Hurricane Risk: Present and Future

1
Hurricane Risk Present and Future

• Kerry Emanuel
• Department of Earth, Atmospheric and Planetary
  Sciences, MIT

With special thanks to Sai Ravela, Emmanuel
Vivant, and Camille Risi
2
Hurricane Risk

• Tropical cyclones account for the bulk of natural
  catastrophe U.S. insurance losses
• Risk assessment is vital to the insurance
  industry and to government disaster preparedness
  programs
• Losses vary roughly as the cube of the maximum
  wind speed
• Katrina caused gt 1300 deaths and gt 130 billion
  in damage

3
Source Roger Pielke, Jr.
4
(No Transcript)
5
Source Roger Pielke, Jr.
6
Cumulative Distribution of Landfall Frequency by
Wind Speed
7
(No Transcript)
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Summary of U.S. Hurricane Damage Statistics

• gt50 of all damage caused by top 5 events, all
  category 4 and 5
• gt90 of all damage caused by storms of category 3
  and greater
• Category 3,4 and 5 events are only 13 of total
  landfalling events only 30 since 1870
• Landfalling storm statistics are grossly
  inadequate for assessing hurricane risk

16
Current Methods of Hurricane Risk Assessment

• Fit standard (e.. Weibull) distribution functions
  to peaks winds within a specified radius of point
  of interest, taken from historical hurricane data
  (Georgiou et al, 1983 Neumann, 1987)
• Find universal distribution functions of wind
  normalized by potential intensity and interpolate
  to specific locations based on historical
  frequency (Darling, 1991 Chu and Wang, 1998)

17

• Generate large database of synthetic storm tracks
  using previous track history and local
  climatology couple to historical intensity data
  (Vickery et al., 2000)

18
Can We Use Physics to Improve Hurricane Risk
Assessment?
19
Our Approach

• Step 1 Use two largely independent techniques to
  generate large (104) numbers of synthetic
  tropical cyclone tracks passing within specified
  radius of point of interest
• Step 2 Run a deterministic coupled tropical
  cyclone intensity model along each synthetic
  track
• Step 3 Directly deduce wind speed exceedence
  probabilities at point of interest

20
Genesis PDFs based on post-1970 historical tracks
21
Synthetic Track Generation,Method 1 Markov
Chains

• Tracks initiated by random draws from space-time
  PDF based on historical genesis data smoothed
  using a three-dimensional Gaussian kernel

22

• Tracks propagated in 6-hour steps by integrating
  in time the rates of change of direction and
  speed, by randomly drawing from the probability
  distribution

Note Probabilities of rates of change based on
previous direction and speed, not their rates of
change. Only last 6-hour step used.
23
Results
60 Markov tracks
60 HURDAT tracks
24
6-hour zonal displacements in region bounded by
10o and 30o N latitude, and 80o and 30o W
longitude
25
6-hour meridional displacements in region bounded
by 10o and 30o N latitude, and 80o and 30o W
longitude
26
Synthetic Track Generation,Method 2 Use of
Synthetic Wind Time Series

• Use genesis technique as in Method 1
• Postulate that TCs move with vertically averaged
  environmental flow plus a beta drift correction
  (Beta and Advection Model, or BAMS)
• Approximate vertically averaged by weighted
  mean of 850 and 250 hPa flow

27
Synthetic wind time series

• Monthly mean, variances and co-variances from
  NCEP re-analysis data
• Synthetic time series constrained to have the
  correct mean, variance, co-variances and an
  power series

28
250 hPa zonal wind modeled as Fourier series in
time with random phase
where T is a time scale corresponding to the
period of the lowest frequency wave in the
series, N is the total number of waves retained,
and is, for each n, a random number
between 0 and 1.
29
The time series of other flow components
or
where each Fi has a different random phase, and A
satisfies
where COV is the symmetric matrix containing the
variances and covariances of the flow components.
30
Example
31
Track
Empirically determined constants
32
Results
33
6-hour zonal displacements in region bounded by
10o and 30o N latitude, and 80o and 30o W
longitude
34
6-hour zonal displacements in region bounded by
10o and 30o N latitude, and 80o and 30o W
longitude, using only post-1970 hurricane data
35
Tropical Cyclone Intensity

• Run coupled deterministic model (CHIPS, Emanuel
  et al., 2004) along each track
• Use monthly mean potential intensity, ocean mixed
  layer depth, and sub-mixed layer thermal
  stratification
• Use shear from synthetic wind time series
• Initial intensity and rate of intensification
  specified as and
• Tracks terminated when v lt

36
Coupled Hurricane Intensity Prediction System
(CHIPS)

• Operational at JTWC
• Unique initialization based on entire storm
  history
• Major improvements in late 2005

37
Southern Hemisphere, 2005-1 April 2006
38
Results
39
Miami
HURDAT 29 events Method 1 3000 events Method 2
2997 events
40
Boston
HURDAT 28 events Method 2 3000 events
41
Boston, worst case of 3000 affecting downtown
Boston, Method 2
42
Washington, D.C.
43
15 worst wind events in Washington
44
Worst wind event of 3000 affecting Washington,
D.C.
45
(No Transcript)
46
Effect of Climate Change on Hurricane Risk
47
Hurricane Intensity Scales with Potential
Intensity The Maximum Intensity Achievable
Based on the Thermodynamic Cycle of the Mature
Hurricane
48
Theory Maximum Storm Intensity Increases with
Tropical SST (Emanuel, Nature, 1987)
49
Global warming scenario Increase potential
intensity by 10
50
Number of events V3
Integrated power dissipation increases 80
51
Annual Storm Count and SST
52
How has Hurricane Activity Actually Changed?
Measure of total amount of energy released by a
hurricane over its life The Power Dissipation
Index, or PDI
53
Global Annual Mean Surface Temperature vs. Global
PDI
54
Northwest Pacific
55
Power dissipation index is highly correlated with
SST
56
Especially true in modern instrumental
record Smoothed with 1-3-4-3-1
Filter
57
Note Correlation of smoothed PDI with smoothed
SST not appreciably improved by including Aug-Oct
MDR vertical wind shear. (1949-2005 r2 increases
from .75 to .8)
58
Decadal Perspective
59
10-year Running Average of Aug-Oct NH Surface T
and MDR SST
60
Do PDI Trends Imply Greater Damage?
Annually accumulated v3 at landfall, U.S. East
and Gulf coasts
61
Note Open Ocean PDI uses 80 times more data
than landfall PDI. Can we detect an existing
trend in landfall PDI? Method Holding Atlantic
genesis rates fixed at 10 - 3.5 per year, draw
randomly from present and increased PDI synthetic
distributions, linearly transitioning from one to
the other over 53 years.
62
Example
X 107
X 105
Total PDI
Landfall PDI
63
Run algorithm 100 times and for each find the
regression slope and correlation coefficient.
Define a simplified slope as f(2003)-f(1950)/f(
1950) where f is the linear regression line.
Result LPDI mean slope 2.77 standard
deviation 12.67 PDI mean slope 0.69
standard deviation 0.41. Observed increase of
PDI 0.75. Conclusion PDI trend barely
detectable, LPDI trend not detectable
64
Summary

• Stochastic hurricane tracks coupled with
  deterministic intensity models are effective for
  assessing hurricane risk
• Modest increases in potential intensity portend
  greatly increased hurricane destructive potential
• Global tropical cyclone power dissipation has
  nearly doubled in the last 30 years
• Little implications for U.S. hurricane damage for
  next 50 years more implications for global
  damage

65
Future Work

• Account for natural variability of upper ocean
  thermal structure
• Bin re-analysis statistics by phase of Natural
  Oscillations such as ENSO
• Use GCM output to estimate tracks/intensity in
  global warming scenarios (genesis PDFs
  problematic)

66
Radial structure
Translation speed added to circular
theoretical/empirical wind field given by
rm and Vm given by intensity model
67
(No Transcript)
68
Miami, 30 random tracks, Method 2
69
Miami, 30 HURDAT
70
Miami, 30 worst tracks, Method 2
71
Miami, 30 worst tracks, Method 1
72
Miami, worst of 3000 storms, Method 2
73
Boston, 30 random tracks, Method 2
74
Boston, 30 HURDAT tracks
75
Boston, 30 worst tracks, Method 2
76
New York City
HURDAT 20 events Method 1 2985 events Method 2
3059 events
77
New York, worst event, Method 1
78
(No Transcript)
79
Comparison of MethodsMethod 1

• Advantages
• Excellent track statistics
• Nonlinear effects of extratropical transition
  accounted for
• Disadvantages
• Tracks good only where historical data plentiful
• Shear largely independent of track
• High latitude track statistics biased by
  surviving events Overestimate of high latitude
  frequency (need a death algorithm)

80
Comparison of MethodsMethod 2

• Advantages
• Only depends on historical genesis data can be
  run anywhere
• Tracks consistent with shear
• Could potentially account for ENSO, NAO, PDO,
  etc.
• Disadvantages
• Does not account for nonlinear processes in
  extratropical transition Underestimates extreme
  events at high latitude

81
Appendix A few other cities
82
Halifax
HURDAT 32 events Method 1 3115 events Method 2
3580 events
83
Halifax 20 random tracks, Method 2
84
Halifax 20 worst tracks, Method 2
85
Worst event, Method 2
86
(No Transcript)
87
Worst event, Method 1
88
San Diego
89
Top 30 tracks
90
Worst event, Method 2
91
Honolulu
92
Honolulu, 30 top tracks
93
Honolulu, worst event, Method 2