A Geostatistical Approach to the Characteristic Values
Horatiu V. Corbeanu Halcrow/CH2M HILL
07/11/12 – 12:15
Pro Statistics
• “Statistical methods . . . constitute the science of collecting, analysing and interpreting data in the best possible way” C. Chatfield (Statistics of Technology, 3rd ed. 1983)
• “The product of an arithmetical computation is the answer to an equation; it is not the solution to a problem” - G.O. Ashley
• “Years ago a statistician might have claimed that statistics deals with the processing of data. . . today’s statistician will be more likely to say that statistics is concerned with decision making in the face of uncertainty” – H.Chernonff & L.E. Moses (Elementary Decision Theory, 1959)
Abuse of Statistics
• “Attempts by statisticians to tackle geotechnical design have often ended in ridicule, and it is very difficult for one person to have sufficient grasp of both disciplines that he can use them sensibly”, Decoding Eurocode 7, A.Bond, A.Harris, 2008
• “There are lies, damned lies, and statistics”, Benjamin Disraeli
ba
ba
ba
baba
bababa
baba
ba
ba
ba
1
1
1)(
)(*)(
)()(*)(
0)(*)(
022
22
Twisted Example
Objective
• The Problem
• The Dataset
• General presentation of geotechnical data
• Distribution of geotechnical data
– Descriptive statistical techniques
– Transformations to a normal distribution
• Conclusions
1) Provide a reliable characteristic value
2) Identify the mathematical tool that can provide confidence to the engineering judgement.
The Problem
The Dataset HOLE EASTING NORTHING GEOL_GEOL ELEVATION ROCK_UCS ROCK_E ROCK_MU ROCK_MC ROCK_BDEN
01.BH-353 228,131 393,303 SL -6.09 13.851 13100 0.265 1.7 21
01.BH-353 228,131 393,303 RUS -19.59 5.678 8300 1.4 17.6
01.BH-536 229,228 393,111 RUS -15.52 7.18 1.8 18.9
01.BH-536 229,228 393,111 RUS -21.52 41.884 26900 0.172 1.2 26.2
01.BH-570 225,071 394,022 MSH -3.80 25.599 22800 0.6 18
01.BH-570 225,071 394,022 RUS -12.80 18.3 17600 1.1 17.4
01.BH-570 225,071 394,022 RUS -23.82 12.85 11700 0.206 1.3 20.2
01.BH-572 226,451 393,594 RUS -10.49 7.957 1.6 20.4
01.BH-572 226,451 393,594 RUS -13.49 8.156 7700 0.212 1.9 20.5
01.BH-572 226,451 393,594 RUS -22.49 12.174 11800 0.238 2.1 21.6
01.BH-755 224,397 394,360 SL -6.49 11.736 2 22.9
01.BH-755 224,397 394,360 MSH -14.54 40.022 1.4 24
01.BH-755 224,397 394,360 RUS -17.54 15.158 2.8 21.9
01.BH-756 220,978 395,899 RUS 0.81 15.197 19500 2.6 17
01.BH-756 220,978 395,899 RUS -5.19 12.452 16400 2.1 17.1
01.BH-756 220,978 395,899 RUS -12.69 10.523 14100 0.215 2.5 25.1
01.BH-757 221,265 395,957 MSH 1.38 16.331 17000 0.200 3.9 22.6
Count :
Minimum :
Maximum :
Mean :
5% Fractile :
226
2.83
12,567
2,531
240
MPa
MPa
MPa
MPa
-42
-40
-38
-36
-34
-32
-30
-28
-26
-24
-22
-20
-18
-16
-14
-12
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
18
20
0 5000 10000 15000 20000 25000
Intact Elastic Modulus (MPa)
Ele
va
tio
n (
m Q
NH
D)
WSL Lab SL & BSL Lab M SH Lab RUS Calcareous Lab RUS Clayey Lab RUS Gypsum Lab
WSL Derived SL & BSL Derived M SH Derived RUS Calcareous Derived RUS Clayey Derived RUS Gypsum Derived
-42
-40
-38
-36
-34
-32
-30
-28
-26
-24
-22
-20
-18
-16
-14
-12
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
18
20
1 10 100 1000 10000 100000
Intact Elastic Modulus (MPa)
Ele
va
tio
n (
m Q
NH
D)
WSL Lab SL & BSL Lab M SH Lab RUS Calcareous Lab RUS Clayey Lab RUS Gypsum Lab
WSL Derived SL & BSL Derived M SH Derived RUS Calcareous Derived RUS Clayey Derived RUS Gypsum Derived
General Presentation of Data
Shapes of Data Distribution
Triangular
Data
Fre
qu
en
cy (
%)
Uniform
Data
Fre
qu
en
cy (
%)
Exponential
Data
Fre
qu
en
cy (
%)
Normal
Data
Fre
qu
en
cy (
%)
Lognormal
Data
Fre
qu
en
cy (
%)
Fact for Log Normal Distribution
c = a + b Normal
C
Fre
qu
en
cy (
%)
Two geological processes: a & b (i.e. rate of deposition, duration of deposition)
Log (c) = Log (a*b)
Log (c) = Log (a) + Log (b)
c = a * b Lognormal
C
Fre
qu
en
cy (
%)
How does it look like?
-42
-40
-38
-36
-34
-32
-30
-28
-26
-24
-22
-20
-18
-16
-14
-12
-10
-8
-6
-4
-2
0
2
4
6
8
10
12
14
16
18
20
1 10 100 1,000 10,000 100,000
Intact Elastic Modulus (MPa)E
lev
ati
on
(m
QN
HD
)
WSL Lab SL & BSL Lab M SH Lab RUS Calcareous Lab RUS Clayey Lab RUS Gypsum Lab
WSL Derived SL & BSL Derived M SH Derived RUS Calcareous Derived RUS Clayey Derived RUS Gypsum Derived
The distribution of Data
- +
68.26%
99.95%
- +
Fre
qu
en
cy (
%)
Parameter
Count :
Minimum :
Maximum :
Mean :
Standard Deviation :
226
2.83
12,567
2,531
2,445
MPa
MPa
MPa
MPa
- = 86 MPa + = 4976 MPa
- 2 = -2359 MPa + 2 = 7421 MPa
- 3 = -4804 MPa + 3 = 9866 MPa
95.44%
- +
-15
-13
-11
-9
-7
-5
-3
-1
1
3
5
7
9
11
13
15
17
19
1 10 100 1,000 10,000 100,000
Intact Elastic Modulus (MPa)
Ele
vati
on
(m
QN
HD
)
SL & BSL Lab SL & BSL Derived
-15
-13
-11
-9
-7
-5
-3
-1
1
3
5
7
9
11
13
15
17
19
-5 -4 -3 -2 -1 0 1 2 3 4 5
Z-Score
Ele
va
tio
n (
m Q
NH
D)
SL & BSL Lab SL & BSL Derived
Z = (x - ) /
- +
68.26%
+
Fre
qu
en
cy (
%)
Parameter
+
Z-Score
Log-normal Distribution
0
10
20
30
40
50
60
70
3
90
0
1,7
98
2,6
95
3,5
93
4,4
90
5,3
88
6,2
85
7,1
82
8,0
80
8,9
77
9,8
75
10
,77
2
11
,67
0
12
,56
7
Intact Elastic Modulus
Nu
mb
er
of
Sa
mp
les
Ways Forward
• Ignoring the mathematical tools, introducing further descriptive statistical parameters (median, skewness, kurtosis, coefficient of variation, etc.)
• Applying transformations (rotations, translations, multiplications, etc.) to the existing data in such a way to be represented as a normal distribution
Descriptive Statistics Approach • Median
The midpoint of the observed values if they are arranged in increasing order (ex. No of Samples = 226 – median = sample no 113)
0
10
20
30
40
50
60
70
3
90
0
1,7
98
2,6
95
3,5
93
4,4
90
5,3
88
6,2
85
7,1
82
8,0
80
8,9
77
9,8
75
10
,77
2
11
,67
0
12
,56
7
Intact Elastic Modulus
Nu
mb
er
of
Sa
mp
les
Count :
Minimum :
Maximum :
Mean :
Median :
226
2.83
12,567
2,531
1,710
MPa
MPa
MPa
MPa
Descriptive Statistics Approach • Median
• Deciles, Percentile, Quartiles
• Interquartile Range (IQR)
Splitting the data into tenth (deciles), hundredths (percentile) or any other fraction (quartiles)
IQR = difference between q0.75 and q0.25
Attention: Excel works a bit different – percentile for any fraction, quartile for min, 25%, median, 75%, max
Box Plot
1.5 * IQR 1.5 * IQR
IQR = q0.75 – q0.25= 2400
q0.25 = 880 q0.75 = 3280
q0.50 = 1710
880-1.5*2400 = -2720
q0.00 = 3 Q1.00 = 12567
3280+1.5*2400 = 6880
6,876.88
3.00
1,711.50
878.75
3,278.00
9,758
9,170
10,188
8,366
10,256
12,153
12,567
11,724
8,489
9,932
8,579
8,885
7,0017,091
9,211
9,049
7,281
-15
-13
-11
-9
-7
-5
-3
-1
1
3
5
7
9
11
13
15
17
19
1 10 100 1,000 10,000 100,000
Intact Elastic Modulus (MPa)
Ele
vati
on
(m
QN
HD
)
SL & BSL Lab SL & BSL Derived
Distribution of Data
Descriptive Statistics in Excel
Descriptive Statistics Approach
0
10
20
30
40
50
60
70
3
90
0
1,7
98
2,6
95
3,5
93
4,4
90
5,3
88
6,2
85
7,1
82
8,0
80
8,9
77
9,8
75
10
,77
2
11
,67
0
12
,56
7
Intact Elastic Modulus
Nu
mb
er
of
Sa
mp
les
Transformations to a Normal Distribution
• Applying any mathematical calculation over the existing dataset (square, log, exponential, etc.)
• Applying rotation and translation matrices
Transformations
0
5
10
15
20
25
30
35
40
1.37 2.48 3.60 4.71 5.83 6.94 8.06 9.17 10.28 11.40 12.51 13.63 14.74 15.86 16.97
Nu
mb
er
of
Sam
ple
s
0
10
20
30
40
50
60
1.04 1.64 2.24 2.84 3.44 4.04 4.64 5.24 5.84 6.44 7.04 7.64 8.24 8.84 9.44
Nu
mb
er
of
Sam
ple
s
Descriptive Statistics
Count 226.000
Minimum 1.041
Maximum 9.439
Range 8.398
Mean 7.392
Median 7.445
Mode 7.706
Standard Deviation 1.060
Sample Variance 1.123
Kurtosis 5.192
Skewness -1.184
Confidence Level (95%) 0.004
x1 = ln(x) x2= x0.3
Descriptive Statistics
Count 226.000
Minimum 1.366
Maximum 16.973
Range 15.607
Mean 9.613
Median 9.333
Mode 10.092
Standard Deviation 2.777
Sample Variance 7.709
Kurtosis 0.098
Skewness 0.336
Confidence Level (95%) 0.012
Box Plots 9.44
4.80
7.45
6.78
8.10
4.58
1.04
3.81
x1 = ln(x) 16.89
2.09
9.33
7.64
11.34
1.37
17x2= x0.3
0
5
10
15
20
25
30
35
40
1.37 2.48 3.60 4.71 5.83 6.94 8.06 9.17 10.28 11.40 12.51 13.63 14.74 15.86 16.97
Nu
mb
er
of
Sam
ple
s
0
10
20
30
40
50
60
1.04 1.64 2.24 2.84 3.44 4.04 4.64 5.24 5.84 6.44 7.04 7.64 8.24 8.84 9.44
Nu
mb
er
of
Sam
ple
s
x1 = ln(x)
x2= x0.3
-15
-13
-11
-9
-7
-5
-3
-1
1
3
5
7
9
11
13
15
17
19
1 10 100 1,000 10,000 100,000
Intact Elastic Modulus (MPa)
Ele
vati
on
(m
QN
HD
)
SL & BSL Lab SL & BSL Derived
Histograms
6,876.88
3.00
1,711.50
878.75
3,278.00
9,758
9,170
10,188
8,366
10,256
12,153
12,567
11,724
8,489
9,932
8,579
8,885
7,0017,091
9,211
9,049
7,281
9.44
4.80
7.45
6.78
8.10
4.58
1.04
3.81
Exp(1.04) = 2.82 MPa
Exp(4.58) = 97 MPa
Exp(4.80) = 120 MPa
Exp(3.81) = 45 MPa
Exp(7.45) = 1,711.5 MPa
Exp(6.78) = 878.75 MPa
Exp(8.10) = 3,278 MPa
Exp(9.44) = 12,580 MPa
Back calculation
Median Ln : 7.45
Back calculated median : Exp(7.45) = 1,711.5 MPa
Conclusion
Statistical assessment:
Minimum = 120 MPa
Maximum = 12,567 MPa
Characteristic value = 1,600 – 1700 MPa
5% Fractile = 380 MPa
95% Fractile = 8,570 MPa
-15
-13
-11
-9
-7
-5
-3
-1
1
3
5
7
9
11
13
15
17
19
1 10 100 1,000 10,000 100,000
Intact Elastic Modulus (MPa)
Ele
vati
on
(m
QN
HD
)
SL & BSL Lab SL & BSL Derived
Outliers
Original assessment:
Minimum = 2.83 MPa
Maximum = 12,567 MPa
Characteristic value = 2,531 MPa
5% Fractile = 240 MPa
95% Fractile = 8,570 MPa
Simsima Limestone
• Earth data are generally characterized by log normal distributions
• The basic statistical tools available are used to quantify our engineering judgment and to provide confidence in our assessment
• Applying blind mathematical tools without the understanding of our dataset can result in misleading answers.
Conclusion
0
5
10
15
20
25
30
200
2,1
07
4,0
14
5,9
21
7,8
29
9,7
36
11,6
43
13,5
50
15,4
57
17,3
64
19,2
71
21,1
79
23,0
86
24,9
93
26,9
00
Midra Shale
0
10
20
30
40
50
60
70
100
2,8
00
5,5
00
8,2
00
10,9
00
13,6
00
16,3
00
19,0
00
21,7
00
24,4
00
27,1
00
29,8
00
32,5
00
35,2
00
37,9
00
Midra Shale and Rus Formation
Rus Formation
-40
-35
-30
-25
-20
-15
-10
-5
0
5
10
15
1 10 100 1000 10000 100000
Intact Elastic Modulus (MPa)E
lev
ati
on
(m
QN
HD
)
MSH Lab RUS Lab
0
1
2
3
4
5
6
7
8
9
2.30
1
2.45
3
2.60
5
2.75
7
2.90
9
3.06
1
3.21
3
3.36
5
3.51
7
3.66
9
3.82
2
3.97
4
4.12
6
4.27
8
4.43
0
Log (E)
Nu
mb
er o
f sa
mp
les
Midra Shale
200 MPa 26,910 MPa
Cluster 1
0
1
2
3
4
5
6
7
2.3
0
2.3
9
2.4
7
2.5
5
2.6
4
2.7
2
2.8
1
2.8
9
2.9
7
3.0
6
3.1
4
3.2
3
3.3
1
3.3
9
3.4
8
Log (E)
Nu
mb
er
of
Sam
ple
s
Cluster 2 - 3
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
3.7
7
3.8
2
3.8
6
3.9
1
3.9
6
4.0
1
4.0
5
4.1
0
4.1
5
4.1
9
4.2
4
4.2
9
4.3
4
4.3
8
4.4
3
Log (E)
Nu
mb
er
of
Sam
ple
s
0
0.5
1
1.5
2
2.5
3
3.5
3.4
91
3.5
69
3.6
47
3.7
24
3.8
02
3.8
80
3.9
57
4.0
35
4.1
13
4.1
90
4.2
68
4.3
46
4.4
23
4.5
01
4.5
79
0
1
2
3
4
5
6
2.0
00
2.1
00
2.2
00
2.3
00
2.3
99
2.4
99
2.5
99
2.6
99
2.7
99
2.8
99
2.9
99
3.0
98
3.1
98
3.2
98
3.3
98
5,900 MPa 10,100 MPa
200 MPa
3,000 MPa
Midra Shale
17,400 MPa 26,910 MPa
MEAN : 905 MPa
MEDIAN: 850 MPa
MEAN : 21,380 MPa
MEDIAN: 22,800 MPa MEAN : 7,465 MPa
MEDIAN: 7,160 MPa
0
2
4
6
8
10
12
14
16
2.00
0
2.18
4
2.36
8
2.55
3
2.73
7
2.92
1
3.10
5
3.28
9
3.47
4
3.65
8
3.84
2
4.02
6
4.21
0
4.39
4
4.57
9
Log (E)
Num
ber o
f Sam
ples
Rus Formation
100 MPa 37,930 MPa
Conclusion Original assessment:
Minimum = 200 MPa
Maximum = 26,900 MPa
Characteristic value = 5.813 MPa
5% Fractile = 275 MPa
95% Fractile = 23,555 MPa
Confidence Level (95) = 2,370 MPa
Midra Shale
Statistical assessment:
Lithology 1 (Possible Shale)
Minimum = 200 MPa
Maximum = 3,000 MPa
Characteristic value = 850 - 1170 MPa
5% Fractile = 260 MPa
95% Fractile = 2,810 MPa
Confidence Level (95) = 295 MPa
Rus Formation
Lithology 2 (Possible Dolomitic Limestone)
Minimum = 16,400 MPa
Maximum = 26,900 MPa
Characteristic value = 21,655 – 22,800 MPa
5% Fractile = 16,640 MPa
95% Fractile = 26,140 MPa
Confidence Level (95) = 2,833 MPa
Original assessment:
Minimum = 100 MPa
Maximum = 37,900 MPa
Characteristic value = 4,750 MPa
5% Fractile = 205 MPa
95% Fractile = 18,550 MPa
Confidence Level (95) = 1,430 MPa
Statistical assessment:
Lithology 1 (Possible Chalk/Calcisiltite)
Minimum = 100 MPa
Maximum = 2,500 MPa
Characteristic value =1,000 – 1,050 MPa
5% Fractile = 200 MPa
95% Fractile = 2,200 MPa
Confidence Level (95) = 160 MPa
Lithology 2 (Possible Limestone)
Minimum = 3,100 MPa
Maximum = 37,900 MPa
Characteristic value = 9,600 – 11,900 MPa
5% Fractile = 3,600 MPa
95% Fractile = 25,450 MPa
Confidence Level (95) = 2,848 MPa
-20
-15
-10
-5
0
5
10
1 10 100 1000 10000 100000
Intact Elastic Modulus (MPa)E
lev
ati
on
(m
QN
HD
)
-40
-35
-30
-25
-20
-15
-10
-5
0
5
10
1 10 100 1000 10000 100000
Intact Elastic Modulus (MPa)
Ele
va
tio
n (
m Q
NH
D)
Conclusion
Midra Shale Rus Formation