Probabilistic Modeling

27th June 2023

 Probabilistic Modeling

•Probabilistic model: Statistical analysis tool that

estimates, on the basis of past (historical) data,
the probability of an event occurring again.
•In practical applications, the distribution
function is required for calculating
exceedance and non-exceedance probabilities
and return periods

•Return periods is defined as the average

time elapsing between successive
occurrences of some events
 RECALL
• The cumulative distribution function (CDF) is the cumulative form of the
probability function (PDF) and expresses the probability of non-exceedance
Pr (X< x).

• Its largest value is unity. The complement of F(x) is called the exceedance
probability of x, i.e. I - F(x) .

• The reciprocal of the exceedance probability is the return period T, i.e.

1 1
 T or F(X T ) 1
1 F(X ) T

• Where XT has a return period of T

Probability Modelling using the Normal
Distribution

Example 1
In the standardized N (0,1) distribution what value of Z has
(a) Exceedance probability 0.1
(b) Non exceedance probability 0.3

Soln: (a) 0.1=1- (Z)   (Z)=0.9 , hence Z=1.28 (From Table)

(b)  (Z)=0.3, Z=-0.52 (as 0.3<0.5)
Note:
• Z is the standardized variate or reduced variate.
• Any variable which has an associated probability distribution is called a
random variable (r.v) or variate
Probability Modelling Using the Normal
Distribution

Example 2
If the annual flow volumes in the River Ndarugo are normally distributed N(100,35)
(a) What value of Q has exceedance prob. = 0.1
(b) What is the non-exceedance prob. of Q = 180

Solution:
a) First find Z value having 1- (Z) =0.1 i.e. (Z)=0.9
 Z= 1.28
Then using the relation Q= +  Z
Q=100+351.28=144.8 m3/s

180-100
(b) Q = 180 implies Z = =2.29 and  (Z) =  (2.29) = 0.989
35
1
This has a return period T= = 91yrs
1-0.989
 Exercise
• Twenty four concrete cubes are prepared and cured under identical
processes. The crushing strength of the concrete cubes are then tested. The
observed strengths are shown below
• (a) What is the mean and the standard deviation of the strength of the
cubes? (b) What is the 95% dependable strength of the concrete cubes?

Obs Strength Obs Strength Obs Strength Obs Strength

No. (N/mm2) No. (N/mm2) No. (N/mm2) No. (N/mm2)
1 19.3465 7 20.0212 13 24.3619 19 22.0988
2 26.3294 8 28.8753 14 20.3917 20 28.0456
3 25.0496 9 27.5152 15 26.4899 21 26.3734
4 33.2465 10 22.8869 16 22.6434 22 23.2713
5 22.8927 11 32.0763 17 24.5022 23 27.1044
6 18.368 12 27.8196 18 18.0844 24 30.4283
 Solution

(a) Let the crushing strengh of the cubes be normally distributed r.v
n

x i
Mean strength for the sample is X= i=1
 99.2227  24.9676
n 24
1 1
 1 n
  1 24
22
 
2
 
2
Standard deviation for the sample is s   xi  X    i
x  24.9676    4.1594N / mm 2
 n 1 i1   24 1 i1 
(b) Let x95 be the 95% depandeble crushing strength
 x  24.9676 
So P( X  x95 )  0.95 or P  Z  95  0.95
 4.1594 
 x  24.9676   x  24.9676 
or 1 P  Z  95   0.95  P  Z  95   1 0.95  0.05
 4.1594   4.1594 
From the standard normal table, the value of Z corresponding to 0.95 is 1.645, and that corresponding to 0.05 is -1.645
 x  24.9676 
So,  95   1.645 or x95  18.125 N / mm2
 4.1594 
 Other distributions
• Exponential
• Gamma
• Pearson Type III
• Lognormal
• Log Pearson Type III
• Extreme value type I (Double exponential= Gumbel)
• Extreme value type II (Frechet= inverse Weibull distribution)
• Extreme value type III (related to Weibull distribution)
• General Extreme value
• Etc.
Exponential distribution
• Unsymmetrical p.d.f with peak at the origin and with a slowly decreasing
tail [Insert Figure for pdf and cdf]

 (x xo)
1 
f (x)  e


• The mean or expected value of an exponentially distributed variable X

with rate parameter β is given by E[X]=β and

• The Var [X]=β2, Such that the standard deviation is equal to the mean
 Example (Workout and Submit by 4 th July 2023)
Laptops produced by “Intel” last on average for 5 years. The lifespan of
each laptop follows an exponential distribution

(a)Calculate the rate parameter

(b)Write the probability density function and graph it
(c)What is the probability that a laptop will last less than three years
(d)What is the probability that a laptop will last more than 10 year
(e)What is the probability that a laptop will last between 4 and 7 years
 Shifted Exponential Distribution
• The PDF and CDF of the exponential distribution starts at x = 0
• In general , the distribution can start at any positive value of x;
the resulting distribution may be called the shifted exponential
distribution
• The corresponding PDF and CDF starting at x0 would be as
follows

 1  ( x x0 )
 e for x  x0


f ( x)  


0 for x  x0
Shifted Exponential Distribution

 (xxo)
1 
f (x)  e

Exponential distribution
 x xo
1 
f (x)  e

x  xo
Let  y or x  xo   y

Non-Exceedance probability G( y)  1 e y
Exceedance probability e y  1 G( y)
1 1
Exceedance probability =1 G( y)  or G( y)  1
T T
1 1
Hence 1 e y  1 and  y  ln  ln T
T T
yT  ln T where yT denotes y value of return period T
and xT  xo   ln T  xo   yT
 Example
• X is exponential with Xo=50 and β=16
• (a) What exceedance probability and return period has x=90 (b) What is
the X value of return period 18?
Solution
( x xo) (9050)
 

(a) F (90)=1- e  1- e 16
 1 e2.5  1 0.082  0.912
Exceedance probability =1-0.912 = 0.082
1
Return period =  12.195 say12
0.082
(b) T=18; xT  xo   ln T  50 16  2.89  96.2
N.B :This can also be computed with the aid of exponential tables
Exercise (Submit by 4th July 2023)

• In a stream, the daily concentration of a pollutant follows

an exponential distribution. The mean daily
concentration of the pollutant is 0.0025 mg/L

• (a) Considering that pollution problem is said to occur if the

concentration is greater than 0.005 mg/L, what is the
probability of pollution problem on a particular day

• (b) What s the return period associated with the

pollution concentration of 0.005 mg/ L
 Extreme Value Distribution

• Extreme value distribution are used to model the maximum and

minimum limits of a normalized set of independent and identically
distributed random variables

• EV distributions can be categorized in to three types

• Type I: The Gumbel distribution-Widely used to model maximum and minimum
daily flows in a stream in a particular year. It is useful in predicting the chance that
an extreme earthquake, flood or other natural disaster will occur.
• Type II: Frechet distribution: Also applied to extreme events such as annually
maximum one-day rainfalls and river discharges-also known as inverse Weibull
distribution
• Type III: Weibull distribution-can be used to model the failure time of structures
 Extreme Value Type I Distribution
• A random variable is said to follow Gumbel distribution if its pdf can be
expressed as
 xu 
 xu    
  xu  
1 
  
e  
 
f (x)  and F (X )  e e 
e

x u
Let y  or x  u   y

y

G( y)  e e
 non  exceedanceprobability
-y

Exceedance probability: 1-G(y)=1- e -e

EV-Type I: Return Period
The value of y which has a return period T satisfies the equation
1 1 y 1
1 G( y)   G( y)  1 or e e  1
T T T
Taking natural logs
y  1 y  1
e  ln  1  or e   ln  1 
 T  T
Taking natural logs again
 1 
 y  ln  ln 1 
  T 
 1 
yT   ln  ln 1   ln T   when T>5
1
  T   2
EV-Type I: Return Period
Value of x which has a return period T satisfies
 xu  
1 
e    1

1- F (x)  or F(x)= e  1
T T
On taking logs twice and re-arranging signs each time
   1 
xT  u    ln  ln 1 
   T  
xT  u   yT
 Example
• X is distributed as EVI with u=50, α=30
(a) What value of exceedance probability has x = 100 (b) What is x value of
return period 30?
 xu   10050 
 
   
e  30  e 1.666
e e

(a) F(100) = e e 
 0.8279
Exceedance probability 1- F(x)  0.1721
Return period =5.81 say 6
   T 1     29 
(b) xT  u    ln ln     50  30  ln ln  30 
   T       
50  30ln 0.0339  50  30 3.3843  151.529
Testing of Goodness-Of-Fit of Distribution
Models
• When a particular probability distribution has been specified to
model a random phenomenon, the validity of the specified or
assumed distribution may be verified or disapproved statistically
by Goodness-Of-Fit tests

• Three such tests for distribution are available and used widely
• (a) the Chi-square
• (b) the Kolmogorov-Smirnov (or K-S)
• (c) the Anderson-Darling (or A-D)

• Write short notes on the THREE test and submit by 4th July 2023