don resio, senior scientist, us army erdc-chl data needs for coastal storm surge estimation ibtracs...
DESCRIPTION
Two Major Types of Data Needs Forecasts: Go/No-Go Evacuation Decision - long lead time (3-5 days) - must be conservative - uncertainty “factored in” Storm-Approach Operations - time-phased information - late evacuation routing - gate/spillway decisions - uncertainty quantified Post-Storm Operations - time-phased information - accurate damage assessment - accurate systems assessment - critical recovery decisions Planning/Risk Mitigation: Accurate Hazard Climatology - consistent data set - long period of record - uncertainty quantified - climatic variability Accurate Response Specification - human response - system response Quantified Risk/Alternatives - time-phased options **** - climatic variability - uncertainty quantified - Objective “cost” estimates Current forecast systems focus on this problemTRANSCRIPT
Don Resio, Senior Scientist, US Army ERDC-CHL
Data Needs for Coastal Storm Surge Estimation
IbTrACS MeetingHonolulu, HI
April 12, 2011
Outline of Talk
• Beyond “evacuation forecasting” and climate change
• Some existing needs• Coastal Flood Protection & Risk Estimates (USACE)• Flood Insurance Rates (FEMA)• Licensing of sites for nuclear power plants (NRC)
• Sufficient information for surge risk estimation
Two Major Types of Data NeedsForecasts:
• Go/No-Go Evacuation Decision- long lead time (3-5 days)- must be conservative- uncertainty “factored in”
• Storm-Approach Operations- time-phased information- late evacuation routing- gate/spillway decisions- uncertainty quantified
• Post-Storm Operations- time-phased information- accurate damage assessment- accurate systems assessment- critical recovery decisions
Planning/Risk Mitigation:
• Accurate Hazard Climatology- consistent data set- long period of record- uncertainty quantified- climatic variability
• Accurate Response Specification- human response- system response
• Quantified Risk/Alternatives - time-phased options ****- climatic variability- uncertainty quantified- Objective “cost” estimates
Current forecast systems focus on this problem
Climate Variability
• Whereas tropical storm frequency alone may be key to understanding and quantifying some aspects of climate change, major impacts may relate more to and variations in storm surge potential
Tropical Storms and Hurricanes(23 = $334B K=$133B)
Heat/Drought(15 = $130B)
Interior Flooding(15 = $74B)
Severe Weather(19 = $31B)
Fires (10 = $19B )
Blizzard (2 = $12B)
Ice (2 = $6B)
Northeasters(1 for $2B)
Gulf Oil Spill (2010) 9/11 Trade Center $20 – $40B $20 - $30BThree Mile Island $1B
EXTREMES AREWHAT CREATES
DISASTERS
Three steps in understanding/quantifying risk :1. Hazard identification/quantification2. Vulnerability Assessment3. Risk Analysis
Storm frequency is insufficient to quantify surge levels at a coast.
For example:
Storm size is critical to storm surges along a coast
No “Cat 5” design for New Orleans!
For single storms we typically use For single storms we typically use HWIND-type of products to force models HWIND-type of products to force models
Reconstructed Winds, 1200 UTC 29
Aug 2005
HWINDS
But for Hazard/Risk Assessment, we need statistics of 6 basic parameters
max 0( , ) ( , | , , , , , , ( ), )
where( , ) is the storm surge at location x and time t, is a numerical model used to generate surges over a grid, is a time invariant grid of bathymetry/topogr
p fx t G W c R v B x S t t
x t
G
max
aphy, is a wind field over the grid at time t, is the central pressure,
is the radius to maximum wind speed from the center of the storm, is the forward velocity of the storm,
is the geograp
p
f
Wc
Rv
0
hic angle of the track, is the Holland "B" parameter, is the landfall location, and
( ) is the position of the storm along the track at time t,
BxS t
General form for surge response at location x and time t:
Bottom line: There are 5 important storm parameters plus storm track and the changes in near-coast storms that have to be considered in JPM.
LAND WATER
Based on data fromOceanweather, decayduring approach to land is about the sameas post-landfall decay.
BUT: Variations near a coast must also be recognized/estimated.
Relationship is specific to northern GOM
Similar results in the time domain have been found by NHC in its own studies.
In addition to the necessity of properly modeling physical phenomena for critical events –
It is also necessary to understand and quantify their probabilities.
To accomplish this, we must accurately specify all of the myriad forcing processes within a probabilistic framework.
Statistical Approach – Fitting the Modeling into RiskJPM with Optimal Sampling (JPM-OS)
2
2
1 2 3 4 5
0 1 01
1
( ( ) )
2 ( )2
( (
3
( , , , , )
[ ( ), ( )] ( )( | ) exp exp (Gumbel Distribution)( | ) ( )
1( | )( ) 2
1( | )2
p p
f l
p p f l
pp
R P R
Pp p
v
f l
p c R v x
F a x a x p a xp c xp c x a x
p R c eP
p v e
2
2
2
2
) )
2
( ( ) )2 ( )
4
5
1( | )( ) 2
( )
f
l l
v
xx
lp x ex
x
Joint probability matrix:
All the surges are put into appropriate bins.Different storms can produce results in the same bin.
, , , , *(| | / 2)i j k l mp
Must include uncertainty!!• Uncertainty is model/system specific• Reduction in uncertainty reduces cost• Adding uncertainty increases cost
Uncertainty added in each mapping
1 2 1 2 1( ) ( , ,..., ) [ ( , ,..., ) ] ...
is a parameter affecting hurricane surge levels(.) is the pdf is the surge level is an analytical operator (modeling system) that coverts
n n n
i
p p x x x x x x dx dx
xp
i a specific set of values of x to a surge
1 2 1 2 1 2( ) ... ( , ,..., ) [ ( , ,..., )] ...n n nF p x x x H x x x dx dx dx
In any dimension we have for the pdf the ability to map from an n-dimensional space into a 1-dimensional space via a Dirac delta-function δ
And the CDF which uses the Heaviside Function (an integral of the delta function)
1 2 1 2 1 2( ) ... ( , ,..., , ) [ ( , ,..., ) ] ...
is the uncertainty in the surge level from the modelingn n nF p x x x H x x x dx dx dx d
Since we remain imperfect, we need to consider an error term also!!!!
This means that we can leave some degree of randomness in our solutions – as long as we can estimate the statistical characterization of this term – which also includes tides, wind field errors, errors in physics, other omissions, etc.
The expected return period can be estimated from the CDF via the assumption that the storm occurrence is governed by a stationary Poisson process, with an average frequency of occurrence of λ.
1( )[1 ( )]
TF
Unfortunately, nature often deviates from this simplistic assumptions – with years containing many storms not following the same distribution as years with few storms
Present Status• Insufficient data exists in “public” data
sources• Data from Oceanweather and Applied
Research Associates is used for driving surge models
• Updates/improvements of these data are difficult to accomplish
• An extension to IbTrACS would go a long way to help improve our ability to estimate hazards/risks worldwide
QUESTIONS??