Professional Documents
Culture Documents
Lesson 2
Return period, cdf and pdf
Stefania Tamea
(FLOOD) FREQUENCY ANALYSIS
With an infinitely long record of measurements, a frequency distribution would
tell us all information about the occurrence of extreme events (eg. frequency of
exceedance of any value)
Q τ1 τ2 τ3 τ4
QT
t
RETURN PERIOD, T, is the expected value of τ, i.e. the mean recurrence interval
measured over an infinite number of occurrences, or
∞
T = E[τ ] = ∑τ ⋅ prob(τ )
τ =1
Return Period
Let’s call p = Pex (Q ≥ QT )
The probability p(τ) of a recurrence after τ years is the product of probabilities of
independent events:
- probabilities of not having the event (ie. 1- p) for (τ -1) years
- probability of finally having the event (ie. p) :
prob(τ ) = (1 − p )τ −1 ⋅ p
∞
The return period is thus (with an infinite sum of terms): T = ∑τ ⋅(1 − p )τ −1 ⋅ p
τ =1
No events in N years (1 − p) N
At least once in N years 1 − (1 − p) N N (years) P occurrence
(for T=100)
N 1 0.01
1 10 0.096
P = 1 − 1 −
T 100 0.634
STATISTICAL INFERENCE
GOAL: estimate a design value associated to a given return period or exceedance
probability, based on a finite number of observations
MEANS: from finite data sample, infer information about the population (infinite
possible values the random variable may assume)
FOCUS DATA: extreme (largest) events at annual scale (return period in years)
Increasing the number of elements (sample size), height of steps will decrease
and curves get close: SAMPLE VARIABILITY DECREASES
F (x)
1 Increasing the size Population
i
n→∞ F ( x(i ) ) = →0
n