefficient can be used to account for
both surface roughness and stability). In the constant stress layer, change in wind speed with
height is a function of the wind speed above this layer and the friction at the surface, with the
greater impact of friction near the surface causing time averaged wind speed to increase
approximately logarithmically with height above the surface (Taylor, 1969). This layer typically
extends from very near the surface to 50 to several hundred meters above the surface. For neutral
stratification, which often a good estimate for very strong winds, the profile changes
logarithmically with height.
When the surface roughness changes abruptly the horizontal flow can separate into two
log-layers, the lower layer being characteristic of the new surface, and the higher layer having
properties of the prior surface (Fig. 9). Work by Arya (1998) details this, where the PBL
encompasses the depth of the lowest portion of the atmosphere in contact with the surface in
which convective and turbulent processes occur, which also includes the equilibrium boundary
layer (EL) and internal boundary layer (IBL). The EL in this case is characterized by regions in
which the PBL is in equilibrium with previous marine surface characteristics. Commonly, this
distance over which the wind passes marine surfaces uninhibited is referred to as fetch, but this
this term can be applied to the distance of any homogenous field of surface roughness with
which wind may interact. It is suggested for winds encountering a step change in surface
roughness, the distance and subsequent time it takes for the system to come into equilibrium over
the step change in surface roughness can be over 10-20 km (ESDU, 2012). Despite what may
appear to be a significant distance over which winds come back into equilibrium over a new
surface roughness, it is assumed this process begins from the surface up (Garratt, 1989).

17
Therefore, the IBL develops more quickly at the surface and winds near the surface winds are
expected to be slower as compared to marine conditions. The rate of change of the winds in the
IBL must decrease as the IBL grows, and it is an ongoing research question as to how quickly an
IBL grows to fill the EL. The current analysis, however, takes the modeled wind speeds at 10 m,
so we will assume a faster transition to equilibrium conditions that should be reflected in the
analysis over varying surface roughness transitions.
As winds transition from marine to land conditions and begin to interact with higher
surface roughness lengths, the IBL begins to form, which is another boundary layer that grows
within the PBL. The IBL is characterized by a change in the vertical wind profile as a result of
winds interacting with regions of different friction, where winds at the lowermost portion of the
IBL response first, and the adjustment gradually moves up the wind profile as the fetch
increases. Fetch is the distance down winds from where the surface roughness changed. If the
surface becomes rougher, then the stress near the near the surface increases and the surface
winds are slower than they were over the smoother surface. Conversely, the IBL will have lower
stress and stronger winds over areas of lower surface friction. Above the IBL, wind
characteristics are observed to be similar to those found in the EL. Near the top of the IBL there
is a transition zone where the wind shear increases, causing momentum from the IBL mix with
momentum from the EL, causing the IBL to grow in height as fetch increases. If the flow is from
a smooth surface to a rough surface, then the IBL from the rough layer will be eroded from the
bottom. Within the modified flow of the IBL, the greater turbulence generated from wind
interaction with the rough surface is perpetuated with altitude through eddies. These eddies and
other circulations generated within this turbulent layer contribute to downward momentum
transfer through the IBL from the faster wind speeds aloft, resulting in the development of gusts
at the surface (Carpman, 2011). Another perspective on the IBL is that there is a non-zero
vertical convergence of horizontal momentum in the IBL. A surface with greater roughness
extracts more momentum, resulting in slower winds. An IBL over a surface with reduced
roughness would have a positive vertical convergence of horizontal momentum, resulting in
greater wind speeds in the IBL.

18
Figure 9: Vertical wind profile as seen over the transition from smooth to rough surface features,
leading to surface wind speed reduction as a result of the modified flow within the IBL (dotted
line). The IBL grows in height with time over the rough surface until equilibrium is reached,
while the undisturbed PBL above the IBL remains at similar magnitude to marine EL conditions
(Raupach and Thom, 1981).

In order to model the wind fields within each of these TCs, the FPLHM also takes into
consideration the radius of maximum winds, which is particularly important for the study of the
extreme tails of the distribution as we examine here. The FPLHM models this first through the
implementation of a wind field model once a TC wind field begins to impact land, first assuming
a boundary layer vortex in gradient wind balance while iteratively solving for the mean PBL
wind with the Navier-Stokes equation. That is, a wind field in balance with the pressure gradient,
Coriolis, and centripetal forces with a counterclockwise radial inflow component for the
Northern Hemisphere around a low pressure system. The resulting wind field is then modeled as
a function of a number of variables, notably for the radius of maximum winds with the Holland
B parameter. This value is used within a separate function described by Holland (1980), marking
the sharpness of the horizontal pressure profile extending from the center of the TC, which
correlates then to the sharpness of the gradient wind profile and radius of maximum winds. The
gradient wind 𝑉7 at radius r is described by the following:

19
𝐴 ,/"
O N−
𝑉7 = Y𝐴𝐵(𝑝8 − 𝑝9 ) exp 𝑟 : ` (6)
𝜌𝑟 :

(𝑝8 − 𝑝9 )
𝐴 = 𝑟 : ln d e (7)
(𝑝 − 𝑝9 )

𝐵 = 1.74425 − 0.007915 ∗ 𝐿𝑎𝑡 + 0.0000084(𝑝8 − 𝑝9 )" − 0.005024 ∗ 𝑅𝑚𝑎𝑥 (8)

Where:

A = scaling parameter; defines location of pressure profile relative to TC origin


B = scaling parameter, known as Holland B parameter; defines shape of the pressure profile
r = radius (kilometers)
p = pressure at radius r
𝑝8 = ambient pressure; described as the first anticyclonically curved isobar outside the TC
𝑝9 = central pressure of the TC
Lat = current latitude (degrees) of the TC center
Rmax = radius (kilometers) of maximum calculated wind speeds

Figure 10 (Holland, 1980) shows the subsequent effects of varying Holland B parameter
values on calculated pressure and wind profiles as a function of radial distance from the TC
center with the remaining parameters fixed. Most pertinent to this current study and the methods
used by the FPLHM is the radius of maximum winds as dictated by the Holland B parameter.
Calculated within the FPLHM with Equation (8), this parameter is important in determining the
magnitude and shape of the most extreme TC wind fields over regions in which we are looking
at corresponding TRP values. These wind fields are basis of the FPLHM that are then subjected to
complex analysis as a function of surface roughness to calculate the final modeled wind speeds.

20
Figure 10: Effects of varying Holland B parameter values on calculated pressure and wind speed
profiles as a function of radial distance from the TC center (Holland, 1980).

Now with the meteorological factors established, this model begins by deriving TC seeds
for a 59,000-year model period from the empirical distributions of HURDAT2 data that records
observed TC tracks and intensities from 1900-2016. Changes in track positions and intensities
are continuously sampled from the empirical distribution. When a modeled TC impacts Florida,
a slab layer wind model solves for the mean PBL wind through the Navier-Stokes equation that
also includes a surface friction term. This wind model converts the mean slab wind to a 10m
surface wind using factors based on GPD dropsonde data and considers other parameters such as

21
radius of maximum winds and the shape of the TC pressure profile (Holland B pressure profile
parameter). The solution of the wind field is then solved numerically through an iterative process
where the wind field is first assumed to be in gradient wind balance for the first iteration of the
solution. Winds are also assumed to be in equilibrium with marine conditions until coming into
contact with land where a surface roughness conversion is done for terrain conditions. Land use
data is sourced from the MRLC 2011 NLCD and Florida water management districts where a
number of previous assessments are used to assign a given surface roughness factor to each
respective land use type. Each of these land use types are marked by a code to which then a
roughness length can be assigned. As winds pass from marine to terrain conditions, a coastal
transition model is implemented where winds are in partial equilibrium with marine conditions
for a short distance inland. Over land, the reduction of TC wind fields is modeled at ~90-meter
resolution where at each of these points effective surface roughness is calculated from eight
different directions. Upstream effects on winds at each location are then calculated with an
exponential weighting function Powell et al., 2005).
The domain covers the entire state of Florida and winds are calculated at 90 m resolution;
however, the dataset in this study uses points of interest selected at a resolution of about 1 km.
Winds below 80 kph/50 mph are not retained as they are below the damage threshold of
FPLHM’s vulnerability model. For the purposes of our return period assessment a 1 km
resolution grid is created and populated by the FPLHM wind data similar to the EBTRK
analysis, the peak wind speed per grid cell is kept for each year of the dataset. A grid at this
resolution is chosen for the purposes of retaining a high enough resolution to retain overall
consistency within local land usage and to demonstrate the change in wind speeds as they come
onshore.

Methodology for FPLHM Winds

Due to the high-resolution nature of this dataset, rather than the examination of an entire
ocean basin and adjacent land, the study area for this chapter is narrowed down to areas around
Tampa and Clearwater, FL, as this region includes a complicated coastline and substantial
differences in land use. Similar to the EBTRK analysis, the maximum wind speed per modeled
year is taken at each grid cell. Smoothing is not required with this dataset due to the high
temporal resolution of the model and its focus on areas related to insurance, so therefore gaps in

22
wind fields at each time step are relatively miniscule. Our TRP analysis empirically derives an
exceedance probability and a value for the corresponding wind speed associated with a value of
TRP from the dataset. We examine return periods of 10, 50, 100, and 250 years. For each grid cell
location, the maximum annual wind speeds are organized by descending rank order and years
without any wind data are removed after ranking. TC winds for each TRP are then calculated as
follows:
,
𝑒𝑥𝑐𝑒𝑒𝑑𝑎𝑛𝑐𝑒 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = < (6)
=>

𝑟𝑎𝑛𝑘(𝑇?@ ) = 𝑖 = 59,000 ∗ 𝑒𝑥𝑐𝑒𝑒𝑑𝑎𝑛𝑐𝑒 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 (7)

𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑇𝐶 𝑤𝑖𝑛𝑑 = 𝑤𝑖𝑛𝑑 𝑎𝑡 𝑐𝑎𝑙𝑐𝑢𝑙𝑎𝑡𝑒𝑑 𝑟𝑎𝑛𝑘 (𝑖) (8)

This procedure is implemented for each location within the grid for each of the previously listed
TRP values and plotted for all wind speeds greater than 80 kph/50 mph (Fig. 8).
Empirically deriving the TRP values from the data is beneficial for an estimation based
purely on the modeled values themselves, and it is effective for cases with sufficient data as we
do with the FPLHM. However, fitting the data to a distribution as was done with EBTRK is
beneficial for extrapolating values across the entire range of wind speeds and further TRP values
when data may not be as plentiful. The same process that was implemented with the EBTRK
data with fitting the data to a given EVD is done with the FPLHM for six locations around
Clearwater/Tampa, however 𝑥! is now taken at the lowest wind threshold for this data at 80 kph
(50 mph). These regions were selected because of land use and varying proximity to the
coastline. We examine the following locations, each designated as a single 1-kilometer grid cell:
3 km (~2 mi) offshore island, coastal residential, bayside city (Tampa cityscape), 8 km (~5 mi)
inland residential, 24 km (~15 mi) inland with proximity to wetland, and 32 km (~20 mi) inland
residential (Figure 11). These specific locations are chosen to directly examine the effects of
upstream land use on extreme TC winds, particularly as those change over short distances
between the marine EL to IBL transition. The offshore location serves as a proxy for winds as
close to marine equilibrium as this dataset will allow given the geographic constraints to
landmasses.

23
Figure 11: Selected locations of varying land use (stars) for the gridded TRP analysis of the
FPLHM. The gridded analysis is taken at one-kilometer resolution and each discrete location
encompasses one of these cells. These marked locations are overlaid on a land use reference map
of the Tampa/Clearwater area developed by Russell et al. (2013).

24
In contrast, the two chosen locations along the coastline, coastal residential and Tampa
cityscape, are intended to focus on the downstream effects of modified wind flow immediately
after the initiation of the IBL from theoretically lower surface roughness conditions with
proximity to marine conditions. The 5-mile inland location is intended to capture wind
conditions within a more fully developed IBL with upstream fetch over a longer distance of
urban development. The two locations father inland also intend to capture similar features within
the more mature IBL while examining the magnitude of winds with a significantly longer
upstream fetch over varying terrain. One of these inland locations is situated by a wetland with
lower surrounding surface roughness. Despite the greater distance inland, the lower surface
roughness may allow for greater sustained wind speeds as a result of winds more rapidly coming
back into a new equilibrium over the relatively smoother surface as compared to the urban
developments. Where there are more developments and higher surface roughness, it is likely the
extent of the modified flow will be greater and corresponding winds will be at a lower
magnitude, particularly within the regions of longest upstream fetch. Examining the effects of
varying surface roughness coefficients at a range of fetch distances will help answer our question
regarding the potential changes of TC wind intensity and related TRP values over finer shorter
distances than have been previously studied.
The two coastal locations, coastal residential and bayside cityscape, are both set back one
kilometer inland from the immediate marine to land surface transition. This selection is designed
to highlight any potentially dramatic wind speeds changes as a result of the upstream fetch over
land. In other words, we will address the question ‘over the shortest distance (1-kilometer) for
which we can analyze saved data for effects of surface friction, how significant are these changes
in expected wind speeds?’ Our metric for this change is the return period. To examine this, the
maximum wind speed per modeled year for each location is retained and years with no data (i.e.
years without a storm or a windspeed <50 mph/80kph) are removed. Modeled years are then
ranked and binned for those exceeding the 50 mph/80 kph threshold in the same manner as was
done for the HURDAT dataset. This sorted data is then fit to the same EVDs as were used with
HURDAT to find the best fit as given by the 𝑅" values quantifying the correlation between the
PDF of observations and the fitted distribution. Again, the GPD was the best fit for every
location, with 𝑅" values at or greater than 0.98, indicating statistical significance at a nearly 99%
confidence level for the GPD’s ability to model the data (Fig. 16). We can take this a step further

25
using the calculated PDF and CDF for the GPD, as previously described with equations (1) and
(2) to then create a scatter plot of the specific RP values and corresponding expected winds for
each of those individual locations (Fig. 17, Table 1). These show a slight underestimation from
the GPD fits for return periods ranging from 10 to 150 year, and sometimes underestimates
return periods for the most extreme events. This rather systemic result suggests that winds from
storm eyewalls might be better described as a different distribution that is added to the
distribution for the passage of non-eyewall winds; however, the limited number of observations
complicates this assessment.

26
CHAPTER 4

RETURN PERIODS NEAR COASTAL LOCATIONS

Before analyzing the expected wind speeds at given TRP values, it is worth emphasizing
the nature of the winds for this dataset as being one-minute sustained winds rather than gusts. By
definition, wind gusts are a small temporal scale phenomenon (<1 minute) that exceed the mean
of the winds over a given time interval (Kristensen, 1991). In brief, numerous factors contribute
to gusts at the surface such as convective overturning, turbulent momentum transfer over
increased surface roughness, organized large eddies, or any process that may allow for the
transfer of gradient winds aloft towards the surface (Rotach, 1993; Vickery & Skerlj, 2005).
While relevant, gusts are not explicitly examined within the FPLHM data used in this analysis
and we will focus more on overall environmental factors contributing to those corresponding 1-
minute sustained modeled values. There are, however, methods that consider mean wind and
atmospheric turbulence in order to estimate wind gusts (Brasseur, 2001) in addition to gusts
calculated with other FPLHM datasets.
Evident in Figure 13, lower TRP for the highest extreme TC winds are seen along the
west coast of Clearwater along the Gulf of Mexico, while within Tampa Bay we see those
occurring predominantly on the south and east facing shorelines. Tampa/Clearwater is on the
west coast of Florida and is locally the point furthest west, therefore it is surrounded by open
ocean on the west side of the area as well as to the south in some respect. TCs form and
strengthen over warm ocean waters due to initiation and support of convective processes through
warm and moist surface conditions (Shay et al., 2000), in this case the Gulf of Mexico, and
sometimes make their way towards land, upon which they begin to weaken due to a lack of these
same conditions (Powell & Houston, 1996). Should a TC be present within the Gulf of Mexico
and track towards Tampa/Clearwater, the scenario in which the TC would remain over warm
ocean water for the longest period of time prior to impacting this region means it would come
from the west or the south. Assuming all other factors contributing to TC strength are sufficient
and isolating the viability of the storm to just the existence of sufficiently warm ocean waters,
the TC should then weaken as soon as that fuel source is removed. In other words, the TC will
begin to weaken as those warm and moist conditions supporting the core of the storm containing
the strongest convection diminish over the comparatively cooler and drier land conditions.
27
Powell & Houston (1996) find results in an analysis of Hurricane Andrew indicating a more
rapid deintensification of the inner core (in this case within a 50-60 km of the circulation center)
containing the strongest winds immediately upon landfall as compared to the wind fields
extending outside of that radius. They also found this weakening trend even despite the moist
conditions Hurricane Andrew passed over in the Everglades, in addition to being unrelated to
other TC dynamical changes such as an eyewall replacement cycle. With this suggestion of a TC
requiring both warm and moist conditions in order to sustain the region of greatest wind speeds
in conjunction with more rapid weakening of the core upon landfall, we should expect general
patterns of the strongest TC wind fields impacting Tampa/Clearwater from the west or south.
TCs formed to the east or north would pass over significantly more landmass than from the other
directions, thereby reducing both the fuel source supporting convective processes and likely
significantly reducing the magnitude of the strongest winds within the core as those make
landfall away from Tampa/Clearwater. Further reduction in wind field magnitude should be
expected by the interaction with greater surface roughness conditions upon landfall that then
induce the formation of the IBL, where the induced turbulence reduces the surface winds as
previously described. It is likely the storms originating from the west or south are bringing the
most destructive winds at the upper end of the distribution, therefore being a strong determining
factor in the distribution of those corresponding TRP values. This has important implications with
steering and climatological factors for this region, which are elaborated upon in Section XX with
future considerations.
Tampa/Clearwater is located in the Northern Hemisphere, which means TCs and their
respective wind fields rotate counterclockwise. This leads to a few implications for which areas
should see greatest expected winds in the absence of other factors that would change the wind
pattern (i.e., different types of surfaces). First, this direction of rotation means the winds in the
right, forward-direction quadrant of the TC are typically going to be strongest due to the
cumulative effect of the storm’s forward motion in conjunction with the TC winds themselves
moving in the same direction. In the case of Tampa/Clearwater where we have already
established, theoretically, the strongest TCs are going to come from the west or south, this means
those shorelines facing either of those directions should experience the highest TC winds. After
analysis, this pattern of western facing shorelines experiencing the highest winds is reflected
along the west coast of Clearwater directly on the Gulf of Mexico, particularly along the barrier

28
islands. However, as noted in Tampa Bay, the shorelines showing the highest modeled
windspeeds are the south and the east. This contradicts what we would initially expect as far as
where the strongest winds should be since an eastern facing shoreline would not experience the
aforementioned cumulative TC wind effects to the same degree. Perhaps even more peculiar at
first glance is that the west-facing shorelines actually experience some of the lowest modeled
wind speeds within the entirety of Tampa Bay coastline, although those shorelines still have a
greater wind speed than locations further inland. This is where the effects of land use and surface
roughness come into play in conjunction with the expected TC wind fields. As the core of the TC
containing the strongest winds comes ashore, we should expect to see more rapid decay of this
portion of the storm as described by Powell & Houston (1996) compared to the regions of the TC
containing relatively lower winds. That inherent reduction of the wind field is also paired with
the effects of greater surface roughness resulting in further reduction of those strongest winds as
we move inland. Locations directly along the coastline with sufficient upstream fetch over
marine conditions and within the closest proximity to the Gulf of Mexico should therefore see
the lowest TRP for the highest winds within the study domain. In contrast, the combined effects of
the more rapid wind field reduction for portions of the storm with higher winds and the effects of
surface roughness should result in lower expected wind speeds. These changes could be
potentially significant based on IBL conditions as we move further inland. This should present
itself as a sharp gradient of expected TC winds at a given TRP value between coastal locations
that satisfy the conditions for the greatest TC winds and inland locations. Locations where the
IBL is more highly influenced by greater upstream surface roughness conditions are where we
should expect to see the strongest gradient between expected TC winds. Greater surface
roughness results in the greater turbulence within the IBL, leading to lower sustained wind
speeds at the surface as the IBL attempts to come into equilibrium over the new terrain. While
inherent decay of the wind field over land is likely to be shown in the analysis, this sharp
gradient between the first few kilometers inland from the shoreline is more likely influenced by
surface wind field reduction over increased surface roughness conditions. Powell & Houston
(1996) note that while the most rapid winds reduce more rapidly, they may not immediately
disappear and can continue inland with similar intensity, yet with a smaller footprint. Therefore,
the regions furthest inland within the study domain are likely to see much lower expected TC

29
winds with the greater influences of both the wind field decay and reduction associated with
significant fetch over inland terrain.
Tampa/Clearwater was chosen for its wide range of terrain and land use (Fig. 11, Fig.
19). It is also unique in its geographical features, particularly within Tampa Bay, in that much of
the bay area is surrounded by land mass with little room (~20 km, 12 mi) for uninterrupted TC
winds to come in directly off the Gulf of Mexico. In order for the strongest TC winds to enter
Tampa Bay uninhibited, the most intense portion of the TC would have to come directly towards
the area at a very limited range of possible angles given the size of the opening into the Gulf of
Mexico (~20 km). The limited size of the opening means the strongest core of a TC is less likely
to directly impact that exact area, which can be shown by the lower intensity expected winds
within the vicinity of the bay opening into the Gulf of Mexico for each TRP value in Figure 15.
The largest west-facing shoreline within Tampa Bay is also oriented facing more northwesterly,
of which not only does the majority directly face landmass, i.e. blocking direct TC winds coming
from the Gulf of Mexico, but under the ‘best’ conditions of an intense TC entering Tampa Bay,
would still likely not receive the strongest winds despite its coastal location. This is mostly
caused by the traditional TC tracks within this area (Fig. 12) that when observed
climatologically, do not favor taking this kind of path towards landfall in this region. One of the
main sources of TC steering mechanisms within this area of the Atlantic Basin is the Bermuda
high, where a strong high pushes TCs in a straighter track westward across areas such as Florida,
while a weak high results in more dominant recurvature of the system back towards the east
(Elsner et al., 2000). This northerly component to TC steering is furthered through the effects of
beta (𝛽), where the interaction between the TC and the planet’s vorticity results in advection of
the TC north and westward (Holland, 1984). With the vast majority of TC tracks moving with
some northward component and referring back to our analysis of the strongest potential TCs
originating from the south or west, we know Tampa/Clearwater is most likely to see the TCs
with the strongest wind fields moving northward or northeastward. Given the orientation of the
northwest-facing portion of the Tampa Bay coastline and the curvature of the wind field,
especially around the centermost, and more often most intense portion of a TC, it is likely winds
will pass over land to the south before reaching that area. Noting the surface-up changes in the
IBL over new terrain and the immediate reduction in wind intensity from off to onshore
locations, we expect to see TC winds moving over any portions of land to reflect the same

30
patterns in surface wind reduction. Then, any locations with upstream landmass interrupting
marine fetch and strongest TC winds should have a reduction the greatest TC wind intensity,
such as is seen with the western facing shorelines within Tampa Bay. In summary, despite the
assumption that generally west-facing coastlines within Tampa Bay would be experiencing more
extreme TC winds, the orientation and location of these coastlines within the bay in conjunction
with TC track climatology and wind field direction results in much more land interference with
TC wind fields and subsequently lower expected winds at a given TRP.

Figure 12: TC tracks through Tampa/Clearwater, Florida from the mid 1850’s-present by
NOAA (2021).

In the case then of the southeasterly facing shorelines within Tampa Bay showing
relatively lower TRP for high TC wind speed values, we have to look at the effects of land use
and consider the TC climatological assumptions already established. Again, referencing the
expected direction of the most intense theoretical TC winds in this area, we would expect the
upstream flow of those to originate from the northwesterly facing coastlines on the south side of
Tampa Bay and beyond. What is beyond that is mostly areas of land dedicated to agriculture
with some grasslands and limited suburban developments interspersed throughout. In short, the
land use south of Tampa Bay is largely low profile, meaning less surface roughness and therefore
less frictional effects to slow down TC winds. Not only does this mean stronger expected TC

31
winds expected more frequently just within this area based on the TRP analysis, but this also has
downstream effects as to how that wind is treated once it again moves over the relatively
frictionless surface of Tampa Bay. With a reduction in surface roughness coefficients over
agricultural swaths of land as compared to more urban developments, the IBL requires a greater
spatial extent to come into equilibrium with the new surface. This results in an overall larger
magnitude of expected winds over this area, which also helps with the return to marine
equilibrium as those already faster winds come back over open marine fetch in Tampa Bay. For
what we have established as the theoretically strongest TC winds that would be moving
northwestward over this region, those would then emerge back over the open waters of Tampa
Bay upstream of the southeastern facing shorelines. These winds may then have the opportunity
to speed up further as the surface returns to low-friction, marine conditions over a fetch of 5-15
kilometers (~4-10 miles) between the northern and southern shores of Tampa Bay. Multiple
changes in surface roughness conditions results in complex downwind interactions that may
retain elements of each previous step change (Blom & Wartena, 1968). Despite this change seen
through the entire vertical wind profile, the transitions to a new surface roughness still results in
a bottom-up IBL formation that results in near-surface conditions being the closest to
equilibrium following the step change (Hanson & Ganapathisubramani, 2015). Again, winds in
this study are modeled at 10m, meaning the IBL formation should be seen very shortly after any
step changes since they are near the surface. Thus, modeled TC winds that first encounter a step
change in surface roughness and then return to marine conditions, such as may be seen on the
southern side of Tampa Bay, should have sufficient time over the 5-15 km distance to reach
near-equilibrium conditions over the new marine surface roughness layer. This should manifest
as southern/southeastern facing shorelines with less interrupted upstream conditions retaining
higher intensity expected TC winds similar to those found on the immediate Gulf coast.
Conversely, areas beyond the Tampa Bay coastline remain within the single step change to larger
surface roughness over urban or forested regions. The IBL over these areas begins to slowly
come into equilibrium with the greater surface roughness, resulting in continually slower
expected wind speeds as the analysis progresses inland. This then corresponds to higher TRP
values for these inland locations for TC wind speeds at any given threshold.

32
CHAPTER 5

LAND AND LAND USAGE IMPACTS ON RETURN PERIOD

To emphasize the impact of increasing the resolution and accounting for effects of terrain
on the TC wind fields, we can compare the general patterns of wind swaths between FPLHM
(Fig. 13) to HURDAT (Fig. 6) and see the vastly different features. This particular modeled wind
swath from FPLHM depicts a TC tracking generally from south to north, running about parallel
to the coastline with hurricane force wind fields moving directly through the study area.
One notable observation is the drastic reduction in windspeed from offshore and coastal
locations to even just 1-kilometer inland, or the smallest resolution at which the wind speeds are
modeled in this case. There is very little homogeneity here as might be incorrectly inferred from
HURDAT, indicating the boundary layer interactions with the TC winds are depicted having
much more sensitivity to the variety of terrain over which they occur. In some locations, we see
wind reduction by up to 25% immediately after reaching mainland Florida or interacting within
roughly 1- km of coastal features. This is particularly evident around Tampa proper as winds
transition from marine to land conditions, particularly within the areas of greatest urban
developments downtown. The winds directly along the southern facing bay coastline of
downtown Tampa hold up to the hypothesized greater relative magnitude within Tampa Bay, yet
diminish quickly as they moved inland over some of the areas of greatest surface roughness
within the cityscape. As exemplified in Figure 15, this noted reduction in windspeeds is reflected
in the calculated TRP, representing the entirety of the modeled data rather than just the one year
shown, with a clear distinction between expected winds along the coastline and winds just
inland. Additionally, Figure 14 depicts the TRP of 75 mph (threshold for hurricane force winds)
for both the FPLHM and HURDAT, showing the vast differences in not only resolutions, but TRP
values themselves. As we expected with some of the previously HURDAT limitations, both
sampling and lack of boundary layer interactions completely change TRP patterns.

33
Figure 13: Modeled wind swath from FPLHM dataset of a south to north moving TC along the
west coast of Florida over Tampa/Clearwater.

34
Figure 14: Return periods for 121 kph (75 mph) TC winds around Tampa/Clearwater as
calculated by modeled FPLHM at 1 km resolution (left) and observed HURDAT BestTrack
Extended at 27.75 km resolution.

We can note for empirically calculated return periods (Fig. 15) the TC wind extremes for
coastal locations have a wide range over the 10-250-year TRP where the highest expected winds
in these locations are around 150% greater than the lowest expected winds at the lowest
examined TRP. However, inland locations influenced by greater amounts of surface roughness
have a lower range of wind speeds corresponding to this same range of TRP values with about a
120% increase, suggesting these inland locations are less prone to extremes experienced directly
along the coastline and have winds strongly moderated by the surface roughness interactions
within the IBL. This leads to a problem where the large amount of data per grid cell actually
hinders the current GPD analysis in some respect because too great a portion of the distribution
is focused around lower values. This is a notable artifact for the distributions at locations furthest
inland with the surface roughness effectively results in a distribution more heavily weighted
towards the lower end of the extreme tail (Fig. 16). The GPD is sensitive to the fraction of years
with values exceeding xo, resulting in the function becoming more linear in nature that then has
difficulty rounding out towards the more sporadically occurring highest extremes.

35
Figure 15: Expected TC wind speeds at a logarithmic TRP scale in Tampa/Clearwater, Florida at
a 80 kph/50 mph threshold.

36
Figure 16: GPD fit to PDF of modeled FPLHM winds at select locations around Tampa,
Clearwater.

37
Thus, while including all of the data for the theoretically most robust fit, the GPD does
very well at fitting the more numerous lower windspeed thresholds yet does not fit well to every
distribution for the 95th or greater percentile. For the 15 and 12-mile inland locations in Figure
16, the GPD does not do well with extrapolating out to the most extreme winds speeds values at
the current threshold due to the relatively small fraction of larger extreme winds. Should we
attempt to mend this through increasing the threshold for which we calculate fitting parameters,
we risk running into the original problem as was seen with EBTRK with a lack of data. The
description of the full distribution of extremes in terms of two distributions, as mentioned earlier,
might resolve this problem. If sampling is causing this problem, then the 95th or greater
percentile winds might simply be acknowledged as unreliable.
While examining the statistical robustness at a given location, we should consider the
relationship that has with the local surface roughness conditions. In areas where upstream surface
roughness has a greatest effect, we expect to see a shift towards lower expected windspeeds with
few, if any, records at higher extremes, such as would be seen with far-inland, dense, urban
developments. Here, the IBL has developed sufficiently such that the winds have greater
difficulty passing a given threshold and therefore produces a limited number of expected
sustained winds greater than 0 kph/0mph, particularly at lower TRP. The GPD then has more data
from low end extremes and therefore does a better job at modeling the entire range of expected
winds. Interestingly enough, the GPD also does well modeling the offshore location despite its
much higher range in expected winds and greater extremes. Here, the mostly uninhibited marine
roughness conditions produce a much more populated spread of the data towards the highest
extremes due to the limited capping of the winds. Therefore, the GPD has more data to work
with across the entire range of wind observations, similar to what we see with locations inland
that are also more equally spread across the distribution, just within their smaller range of wind
speeds. With this more equal spread across the distribution for both significant and limited
mitigation of expected winds, the GPD then does a better job at modeling the full spread of the
data. In short, the GPD analysis is more successful in cases where there is a greater number of
data points at a given threshold within the PDF of extreme winds due to its bias towards the bulk
of the distribution. In cases where that is more equally dispersed throughout the range of the
distribution, the GPD performs well, according to our results. It is when we begin looking at
more modest surface roughness conditions that the GPD struggles capturing the full range of

38
expected winds. In these locations, such as shortly inland from the coast within urban
development, the IBL has begun to develop as winds attempt to come into equilibrium over the
new, rougher surface. However, this process takes time and so while capping of the winds does
occur shortly inland within 1-kilometer immediately as the IBL forms (Fig. 13, Fig. 15), the
gradient of marine EL winds to IBL winds spreads further inland as this process fully transitions
over a greater spatial extent. Over the transition from marine EL to IBL conditions, the winds
beyond the first step change will then, for a brief period, possess characteristics of both boundary
layers as those begin to come into equilibrium with IBL conditions. Work by Taylor (1969) also
indicates the growing temporal and spatial extent of this transition as the intensity of the winds
over this step change grows. Therefore, some of the highest intensity winds will persist further
inland despite the surface-up formation of the IBL while this transition takes more time to occur.
In theory, this process is faster for the lower windspeeds, resulting in greater mitigation for those
lesser thresholds. This combined with the inherent lesser number of most extreme modeled
storms results in some ‘outliers’ for the highest modeled winds in locations within this transition
zone, such as with the location 5-miles inland within urban terrain. We hypothesize while a bulk
of the winds are sufficiently mitigated through the IBL transition for weaker TCs, the less
occurring, yet highest modeled winds persist further inland over this extended IBL transition. It
is also possible these most extreme cases may be an artifact of less frequently occurring optimal
physical conditions such as angle of wind approach. This may explain why, particularly when
fitted to the GPD, a small group of remaining events (>99th percentile) are not fully captured.
These few cases that do not fit the remaining distribution make for difficulty in capturing them
with the GPD, which does well for the tail of distributions, yet becomes less effective with a
fewer number of observations from which the distribution is calculated. This effectively splits
the distribution into two that can both be described by the GPD; however, the fit needs to be
taken at different thresholds in order to do so and cannot be accurately described by a single
GPD function (Fig. 13, Fig. 15).
Not accounting for the most extreme few modeled TCs for inland locations, the GPD
does a sufficient job at capturing the distribution of the bulk of the remaining data as exhibited
by the >99% confidence levels for this range. This is also well within the range of damaging
winds, so their representation is still crucial. Additionally, the modeled tails for extremes
offshore and in coastal locations is done very well with the GPD due to the higher prevalence of

39
those events, so we have a much better picture of what to expect within those locations for the
occurrence of the strongest TCs. For example, the PDF of the location 2-miles offshore depicted
in Figure 16 contains the most data of the selected points within the uppermost extreme portion
of the tail due to limited surface roughness. As a result, the GPD fits the entire range of the data
very well, allowing for a more accurate extrapolation of the most intense winds as observed
within the model. On the contrary, a location such as the 15-mile inland residential with
proximity to wetlands has a bias towards lower TC winds due to the effects of surface roughness.
However, this location is unique in that the surface roughness is not high enough such that it caps
the expected TC winds as we see with the effect of the skyscrapers limiting maximum winds in
Tampa proper. Rather, this location still has a select few high intensity values that are outlying
modeled TCs that are not in line with the PDF trends or GPD fit. Because the GPD is fit through
the method of moments that calculates fitting parameters from the mean and variance, this fit
does not do a good job at capturing outliers when including the full spread of the data as we see
in this example. Efforts to reduce this bias are discussed in Section 6.1.

Expected TC Winds vs. Terrain: Features of Case Study Locations

TRP is calculated at each of the six locations for 10, 50, 100, and 25-years using the CDF
for the GPD (Fig. 15) and it is evident how dramatically TC winds can vary over short distances
between these chosen locations. The entire domain for this case study is only about 48 km (30
miles) wide and with 59,000 modeled years of TCs, these subsequent wind fields should be well
sampled throughout the domain, with a large emphasis on boundary layer effects. We have also
noted through other work (Powell & Houston, 1996) how TC wind field decay after landfall
independent of boundary layer effects is a more gradual process than is shown through the sharp
gradients seen in results for this study (Fig. 13, Fig. 15). This implies then the vast majority of
the variation seen in this study is directly caused by differences in terrain (including land usage)
and those effects on TC winds, rather than from the inherent gradual reduction of the entire TC
upon landfall.
Since we can attribute a large portion of these spatial changes in wind speeds to boundary
layer effects and have established much of the origin of those through analysis of the FPLHM,
we can examine some of these case study locations to see if those results are consistent with our

40
qualitative assumptions. Specifically, we expect surfaces with greater roughness to have greater
drag which will slow the winds, but also be more efficient for the transfer of momentum down to
the surface. This process may increase the height averaged wind speeds over the boundary layer.

Figure 17: Expected winds for a continuous TRP distribution as fit to GPD for six locations
around Tampa/Clearwater. Empirical rank-calculated return periods of FPLHM data plotted as
blue dots; GPD fit plotted as black line with 95% confidence intervals as calculated by the Delta
method plotted as dotted red lines.

41
Figure 18: GPD calculated at 𝑥! = 50; 95% confidence intervals (red) within ±1mph (top) and
GPD calculated at 𝑥! = 85; 95% confidence intervals within ±18 mph (bottom)

Despite these greater average wind speeds through the entire boundary layer, we expect
greater surface roughness to reduce both the change in wind speed with height well above the
surface, and increase it very near the surface, resulting in lower surface winds (10m) as we study
42
here. Wind observations for this study are also taken at 1-minute averages that do not capture
stronger gusts. Thus, we expect the surface wind speed over rougher surfaces to be reduced
because of greater drag upstream of a given location based on a given TC wind field. It is
assumed based off these conditions the highest expected wind speeds should be located along the
immediate coastline of the Gulf of Mexico and on much of the southeast facing shorelines within
Tampa Bay. The location 2-miles offshore near the opening of Tampa Bay is, in fact, locally
modeled as having some of the highest wind speeds. It is consistently projected, based on the
GPD fit, to have winds at all TRP ≥ 10 years within the range of at least equivalent high-end
Category-1 (119-153 km/h, 74-95 mph) TC and is the only selected location to reach equivalent
major Category-3 (178-251 km/h, 111-129 mph) wind speeds for any TRP. We expect similar
intensity wind speeds over areas with the same local and upstream terrain. This is a general
assumption, however, and should be considered with application to other points in this area. We
should note this location is offshore as far away from landmasses as available data would allow
for this region, but may not be far enough away to negate any wind fields that may originate
from over landmass immediately east. This may result in some influence within those respective
IBLs that develop over regions of greater surface roughness. Being that this offshore location
had intensities similar to locations directly onshore within Tampa Bay with the strongest
expected TC winds, this could mean two things: the offshore location is modeled to experience
influences of the upstream IBLs that have originated over land masses, and/or the fetch over
Tampa Bay is sufficient such that wind speeds come back to near marine EL conditions similar
to the offshore location. At this point we cannot say with certainty which or perhaps if both play
some kind of role in the calculated TC winds in these areas, which could affect what kind of
conditions may be seen offshore for other locations around this region.
Both the coastal residential and bayside cityscape locations are set back one kilometer
inland from the immediate coastline so we can observe the effects of TC wind interaction within
the first step of IBL development. This should highlight any drastic changes in modeled wind
speeds between locations on the coastline and those set back just inland behind the first step
change in surface roughness. Within the coastal residential area, there is extensive development;
however, these homes are generally 1-2 stories as compared to the coastal bayside location set
back within 10-20+ story skyscrapers. Therefore, we would expect greater upstream surface
roughness features within the cityscape, resulting in reduced expected TC wind speeds despite

43
equal distance from shorelines that may experiencing similarly high-intensity winds. The
analysis shows this is the case where winds in the coastal residential area not only stays
consistently above the winds in bayside cityscape for projected wind intensities, but also has a
greater range of wind speeds on the upper tail of the distribution with respect to TRP values. The
significantly higher surface roughness within the cityscape is projected to both mitigate potential
TC surface winds and effectively cap the maximum strength to well within equivalent Catgeory-
1 intensity. To further emphasize the effect a cityscape and therefore highest surface roughness
conditions has on the maximum TC wind potential just 1-kilometer inland, of the six chosen
study locations with varying coastal proximity and upstream surface roughness, the only location
with lower expected wind speeds is the one 20-miles inland within dense urban development.
This also further emphasizes the role distance from the initiation of the IBL plays in mitigating
downstream wind conditions. While not surprising to find modeled results that reflect the
physical conditions previously described, the magnitude of the changes is quite substantial. This
highlights the need to consider the impacts of surface roughness on TC winds and wind statistics
over higher resolutions that have been previously considered.
The last two locations furthest inland are distinguished by their surrounding terrain, the
24 km (15-miles) inland location being in a sparsely developed area near wetlands (low surface
roughness) while the 32 km (20 miles) inland location is surrounded by residential developments
similar to the 5-mile inland location. The 5-mile and 20-mile inland locations end up sharing
very similar TRP wind projections despite their varying coastal proximities, suggesting a limit to
the effect a given surface roughness has on TC winds independent of other factors. It may be the
case, after a certain period of time over a given, homogenous surface roughness that TC winds
come into equilibrium and remain generally steady at that given intensity. Being that this model
shows this same intensity for both the 5 and 20-mile inland residential locations suggests TC
winds come into equilibrium over a given surface roughness within 5-miles. We can also
compare the 15 and 20-mile inland locations because while they have a similar coastal
proximity, the reduced surface roughness of the 15-mile inland location with wetland proximity
should suggest higher expected TC winds. Likely due to that reduced surface roughness, this
location ends up having the third highest projected winds, only behind the offshore and coastal
residential locations. The expected TC winds for all TRP values meet the threshold of equivalent

44
Category-1 strength and it is the only other location aside from the offshore and coastal
residential locations to meet the Category-2 strength threshold.
In summary, the variation between the chosen study locations reveals some substantial
changes in expected TC winds over short distances that are not explicitly captured through
previous analyses. Between locations that are more influenced by marine EL conditions versus
locations within an IBL resulting from significant surface roughness conditions, both the range
of the expected winds and their respective magnitudes are found to decrease with greater surface
roughness. For the offshore location, the range between the expected TC surface winds at 10 and
250-year TRP is about 48 kph (30 mph), while the locations furthest inland within developed
IBLs have a range of only about 16-24 kph (10-15 mph). This indicates winds more influenced
by upstream marine EL conditions are more likely to experience the highest intensity TC surface
winds, while locations within a developed IBL will have winds being effectively capped at a
given intensity based on those respective conditions. The most notable case of this difference for
the chosen study areas is the Tampa cityscape set back 1-kilometer inland from the highest
intensity expected TC winds of similar intensity to the offshore location. Of the six locations, the
Tampa cityscape set back from the immediate coastline had the lowest expected TC surface
winds at any given TRP, while those TC wind intensities were modeled at values similar to those
found at the offshore location. At longest TRP value for which expected TC winds were
calculated (250 years), the difference in expected winds between these locations just 1-kilometer
apart was 64 kph (40 mph). Particularly in comparison to our expected values with the HURDAT
analysis, the variations are much higher over shorter distances when examining TC winds as a
function of surface roughness and at higher resolution. Additionally, in comparison to similar
previous assessments by other researchers, the highest resolution for a continuously assessment
was found at 10-kilometers. The results of this study, however, indicate risk analyses are the
most accurate at higher resolution of what we have found to be variation over scales at least as
small as 1-kilometer. Based on the sharp gradients such as were found between coastal Tampa
Bay and the Tampa cityscape it is likely this variation is seen across even smaller spatial scales
this study did not yet capture. The overall results of this assessment put great emphasis on the
possible significant variation of expected TC winds over small spatial scales that have not yet
been extensively studied, particularly as those are resolved over changing surface roughness
conditions.

45
CHAPTER 6

FUTURE CONSIDERATIONS AND CONCLUDING REMARKS

Future Considerations and Areas of Improvement

Implementation of effective surface roughness in this study shows the substantial


sensitivity to surface roughness. Such a consideration in modeling TRP values appears to be vital
in the development of more accurate risk assessments. Despite its potential, there are important
drawbacks of the current assessment that should be addressed in any future models. The FPLHM
stochastically models and sources from a historical database (HURDAT2), thus trends in TC
strength, steering, etc. are only captured from that time period covered by HURDAT2. Current
literature suggests the possibility of not only changes in TC tracks, but also changes in the
distribution of TC intensities, favoring a potential for a greater number of high-intensity storms
as a result of climate change (Emanuel et al., 2008). Future assessments should consider future
trends in TC numbers, tracks, and intensities, accounting for these climatological changes in
order to remain valid to a longer period of time.
While the GPD sufficiently captures the distribution of winds in locations with a greater number
of modeled occurrences of winds above the given threshold parameter, the fit lacks in locations
where surface roughness plays a greater role in mitigating TC winds, particularly at the highest
extremes. This could be mended through the combination use of more than one kind of
distribution, similar to the methods as described by Jagger et al. (2010). In theory, it is also
possible to reduce this bias by implementing a separate analysis while still using the GPD,
instead taking a higher threshold for the data analysis (as compared to the 80 kph/50 mph
threshold for this study of the FPLHM). This would allow the GPD to capture only the features
of the upper most extreme portion of the tail, resulting in a better fit for those values alone. One
caveat of that is the reduction in data, which may actually further hinder the analysis depending
on the distribution above that threshold. While the confidence intervals for this analysis leave
only a margin of error of a few miles per hour for the GPD fit, we have only examined a handful
of locations and so it is better practice to ensure a more widely applicable EVD fit rather than
assuming practicality for the GPD at all locations outside of this study area. Since the GPD

46
results suggest some uncertainty in extrapolation over the full range of data, this may be
exacerbated in other locations so alternative EVD options should be considered. This also leads
us to consider the caveats of calculating TRP empirically as was done for Figure 15, versus fitting
data to an EVD as was done for the full extrapolation of continuous TRP values for any
theoretical TC wind speed. Using a stochastic model such as the FPLHM means one is not
confined to the assumptions of a distribution. Therefore, at a sufficiently high number of
modeled records for a given location, estimating TRP can be sufficiently accurate. Though, for
cases where data is insufficient, which may still be the case for estimating TRP at some locations
past a given threshold, it may be better to assume a fitted distribution and extrapolate values
from there. There is no one clear answer for calculating TRP values and so these caveats must be
considered in future studies.
At this time, this effective surface roughness analysis provided by the FPLHM is only
calculated for the domain of the state of Florida and for locations of interest due to its specific
use for insurance purposes. To account for future use it would be beneficial to analyze a larger
number of points within Florida not just because it is one of the areas most frequented by strong
TCs, but also because calculating TRP at these locations, rather than implementing interpolation
techniques, will presumably give more accurate estimates for TC wind speeds in potential areas
of future development not limited to only locations that are already developed. It would also be
highly desirable to test these modeled changes in surface winds as a function of fetch and
effective surface roughness in comparison to observations. Technology has recently been
developed to make such tests in a laboratory setting. This information seems to be critical to
making this assessment. This approach is much more practical the thickly instrumenting and area
that is at relatively high risk of high winds, however, observations will be necessary to reduce
biases in the model. The hardening of such an observing system to withstand the conditions for
which the observations are intended is quite costly.
It may also be beneficial to examine the effect of modes of climate variability on
interdecadal patterns of TRP. Long-term analyses of TRP values at varying wind speeds, such as of
the FPLHM, only provide the statistical likelihood over that entire period, but does not capture
the smaller variations within that time frame. Particularly within the Northern Atlantic basin, the
El Niño Southern Oscillation, Madden Julian Oscillation, and the North Atlantic Oscillation, to
name a few notable contributors, all affect TC genesis and steering patterns over varying time

47
scales (Camargo et al., 2010). Accounting for these climatological differences that vary from
yearly to decadal may reveal patterns of seasons that differ from the overall distribution of TC
winds, effectively increasing or reducing the TRP of expected TC winds within the reference
frame of that particular phenomena’s time scale.

Summary

This study preliminarily examines the HURDAT BestTrack Extended (EBTRK) dataset
(1988-2017) in specific regard to calculating return periods of extreme TC winds within the
Northern Atlantic basin and surrounding landmasses. We suspect a need for higher spatial and
temporal resolution in order to portray more accurate return period projections, which is
confirmed with a gridded analysis of calculated expected wind speeds and corresponding return
period values as fitted to the Generalized Pareto Distribution (GPD). Despite the statistically
significant (𝜎 < 0.05) GPD probability density function fit to EBTRK, the limited data failed to
fully represent the upper tail of extremes wind speed values largely due to a lack of sampling.
This limited analysis resulted in some areas with a high bias towards stronger expected wind
speeds at a given return period value where TC tracks happened to frequent during the 30-year
study period and vice versa for areas over which few TCs occurred, unrelated to direct
climatology. We also come to the conclusion the spatial resolution of EBTRK is too coarse to
capture the smaller scale variability attributed to local terrain differences, which is one of
primary topics of the study in order to expand upon previous research that tends to generalize
return periods to larger geographic regions.
In order to mend some of these problems, we then implement the use of the FPLHM that
stochastically seeds TCs in the model from the PDF of TCs as recorded by EBTRK for the
period of 1900-2016, resulting in a much higher sample of 59,000 modeled years. The FPLHM
also models wind speeds at a resolution of about 1-kilometer, accounting for upstream surface
features within a 90-meter resolution. We narrow down a study area to the Tampa/Clearwater
area in Florida to examine some of the calculated patterns of smaller scale TRP using the same
techniques to analyze the HURDAT BestTrack. Six locations of varying terrain and proximity to
the coastline are chosen to examine the effect of TC winds as a function of these two factors and
fit the PDF of the FPLHM data to a number of EVDs. Again, the GPD remains the most

48
statistically significant (𝜎 < 0.05) at each of the six locations and using that fit, the projected
wind speeds at a number of TRP values are estimated. Results confirm a relationship with features
that correspond to the reduction of TC winds with respect to both upstream surface roughness
and distance inland. These relationships in conjunction with general hurricane climatology
appear to explain a number of features observed with the calculated TRP within notable locations
of the study area.
Around Tampa/Clearwater, offshore and southeastern facing shorelines within Tampa
Bay experience a notably higher trend in expected TC winds. Within 1-kilometer inland of
nearly all coastal locations, projected wind intensity, particularly at higher TRP, reduce
significantly due to shoreline and surface feature interaction. Modeled features with a high
surface roughness coefficient such as skyscrapers within Tampa proper both mitigate expected
TC wind speeds while capping the overall peak expected winds. This phenomenon is observed to
a lesser extent among regions of lower surface roughness. Locations within the agricultural
regions to the south of Tampa/Clearwater with little to no features contributing to surface
roughness are modeled to have higher wind speeds than residential areas in the most urban areas.
Comparisons between land use maps and projected return period values validate qualitatively
expected results and indicate these small-scale features should be accounted for in other return
period analyses.

49
Table 1: Expected wind speeds from a fitted GPD to modeled FPLHM TC winds at varying TRP.
Confidence intervals for 𝜎 < 0.05 for all locations are within ±1 𝑚𝑝ℎ.
10-year RP 50-year RP 100-year RP 250-year RP
2-miles Offshore 93 mph 114 mph 118 mph 124 mph
Coastal Residential 76 mph 92 mph 95 mph 100 mph
(1-km inland)
Bayside Cityscape 68 mph 78 mph 81 mph 83 mph
(1-km inland)
5-miles Inland 73 mph 84 mph 86 mph 90 mph
Residential
15-miles Inland - 81 mph 91 mph 94 mph 96 mph
Wetland Proximity
20- miles Inland 73 mph 80 mph 83 mph 86 mph
Residential

50
Figure 19: Land use map around Tampa/Clearwater by Russell et al. (2013)

51
REFERENCES

Arya, P. S. (1988). Introduction to Micrometeorology. International Geophysics.


doi:10.1016/s0074-6142(08)x6024-3

Bloemendaal, N., de Moel, H., Muis, S., Haigh, I. D., & Aerts, J. C. (2020). Estimation of global
tropical cyclone wind speed probabilities using the STORM dataset. Scientific Data, 7(1).
doi:10.1038/s41597-020-00720-x

Blom, J., & Wartena, L. (1969). The Influence of Changes in Surface Roughness on the
Development of the Turbulent Boundary Layer in the Lower Layers of the Atmosphere.
Journal of the Atmospheric Sciences, 26(2), 255–265. doi:10.1175/1520-
0469(1969)026<0255:tiocis>2.0.co;2

Camargo, S. J., Sobel, A. H., Barnston, A. G., & Klotzbach, P. J. (2010). The Influence of
Natural Climate Variability on Tropical Cyclones, and Seasonal Forecasts of Tropical
Cyclone Activity. World Scientific Series on Asia-Pacific Weather and Climate, 325–360.
doi:10.1142/9789814293488_0011

Carpman, Nicole. (2011). Turbulence Intensity in Complex Environments and its Influence on
Small Wind Turbines.

Christopeit, N. (1994). Estimating parameters of an extreme value distribution by the method of


moments. Journal of Statistical Planning and Inference, 41(2), 173–186.
doi:10.1016/0378-3758(94)90162-7

Elsner, J. B., Liu, K.-biu, & Kocher, B. (2000). Spatial Variations in Major U.S. Hurricane
Activity: Statistics and a Physical Mechanism. Journal of Climate, 13(13), 2293–2305.
doi:10.1175/1520-0442(2000)013<2293:svimus>2.0.co;2

Emanuel, K., & Jagger, T. (2010). On Estimating Hurricane Return Periods. Journal of Applied
Meteorology and Climatology, 49(5), 837–844. doi:10.1175/2009jamc2236.1

Emanuel, K., Sundararajan, R., & Williams, J. (2008). Hurricanes and Global Warming: Results
from Downscaling IPCC AR4 Simulations. Bulletin of the American Meteorological
Society, 89(3), 347–368. doi:10.1175/bams-89-3-347

Garratt, J. R. (1990). The internal boundary layer – A review. Boundary-Layer Meteorology,


50(1-4), 171–203. doi:10.1007/bf00120524

Giammanco, I. M., Schroeder, J. L., & Powell, M. D. (2012). Observed characteristics of tropical
cyclone vertical wind profiles. Wind and Structures An International Journal, 15(1), 65–
86. doi:10.12989/was.2012.15.1.065

52
Hansen, F. V. (1993). Surface Roughness Lengths. Army Research Laboratory.

Hanson, R. E., & Ganapathisubramani, B. (2016). Development of turbulent boundary layers


past a step change in wall roughness. Journal of Fluid Mechanics, 795, 494–523.
doi:10.1017/jfm.2016.213

Historical Hurricane Tracks. https://coast.noaa.gov/hurricanes/#map=4/32/-80. Accessed 3


March 2021

Holland, G. J. (1980). An Analytic Model of the Wind and Pressure Profiles in Hurricanes.
Monthly Weather Review, 108(8), 1212–1218. doi:10.1175/1520-
0493(1980)108<1212:aamotw>2.0.co;2

Holland, G. J., 1984: Tropical cyclone motion: A comparison of theory and observation. J.
Atmos. Sci., 41, 68–75, doi:10.1175/1520-0469(1984)041<0068:TCMACO>2.0.CO;2.

Keim, B. D., Muller, R. A., & Stone, G. W. (2007). Spatiotemporal Patterns and Return Periods
of Tropical Storm and Hurricane Strikes from Texas to Maine. Journal of Climate, 20(14),
3498–3509. doi:10.1175/jcli4187.1

Kristensen, L., Casanova, M., Courtney, M. S., & Troen, I. (1991). In search of a gust definition.
Boundary-Layer Meteorology, 55(1-2), 91–107. doi:10.1007/bf00119328

Mouche, A., Chapron, B., Knaff, J., Zhao, Y., Zhang, B., & Combot, C. (2019). Copolarized and
Cross-Polarized SAR Measurements for High-Resolution Description of Major Hurricane
Wind Structures: Application to Irma Category 5 Hurricane. Journal of Geophysical
Research: Oceans, 124(6), 3905–3922. doi:10.1029/2019jc015056

NOAA National Centers for Environmental Information (NCEI) U.S. Billion-Dollar Weather
and Climate Disasters (2021). https://www.ncdc.noaa.gov/billions/, DOI: 10.25921/stkw-
7w73

Powell, M. D., & Houston, S. H. (1996). Hurricane Andrew's Landfall in South Florida. Part II:
Surface Wind Fields and Potential Real-Time Applications. Weather and Forecasting,
11(3), 329–349. doi:10.1175/1520-0434(1996)011<0329:halisf>2.0.co;2

Powell, M. D., Soukup, G., Cocke, S., Gulati, S., Morisseau-Leroy, N., Hamid, S., et al.
(2005).State of Florida Hurricane Loss Projection Model: Atmospheric Science
Component. Journal of Wind Engineering and Industrial Aerodynamics, 93, 651-674.

Raupach, M. R., & Thom, A. S. (1981). Turbulence in and above Plant Canopies. Annual Review
of Fluid Mechanics, 13(1), 97–129. doi:10.1146/annurev.fl.13.010181.000525

Robertson, I. N., Riggs, H. R., Yim, S. C., & Young, Y. L. (2007). Lessons from Hurricane
Katrina Storm Surge on Bridges and Buildings. Journal of Waterway, Port, Coastal, and
Ocean Engineering, 133(6), 463–483. doi:10.1061/(asce)0733-950x(2007)133:6(463)

53
Rotach, M. W. (1993). Turbulence close to a rough urban surface part I: Reynolds stress.
Boundary-Layer Meteorology, 65(1-2), 1–28. doi:10.1007/bf00708816

Russell, M., A. Teague, F. Alvarez, D. Dantin, M. Osland, J. Harvey, J. Nestlerode, J. Rogers, L.


Jackson, D. Pilant, F. Genthner, M. Lewis, A. Spivak, M. Harwell, and A. Neale. 2013.
Neighborhood scale quantification of ecosystem goods and services. U.S. Environmental
Protection Agency, Office of Research and Development, Gulf Ecology Division, Gulf
Breeze, Florida. EPA/600/R-13/295. November 2013

Shay, L. K., Goni, G. J., & Black, P. G. (2000). Effects of a Warm Oceanic Feature on Hurricane
Opal. Monthly Weather Review, 128(5), 1366–1383. doi:10.1175/1520-
0493(2000)128<1366:eoawof>2.0.co;2

Taylor, P. A. (1969). The Planetary Boundary Layer above a Change in Surface Roughness.
Journal of the Atmospheric Sciences, 26(3), 432–440. doi:10.1175/1520-
0469(1969)026<0432:tpblaa>2.0.co;2

Vickery, P. J., & Skerlj, P. F. (2005). Hurricane Gust Factors Revisited. Journal of Structural
Engineering, 131(5), 825–832. doi:10.1061/(asce)0733-9445(2005)131:5(825)

Wong, M. L., & Chan, J. C. (2007). Modeling the Effects of Land–Sea Roughness Contrast on
Tropical Cyclone Winds. Journal of the Atmospheric Sciences, 64(9), 3249–3264.
doi:10.1175/jas4027.1

54
BIOGRAPHICAL SKETCH

Walker Whitfield grew up in Chapel Hill, North Carolina, where his interest in
meteorology was spurred through the wide range of weather events that impact this region from
winter storms to tropical cyclones. Being that he found meteorology so enjoyable and saw it as
nothing more than just a side hobby, he did not consider a career in the field until after beginning
his undergraduate career at University of North Carolina – Wilmington, where he had intended
to study marine biology. While taking an elective weather and climate course, the professor
noted Walker’s inherent interest and knowledge of the subject, leading the professor to reach out
to him about his thoughts on a path towards studying meteorology. A couple of meetings later,
he had switched majors and graduated four years after with a degree in geography and a minor in
mathematics, setting him up for meteorology in graduate school.
He now attends Florida State University where he is pursuing a master’s degree in
meteorology, which also opened the door to other academic opportunities such as interning at
NASA-JPL. There, he spent a summer working on updating databases of tropical cyclone wind
and precipitation measurements for NASA-JPL’s Tropical Cyclone Data Archive. Based on his
interest in extreme weather such as tropical cyclones, particularly as they pertain to real-world
impacts, he plans to pursue a career in topics related to risk and risk management.

55

You might also like