EXPLORATORY ANALYSIS &
INTENSIFICATION (AND DECAY) RATES
EXAMINING INTENSITY TRENDS OF TROPICAL CYCLONES IMPACTING THE LOUISIANA DELTAIC REGION:
AN EXTREME VALUE ANALYSIS APPROACH
Savannah Collins-Key, University of Tennessee, Knoxville
Enhanced understanding of future interannual and seasonal variability
risk of tropical cyclone hazards in hurricane-prone regions is crucial in an era of
fluctuating climate. Tropical cyclones (TCs) are massive, multi-hazard systems
which constitute an annual threat to millions of people living in coastal
communities. As the intensity of a hurricane determines its damage potential,
appropriately demonstrating the threat level of extreme hurricanes is essential
for instigating effective coastal hazard planning and policy. Southeastern
Louisiana is a frequent target for TCs, but is also vulnerable to other types of
flood events due to increasing population growth, subsidence, and low elevation.
Delineating the stochasticity of tropical cyclogenesis and defining features of
a TC during its lifetime is pertinent to hurricane climate prediction. A widely-
embraced technique in the applied sciences is extreme value analysis, which
quantifies the probability of a high-intensity or ‘extreme’ event owing to
stochastic processes. Multiple studies have endeavored to project return periods
for TC events of var ying intensity (though mostly extreme) along the Gulf Coast.
The intent of this study is to model hurricane intensity temporally over the
southeastern Louisiana flood delta and evaluate whether return periods can be
projected into the mid-21st century using a satellite-era dataset that includes the
2017 North Atlantic hurricane season.
STUDY SITE & DATA
INSTABILITY IN THE RECORD
“Why did we rush things? We’re so interested in people, as they plan to rebuild and in the wake of these
terrible disasters, not use the past as a guide to the future. [Perhaps] the past is no longer, in many
respects, a very go od guide to the future.”
Kerry Emanuel, on the swiftness of publications following 2017 Atlantic hurricane
season, AMS Annual Meeting, January 2018
Placing emphasis on the tail distribution of the updated dataset is imperative to
determining levels of risk in these areas and how risk could be changing over time.
The risk v alue affiliated w ith the tail weig ht is why we focu s on extremes: al though in
essence they are ‘extreme’ in their appearance, these events account for the
greatest amount of loss. The values of the tail are deemed ‘low frequency, high
severity’: they don’t happen often—yet when they do, they can be devastating.
Therefore, it is es sential to establish the likelihood the se extremes could occ ur and
the magnitude level they could attain. Additionally, climate change may be
influencing the recurrence frequencies of extreme events. In records of paleofloods,
the frequenc y of large flood events rose and fell parallel to the changing climate
conditions. To ignore the issue of non-stationarity in return period estimates of TC
magnitude would significantly reduce the accuracy of projected future return
Hurricane trajectories within study area for the period 1967–2017.
DEFINING HURRICANE INTENSITY
Hurricane intensity is represented by the maximum wind speed detected within a
tropical cyclone, typically occurring just within the eyewall of the storm and less
than 60% of the forward (translational) speed (Landsea et al. 2006). Storm
strength has also been measured by storm surge height, storm size, central
pressure of the storm, and the total energy (real and potential) estimated to
have been discharged from the system over its lifetime (Ellis et al. 2014).
However, maximum sustained wind speeds (MSW) are highly correlated with
several destructive aspects of a hurricane, included all of the above
characteristics. The maximum wind speed observed over the life cycle span of a
TC, i.e. f rom ge nes is to dis sip ati on, is e xpr ess ed as the l ife time m axi mum in ten sit y
(LMI) (Wal sh et al. 2015). The LMI and intensity rate (i.e. how the storm reached
LMI then decayed from that point) are features focused upon in this study.
The a priori knowledge of the
thermodynamic systems which largely
influence TC intensity allows
researchers better assess intensity
trends, although the nature of
dynamical atmospheric processes that
dominate TC frequency is less well-
established (Vec chi et al. 2008; Fraz a et
al. 2016). Furthermore, relatively quiet
hurricane seasons are still capable of
producing highly destructive and lethal
storms (Holland 2007; Hart et al. 2016).
WHY SOUTHEASTERN LOUISIANA?
In a recent study published by The New York
Times on the cost of natural disasters from 2002–2017,
11% of all losses occurred in 10 zip codes near
New Orleans (Chinoy 2018).
Another study explored
effectively inundated communities, or districts where at 10% of useable land is
flooded at least 26 times annually; out of the 91 of these communities found in
the US, over half are located solely in Louisiana (Dahl et al. 2017).
Coastal Louisiana also experiences one of the highest
rates of relative sea-level rise (RSLR) in the world,
at 12 (+/– 8) mm per year (Jankowski et al. 2017).
The Gulf C oast is also believed to exp erience tropical
cyclone patterns that differ from temporal trends of other
North Atlantic regions (Cohen 2010). One study observed
over the e ntire TC record (which begin s in 1851) up to 2006,
50 hurricanes have made landfall in southeastern Louisiana;
19 of these storms were considered major (≥ Cat 3)
hurricanes (Chapman et al. 2008). The Pearl River Basin has
observed the landfall of several of these storms: the basin
experienced 28 TCs just in the last two decades, including
Hurricane Katrina in 2005, which made its third (and final)
landfall at the mouth of the Pearl River (Stone & Cohen 2017).
The National Hurricane Center
(NHC) Hurricane Research Division’s
Hurricane Database (HURDAT2) is a
continually updated best-track
dataset of past tropical cyclones,
both observed within the satellite era
and detected through historical
records. HURDAT2 displays TC
observations at 6-hour intervals,
which is routine in meteorological
forecasts (Jarvinen et al. 1984).
Major TCs show potential for
displaying TC-force wind speeds up to
360 km from storm center, although the
maximum sustained winds (MSW) more
than likely occurred earlier in the TC
life cycle before landfall (Keim et al.
2007). A buffer of 300 km was chosen to
include most of the Mississippi deltaic
region, with the Pearl River Wildlife
Management Area as the center point.
The period 1967–2017 saw 59 TCs co me
within the 300-km buffer zone.
A precursory extreme value analysis
was run on data obtained from the
NHC that included tropical cyclones
impacting the Louisiana delta during
the period 1967–2017. Initial results
indicate a shift in return periods as well
as changes in the intensification/decay
rates of storms. The objectives of the
proposed study are to 1) employ a
quantile regression function to
evalua te how th e time variabl e
influences the intensity values of SE
Louisiana landfalling TCs; 2) estimate
the annual probability of a major
hurricane striking SE Louisiana using
extreme value analysis; and 3)
determine potential increasing or
decreasing trends in hurricane intensity.
Model fit of LMI da ta
The histogram rug for the LMI values
shows a break in data, contributing to a
distinct bimodal distribution. Previous
studies have also observed bimodality in
the distributions of datasets depicting
global LMI of t ropical cyclon es. This
“secondary maximum” is usually distinct
and suggests that what we consider
major hurricanes may not be as
uncommon as we formerly assumed. Lee
et al. (2016) presents the process of
rapid intensification (RI) as the factor
producing the bimodal tendency.
A quantile regression is employed for the
exploratory analysis. A semi-parametric
technique, it assesses multiple covariate
relationships as well as extreme events in
the dataset, but relies on quantiles, thus it
isn’t held to rigid assumptions and displays
less sensitivity toward extreme values
(Elsner & Villarini 2011).
The qua ntile regression mo del can ref lect the sp ecifics of the relationship that we are
evalua ting by a nalyzing for condi tional qu antiles a s opposed to a con ditional mean, as
is such in an OLS analysis, where estimating the mean of the response variable Y is
conditional on the explanatory variable X.
The rates of intensification/decay were produced with the Savitzky-Golay smoothing
filter, and depict the estimated growth/weakening of each storm for every observed 6
hr interval. Computed as a derivative value, the rate of change is found through the
first derivative of intensity using the smoothing filter and a weighted differencing
scheme, while preserving the maximum and minimum intensity values (Savitzky & Golay
1964). A 3rd-degree polynomial regression approximates the smoothed wind speed
value at a given location; this polynomial is capable of exhibiting the most variability
without overfitting the filter (Elsner & Jagger 2013).
The highest intensification and decay rates along
the track of each TC are then obtained, so one
maximum and one minimum value represent the
storm system. This is accomplished for two datasets
organized by each storm’s utmost rate of
strengthening (values above zero) and rate of
weakening (values below zero) (Malmstadt et al.
2010). Decay rate is important to assess, as some
studies have detected a decrease in wind speeds in
the final twelve (12) hours before making landfall.
The 50th, 75th, and 90th percentile trends for both
intensification and decay rates are listed in Tab le 1.
POTENTIAL RETURN PERIODS
Keim et al. (2007) calculated return periods for the coastline region encompassing
New Orleans as 3 years for a TC of tropical storm or hurricane intensity; 7 years for
a hurricane; and 26 years for a major (≥ Cat 3 or wind speeds of 50 m s–1) hurricane.
More recently, Tr e pa n ie r & S c he i tl i n (2014) calculated a
shorter return period of 21 years for a tropical cyclone
experiencing wind speeds of 50 meters per second (m s–1)
—projecting an event of this magnitude will occur in New
Orleans every 21 years. New Orleans was found to have
the highest risk for a 100-yr return period compared to
several other cities along the Louisiana coastline,
expecting a TC event with wind speeds of at least 64 m s–1.
Overall, results suggest the most extreme TC events
are becoming slightly stronger over time, and the
subsequent return-level calculations for the longest
return periods may prove to be somewhat
conservative; these results show that, as time goes
on, values may become more extreme compared to
past studies (Elsner et al. 2008; Knutson et al. 2010).
Elsner JB & Jagger TH (2013) Hurricane Climatology: A Modern Statistical Guide Using R. Oxford University Press, New York.
Elsner JB, Kossin JP, & Jagger TH (2008) The increasing intensity of the strongest tropical cyclones. Nature 455(7209): 92–95.
Fraza E, Elsner JB, & Jagger TH (2016) A space-time statistical climate model for hurricane intensiﬁcation in the North Atlantic basin. Advances in Statistical Climatology, Meteorology and Oceanography 2: 105–114.
Hart RE, Chavas DR, & Guishard MP (2016) The Arbitrary Deﬁnition of the Current Atlantic Major Hurricane Landfall Drought. Bulletin of the American Meteorological Society 97(5): 713–722.
Keim BD, Muller RA, & Stone GW (2007) Spatiotemporal Patterns and Return Periods of Tropical Storm and Hurricane Strikes from Texas to Maine. Journal of Climate 20(14): 3498–3509.
Knutson TR, McBride JL, Chan J, Emanuel K, Holland G, Landsea C, Held I, Kossin JP, Srivastava AK, & Sugi M (2010) Tropical cyclones and climate change. Nature Geoscience 3(3): 157–163.
Landsea CW, Harper BA, Hoarau K, & Knaff JA (2006) Can We Detect Trends in Extreme Tropical Cyclones? Science 313(5786): 452–454.
Lee C-Y, Tippett MK, Sobel AH, & Camargo SJ (2016) Rapid intensiﬁcation and the bimodal distribution of tropical cyclone intensity. Nature Communications 7: 10625.
Malmstadt JC, Elsner JB, & Jagger TH (2010) Frequency and Intensity of Hurricanes Within Florida’s Threat Zone. In: Elsner JB, Hodges RE, Malmstadt JC, & Scheitlin KN (eds) Hurricanes and Climate Change: Volume 2.
Springer, London: 191–203.
Savitzky A & Golay MJE (1964) Smoothing and differentiation of data by simpliﬁed least squares procedures. Analytical Chemistry 36(8): 1627–1639.
Trepanier JC & Scheitlin KN (2014) Hurricane wind risk in Louisiana. Natural Hazards 70(2): 1181–1195.
Trepanier JC, Ellis KN, & Tucker CS (2015) Hurricane Risk Variability along the Gulf of Mexico Coastline. PLoS ONE 10(3): e0118196.
Walsh KJE, McBride JL, Klotzbach PJ, Balachandran S, Camargo SJ, Holland G, Knutson TR, Kossin JP, Lee T, Sobel A, & Sugi M (2015) Tropical cyclones and climate change. WIREs Climate Change 7(1): 65–89.
Refe re nces