You are on page 1of 5

Supplement: Statistical Methods in Coastal Engineering – Determining the Design

Condition

Short-term hours
Long-term >20 years

Long-term Methods
• Typically the largest value of a parameter is the primary design concern
• Extreme events are highly variable in intensity and sequencing
• Extreme events are rare, but may occurs as chance events during shorter than
predicted periods (i.e. a 100-yr storm may occur within a 50 year period of collection)

Long-term statistical methods must:


1. Extend available information to longer time periods… i.e. use 10 years of
collected data to estimate the 25-year extreme
2. Use a small sample of extreme events to get unbiased estimates of extremes and
some measure of confidence or variability

Basic Methods:
• Historical Method – use historical data to generate recurrence interval
relationships
o graphical approach… construct a cumulative probability density function
(or probability density function) where event magnitude is plotted against
the probability of exceedance.
o analytical approach… select a pdf based on the behavior of the data and
the purpose of the study
• Synthetic Method – a particular event is characterized by several descriptive
components or parameters whose probability can be used to determine a joint
probability of the total event
o JPM (Joint Probability Method) – parameters are independent and the joint
probability is the product of the individual probabilities.
• Empirical Simulation Technique (II-5-47) – statistical resampling scheme that
develops joint probability relationships among various parameters using historical
data
o fewer assumptions than JPM
o site specific (needs historical data)
o directly computed parameter interdependence
Historical Method
Graphical Approach
• Requires a large data set (thumb rule: at least T/2 years of data)
• Plot the event magnitude versus the probability of exceeding that magnitude (i.e. cdf)

Pr{X ≥ xn} ≡ Probability that X will be greater than or equal to a given value (xn)
# times X ≥ x n # times X ≤ x n
Pr{X ≥ x n } = = 1−
total # of X total # of X
T(xn) ≡ The return interval for event xn

Pr{X ≤ x n } = F(x n )
Pr{X ≥ x n } = 1 − F(x n )
where F(x) is the cumulative probability density function (cdf) and n

1
T( x n ) =
1 − F( x n )

Case 1:
• Have ordered sample: y1 < y2 < … < yn (no repeats)
• N = total # of data points
• tm = observation period

• if Y ={ y1, y2, … yn} is the ordered sample, then the probability that any y will be
greater than yn will be 1/n (i.e. only yn ≥ yn)
• likewise Pr{Y ≥ yi} = (n-i+1)/n or Pr{Y > yi} = (n-i)/n and F(yn) = 1 – 1/n

1. Plot ln(n) vs. ln(xn)


2. calculate the number of observations (M) in a given period (T): M = T × tm
3. Extend plot to desired ln(M) (extrapolate with straight line)
4. For desired M, find xM on plot

ln(M)

ln(n) End of data set

ln(yn) ln(yM)
Case 2:
• Sample is not ordered Æ alternate method for determining the cdf (F)
• N = total # of data points or observations
• tm = observation period

1. Construct Histogram distribution plot for the data set, X


fi is the number of occurrences in each bin (range), Y is the mid-points of the bins
n

∑f i
2. F(y n ) = i =1

N
 1 
3. plot ln   vs. ln(yn)
1 − F(y n ) 
4. calculate the number of observations (M) in a given period (T): M = T × tm
5. Extend plot to desired ln(M) (extrapolate with straight line)
6. For desired M, find yM on plot
Example: Estimate the 50-year extreme value from the following 3 years of data.

Sea
bin mid- F(yn) =  1 
Severity # of obs ln(yn) ln  
1 − F(y n ) 
point (yn) Σfi/N
(m)
0-0.5 0.25 1280 0.237 -1.386 0.270
0.5-1 0.75 1549 0.523 -0.288 0.740
1-1.5 1.25 1088 0.724 0.223 1.286
1.5-2 1.75 628 0.840 0.560 1.831
2-2.5 2.25 402 0.914 0.811 2.454
2.5-3 2.75 192 0.950 1.012 2.987
3-3.5 3.25 115 0.971 1.179 3.534
3.5-4 3.75 63 0.982 1.322 4.042
4-4.5 4.25 38 0.989 1.447 4.553
4.5-5 4.75 18 0.993 1.558 4.933
5-5.5 5.25 21 0.997 1.658 5.706
5.5-6 5.75 7 0.998 1.749 6.198
6-6.5 6.25 8 0.999 1.833 7.498
6.5-7 6.75 2 1.000 1.910 8.596
7-7.5 7.25 1 1
Σ= 5412

T = 50 years
m = 5412/3 years = 1804
M = 50 × 1804 = 90,200
ln(90,200) = 11.4
from plot: ln(yn) = 2.1 Æ yn = 8.2 m
12
11 ln(50 years = 90,200) = 11.4
10

9
8
7 ln(y n) = 2.1
ln[1/(1-F)]

5
4
3

2
1
0
0.0 0.5 1.0 1.5 2.0 2.5
ln(yn )
Analytical Method (asymptotic distribution)
• small data set
• must find the distribution (to determine the cdf) that fits the data

Basic idea is to determine a distribution function which fits the data.


Given the data set {y1, y2, … yn} where y is the maximum (or minimum) value in a given
time interval.

  y − B 
Type I (Fisher-Tippet I or Gumbal) F( y n ) = exp − exp − n  -∞ < yn < ∞
  A 

  y − B 
F( y n ) = exp − exp − n 
  A 

6
A= σ ≈ 0.779σ , where σ 2 = var(Y ) , asymptotic value: var[ y ∞ ] = π 6
π
6
B=µ− γσ , where µ = average(Y ) and γ = 0.5772 (Euler's number) Æ
π
B = µ − 0.45σ

see CEM Figures II-8-1 and II-8-2 for more

1. Determine F(y)
2. Calculate 1/[1-F(yn)] = M = T × m Æ F(yn) = 1 – 1/M
3. solve for yn

You might also like