Industrial ecology of earth resources
EAEE E4001
MANAGING THE WATER RESOURCES
INVITED LECTURE BY PROF. UPMANU LALL
Decision Analysis Tools for
Total Maximum Daily Loads
- EPA's Water Quality Management Program
Presentation by Prof. Upmanu Lall,
Earth and Environmental Engineering
Outline
• The Playing Field
– The Clean Water Act, Water Quality and the EPA
• The Problem
– Management or Regulation ? Data, Science,
Problem Scale and Decisions
• An Approach
– Emphasis on Science to support Decision Process
– Bayes Networks + Packaging
Fires plagued the Cuyahoga beginning in
1936 when a spark from a blow torch ignited
floating debris and oils. Fires erupted on the http://www.cnn.com/NATURE/9906/22/saving.cuyahoga/
river several more times before June 22,
1969, when a river fire captured national
attention when Time magazine described the
Cuyahoga as the river that "oozes rather than
flows" and in which a person "does not drown
but decays." This event helped spur an
avalanche of pollution control activities
resulting in the Clean Water Act, Great Lakes
Water Quality Agreement, and the creation of
the federal and state Environmental
Protection Agencies.
Key Legislation mandating EPA’s role in Water
National Environmental Policy Act, 1969: Environmental Assessments (EA's) and Environmental
Impact Statements (EIS's) for all federal activities
Federal Water Pollution Control Act 1972 : Regulates discharges of pollutants to waters
Endangered Species Act, 1973: Conservation of threatened/endangered plants and animals and the
habitats in which they are found
The Safe Drinking Water Act, 1974, 1996: Protect the quality of all waters actually or potentially
designed for drinking use, whether from above ground or underground sources. EPA to establish
safe standards of purity and required all public water systems to comply with primary (health)
standards. State governments, also encourage attainment of secondary standards (nuisance).
The Clean Water Act 1977: Focus on toxics. EPA gets authority to set effluent standards on an industry
basis (technology-based) and water quality standards for all contaminants in surface waters. The
CWA makes it unlawful for any person to discharge any pollutant from a point source into
navigable waters unless a permit (NPDES) is obtained.
Comprehensive Environmental Response, Compensation, and Liability Act, 1980: Federal
“Superfund” to clean up uncontrolled or abandoned hazardous-waste sites as well as accidents,
spills, and other emergency releases of pollutants and contaminants into the environment
The Clean Water Act 1987: authorized citizen suit provisions, and funded sewage treatment plants
(POTW's) under the Construction Grants Program. EPA can delegate many permitting,
administrative, and enforcement aspects of the law to state governments.
Resource Conservation & Recovery Act, 1976, 1986: Underground Storage Tanks, Non-Haz Waste
EPA
Assessed Rivers, Lakes, and Estuaries Meeting All Designated Uses
1994/1996 Using Latest State Information Reported - EPA
Percent of Impaired Waters - 1998
-EPA
Major Water Quality Concerns Today
•
Non-point source Pollution (e.g., Agricultural Sources, Urban Runoff)
– Nutrients (P, N), Sediment, Stream temperature, Dissolved Oxygen, Pathogens,
Pesticides
– Organics in Urban Runoff
– Eutrophication
– Endangered Species/Habitat/Riparian Zone, Recreation Impacts
– Climate Variability and Dynamic Range of Biophysical Processes
– Rapid Urbanization/Shifts in Land Use
– Competition between Environmental and Agricultural/M&I demands
– Paucity of Data in Space and Time
– Limited Understanding of Long Term Impacts in Ephemeral Streams
-EPA
Sediments
Nutrients
Pathogens
Dissolved Oxygen
Metals
Habitat
pH
Suspended Solids
Temperature
Flow Alterations
Pesticides
Noxious Plants
Turbidity
Fish Contamination
Ammonia
0
% of Water Segments
18
Sediment Runoff Potential - 1990-1995
Nitrogen Runoff Potential - 1990 -1995
Pesticide Runoff Potential - 1990 -1995
Fish Consumption Advisories - 1997
Key Points
• TMDLs - legal domain dominates process - policy or science issue ?
•
•
•
•
•
Major Public Sector Expenditures and Urgency - Legal Impetus
• 40,000 TMDLs @ $100k each = $4 billion EPA expense in 10 yrs
There are significant data gaps to even evaluate existing conditions
Natural climatic variability leads to short and long range effects on the
landscape that complicates the assessment of sediment, nutrient, and
other inter-related “loads” that result from modified land use practices
Various “best” management practices have been proposed. Little is
known about their efficacy
How does one develop management/regulation plans for specific
watersheds in this environment ?
– Scale of watershed, location of source, Equity
The “Local” Decision Problem
Collaborative Decision Making as a
Watershed Management Approach
- The Physical Scientists Role
TMDL: Total Maximum Daily Loads
Key aspects:
For each stream reach listed for a specific pollutant
• Identify Beneficial Use
•Identify Sources (point and non-point) & Background Loading
•Allocate Loads to all Point and Non-Point Sources + Margin Of Safety
such that water quality standard(s) relative to beneficial use designation are met
Stakeholders
Regulatory
Constraints
Management
Options
Time Point &
Time Frame
Desired State
Decision
Human
Activities
Values of
Activity
Outcomes
Management
Goals
Natural State
Water
Bodies
Eco-environmental
State
Natural Processes
E.g. Climate, Geology
Information Scale, Relevance, Availability, Source and Cost
Typical “Stakeholders”
• Regional EPA Offices
• State Departments of Environmental Quality, Water Rights, Fish and
Wildlife, Natural Resources, Agriculture, Planning and Budget
• The Western Governors Association (WGA)
• Federal Land and Resource Managers – e.g., USFS, BLM, USBR,
NPS, USFWS, NRCS, USDA, DOE, DOD
• Association of County Governments and County Planners, Irrigation
and Water Conservation Districts, Major Industries and Utilities
• Environmental Organizations and Action Groups, Farm Groups
• Information Suppliers: The scientific community - Universities and
other Researchers
THE TMDL EQUATION as per EPA
TMDL = Sum of WLA + Sum of LA + MOS
• TMDL = WQ Standard for Beneficial Use Designation
• WLA = Waste Load Allocation for each point source
NPDES Permits
• LA = Load Allocation for each non-point source
BMPs
• MOS = Margin of Safety (e.g. 20% of TMDL)
Socio-Economic Factors ? Uncertainty ? Variability ?
Develop for each “reach”. Watershed Sequencing ? Trading ?
Current Modeling Approaches
 Lumped, and spatially distributed simulation models for overland and channel flow
generation given landscape information and climatic time series
 Associated models for modeling the transport, reactions, decay and mixing of
contaminants in a river channel and/or non-point source load generation
 Lumped and spatially distributed discrete time simulation models for groundwater
flow and transport and surface-water exchanges
 "GIS Models" that use spatial landscape data and simple physics or statistical
approaches to estimate mean fluxes at a point or perform discrete time simulations
of the system given assumptions on exogenous climate and pollutant application
 Statistical approaches based on linear or nonlinear (including neural network)
regressions to provide estimates of the mean value (and variance) of loads and
fluxes given specific indicators.
 Statistical indicators of ecological condition of the watershed
 Ecological models of habitat dynamics conditional on exogenous forcings
• Statistical and dynamical models of economic markets and user preferences
Focus on data and decision questions or on unit process modeling ?
Do decision makers react to results of process simulations or to the
chance of key outcomes ?
Source: Jarrell (1999)
Approach
• TMDL Management Plan = translate regulatory needs,
competing stakeholder objectives into monitoring system
design and watershed “operation” strategy given evolving
data and goals
• Focus on Variables/Processes needed for decision analysis and
their interconnections - represent as a directed network
• Recognize that unit process models provide means to connect
variables on network
• Explicitly consider role of natural variability and ignorance in
defining uncertainty - Bayes network
• Communication - Visualization / Presentation Tools
– Simcity/Simearth motif with BayesNet providing SimRules
– Limiting Probabilities vs Time Simulation ?
Margin of Safety - Some thoughts
Example
1 point & 1 Non-point source
Concern: Dissolved Oxygen
Riparian
Buffer
Ranch
Natural
Area 1
Reach 1
Reach 2
Natural
Area 2
Beet
Factory
Natural
Area 3
Reach 3
Secondary
Treatment
MOS = TMDL - Total Loads
FS = DOmean/TMDL
DO pdf w/ Beet factory only
DOmean=6.6
FS=1.34
DO<TMDL
3.5% of time
DO pdf w/ Beet factory + ranch
DOmean=6.2
FS=1.24
DO<TMDL
23% of time
DO pdf w/ Ranch only
DOmean=7
FS=1.4
DO<TMDL
11% of time
The NPS (Ranch) has a higher FS
than Beet Factory, but worse risk
of violation of the standard
For the case with both the beet
factory and the ranch, we have a
FS>1, but the TMDL is violated
23% of the time. Is this acceptable
for ecological impacts ?
DO pdf w/ Beet treatment
DO pdf w/ Riparian Buffer
DOmean=7
FS=1. 4
DOmean=7
FS=1.4
DO<TMDL
12% of time
DO<TMDL
5% of time
DO pdf w/ Beet treatment +
riparian buffer
DOmean=7.2
FS=1.44
DO<TMDL
1% of time
The 2 treatment options
individually lead to the same FS,
but the Riparian buffer leads to a
lower risk of TMDL violation.
Trading ?
Is a 5% risk acceptable or do we
need both options to reduce the
risk to 1% ?
Setting up a Bayesian Network to Determine the Probabilities of Outcomes
Example
1 point & 1 Non-point source
Concern: Dissolved Oxygen
Riparian
Buffer
Ranch
Natural
Area 1
Reach 1
Reach 2
Natural
Area 2
Beet
Factory
Natural
Area 3
Reach 3
Secondary
Treatment
Climate
Conditions
Economic
Conditions
Treatment
NA1
BOD
Ranch
BOD
Riparian Buffer
Beet
BOD
NA2
BOD
R1
DO
R2
DO
R3
DOcrit
Costs and Benefits
Network Diagram of Key Elements of the Physical System
Bayes Network if “Local” Conditional Probability
Distributions are defined for each Parent-Child Link set
Beet
BOD
The Bayes Network can be
reduced to focus on key
items of interest
Ranch
BOD
R3
DOcrit
Conditional Probabilities
are evaluated from data
or simulation models or
can be specified by an
expert
Ranch BOD
(mg/l)
0-30
30-60
60-90
0-30
30-60
60-90
Beet BOD Reach 3 critical DO (mg/l)
(mg/l)
0-5
5-8
>8
0-10
0
0.3
0.7
0-10
0.1
0.5
0.4
0-10
0.2
0.6
0.2
10-20
0.1
0.4
0.5
10-20
0.25
0.65
0.1
10-20
0.35
0.65
0
East Canyon Reservoir, Utah
A case study
Lost Cr. Res.
Lost Cr.
Stoddard
Diversion
Echo Cr.
Chalk Cr.
Echo Res.
East
Canyon
Cr.
East
Canyon
Res.
Rockport Res.
Silver Cr.
Concerns
• Current problems
– loss of cold water fisheries (stream and flat water)
– reduction in use of State Park at East Canyon Reservoir
• Annual Visitors (300K in 1986, now <100K)
– periodic low dissolved oxygen in stream
– eutrophication of reservoir
• Future concerns
– development - population to grow 400% in 20-30 years
– conversion of agricultural to urban lands
– interbasin water transfers that may reduce flushing
Goals
• Control of Contaminant Discharges into East Canyon
Cr. and Reservoir
– Determination of total loading of a particular pollutant into
a water body
• point and non-point sources
• overland flow
• ground water
– Need for discharge management to achieve water quality
goals
Approach
• Analysis of data
– identification of data sources
• U.S. EPA STORET repository, USGS repository, State of
Utah, Results of university and other studies, data held by
stakeholders
– probability distributions for hydrologic, water quality,
and contaminant source inputs
• exploratory data analysis, parametric distribution fitting, nonparametric distribution ‘fitting’
– methods for finding probability distributions of outputs
• Bayes nets, deterministic models
East Canyon Conceptualization
East Canyon
Reservoir
Snyderville
Basin WWTP
Kimball
Junction
Kimball Junction - Headwater
TP
Temp
Snyderville WWTP - Point
Source Temp
TP
Flow
Flow
East
Canyon
Creek
East Canyon
Reservoir
Models to route probabilities
• Input Variables from Historical Data
• QUAL2E/UNCAS - EPA developed and
maintained
– stream water quality and temperature
•
•
•
•
BOD/DO
nutrients
algae/Coliform bacteria
user-defined constituents
– detailed representation of stream hydraulics
– error propagation integral to model
Results - Reservoir Inflow
Fre que ncy Distribution - Total Phosphorus
800
700
Target Value
Frequency
600
500
400
300
200
100
0
0.001
0.01
0.1
1
Reservoir Influent P concentration, mg/L
10
Results
Reservoir Status
KFLOW
L
L
L
L
H
KP
L
M
H
H
H
WFLOW
H
H
H
M
M
WP
H
H
H
H
H
L
0.01
0.01
0.03
0.01
0.12
RP
M
0.01
0.01
0.03
0.19
0.12
High Probability of M
L
H
H
L
L
L
H
L
M
M
L
L
M
M
M
M
M
H
M
H
M
M
M
M
0.10
0.17
0.15
0.30
0.31
0.06
0.75
0.73
0.70
0.68
0.68
0.67
0.15
0.10
0.15
0.01
0.01
0.27
High Probability of L
H
L
M
M
0.42
0.42
0.15
High Probability of H
H
0.98
0.97
0.94
0.80
0.76
A future TMDL Strategy as a Bayes Net
Forecast
Monitoring
Strategy
Monitoring
Cost
Stream
flow
Prior
Water
Quality
Treatment
Strategy
Posterio
r
Water
Quality
Treatment
Cost
Benefit
Total Cost
Net Benefit
An Approach to the Analysis of Sediment Loads
Questions Addressed
• Where are the most erodible soils located within the watershed?
• Where do the eroded materials move to; i.e., what stream
segment?
• What are the high priority areas to address erosion/
sedimentation for the watershed?
• What stream segments are at greatest risk from sedimentation?
• How would implementation of BMPs in selected areas of the
watershed affect the erosion rate, mass, and loading into the
receiving streams?
• How is the amount of eroded material attenuated as it moves
over and through different areas with varying land cover/use,
soils, and slope?
GIS input data
•
•
•
30 m DEM based on USGS data
River Reach Files
SSURGO soils data (USDA-NRCS)
Hierarchical Estimation of Sediment Load
TH1
H1
DH1
TH3
TC1
LH1
DC1
C1
H2
LC1
TH2
LH2
TC2
T: Transport,
DC2
LC2
C3
D: Detachment,
LH3
DH2
H3
C2
DH3
TC3
DC3
LC3
L: Load
Hillslopes
LH1 = Min(DH1, TH1), LH2 = Min(DH2, TH2), LH3 = Min(DH3, TH3)
Channels
LC1 = Min(DC1+LH1, TC1), LC2 = Min(DC2+LH2, TC2),
LC3 = Min(DC3+LC1+LC2+LH3, TC3)
A Bayesian Framework
Distribution of Hillslope Transport Capacity (TH)
1.2
With Riparian Buffer
1
0.8
No Riparian Buffer
0.6
0.4
0.2
Full Bayesian Dependence Network
0
Precip
To all D and
T nodes
0
TH1
DH1
LH1
TH3
TC1
LC1
LH2
To TH3
3
4
5
DH3
LH3
TC2
LC2
DC2
TC3
RB%H1
LC3
DC3
6
Consolidated Bayesian Dependence
Network
Watershed response
evaluated in GIS
DH2
RB%H2
RB%H3
2
DC1
RB%H1
TH2
1
RB%H2
RB%H3
Decision variables
LC3
Precip
Functional Elements of Computer
Package
Database
Spatial, Environmental, Economic and Regulatory
Electronic Library
Case studies, beneficial uses, stakeholder information
Decision Process Guide
Walk through decision process, flag data, prescribe analyses
System Representation Tools
Physical representation of the landscape and control points,
pollution sources and sinks
Process representation using Bayes networks
Analysis Tools
Statistical and Mechanistic Models
Using the System to Develop TMDLs
Project Setup
•Identify sub-basin
•Organize reaches by stream order
•Determine listing status
•Select pollutant for TMDL
•Organize data sets
Sub-basin assessment
For each reach in sub-basin:
•Determine beneficial use designation
•Determine standards for selected pollutant
•Identify sources
•Characterize loads from each source
•Generate probabilities from data
•Identify management options
•Evaluate management options
Load allocation
•Visualize system interconnections
•Display causal network
•Apply management options
•Assess optimal economic load allocation strategy
Information
•Deterministic Models
•Regional statistical analysis
Project Environment
A standalone
Windows-PC
application (no
extra software
required). Live
links to local or
Internet-based
data sources, GIS
capabilities and a
guidance system
that walks the
user through the
data analysis,
modeling and
decision-making
processes.
Internal Data Processing Graphical Scripting
A schematicbased data
processing
interface
provides the user
with an intuitive
“drag and drop”
ability to process
and prepare both
temporal and
spatial data for
use in the system.
External Data ProcessingRegional Analysis
Commercial GIS
products such as
ArcView are used
to conduct a
regional data
analysis of all water
quality stations in
similar watersheds.
Regional Statistical Analysis
A conditional probability
table is generated from
monitoring stations
throughout the adjoining
ecoregions.
Total
Nitrogen
Medium
High
Low
Medium
High
Low
Low
High
High
Medium
High
High
Medium
High
Low
Medium
Low
High
High
High
Medium
Low
Low
Medium
Low
Contributing
Area in
Contributing
Buffer Zone
Basin Area
Medium
Medium
Medium
High
Medium
Low
Medium
Medium
Medium
High
Medium
Low
Medium
Low
Medium
High
Medium
High
Medium
Medium
High
High
Medium
High
Medium
Medium
Medium
High
Medium
Low
Medium
Medium
Medium
Low
Medium
High
Medium
High
High
High
Medium
Medium
Medium
Low
Medium
Low
Medium
Medium
Medium
Low
Land Use
in Buffer
Zone
2
3
1
2
3
1
1
3
3
2
3
3
2
3
1
2
1
3
3
3
2
1
1
2
1
Soil
Type
2
2
3
1
2
2
2
2
3
2
2
2
3
2
2
1
2
2
2
2
2
2
2
3
1
Average
Monthly
Slope in
Precipitation Buffer Zone
High
High
Medium
Medium
High
High
Medium
Medium
High
High
High
High
High
High
High
High
High
High
High
High
High
High
Medium
Medium
High
High
High
High
Medium
Medium
Medium
Medium
High
Medium
High
High
High
High
Medium
Medium
Medium
Medium
Medium
Low
High
High
High
High
Medium
Low
This table is used to
provide estimates of
nonpoint source
loadings.
GIS-Based Interaction
User selects a
sub watershed
by clicking on
the map. A
recursive
algorithm
selects
connected
streams and
builds a
topological
network.
Data Extraction and Manipulation
Extracted data
include:
stream network,
point sources,
monitoring stations, land use,
soil type,
contributing area,
average slope,
mean annual
precipitation and
elevations.
Nonpoint Source Modeling
These data will
be used to
generate
nonpoint source
loadings for the
reaches based on
the conditional
probability table
developed from
the ecoregionwide causal
relationship
analysis.
Bayesian Network Generation
A Bayesian belief
network or causal
network is
generated from
the GIS layers.
The network
includes control
points, point and
nonpoint sources,
management
options,
monitoring
stations and
cost/benefits.