Transcript Selecting an Environmental Laboratory

Understanding
Analytical
Environmental Data
[email protected]
Complex and Confusing
 Interested
in low concentrations of targets
 Heterogeneous
 Matrix
samples - variable results
interference on analysis
 Regulations
don’t address these problems
Complex and Confusing
 What
do you need?
 Why
do you need it?
 How
will you use it?
 Bad
or good decisions can come from it?
Complex and Confusing
 All
data have error.
 Nobody
can afford absolute certainty.
 Tolerable
 Without
error rates (99 % vs. 95 % certainty)
DQOs, decisions are uninformed.
 Uninformed
decisions - conservative and expensive
Appendix IA Parameters

Dissolved Anions Method 300 or 9056 (pay attention to hold times) +
Alkalinity Method 310.1
 48 hour hold on NO3- and NO2- (May need 353.1, 353.2, 353.3)

Dissolved Cations Method 6010B/6020

Field Parameters
 Specific Conductance Method 160.1
 pH Method(s) 150.1 or 9040B
 Temperature Method 170.1
 TOC (Not field parameter) Lab Method 9060

Ask for what you need and want
Appendix IB Parameters

Total Elements Method 6010B/6020

Volatiles
 Method 8260B

Method 624
Ask for what you want
 Communicate, communicate, communicate

DQO Approach: 3 Phases



Planning

Data Quality Objectives (Why sample?)

Quality Assurance Project Plan (“QAPP”)
Implementation

Field Data Collection (Sampling)

Quality Assurance/Quality Control Activities
Assessment

Data Validation

Quality Assurance/Quality Control Activities
Much Work Remains to be Done before We Can Announce
Our Total Failure to Make any Progress
•Implementation
Assessment
Environmental Data:
What does this information tell us?
(Reading between the Regulatory Lines)
Why monitor?
Why do statistical analysis?
Understand the hydrological setting.
Detect and deal with environmental impacts.
Understand risks and liabilities.
Focus resources.
Reduce monitoring costs.
“A” horizon
Topsoil, organic material
Zone of leaching
“B” horizon
Zone of accumulation
“C” horizon
Parent material ( rock,
gravel, sand)
The soil profile of a dark brown
Chernozemic soil formed under native
grassland
Detection Monitoring
Includes all Appendix I parameters (Appendix IA and IB).
May be modified, in consultation with local governing body
to delete any Appendix I parameter on a Site Specific Basis, if
Removed constituents not reasonably expected to be derived from waste
Detection Monitoring
May add parameters, if
Acceptable analytical method,
Commercially available calibration standard,
Analyte is chemically stable,
Reasonable sample collection and preservation technique
Reasonable expectation of detection, and is a good indicator and
possible precursor to other more hazardous constitutents that might
Be released later.
Detection Monitoring
Department considerations in modifying Appendix I parameters:
Types, quantities, and concentrations of constituents in waste
managed at the SWDS and facilities
Mobility, stability, and persistence of constituents, or their reaction
products in the unsaturated zone beneath the MSWLF unit.
Detection Monitoring
Department may specify a monitoring frequency during the active life
and post-closure.
Minimum of semi-annually, unless approved by the Department.
Considerations:
Lithology of the saturated and unsaturated zone
Hydraulic conductivity of groundwater
Groundwater flow rates and minimum distance of travel
Resource value of the groundwater
Background Data
Owner/operator must acquire a minimum of Eight Quarterly Samples
From each well and analyzed for Appendix IA and IB constituents.
Owner/operator must specify in the operating record, one or more
statistical tests for each hazardous constituent.
Changes in these statistical tests shall be reviewed and approved within
two weeks of the request and entered into the operating record.
Background Data
Owner/operator must acquire a minimum of Eight Quarterly Samples
From each well and analyzed for Appendix IA and IB constituents.
Owner/operator must specify in the operating record, one or more
statistical tests for each hazardous constituent.
Changes in these statistical tests shall be reviewed and approved within
two weeks of the request and entered into the operating record.
Statistically Significant Increase over Background
Documentation in Operating Record indicating which constituent is above
Background, and forward the Documentation to the Department and local
Governing Body within 14 days.
Begin Assessment Monitoring, or
Provide an Alternative Source Demonstration
Error in sampling, analysis, or natural variations in water
Certified by a qualified groundwater scientist
If not successfully demonstrated begin Assessment Monitoring in 90 days.
Statistical Methods and Requirements
Trend analysis
Control charts
Prediction interval (tolerance intervals)
ANOVA comparison with background
Other……………………….
------------------------------------------------------------------Regulations……..Type I error = 0.01
99 % Certainty (for each constituent in each well)
Statistical Methods and Requirements
Intrawell Statistics, or
Interwell Statistics
(groups and/or Upgradient – Downgradient)
Analyses of Variance (ANOVA)
Trend Analysis
Nitrate
40.00
35.00
milligrams/Liter
30.00
25.00
Series1
20.00
Linear (Series1)
15.00
10.00
5.00
0.00
0
5
10
15
20
Sampling Event
25
30
35
Control Charts
Family of Charts: Shewhart used 3 sigma (3 standard deviations, 98.5 %
probability, others have used the Standard error of the Estimate, etc.)
1 sd 67 % of data fits within limits
2 sd 95 % of data fits within limits
3 sd 98.5 % of data fits within limits
4 sd 99 % of data fits within limits
“….the fact that the criterion which we happen to use has a fine ancestry in highbrow
statistical theorems does not justify its use. Such justification must come from empirical
evidence that it works. As the practical engineer might say, the proof of the pudding is
in the eating.”
Walter A. Shewhart
Control Charts
Criticisms:
Controversial.
Operators expected to determine if a special case has occurred.
Process in control – 0.27% probability that a point will be out of specs
(1/0.0027 or 1 in 370.4)
Good at detecting large changes, does not detect small changes efficiently
Strengths:
May work well for non-parametric data
Special control chart CUMSUM does detect small changes
Control Charts
Control Charts
Tolerance Interval
A tolerance interval, also known as a tolerance limit, or prediction
interval is an interval within which, with some confidence, a specified
proportion of a population falls. This differs from a confidence interval
in that the confidence interval bounds a population parameter
(the mean, for example) with some confidence, while a tolerance interval
bounds a population proportion.
Criticisms:
Difficult to use and interpret…..takes some experience
Strengths:
Works well on non-parametric data
Tolerance Interval
Tolerance Intervals for the Normal Distribution
Fill in the following information:
If I measured a sample of
and got a mean of
and a standard deviation of
then I can be
that
will be contained…
8
97.07
1.5
99.0%
90.0%
within the interval from: 90.84983
items,
certain
of the population
to
103.2902 (a Two-sided Tolerance Interval)
below the value: 102.7091
(an Upper One-sided Tolerance Interval)
above the value: 91.43086
(a Lower One-sided Tolerance Interval)
You can ignore the following intermediate quantities used in the calculation:
z(1-p): 1.281551
z(1-g): 2.326342
a: 0.613438
b: 0.965889
k1: 3.759429
z((1-p)/2):
ChiSq(g,n-1):
k2:
1.644853
1.239032
4.146782
Analyses of Variance (ANOVA)
Parametric – populations behave as a Normal Distribution
0.01
0.008
Ndist calc
0.006
excel t dist
0.004
0.002
500
375
250
125
0
-125
-250
-375
-500
0
Non-parametric – population does not behave Normally
Can it be mathematically transformed to behave Normally ?
log, antilog, power transformation
Hypothesis Testing – Probability and Inferential Statistics
Hypothesis:
Ho : The Landfill is contributing pollutants in excess of standards, and background.
Ha : The Landfill is not contributing pollutants in excess of standards, and background.
There are two decisions possible:
(1). Accept the null hypothesis (Ho),
(2). Reject the null hypothesis (Ho ), equivalent to “accept the alternate hypothesis (Ha)”.
There are two possible situations either the null hypothesis (Ho ) is true, or it is false.
Because of these facts the possible errors are:
Situation
Ho is True
Ho is False_
Accept Ho
correct
Type II error (Beta)
Reject Ho
Type I error (alpha)
correct
Decision
Hypothesis Testing – Probability and Inferential Statistics
The Type I (alpha) error occurs when Ho is true, but we reject it.
This error would occur when the Landfill is contributing pollutants to water above standards and
background, but we conclude that it is not. The consequences of the Type I (alpha) error are the most
severe. This error would mislead an understanding of the actual impacts to water resources and public
health. In addition, the Type I (alpha) error would be the most embarrassing error to the agency.
The Type II (Beta) error occurs when Ho is false, but we accept it.
This error would occur when the Landfill is not contributing pollutants above standards and background,
but we conclude that it is. The Type II error (Beta) is less embarrassing to the organization, but carries a
large opportunity cost by unnecessarily alarming residents of the area and possibly causing unnecessary
remediation activities.
Hard to imagine good and bad from
Groundwater Statistics !!!!!!!
Hypothesis Testing – Probability and Inferential Statistics
Well Data on Lead
700
600
Probability
500
Upgradient Well
400
Dow ngradient Well
300
200
100
201
191
181
171
161
151
141
131
121
111
101
91
81
71
61
51
41
31
21
-100
11
1
0
Lead Concentration (ug/L)
Ho - The two well populations are not statistically equivalent
Ha - The two well populations are statistically equivalent
90 % Certainty
95 % Certainty
99 % Certainty
Accept Ho
Accept Ho
Accept Ho
Hypothesis Testing – Probability and Inferential Statistics
Well Data on Lead
700
600
Probability
500
Upgradient Well
400
Dow ngradient Well
300
200
100
201
191
181
171
161
151
141
131
121
111
91
81
71
61
51
101
-100
41
31
21
11
1
0
Lead Concentration (ug/L)
Ho - The two well populations are not statistically equivalent
Ha - The two well populations are statistically equivalent
90 % Certainty
95 % Certainty
99 % Certainty
Reject Ho
Accept Ho
Accept Ho
Hypothesis Testing – Probability and Inferential Statistics
Well Data on Lead
700
600
Probability
500
Upgradient Well
400
Dow ngradient Well
300
200
100
199
188
177
166
155
144
133
122
111
89
100
78
67
56
45
34
23
-100
12
1
0
Lead Concentration (ug/L)
Ho - The two well populations are not statistically equivalent
Ha - The two well populations are statistically equivalent
90 % Certainty
95 % Certainty
99 % Certainty
Reject Ho
Reject Ho
Accept Ho
Hypothesis Testing – Probability and Inferential Statistics
Well Data on Lead
700
600
Probability
500
400
Upgradient Well
300
Dow ngradient Well
200
100
201
191
181
171
161
151
141
131
121
111
91
81
71
61
51
41
101
-100
31
21
11
1
0
Lead Concentration (ug/L)
Ho - The two well populations are not statistically equivalent
Ha - The two well populations are statistically equivalent
90 % Certainty
95 % Certainty
99 % Certainty
Reject Ho
Reject Ho
Reject Ho
Injecting Common Sense into Statistic Evaluations
If determination is that constituent concentration is > Background
- Is it consequential ?
- Is result above GW standard, or tending toward > GW standard ?
- Look over the data, is it cogent?
- Is there a failure, or misrepresentation of the statistical protocol?
- Resample, errors happen and GW variations are the norm.
Uggradient Well
MW-1
Parameter
Action Level:
One-tail 0.05
Two-tail 0.10
n-1 = 1
n-1 = 2
n-1 = 3
n-1 = 4
n-1 = 5
n-1 = 6
6.31
2.92
2.35
2.13
2.02
1.94
Avg
Std Dev
95% UCL
95% UCL Test
n samples taken
"n" needed
"n" Test
t n-1 used
Above Std
Metals
Al
5
A
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
PASS
Sb
0.006
P
As
0.050
P
Ba
2.000
P
Be
0.004
P
B
0.750
A
Cd
0.005
P
Cr
0.100
P
Co
0.050
A
Cu
1.000
S
0.258
0.150
0.368
FAIL
7.000
1.322
Pass
1.940
FAIL
0.004
0.010
0.011
Pass
7.000
0.186
Pass
1.940
PASS
0.106
0.102
0.181
Pass
7.000
0.011
Pass
1.940
PASS
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
PASS
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
PASS
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
PASS
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
PASS
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
PASS
0.008
0.015
0.019
Pass
7.000
0.001
Pass
1.940
PASS
Sb
0.006
P
As
0.050
P
Ba
2.000
P
Be
0.004
P
B
0.750
A
Cd
0.005
P
Cr
0.100
P
Co
0.050
A
Cu
1.000
S
0.007
0.009
0.013
FAIL
7.000
41.488
FAIL
1.940
FAIL
#DIV/0!
4.280
#DIV/0!
#DIV/0!
1.800
#DIV/0!
#DIV/0!
0.007
No
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
Pass
#DIV/0!
4.280
#DIV/0!
#DIV/0!
1.800
#DIV/0!
#DIV/0!
#DIV/0!
#DIV/0!
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
Pass
#DIV/0!
4.280
#DIV/0!
#DIV/0!
1.800
#DIV/0!
#DIV/0!
#DIV/0!
#DIV/0!
0.143
0.136
0.243
FAIL
7.000
38.042
FAIL
1.940
FAIL
#DIV/0!
4.280
#DIV/0!
#DIV/0!
1.800
#DIV/0!
#DIV/0!
0.143
No
0.068
0.070
0.120
FAIL
7.000
57.619
FAIL
1.940
FAIL
#DIV/0!
4.280
#DIV/0!
#DIV/0!
1.800
#DIV/0!
#DIV/0!
0.068
No
0.102
0.096
0.172
Pass
7.000
0.043
Pass
1.940
Pass
43.022
4.280
No
-6.266
1.800
No
Yes
-0.121
No
MW-2
Parameter
Action Level:
One-tail 0.05
Two-tail 0.10
n-1 = 1
n-1 = 2
n-1 = 3
n-1 = 4
n-1 = 5
n-1 = 6
F 95%
9.280
6.390
5.050
4.280
6.31
2.92
2.35
2.13
2.02
1.94
Avg
Std Dev
95% UCL
95% UCL Test
n samples taken
"n" needed
"n" Test
t n-1 used
Above Std
F*
F table value
Equiv Variance
t pooled
tpooled table
Statisticallydifferent?
Unequiv Variance
t se/sd
Statisticallydifferent?
Metals
Al
5.000
A
0.000
0.000
0.000
Pass
7.000
0.000
Pass
1.940
Pass
#DIV/0!
4.280
#DIV/0!
#DIV/0!
1.800
#DIV/0!
#DIV/0!
#DIV/0!
#DIV/0!
0.281
0.224
0.466
FAIL
6.000
2.694
Pass
2.020
FAIL
2.234
6.390
Yes
-1.505
1.800
No
No
-2.126
No
0.073
2.636
0.104
2.024
0.150
4.120
FAIL
FAIL
7.000
7.000
74.997
38.148
FAIL
FAIL
1.940
1.940
FAIL
FAIL
102.987 396.614
4.280
4.280
No
No
-6.042
14.058
1.800
1.800
No
Yes
Yes
Yes
-0.025
2.497
No
Yes
Downgradient Well 1
Element Std (mg/L) Std Type
Al
5.000
A
Sb
0.006
P
As
0.050
P
Ba
2.000
P
Be
0.004
P
B
0.750
A
Cd
0.005
P
Cr
0.100
P
Co
0.050
A
Cu
1.000
S
Fe
0.300
S
Pb
0.050
P
Mn
0.050
S
Ni
0.200
P
Se
0.050
P
Ag
0.050
P
Ti
0.002
P
V
0.100
A
Zn
2.000
A
Li
2.500
A
Hg
0.002
P
P or F
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Pass
FAIL
Pass
FAIL
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Pass
Upgradient
Variance = Upgrad
Statistically differ from Upgrad
FAIL
Yes
No
FAIL
Yes
No