A PARADIGM FOR SPATIAL MAPPING OF GROUNDWATER CONTAMINATION IN RURAL SETTINGS: LESSONS FROM ARSENIC

1
A PARADIGM FOR SPATIAL MAPPING OF GROUNDWATER
CONTAMINATION IN RURAL SETTINGS LESSONS FROM
ARSENIC CONTAMINATION IN BANGLADESH

• Faisal Hossain
• Department of Civil and Environmental Engineering
• Tennessee Technological University

2
Collaborating Institutions

• Tennessee Technological University
• Tri-State University Jason Hill
• University of Connecticut Dr. A.C. Bagtzoglou
• Griffith University, Australia Dr. B. Sivakumar
• University of Texas, Austin Dr. Sanjay
  Srinivasan (and Louis Forster)
• Supporting Organizations in Bangladesh
  Institute of Water Modeling, Rajshahi University,
  Bangladesh Council for Scientific and Industrial
  Research (BCSIR), Ministry of Environment
• Others Nurun Nahar, Md. Delawer Hossain, Sayma
  Rahman, Shamshuddin Shahid, Abu Saleh Khan,
  Mafizur Rahman

3
Spatial Mapping 101
4
Groundwater Contamination and Bangladesh
LOW
LOW
HIGH
HIGH
HIGH
From British Geological Survey
5
WHAT MAKES SPATIAL MAPPING OF ARSENIC CHALLENGING?
1. More than 50 of Bangladesh population at risk
due to Arsenic contamination in shallow aquifers
(lt 150m).
2. Total number of shallow operational drinking
wells UNKNOWN (10-18 million).
3. Most shallow wells privately-owned, sunk
randomly at short notice, difficult to be updated
through inventory control.
4. UNICEF/Government has tested about 4-5
million wells using semi-quantitative field kits.
5. Field kits have large false positives and
negatives.
Question? How can we spatially map aquifer
contamination for rural Bangladesh with limited
in-situ sampling data?
6. Accurate and time-varying sampling of
groundwater quality EXPENSIVE. 7. Spatial mapping
accuracy depends of accurate sampling at adequate
resolution
6
Justification for a New Spatial Mapping Scheme
for Developing Nations

• The near impossibility of testing every single
  shallow well in Bangladesh requires a simulation
  methodology that can accurately characterize a
  well as being safe/unsafe without the need for
  extensive and expensive in-situ sampling tests.
• Such a method can then act as a fast-running and
  inexpensive proxy to the time-consuming lab-based
  field campaigns and save considerable testing
  resources by judiciously directing them to those
  wells pre-determined by the simulation method to
  have a high likelihood of being unsafe.
• Furthermore, by flagging a safer cluster of wells
  functional (and an unsafe cluster of wells as
  non-operational), villagers are expected to find
  this approach socially more convenient than the
  more expensive house-hold treatment options
  currently available in Bangladesh.

7
Outline of Seminar
1st Part

• Overview of Groundwater Arsenic Contamination in
  Bangladesh
• Spatial extent and results from a recent
  social survey.
• Development of Paradigm on Spatial Mapping
• Marriage of Non-Linear Chaos Theory with
  Linear Stochastic Dynamics.
• Assessment of conventional geostatistical methods
• Chaos-based analysis of spatial pattern of
  arsenic
• Merit of a chaos-based approach
• Vision for the Future
• A Cost-effective Mapping Scheme that
  integrates the physics of contamination

2nd Part
3rd Part
8
A DISCLAIMER

• Work presented from a data-based perspective
  of spatial mapping.

9
PART ONE

• OVERVIEW

10
Overview of Arsenic Contamination in Bangladesh

• First case of Arsenic in 1993
• Nationwide (BGS-DPHE) survey indicated widespread
  arsenic - shallow ground water (lt150 m)
• 80 of population (gt 100 million) depend on
  shallow ground water.
• Arsenic is geologic in origin Pleistocene-Holocen
  e
• Million health cases per year projected.
• A major Health Disaster in the making for the
  rural poor

11
Large-Scale Remedial Efforts(For Rural
Bangladesh)

• Exploration of the potential of deep aquifers
  High and Long Waiting Time
• Understanding the mechanism of Arsenic
  contamination for long-term structural solution
  High and Waiting Time
• Closing cluster of unsafe shallow wells.
  Replacement with treatment options or drill safe
  shallow wells. Low-Medium and Waiting Time

12
Limitations of Ongoing Efforts (Option 3-Closing
Wells)

• Safe/Unsafe Well detection is by UNRELIABLE FIELD
  KITS (but very inexpensive).
• Total Number of Shallow wells UNKNOWN (10-18
  million). Lack of accurate census.
• Testing every well and flagging it manually
  requires LONG WAITING TIME (only 4-5 million
  wells tested so far) Inventory is dynamic
  (difficult to monitor).
• Not all treatment options (e.g., Filter) appeal
  to the rural public. Social compatibility
  issues.

Villagers would rather travel faraway once a day
to collect their drinking water (after Haque et
al., 2004, Public Health) Our recent surveys
confirm this
13
(No Transcript)
14
Why are Field Kits Unreliable?

• Semi-quantitative based on subjective
  interpretation.
• Detection is probabilistic
• (4 possible outcomes)
• Successful Safe Well Detection
• Unsuccessful Safe Well Detection (False Alarm)
• Successful Unsafe Well Detection
• Unsuccessful Unsafe Well Detection (False Hope)
• Safe/Unsafe according to 50 or 10 ppb limit

Major Field Kit Brands in use in Bangladesh
False Hopes Silent Poisoning False Alarms
Unnecessary and Time wastage
15
Reliability Analysis of Field Kits
Asia Arsenic Network (AAN) kit- After Hossain et
al. (2006) Hydrological Processes
After Rahman et al. (2002)- Env. Sci. Technol.
16
Cost-effectiveness of Field Kits Should we
discard them altogether?
Field Kit test requires minimal training for
staff (unlike AAS)
Field Kit tests are quick and dirty, no
complicated protocols to follow
Field Kits are highly portable and commercially
available in bulk qty
Source NAISU NGO Arsenic Information Support
Unit www.naisu.org
Field Kits can be soft data
17
What type of Spatial Mapping Scheme do We Need
for Rural Settings?

• A rapid and low-cost methodology to identify
  cluster of unsafe shallow wells for immediate
  closure.
• Justifications
• Rapid? Too many wells- status unknown (10-18
  million). Reduce exposure (Time is of the
  essence)
• Low-cost? Rural setting (Money is of the
  essence)
• Closure of Unsafe well cluster? The major
  difficulty/unknown of remediation effort is
  accurate identification-

Implications for Cambodia, Vietnam, Mexico, West
Bengal (India)
18
Attitude of Rural PublicAfter a Decade Results
from a Recent Survey

• Current awareness among villagers is GOOD.
• Haque et al. (2004) survey indicates that not all
  treatment options are socially compatible.
• Villagers seem to prefer minimum maintenance,
  high flow rates, central distribution system.
• Traditional water collection by females is still
  widespread.
• Our Survey (conducted by Nurun Nahar of Japan
  Advanced Institute of Sci and Tech JAIST and
  Ministry of Environment, Bangladesh)

19
Attitude of Rural PublicAfter a Decade Results
from Our Survey
20
Attitude of Rural PublicAfter a Decade Results
from Our Survey
21
Part Two
DEVELOPING THE PARADIGM FOR SPATIAL MAPPING
22
Conventional Spatial Mapping Scheme (Estimating
at Unsampled Locations)
1. Conventional approach is Geostatistical E.g.
Kriging (and many others)
2. Based on linear stochastic dynamics. 3.
Estimate at unsampled location is a weighted
linear combination
23
Assessment of Ordinary Kriging for Arsenic
Contamination
24
Assessment of Ordinary Kriging
1. General trends are picked up satisfactorily at
scales of gt 50 km with large underestimation. 2.
Kriging misses the local hot spots due to its
smoothing function (Conditional Indicator
simulation may be needed)
25
Indicator Kriging/Simulation

• Does not require assumption of normality.
• Good for threshold-based estimation.
• Handles skewness well (zeros, detection limit
  issues etc.).
• Handles scarce data well.
• 3-D Indicator kriging.

Indicator kriging represents the spatial
continuity of higher arsenic concentrations more
accurately than ordinary kriging.
Sanjay Srinivasan and Louis Forster ongoing work
26
Why Search for Alternative Approaches for Mapping?

• Why?
• Conventional Geostatistical methods are two-point
  schemes linear correlation between two points
  separated by a lag h pattern filling approach
• Linear Stochastic in nature makes no
  recognition of deterministic nature of data (the
  physics) (treats uncertainty as irreducible)
• Simplifies spatial pattern manifested by complex
  interactions between geology and time-sensitive
  fluid flow dynamics.

Arsenic in groundwater is not a purely random
occurrence and that there exist distinct
geological and geochemical factors controlling
its variability. It is no longer defensible to
continue to use pure geo-statistical approaches
of pattern filling as stand-alone techniques for
its spatial interpolation in resource poor
settings that are typical of developing nations.
27
Chaos Theory as an Alternative Approach

• Why Chaos Theory?
• Evidence for a number of hypotheses on arsenic
  contamination have been observed in Bangladesh.
• Opinion Poll by Akmam (2002) revealed lack of a
  unifying theory
• 58 support for Oxy-Hydroxide Reduction
  hypothesis
• 33 support for Pyrite Oxidation hypothesis
• 75 support groundwater extraction causes
  arsenic release
• Each hypothesis can be represented as a sum of at
  least 3 partial differential equations
  necessary condition for phenomenon to exhibit
  Chaos
• Chaos theory is based on non-linear deterministic
  theory and can potentially bridge the gap between
  mechanistic understanding (physics) and pure
  stochastic modeling (data-based).
• Correlation Dimension (CD) is one measure of
  chaos Grassberger-Procaccia Algorithm.

28
Evidence of Chaos in Arsenic data in Bangladesh
Correlation Dimension Analysis

• Hossain and Sivakumar (2006), Stochastic Env.
  Res. and Risk Analysis demonstrated Chaos in
  Arsenic spatial variability using BGS data.
• Correlation Dimension of 8-10 observed. Embedding
  Dimension of 10-12.
• Arsenic contamination in space, from the chaotic
  point of view, is a medium- to high-dimensional
  problem.

The Hénon map is given by x(t) a b y(t-1)
x(t-1)2 y(t) x(t) a1.4, b0.3.
Deterministic Randomness
Hypothesis - At least 8 variables/dimensions
needed to optimally model spatial variability
deterministically
29
Assessment of Correlation Dimension
Is Correlation Dimension a Reliable Proxy for
the Number of Dominant Influencing Variables for
Modeling Risk of Arsenic Contamination in
Groundwater?
1. Using Ordinary Logistic Regression Models, the
value of CD as a proxy was assessed
lnp/(1-p) logit (p) a ßx
Where, p probability of a well exceeding a
concentration limit xvector of influencing
variables a is a constant, ß is a vector of
slope coefficients
2. INFLUENCING VARIABLES? CD does not inform on
the choice but only the optimal number of
variables in a deterministic model
30
Assessment of Correlation Dimension
All possible combinations of LR models considered
2048 combinations
Uncertainty associated with prediction of wells
as safe and unsafe by LR model declines
systematically as the total number of influencing
variables increases from 8 to 11.
Sensitivity of the mean predictive performance
also increases noticeably for this range.
31
Part Three
VISION FOR THE FUTURE
32
The Future of GW Contamination Mapping in Rural
Settings
Where are we right now?
Linear Geostatistical techniques two-point only
correlation, smoothing filter, misses local hot
spots, treats contaminant as a pure random
variable with no regard for the physics behind
the spatial variability
Chaos-based non-linear models does not treat
contaminant as a pure random variable
deterministic randomness can be quantified
Correlation Dimension appears to have merit as a
proxy in deterministic models but does not
prioritize influencing variables
Can we use current arsenic geochemistry knowledge
to prioritize influencing variables in
chaos-based non-linear models? Can chaos-theory
be a bridge between linear stochastic techniques
and contamination physics? Can we use multiple
point techniques?
33
New Paradigm for Spatial Mapping

• Combines two paradigms Geostatistical Paradigm
  and Chaotic Paradigm
• Geostatistical Paradigm Pattern Filling
  (Kriging and/or conditional simulation)
• Chaos Paradigm Pattern Recognition (number of
  variables defining spatial variability)
  Physics-based
• Combining both may increase success rate of
  identifying unsafe wells (reducing false hopes).

Success Ratio (Geostatistics.AND.Chaos Theory)
IS GREATER THAN Success Ratio
(Geostatistics.OR. Chaos Theory) ?
Implications for any contaminant variable and
other rural regions Southeast Asia, Mexico,
South America probably the US
34
General Formulation of Our Mapping Scheme

• Theoretical formulation recognizes explicitly the
  complex fluid flow patterns through multiple
  connection statistics
• Mapping scheme explicitly integrates the dominant
  physical knowledge in the parameterization of the
  chaos-based models

Calibration data requirement should stay invariant
Integration of techniques is Bayesian and treated
as a priori for indicator simulation
35
The Future of Mapping in Rural Settings

• Agenda what is needed to move forward?
• Greater Collaboration with community engaged in
  mechanistic understanding of arsenic
  contamination (geologists, soil geochemist,
  groundwater hydrologists etc) to identify
  influencing variables and integrate them
  physically in chaos-based mapping schemes.
• Assess enhanced geostatistical methods
  Conditional Indicator Simulation
• Address the transient nature of the problem
  leverage realtime environmental monitoring
  network
• Implement the proposed scheme (and paradigm) in
  real-world using limited sampling data.
• Search for ways to generalize the approach for
  any contaminant variable under a resource-poor
  setting for a developing country.

36
SALIENT POINTS (Current Progress Report)

• Typical Field Kits used in Bangladesh have large
  false positives and negatives (25-80). Social
  survey indicates villagers willingness to
  walk/pay.
• Mainstream linear geostatistical methods for
  spatial mapping are inadequate for locating local
  scale hot spots/variability at scales lt 50 km.
• Mainstream geostatistical methods smoothen the
  complexities of contamination and should be
  augmented with enhanced methods. Examples are
  non-linear chaos, multi-point and indicator
  kriging.
• Arsenic contamination exhibits clear
  deterministic dynamics in spatial pattern
  sensitive to geology.
• Correlation Dimension analysis indicates 8 or
  higher influencing variables needed to spatial
  model variability optimally.
• Correlation Dimension has information value as a
  rapid proxy (at least for Logistic regression).
• As a path forward, greater collaboration is now
  needed with community on mechanistic
  understanding of contamination to bridge the gaps
  between mapping scheme and integration of physics
  in the interpolation.

37
Relevant Publications (available at
iweb.tntech.edu/fhossain/publications.html)
1. Hossain, F., A.C. Bagtzoglou, N. Nahar and
M.D. Hossain. (2006). Statistical
Characterization of Arsenic Contamination in
Shallow Tube wells of Western Bangladesh.
Hydrological Processes. vol. 20(7), pp. 1497-1510
(doi10.1002/hyp.5946).
2. Hossain, F. and B. Sivakumar. (2006). Spatial
Pattern of Arsenic Contamination in Shallow
Tubewells of Bangladesh Regional Geology and
Non-linear Dynamics Stochastic Environmental
Research and Risk Assessment , vol 20(1-2), pp.
66-76 3. Hossain, F. and B. Sivakumar (2007).
Spatial Interpolation of Contaminantion based on
Linear and Non-linear Paradigms for Developing
Countries, Stochastic Environmental Research and
Risk Assessment (Revised and in review).
4. Hossain, F., A.J. Hill, and A.C. Bagtzoglou
(2006). Geostatistically-based management of
Arsenic Contaminated Ground water in Shallow
wells of Bangladesh. Water Resources Management.
(In press, doi 10.1007/s11269-006-9079-2)
5. Hill, A.J., F. Hossain and B. Sivakumar.
(2006). Is Correlation Dimension a Reliable Proxy
for the Number of Influencing Variables required
to Model Risk of Arsenic Contamination in
Groundwater? Stochastic Environmental Research
and Risk Assessment, (In press doi
10.1007/s00477-006-0098-6).
6. Nahar, N., F. Hossain, and M.D. Hossain
(2007). Health and Socio-economic Effects of
Groundwater Arsenic Contamination in Rural
Bangladesh Evidence from Field Surveys,
International Perspectives Journal of
Environmental Health. (Provisionally accepted)
7. Rahman, S., and F. Hossain . (2007). A
Forensic Look at Groundwater Arsenic
Contamination in Bangladesh, Environmental
Forensics. 8(4), December (In press)
38
Acknowledgements

• Center for Management, Protection, Utilization of
  Water Resources, Tennessee Technological
  University (TTU)
• Department of Civil and Environmental
  Engineering, TTU
• Office of Sponsored Research, TTU
• Ministry of Environment, Bangladesh
• Institute of Water Modeling, Bangladesh (5-year
  MOU with TTU)
• Bangladesh Council for Scientific and Industrial
  Research (BCSIR)
• British Geological Survey and Department of
  Public Health, Bangladesh
• Asia Arsenic Network (Japan)
• Rajshahi University, Bangladesh
• And many other friends and colleagues

39
Thank You!
When a large portion of the rural population
continues to suffer from the arsenic calamity,
we, the more fortunate ones with time to
brainstorm, have the responsibility to critically
assess any novel idea until a long-term
structural solution is in the horizon.
Questions?