Spatiotemporal Analysis of Surface Water Tetrachloroethene in New Jersey

1
Spatiotemporal Analysis ofSurface Water
Tetrachloroethene in New Jersey

• Presentation of the project of Yasuyuki Akita
• Temporal GIS Fall 2004

2
Agenda

• About Tetrachloroethene
• Monitoring Data
• Details of BME Method
• BME Analysis
• Results of BME Analysis
• New Criterion
• Model Comparison
• Conclusion

3
About Tetrachloroethene
4
About Tetrachloroethene

• Tetrachloroethene C2Cl4
• Volatile organic compound
• Nonflammable colorless liquid at room temperature
• Ether-like odor
• Synonym Tetrachloroethylene, Perchloroethylene,
  and PCE

5
Use and Production

• Mainly Used for dry cleaning, chemical
  intermediates, and industrial solvent
• PCE used in dry cleaning industry has been
  declining during 90s
• Recent Demand 763 million lb (1980)
• 318 million lb (1999)

6
End-Use Pattern in 70s and 90s
7
Exposure pathway

• Primary route
• Inhalation
• Ingestion of contaminated food and water
• Widely distributed in environment
• 38 of surface water sampling sites in the U.S.
• 771 of the 1430 National Priorities List sites
• 154 of 174 surface water samples in N.J.
  (19771979)

8
Health Effect of Tetrachloroethene

• Acute Effect (inhalation exposure)
• Dizziness, headache, sleepiness, confusion,
  nausea, difficulty in speaking and walking,
  unconsciousness, and death
• Chronic Effect (oral/inhalation exposure)
• Detrimental effect to kidney and liver

9
Carcinogenicity

• Reasonably anticipated to be a human carcinogen
  (US DHHS)
• Group 2A (Probably carcinogenic to humans) (IARC)
• Animal studies tumors in liver and kidney

10
Quality Standard for Tetrachloroethene

• Maximum Contaminant Level (MCL) in drinking water
  - 0.005 mg/L
• Surface Water Quality Standard in New Jersey -
  0.388 µg/L
• N.J. adopted more stringent standard

11
Monitoring Data
12
Monitoring Dataset for New Jersey

• Data Source
• NJDEP/USGS Water Quality Network Website
• EPA STORET database
• Data used in this study
• 369 measured values
• 171 monitoring stations
• From 1999 to 2003

13
Monitoring Data Histogram
Raw Data
Log-Transformed Data
14
Monitoring Data Statistical Moments
15
Distribution of Data Points
16
Distribution of Data Points
17
Distribution of Data Values
18
What we want to know is

• Challenge of our research
• Assess all river reaches
• Taking into account the space/time variability

Framework for the space/time estimation
Bayesian Maximum Entropy (BME) analysis of TGIS
19
Details of BME Method
20
Space/Time Random Field

• The concentration field is modeled in terms of
  Space/Time Random Field (S/TRF)
• Collection of random variables

S/TRF Collection of all possible realization

• Stochastic characterization of S/TRF is provided
  by multivariate PDF

21
Knowledge Base

• General Knowledge Base G
• Describe global characteristics of the random
  field of interest
• Expressed as statistical moments
• Site-specific knowledge Base S
• Available monitoring data over the space/time
  domain of interest
• Total Knowledge Base K
• K G ? S

22
General Knowledge Base G

• Mean Trend
• Global trend of the S/TRF of interest
• Covariance
• Measure of dependency between two points
• Sill variance covariance(r0)
• Range shows the extent that co-variability exists

23
BME analysis of Temporal GIS

• Prior stage
• Examine all general knowledge base G and
  calculate Prior PDF
• Integration stage
• Update Prior PDF using Bayesian
  conditionalization on the site-specific knowledge
  base S and obtain posterior PDF
• Interpretive stage
• Obtain estimation value from Posterior PDF

24
BME analysis of Temporal GIS

• General KB Prior PDF

• Update prior PDF with Site-specific KB
• Bayesian conditionalization
• Posterior PDF is given by conditional probability

25
Summary of BME analysis of TGIS

• General KB
• Mean trend
• Covariance

• Site-Specific KB
• Hard Data

BME
Estimation Point
Data Point
26
BME Analysis
27
S/TRF for Log-transformed PCE concentration

• S/TRF representing Log-tranformed concentration
• Residual field describes purely stochastic aspect
  of the concentration field

Mean Trend
Residual Field
28
Mean Trend of Log-transformed concentration field

• Mean trend consist of two components
• Purely spatial component
• Purely temporal component
• Each component is calculated by exponential
  smoothing

29
Mean Trend Temporal Component

• Increase from Jan. 1999 to Jan. 2003
• Decrease from Jan. 2003

30
Mean Trend Spatial Component

• Contaminated Area
• Northeastern region
• Southwestern region

31
Homogeneous/Stationary S/TRF
Log-transformed data
Removing the mean trend
Residual data for S/TRF

• Homogeneous/Stationary Random Field
• Its mean trend is constant
• Its covariance is only function of the spatial
  lag and temporal lag

32
Covariance for Residual S/TRF
33
Covariance for Residual S/TRF
34
Covariance Surface
35
Results of BME Analysis
36
BME Estimation Temporal Fluctuation
37
BME Estimation Spatial Distribution
38
BME Estimation Spatial Distribution
39
BME Estimation Spatial Distribution
(Apr. 15, 2002)
40
BME Estimation Contaminated Area
Area above the quality standard 0.388µg/L
(Apr. 15, 2002)

• BME mean estimate
• Upper bound of the BME 68 confidence interval
• Upper bound of the BME 95 confidence interval

41
BME Estimation Along River Stream

• Equidistance points along river stream
• More accurate estimation for surface water

42
BME Estimation Along River Stream

• Fraction of river miles that does not attain the
  quality standard

43
New Criterion
44
Assessment Criterion

• S/TRF is characterized by Posterior PDF
• Area under the curve Probability

ProbPCEgtQSTDArea under the curve
(QSTDltPCElt8)
45
Assessment Criterion

• ProbNon-AttainmentProbPCEgt0.388µg
• Highly Likely in Attainment
• ProbNon-Attainmentlt10
• Highly Likely in Non-Attainment
• ProbNon-Attainmentgt90
• Non-Assessment
• 10?ProbNon-Attainment?90
• More Likely Than Not in Non-Attainment
• ProbNon-Attainmentgt50

46
Fraction of River Miles
47
Identifying Contaminated WMAs

• The state of New Jersey is divided into 20
  Watershed Management Area (WMA)
• Assess which part of the state is contaminated
• Contribution of each WMA to the fraction of river
  miles assessed as
• Highly Likely in Non-Attainment
• More Likely Than Not in Non-Attainment

48
Contribution of WMAs

• Highly Likely in Non-Attainment

49
Contribution of WMAs

• More Likely Than Not in Non-Attainment

50
Fraction of River Miles in WMAs
51
Model Comparison
52
Model Comparison Error Variance
Space/Time Analysis
Purely Spatial Analysis
(Feb. 5, 2000)
53
Model Comparison Cross-Validation
Space/Time Analysis
Purely Spatial Analysis
54
Model Comparison Cross-Validation
55
Model Comparison Fraction of River Miles
Space/Time Analysis
Purely Spatial Analysis
56
Conclusion
57
Conclusion

• About Monitoring Data
• Some high concentration values are observed in
  New Jersey between 1999 to 2003.
• Monitoring data shows high Space/Time variability
  in terms of location of the monitoring point and
  monitoring value
• Application of BME method of TGIS
• It enables us to take into account high
  space/time variability and to estimate the
  concentration all river reaches

58
Conclusion

• New Criterion
• New criterion takes into account the uncertainty
  information of posterior PDF
• It is used to complementary criterion for the
  conventional one
• Fraction of the river miles assessed as Highly
  Likely in Non-Attainment reached about 0.45 in
  2000
• Fraction of the river miles assessed by the
  conventional criterion (More Likely Than)
  reached about 1.8 in 2002

59
Conclusion

• Model Comparison
• Space/Time analysis produces more accurate
  estimation than the conventional purely spatial
  analysis
• Space/Time analysis produced very different
  estimate
• In purely spatial analysis, non-assessment river
  miles reach about 99
• NJ DEP will be able to better assess PCE
  concentration in all river reaches by using this
  method and new criterion