United States
Environmental Protection
Agency
Office of Environmental
Information
Washington, DC 20460
EPA/240/R-02/007
December 2002
Guidance for Quality
Assurance Project Plans
for Modeling
EPA QA/G-5M
-------
FOREWORD
The U.S. Environmental Protection (EPA) Agency has developed the Quality Assurance
(QA) Project Plan as a tool for project managers and planners to document the type and quality of
data and information needed for making environmental decisions. This document, Guidance for
Quality Assurance Project Plans for Modeling (EPA QA/G-5M), contains advice and
recommendations on how to develop a QA Project Plan for projects involving the model
development or application using data acquired from other sources.
EPA works every day to produce quality information products. The information used in these
products are based on Agency processes to produce quality data, such as the quality system described
in this document. Therefore, implementation of the activities described in this document is consistent
with EPA's Information Quality Guidelines and promotes the dissemination of quality technical,
scientific, and policy information and decisions.
This document was designed for internal use and provides guidance to EPA program
managers and planning teams. It does not impose legally binding requirements and may not apply
to a particular situation based on the circumstances. EPA retains the discretion to adopt
approaches on a case-by-case basis that differ from this guidance where appropriate. EPA may
periodically revise this guidance without public notice.
This document is one of the U.S. EPA Quality System Series documents. These
documents describe the EPA policies and procedures for planning, implementing, and assessing
the effectiveness of the Quality System. As described in EPA Manual 5360 Al (EPA, 2000a),
this document is valid for a period of up to five years from the official date of publication. After
five years, this document will be reissued without change, revised, or withdrawn from the U.S.
EPA Quality System Series documents. Questions regarding this document or other Quality
System Series documents should be directed to the Quality Staff at:
U.S. EPA
Quality Staff (2811R)
1200 Pennsylvania Ave., NW
Washington, DC 20460
Phone: (202) 564-6830
Fax: (202) 565-2441
E-mail: quality@epa.gov
Copies of EPA Quality System Series documents may be obtained from the Quality Staff
directly or by downloading them from its Home Page:
www.epa.gov/quality
Final
EPAQA/G-5M i December 2002
-------
ACKNOWLEDGMENTS
This guidance document benefitted greatly from comments from many reviewers with
experience in environmental modeling and in quality assurance in both early internal reviews and
review by the Council on Regulatory Environmental Modeling. The Models 2000 Quality
Assurance and Peer Review Action Team at EPA was especially instrumental in developing the
first working draft of this guidance.
Final
EPAQA/G-5M ii December 2002
-------
TABLE OF CONTENTS
Page
CHAPTER 1. INTRODUCTION 1
1.1 WHAT ARE EPA'S QUALITY-RELATED POLICIES REGARDING
ENVIRONMENTAL MODELING? 1
1.2 WHAT INFORMATION DOES THIS GUIDANCE PROVIDE? 1
1.3 WHY IS PLANNING FOR MODEL DEVELOPMENT AND APPLICATION
IMPORTANT? 2
1.4 WHAT QUESTIONS DOES THIS GUIDANCE ADDRESS? 3
1.5 WHO CAN BENEFIT FROM THIS DOCUMENT? 4
1.6 WHAT IS THE GRADED APPROACH TO QA PROJECT PLANS,
AND HOW DO I DETERMINE WHAT QUALITY ASSURANCE
AND QUALITY CONTROL ACTIVITIES ARE NEEDED
FOR A SPECIFIC PROJECT? 4
1.7 HOW DOES THIS GUIDANCE FIT IN THE EPA QUALITY SYSTEM? 6
1.8 HOW IS THIS DOCUMENT ORGANIZED? 7
CHAPTER 2. QA PROJECT PLAN PROCEDURES 9
2.1 GENERAL EPA POLICY ON QA PROJECT PLANS 9
2.2 QA PROJECT PLAN RESPONSIBILITIES 9
2.3 QA PROJECT PLAN ELEMENTS FOR ENVIRONMENTAL MODELING ... 10
2.4 REVISIONS AND RELATED GUIDANCE 11
2.4.1 Revisions and Review 11
2.4.2 Related Guidance 12
CHAPTER 3. THE MODEL DEVELOPMENT AND APPLICATION PROCESS:
RELATIONSHIP TO QA PROJECT PLAN ELEMENTS 13
3.1 MODELING NEEDS AND REQUIREMENTS ANALYSIS 15
3.1.1 Needs Assessment 15
3.1.2 Define Purpose, Objectives, and Output Specifications 15
3.1.3 Define Quality Obj ectives, Desired Performance Criteria, and
Documentation Needs for Model Output 16
3.1.4 Is a New Model Needed? 17
3.2 MODEL DEVELOPMENT 18
3.2.1 Model Design 19
3.2.2 Model Coding 20
3.2.3 Model Testing 21
3.2.4 First Application 23
3.3 MODEL APPLICATION 24
3.3.1 Select the Most Appropriate Model 24
3.3.2 Consider Site-Specific Information 25
3.3.3 Are Performance Criteria Met? 25
Final
EPA QA/G-5M iii December 2002
-------
Page
3.3.4 Generate and Assess Model Outputs 26
3.3.5 Summarize Results and Document 26
CHAPTER 4. ELEMENTS OF QUALITY ASSURANCE PROJECT PLANS
FOR MODELING PROJECTS 27
4.1 GROUP A: PROJECT MANAGEMENT 27
4.1.1 Title and Approval Sheet (Al) 27
4.1.2 Table of Contents and Document Control Format (A2) 28
4.1.3 Distribution List (A3) 29
4.1.4 Project/Task Organization (A4) 29
4.1.5 Problem Definition/Background (A5) 30
4.1.6 Project/Task Description and Schedule (A6) 32
4.1.7 Quality Objectives and Criteria for Model Inputs/Outputs (A7) 33
4.1.8 Special Training Requirements/Certification (A8) 35
4.1.9 Documentation and Records (A9) 36
4.2 GROUP B: MEASUREMENT AND DATA ACQUISITION 39
4.2.1 Calibration (B7) 40
4.2.2 Non-direct Measurements (Data Acquisition Requirements) (B9) 45
4.2.3 Data Management and Hardware/Software Configuration (BIO) 49
4.2.3.1 Data Management (Element BlOa) 50
4.2.3.2 Hardware/Software Configuration (Element BlOb) 53
4.3 GROUP C: ASSESSMENT AND OVERSIGHT 59
4.3.1 Assessment and Response Actions (Cl) 61
4.3.2 Reports to Management (C2) 69
4.4 GROUP D: DATA VALIDATION AND USABILITY 70
4.4.1 Departures from Validation Criteria (Dl) 71
4.4.2 Validation Methods (D2) 73
4.4.3 Reconciliation with User Requirements (D3) 74
CHAPTER 5. EXAMPLES 77
5.1 COMPLEX MODELS - LAKE MICHIGAN MASS BALANCE PROJECT .... 77
5.1.1 Part A: Project Management 77
5.1.2 PartB: Data Acquisition and Management 80
5.1.3 Part C: Assessments and Response Actions 81
5.1.4 PartD: Output Assessment and Model Usability 81
5.2 INTERMEDIATE MODELS - PORT RIVER NITROGEN AND PHOSPHATE
TOTAL MAXIMUM DAILY LOAD (TMDL) PROJECT 82
5.2.1 Part A: Project Management 82
5.2.2 Part B: Data Acquisition and Management 85
Final
EPA QA/G-5M iv December 2002
-------
Page
5.2.3 Part C: Assessments and Response Actions 88
5.2.4 Part D: Output Assessment and Model Usability 89
5.3 Single Module - Land Application Module 90
5.3.1 Background 90
5.3.2 Part A: Project Management 90
5.3.3 Part B: Data Generation and Acquisition 93
5.3.4 Part C: Assessment and Response Actions 94
5.3.5 Part D: Output Assessment and Model Usability 98
APPENDIX A: BIBLIOGRAPHY A-l
APPENDIX B: TERMS AND DEFINITIONS B-l
APPENDIX C: CHECKLISTS FOR QA PROJECT PLAN ELEMENTS C-l
Final
EPA QA/G-5M v December 2002
-------
LIST OF FIGURES
Page
Figure 1. EPA Quality System Approach to Addressing Data Generated from
Model Application 6
Figure 2. Typical Life-Cycle of a Three-step Modeling Project 14
Figure 3. Map of QA Project Plan Elements to Modeling Tasks 16
Figure 4. Map of the QA Elements to Modeling Tasks 20
Figure 5. Propagation of Uncertainty and Variability 22
Figure 6. Map of QA Project Plan Elements to Modeling Tasks 24
Figure 7. Example of a Document Control Format 28
Figure 8. Example of an Organizational Chart for a Modeling Project 30
Figure 9. The Systematic Planning Process 34
Figure 10. Example of a Process Flow Diagram for a Modeling Project 51
Figure 11. Modeling Framework for Lake Michigan Mass Balance Project 78
Figure 12. Module Components of 3MRA 91
Figure 13. Data Flow Diagram for the LAU Module 94
LIST OF TABLES
Table 1. Questions this Guidance Addresses 3
Table 2. Examples of Modeling Projects with Differing Intended Use 5
Table3. Examples of Important Reporting and Documentation on a Modeling Project 37
Table 4. Typical Activities and Documentation Prepared Within the System
Development Life Cycle of a Modeling Project that Should be
Considered When Establishing the QA Program for the
Hardware/Software Configuration 54
Table 5. Description of Modeling Project Tasks 92
Final
EPAQA/G-5M vi December 2002
-------
CHAPTER 1
INTRODUCTION
1.1 WHAT ARE EPA'S QUALITY RELATED POLICIES REGARDING
ENVIRONMENTAL MODELING?
The EPA Quality System defined in EPA Order 5360.1 A2 (EPA, 2000d), Policy and
Program Requirements for the Mandatory Agency-wide Quality System, includes coverage of
environmental data produced from models. Environmental data includes any measurement or
information that describe environmental processes, location, or conditions; ecological or health effects
and consequences; or the performance of environmental technology. For EPA, environmental data
includes information collected directly from measurements, produced from models, and compiled from
other sources such as databases or literature. The EPA Quality System is based on an American
National Standard, ANSI/ASQC E4-1994.
Consistent with the National Standard, E4-1994, Section §6.a.(7) of EPA Order 5360.1 A2
states that EPA organizations will develop a Quality System that includes "approved Quality Assurance
(QA) Project Plans, or equivalent documents defined by the Quality Management Plan, for all
applicable projects and tasks involving environmental data with review and approval having been made
by the EPA QA Manager (or authorized representative defined in the Quality Management Plan).
More information on EPA's policies for QA Project Plans is provided in Chapter 5 of the EPA
Manual 5360 A1 (EPA, 2000a), EPA Quality Manual for Environmental Programs and
Requirements for Quality Assurance Project Plans (QA/R-5) (EPA, 2001). This guidance helps to
implement the policies for models defined in Order 5360.1 A2.
1.2 WHAT INFORMATION DOES THIS GUIDANCE PROVIDE?
This guidance provides information about how to document quality assurance planning for
modeling (e.g., model development, model application, as well as large projects with a modeling
component). A "model," for the purpose of this guidance, is something that creates a prediction. The
elements of QA Project Plans for data collection are described in EPA Requirements for Quality
Assurance Project Plans (QA/R-5) (EPA, 2001) and EPA Manual 5360 A1 (EPA, 2000a).
Further details on developing these QA Project Plans are described in EPA Guidance for Quality
Assurance Project Plans (QA/G-5) (EPA, 1998a).
This guidance is based on the recommendations and policies from the above EPA documents,
but is written especially for modeling projects, which have different quality assurance concerns than
traditional environmental monitoring data collection projects. If you are familiar with the QA Project
Plan structure described in these documents, you will find that the structure for the QA Project Plan for
Final
EPAQA/G-5M 1 December 2002
-------
modeling is consistent with those in EPA Requirements for Quality Assurance Project Plans
(QA/R-5) (EPA, 2001) and EPA Guidance for Quality Assurance Project Plans (QA/G-5) (EPA,
1998a), though for modeling not all elements are included because not all are relevant.
1.3 WHY IS PLANNING FOR MODEL DEVELOPMENT AND APPLICATION
IMPORTANT?
Planning for modeling projects is just as important as planning traditional environmental
measurements for data collection projects. In order to be able to use model output for anything from
regulatory purposes to research, you should be sure that the model is scientifically sound, robust, and
defensible. The way to ensure this is by following a thorough planning process that incorporates the
following elements:
a systematic planning process including identification of assessments and related
performance criteria;
peer reviewed theory and equations;
a carefully designed life-cycle development process that minimizes errors;
documentation of any changes from the original plan;
clear documentation of assumptions, theory, and parameterization that is detailed
enough so others can fully understand the model output;
input data and parameters that are accurate and appropriate for the problem;
output data that can be used to help inform decision making.
These features lead to confidence in results. The steps for documenting these processes will be
described in an EPA QA Project Plan for modeling efforts.
A QA Project Plan and good project management in modeling projects are closely tied
together. A good QA Project Plan is valuable to a modeling project in the following ways:
A QA Project Plan documents all the criteria and assumptions in one place for easy
review and referral by anyone interested in the model.
Final
EPAQA/G-5M 2 December 2002
-------
A QA Project Plan can be used to guide project personnel through the model
development or application process and helps ensure that choices are consistent with
the established objectives and project-specific requirements.
Using a consistent format makes it easy for others to review information on models and ensures that no
steps are overlooked in the planning phase.
1.4 WHAT QUESTIONS DOES THIS GUIDANCE ADDRESS?
For quick reference to the information in this document, Table 1 provides a summary of the
main questions addressed by this guidance and indicates the chapter and sections containing this
information.
Table 1. Questions this Guidance Addresses
Questions
What are the guidelines for preparing QA Project Plans for projects that
involve model development or model application?
Who is responsible for developing and reviewing these QA Project Plans?
How do I document changes from the planned process described in the QA
Project Plan?
How do the model development and model application processes relate to
the QA Project Plan contents?
How does peer review relate to quality assurance activities?
How are acceptance and performance criteria used in a modeling project?
Where are the records and documentation planned for this modeling effort
described?
Where are the model calibration plans described?
What information on non-direct measurements (i.e., secondary use of
existing data) should be included in a QA Project Plan for model
development or application?
Where are the model evaluation plans described?
Where are the assessment plans for acceptance of final model results
described?
Where are assessment activities documented in a QA Project Plan?
Chapter/Section
2.1 (overview)
Chapter 4 (details)
2.2
2.4
Chapter 3
(Figure 4)
3.2.2
3.1.3,3.2.2,4.1.7,
4.3.1,4.4.1
4.1.9
4.2.1
4.2.2
4.3
4.4
4.3,4.4
EPA QA/G-5M
Final
December 2002
-------
1.5 WHO CAN BENEFIT FROM THIS DOCUMENT?
This guidance has many potential users, including all those involved in planning and
implementing modeling projects or in reviewing modeling projects. Specifically,
Modelers can use this guidance to see how the modeling development and application
processes are linked to the various elements that are documented in the QA Project
Plan. This guidance will help them to prepare and review a QA Project Plan.
Managers of model development and application projects will find this guidance useful
for understanding how planning is related to developing good models, and how the
EPA policies defined in EPA Order 5360.1 A2 can be met.
QA reviewers and QA officers will find this guidance useful for understanding the steps
and details behind the planningespecially those who might not be familiar with the
model development process. This guidance will help these users clearly understand
how the quality system is applied to model development.
1.6 WHAT IS THE GRADED APPROACH TO QA PROJECT PLANS, AND HOW DO I
DETERMINE WHAT QUALITY ASSURANCE AND QUALITY CONTROL
ACTIVITIES ARE NEEDED FOR A SPECIFIC PROJECT?
EPA defines the graded approach as "the process of basing the level of application of
managerial controls applied to an item or work according to the intended use of the results and degree
of confidence needed in the quality of the results" (EPA, 1998a). This is an important element of the
Quality System because it allows the application of quality assurance and quality control activities to be
adapted to meet the rigor needed by the project at hand. Models that provide an initial "ballpark"
estimate or non-regulatory priorities, for example, would not require the same level of quality assurance
and planning as models that will be used to set regulatory requirements. There are no explicit
categorizations or other specific guidelines for applying the graded approach, but the information in this
section and in the examples (Chapter 5) should provide some insight into how to do so.
In applying the graded approach, two aspects are important for defining the level of QA effort
that a modeling project needs: intended use of the model and the project scope and magnitude.
The intended use of the model is a determining factor in the level of QA needed
because it is an indication of the seriousness of the potential consequences or impacts
that might occur due to quality problems. For example, Table 2 shows that higher
standards would be set for projects that involve potentially large consequences, such as
Congressional testimony, development of new laws and regulations, or the support of
Final
EPAQA/G-5M 4 December 2002
-------
litigation. More modest levels of defensibility and rigor would be acceptable for data
used for technology assessment or "proof of principle," where no litigation or regulatory
action are expected. Still lower levels of defensibility apply to basic exploratory
research requiring extremely fast turn-around, or high flexibility and adaptability. In
such cases, the work may have to be replicated under tighter controls or the results
carefully reviewed prior to publication. By analyzing the end-use needs, appropriate
QA criteria can be established to guide the program or project. These examples are
presented for illustration only, and the degree of rigor needed for any particular project
should be determined based on an evaluation of the project needs and resources.
Table 2. Examples of Modeling Projects with Differing Intended Uses
Purpose for Obtaining
Model-Generated
Information (Intended Use)
Regulatory compliance
Litigation
Congressional testimony
Typical Quality Assurance Issues
Legal defensibility of data sources
Compliance with laws and regulatory mandates
applicable to data gathering
/
Regulatory development
State Implementation Plan
(SIP) attainment
Verification of Model
Compliance with regulatory guidelines
Existing data obtained under suitable QA program
Audits and data reviews
/
Trends monitoring (non-
regulatory)
Technology development
"Proof of principle"
Use of accepted data-gathering methods
Use of widely accepted models
Audits and data reviews
/
Basic research
Bench-scale testing
QA planning and documentation at the facility level
Peer review of novel theories and methodology
Level
ofQA
|
f
EPA QA/G-5M
Final
December 2002
-------
Other aspects of the QA effort can be established by considering the scope and
magnitude of the project. The scope of the model development and application
determines the complexity of the project; more complex models need more QA effort.
The magnitude of the project defines the resources at risk if quality problems lead to re-
work and delays.
The scope and magnitude of the project should be considered when defining:
type and extent of programmer's and end user's documentation,
testing effort and documentation of test results,
rigor of quality objectives and criteria,
cost and schedule control.
Specific examples of how the considerations described above may be used to define a project's QA
effort are provided in Chapter 5.
1.7 HOW DOES THIS GUIDANCE FIT IN THE EPA QUALITY SYSTEM?
The EPA Quality
System structure was
designed for data collection
activities. However, because
modeling can also produce
data that will be used in
decision making, quality
issues are also relevant for
these data. The paradigm
needs to be modified, as
indicated in Figure 1, for
application to data produced
by models rather than data
produced by sampling and
analytical measurement of
environmental media.
Figure 1 portrays the project
portion of EPA' s Quality
System. Systematic planning, an important first step, calls for stated and clarified objectives. In this
step, objectives are clarified. Quality control will include peer review of theory and approach, code
evaluation, and/or procedures for model calibration. Quality assurance of input data and parameter
values are important to model quality. Because the input data will most likely be obtained from other
. IMPLEMENTATION
ESSME
Figure 1. EPA Quality System Approach to Addressing
Environmental Data Generated from Model Application
EPA QA/G-5M
Final
December 2002
-------
sources, data quality procedures for secondary data use should be followed. After the process of
model development or application is complete, evaluation of the model output is completed.
Documentation is completed or updated at the end of the model development or application process.
1.8 HOW IS THIS DOCUMENT ORGANIZED?
This document will best be used as a reference document rather than a "how-to" guide to be
read front-to-back. To that end, Chapter 2 contains relevant information about EPA policies regarding
QA Project Plans. Chapter 3 describes the typical planning processes a modeler would use to develop
or apply a model and provides linkages to the elements in a QA Project Plan for modeling. Chapter 4
contains a description of the QA Project Plan elements, interpreted for a modeling project. Finally,
Chapter 5 contains examples of the content of a QA Project Plan that illustrate the planning process for
model development and model application. Three types of projects are described to give an indication
of the graded approach and how different problems might be addressed with different levels of detail
and description. These are not full QA Project Plans, but are meant to give an idea of the information
that would be included in such a plan.
A checklist is provided in Appendix C that can be used when writing a QA Project Plan for
modeling. This checklist indicates the basic pieces of information that may be relevant for each element.
A glossary and a list of references are also included in the appendices.
Final
EPAQA/G-5M 7 December 2002
-------
Final
EPAQA/G-5M 8 December 2002
-------
CHAPTER 2
QA PROJECT PLAN PROCEDURES
This chapter provides a summary of guidelines for preparing QA Project Plans in projects that
involve model development, modification, or application.
2.1 GENERAL EPA POLICY ON QA PROJECT PLANS
The QA Project Plan is a key component of the EPA Quality System and is the "blueprint" by
which individual projects are implemented and assessed. EPA Order 5360.1 A2 (EPA, 2000d)
establishes a mandatory, Agency-wide Quality System that applies to EPA and all organizations
performing work for EPA. These organizations need to ensure that data collected and used for the
characterization of environmental processes and conditions are of the appropriate type and quality for
their intended use. The applicability of this policy to modeling is based on the "use" of environmental
data (as stated in the policy).
A QA Project Plan is necessary in modeling projects even when no monitoring or other
environmental data measurements are performed because modeling results frequently serve as a
surrogate for these data, or are used for their interpretation. EPA allows flexibility in the organization
and content of a QA Project Plan to meet the unique needs of each project or program. Note that
while most QA Project Plans will describe project- or task-specific activities, there may be occasions
when a generic QA Project Plan may be more appropriate. A generic QA Project Plan describes, in a
single document, the information that is not site-, process-, or time-specific but applies throughout the
program (e.g., of a specific model for a defined use). Application-specific information is then added to
the approved QA Project Plan as that information becomes known or completely defined. A generic
QA Project Plan is reviewed periodically to ensure that its content continues to be valid and applicable
to the program over time.
2.2 QA PROJECT PLAN RESPONSIBILITIES
The QA Project Plan for a program or project may be prepared by an EPA organization, a
contractor, an assistance agreement holder, or another Federal agency under an interagency agreement.
Except where specifically delegated, all QA Project Plans prepared by non-EPA organizations need to
be approved by EPA before implementation. However, EPA may grant conditional approval to a QA
Project Plan to permit some work to begin while non-critical deficiencies in the QA Project Plan are
being resolved. EPA Order 5360.1 A2 also states that all QA Project Plans shall be implemented as
approved for the intended work. All personnel should understand the project-specific requirements
prior to the start of data generation activities (including generation or interpretation of environmental
data using modeling techniques).
Final
EPAQA/G-5M 9 December 2002
-------
For model development projects, one or more staff members who are intimately familiar with
the scientific and technical details of the modeling (scientist, meteorologist, programmer, etc.) are
usually designated as lead modeler of the QA Project Plan. Input from the modeling staff is critical in
preparing sections of the QA Project Plan that list the critical model structure and parameters to be
tested, the criteria that the tests need to meet, and the methods to evaluate the test results. In larger
projects involving a staff of model developers, it is appropriate to include one or more individuals with
software development, testing, and QA expertise to contribute to the QA Project Plan in areas of
documentation, configuration control, test plan development, coding standards, etc. The project
manager is ultimately responsible for assessing whether the performance and acceptance criteria for the
intended modeling use were met and works iteratively with the intended users of the results.
In projects involving routine use of well-accepted environmental models, the project staff may
lack personnel who are familiar with the software's internal operations. In this situation, a senior
scientist who understands the quality objectives and performance criteria and who can assess the
model's performance in the context of the project is a good candidate for preparing or reviewing the
QA Project Plan and deciding if model documentation is sufficient (e.g., proprietary software). The
project manager is ultimately responsible for assessing the appropriateness of the chosen model(s) for
the particular application and assessing whether the performance and acceptance criteria were met and
works iteratively with the intended users of the results.
2.3 QA PROJECT PLAN ELEMENTS FOR ENVIRONMENTAL MODELING
This section lists the elements of a QA Project Plan defined in EPA Requirements for Quality
Assurance Project Plans (QA/R-5) (EPA, 2001). Subsequent chapters of this document will provide
detailed information about these elements with the emphasis on modeling projects. However, some
titles of the QA Project Plan elements listed below are slightly different in subsequent chapters to
emphasize the application to modeling.
Elements of a QA Project Plan
Defined in the EPA Requirements for Quality Assurance Project Plans (EPA, 2001)
A. Project Management
Al
A2
A3
A4
A5
Title and Approval Sheet
Table of Contents
Distribution List
Project/Task Organization
Problem Definition/Background
A6 Project/Task Description
A7 Quality Objectives and Criteria
A8 Special Training/Certification
A9 Documents and Records
EPA QA/G-5M
10
Final
December 2002
-------
B. Measurement/Data Acquisition
Bl
B2
B3
B4
B5
Cl
Sampling Process Design
Sampling Methods
Sample Handling and Custody
Analytical Methods
Quality Control
B6
B7
B8
B9
BIO
C. Assessment and
Assessment and Response Actions
C2
Instrument/Equipment Testing,
Inspection, and Maintenance
Instrument/Equipment Calibration and
Frequency
Inspection/ Acceptance Requirements for
Supplies and Consumables
Non-direct Measurements
Data Management
Oversight
Reports to Management
D. Data Validation and Usability
Dl
D2
Data Review, Verification, and
Validation
Verification and Validation Methods
D3
Reconciliation with User Requirements
2.4 REVISIONS AND RELATED GUIDANCE
2.4.1 Revisions and Review
Because of the complex and diverse nature of environmental data operations, changes to QA
Project Plans, methods, and objectives are sometimes needed. When a substantive change is
warranted, the QA Project Plan needs to be modified to reflect the change and those changes needs to
be submitted for approval. According to EPA policy, a revised QA Project Plan needs to be reviewed
and approved by the same authorities that performed the original review.
Changed procedures may be implemented only after the revision has been approved. Changes
to the technical procedures should be evaluated by the EPA QA Manager and Project Officer to
determine if they significantly affect the technical and quality objectives of the modeling project. If the
procedural changes are determined to have significant quality impacts, the QA Project Plan should be
revised and re-approved, and a revised copy should be sent to all persons on the distribution list. Only
after the revision has been received and approved (at least verbally with written follow-up) by project
personnel, can the change be implemented.
EPA QA/G-5M
11
Final
December 2002
-------
For programs or projects of long duration, QA Project Plans should be reviewed at least
annually by the EPA Project Manager. Guidance for Quality Assurance Project Plans (QA/G-5)
(EPA, 1998a) and EPA Requirements for Quality Assurance Project Plans (QA/R-5) (EPA, 2001)
contain additional information applicable to QA Project Plan revisions.
2.4.2 Related Guidance
EPA QA policy and guidelines are based on the National Consensus Standard, ANSI/ASQC
E4-1994, Specifications and Guidelines for Environmental Data Collection and Environmental
Technology Programs (ANSI/ASQC, 1995). This standard describes the quality management
elements that are generally common to environmental problems, regardless of their technical scope. It
also discusses the specifications and guidelines that apply to project-specific environmental activities
involving the generation, collection, analysis, evaluation, and reporting of environmental data.
EPA Requirements for Quality Assurance Project Plans (EPA, 2001) and Guidance for
Quality Assurance Project Plans (EPA, 1998a) have been developed consistent with ANSI/ASQC
E4-1994. These documents are the primary references for preparing, approving, and revising QA
Project Plans for EPA projects and programs. Other guidance documents are available that provide
background information and techniques that may be helpful for preparing QA Project Plans for
modeling projects, including: Quality Assurance of Multi-media Model for Predictive Screening
Tasks (EPA, 1999c) and Systems Development Life Cycle Methodology (SDLCM) Statement of
Policy, Standards and Guidelines (EPA, 2000f).
Final
EPAQA/G-5M 12 December 2002
-------
CHAPTER 3
THE MODEL DEVELOPMENT AND APPLICATION PROCESS:
RELATIONSHIP TO QA PROJECT PLAN ELEMENTS
What is the purpose of this chapter? This chapter shows the modeler which elements of a QA
Project Plan are linked to specific quality aspects or features of a typical model development or
application project. In this chapter, the general process for conducting a model development or
application project is presented. The process is divided into three basic steps, which are further
subdivided into tasks. These tasks are then mapped to one or more elements of the QA Project Plan
according to the types of procedures conducted and the data quality issues that need to be addressed
within each task. Thus, this chapter introduces the modeler to the relevant elements of a QA Project
Plan for a modeling project, within the context of a general modeling process with which the modeler is
familiar.
What is the general process for conducting a model development or application project? While
each modeling project is unique, most projects that involve model development and application typically
follow a process similar to the one illustrated in Figure 2. The three steps in this process are the
following:
1. Modeling Needs and Requirements Analysis: Use a systematic planning process to
determine modeling needs and project-specific requirements and to decide whether or
not a model currently exists that can be used to achieve these needs and requirements.
2. Model Development: Develop a new model in situations where no existing model can
be used to address the particular study objective.
3. Model Application: Apply an existing model (identified from a model selection
procedure when more than one model is relevant) or a newly developed model.
For projects in which modeling is only a part of the total effort, the modeling process may be
preceded by a systematic planning process as indicated in Figure 2, in which quality objectives for the
entire project are established. The process in Figure 2 is iterative in nature, as reviews and checks
occur within the process to ensure that the model outputs will address all necessary project objectives
and meet necessary performance criteria. The iterative aspect also provides flexibility to accommodate
the exploratory and evolutionary nature of many model development projects.
The individual tasks performed within each step in Figure 2 are linked to one or more elements
of the QA Project Plan. The QA Project Plan elements associated with a given task specify the type
and quality of information that is needed to execute that task. This linkage (or "cross-walk") between
Final
EPAQA/G-5M 13 December 2002
-------
Systematic Planning
[see Guidance for the Data Quality Objectives Process (EPA QA/G-4)]
Step 1: Modeling Needs
and Requirements Analysis
Step 2: Modeling Design
Model Testing
Model Design
Model Coding
Needs Assessment
Define purpose,
objectives, and output
specifications
meets performance
Define Quality Objectives,
desired performance
criteria, and documentation
needs for model output
Is a
new model
needed?
First Application
Step 3: Model Application
Select the Most
Appropriate Model
Run the Computer
Code
Do Data/Models/
Parameters for this
Application Meet
Desired
Performance
Criteria?
Model Output Testing
and Peer Review
Data Development,
Model Parametrization,
and Model Calibration
for the Application Site
Summarize Results and
Document
Figure 2. Typical Life-Cycle of a Three Step Modeling Project
EPA QA/G-5M
14
Final
December 2002
-------
the modeling tasks and the elements of the QA Project Plan is described in Sections 3.1 through 3.3 of
this chapter.
3.1 MODELING NEEDS AND REQUIREMENTS ANALYSIS
The first step of the modeling process consists of the following three tasks:
assess the need(s) of the modeling project;
define the purpose and objectives of the model and the model output specifications;
define the quality objectives to be associated with the model outputs as described in
Section 4.1.7, and the examples in Chapter 5.
The modeling planning team uses the outcomes of these three tasks to determine whether an existing
model can be used or a new model needs to be developed. Figure 3 indicates how information
generated in these tasks is documented in the specific QA Project Plan elements.
3.1.1 Needs Assessment
The regulatory or scientific need for using a model (versus, for example, using existing data or
collecting new measurements) and the necessary features of the model should be specified within
Element A5 (Problem Definition/Background) of the QA Project Plan. The individuals or organizations
that would address these needs (e.g., members of the planning team, decision makers, statisticians,
principal data providers and users, computer hardware and software development personnel, statistical
expertise, QA Manager), along with their specific roles and responsibilities and their relationships and
organizational structure within the project, should be documented in Element A4 (Project/Task
Organization).
Expertise or certification requirements within the study team should be documented in Element
A8 (Special Training Requirements/Certification) of the QA Project Plan. Examples are specialized
computer hardware/software programmers and analysts, certified quality assurance managers, project
managers, and statisticians experienced in performing goodness-of-fit tests, characterizing uncertainty,
and conducting sensitivity analyses.
3.1.2 Define Purpose, Objectives, and Output Specifications
Element A6 (Project/Task Description and Schedule) of the QA Project Plan describes the
approach that will be used to meet the needs identified in the needs assessment. This element also is
Final
EPAQA/G-5M 15 December 2002
-------
used to provide an overall understanding of the modeling project, the work to be performed, the model
outputs to be obtained, and the project implementation schedule.
Step 1: Modeling Needs
and Requirements Analysis
Needs Assessment
Define purpose, objectives,
and output specifications
Define Quality Objectives,
desired performance criteria,
and documentation needs for
model output
Relevant QA Project Plan Elements
A4: Project/Task Organization
A5: Problem Definition/Background
A8: Special Traning Requirements/
Certification
A6: Project/Task Description and
Schedule
A7:Quality Objectives and Criteria for
Model Inputs/Outputs
A9: Documentation and Records
3
}
Figure 3. Map of QA Project Plan Elements to Modeling Tasks. Information for the QA
Project Plan Elements is generated in the indicated modeling task in the Modeling Needs and
Requirements phase of the modeling process.
3.1.3 Define Quality Objectives, Desired Performance Criteria, and Documentation Needs
for Model Output
On a model development or application project, quality objectives and the qualitative and
quantitative model performance criteria to be associated with the model output are determined by
implementing a systematic planning process. This process addresses questions such as the following:
EPA QA/G-5M
16
Final
December 2002
-------
How accurately and precisely does the model need to predict a given quantity at the
application site or in the process step of interest in order to satisfy regulatory or
scientific objectives?
What are appropriate criteria for making a determination of whether the model estimate
is accurate and precise enough (e.g., based on past general expertise combined with
site or process specific knowledge)?
How would these criteria be used to determine whether model outputs achieve the
needed quality? For example, what goodness-of-fit test [e.g., chi-square test on
predicted and observed frequency distributions (Snedecor & Cochran, 1989); t-test on
predicted and observed Robust Highest Concentrations (Cox and Tikvart, 1990); t-test
of differences in normalized means squared errors of regime averages from two models
(ASTM D6589)] would be performed (e.g., to compare predicted and observed
frequency distributions) and what significance level and power does this test need to
achieve and what other program specific criteria will be considered (e.g., Guideline on
Air Quality Models 40 CFR Part 51 Appendix W)?
The quality objectives and performance criteria that are established from implementing the
systematic planning process are to be documented in QA Project Plan Element A7 (Quality Objectives
and Criteria for Model Inputs/Outputs). The possible negative consequences of making inappropriate
decisions due to poor model prediction ability are a key concern.
Element A9 (Documentation and Records) of the QA Project Plan will specify those project
records to be included in reports and other documents, the model testing or data assessment reporting
formats, document control procedures, documentation of response actions during the project to correct
model development or implementation problems, and procedures for the final disposition of records and
documents, etc.
3.1.4 Is a New Model Needed?
In general, a new model is developed for a given application when an existing model cannot
meet the needs and requirements for the new application without major adjustments. The process of
identifying and assessing existing models for a new application begins with the analysis of modeling
needs and requirements and makes use of the information documented in Elements A5 (Problem
Definition/Background), A6 (Project/Task Description and Schedule), and A7 (Quality Objectives and
Criteria for Model Inputs/Outputs) of the QA Project Plan, as discussed in the previous sections, along
with information from Element B9 (Non-direct Measurements). Given this information, the types of
questions that are generally asked when assessing the need for a new model include:
Final
EPAQA/G-5M 17 December 2002
-------
For what specific tasks will the model be used in the given application?
What data will be collected or obtained to characterize the application site and to
develop a site conceptual model that will be compared with existing models? What is
the needed spatial and temporal scale of the model inputs?
What model outputs are needed? What is the spatial and temporal scale needed for the
outputs?
What levels of uncertainty are acceptable in model outputs?
What are the strengths and weaknesses of existing models?
If an existing model is available, are its parameter default values, input data, boundary
conditions, and underlying assumptions acceptable?
Is the existing model software compatible with the modeler's hardware/software
configuration requirements for the new application?
Are any improvements in the existing model's computer code operating characteristics
(e.g., run time) needed?
Do the quality and documentation associated with the existing model meet the project-
specific requirements?
In addition, if multiple options for a model are being evaluated, model selection may involve
evaluating the needed complexity of the model relative to the given project. Using a complex model
over a simpler model may have certain drawbacks if the project does not need such complexity, while
using a simpler model to address a more complex problem may result in an inability to meet certain
performance criteria.
3.2 MODEL DEVELOPMENT
If the modeling needs and requirements specify that a new model is needed, then the model
development process (Step 2) in Figure 2 is executed. There are four phases to Model Development:
Model Design, Modeling Coding, Model Testing, and First Application, which are shown in Figure 4
and described in the sections below.
Peer review can be incorporated in many different parts of the model development, as
appropriate for the level of complexity of the model and the degree of confidence needed in the model
Final
EPAQA/G-5M 18 December 2002
-------
output. There are different types of peer review, from a simple review by a few easily accessible
internal experts, to an elaborate peer review panel of independent experts who provide detailed written
evaluations. The places in the process in which peer review can typically take place are described in
subsequent sections, and is indicated by an underlining of the phrase "peer review" so that these steps
can be easily found in the descriptions of the model development processes.
3.2.1 Model Design
As shown in Figure 4, the model design process consists of three phases: theoretical
development, mathematical formulation, and identification of input data. As the first step in theoretical
development, in order to develop a model which accurately reflects the process that takes place, a
conceptual model of this process may be developed (e.g., identifying what is known or hypothesized
to be the source, environmental transport, and fate of a contaminant) or assemble information on a
lexicological pathway. This conceptual model or information should be described in Elements A5
(Problem Definition/Background) and A6 (Project/Task Description and Schedule) of the QA Project
Plan. The conceptual model is then used as a starting place for the construction of explanations behind
the process to be modeled. The theoretical formulation of the model would include any formulation
known about the processes being modeled that will link it to the desired output.
Next, the model theory is transformed into mathematical algorithms and equations, including
sub-models (modules) and model parameters, so that appropriate model outputs (predictions) can be
computed. This mathematical formulation of the theory involves identifying algorithms and
equations that have been developed to support this model theory or, if such equations are not already
available, developing these equations. The sources of these algorithms, including assumptions and
limitations, should be documented according to project-specific requirements contained in Element A9
(Documentation and Records) of the QA Project Plan.
Once the algorithms are developed, the identification of needed input data and parameter
values is undertaken. The data sets that are identified as possible sources should be evaluated for
quality, including relevance and appropriateness for the scenario this model is being developed to
address. If non-direct measurements (i.e, existing data from computer databases, programs, literature
files, or other historical sources) may be used as input to the model or as values for model parameters,
the methods to be used to evaluate these data for quality should be described in Element B9 (Non-
direct Measurements) of the QA Project Plan.
The scientific concepts supporting model design can be assessed in a science peer review
during this phase. A science peer review would evaluate the soundness of the theoretical approach, the
relevance of the theory to the problem at hand, and the appropriateness of the translation of the theory
into mathematical formulation. The necessary documentation from this peer review process should be
described in Element A9 (Documentation and Records), and might also be subject to project-specific
Final
EPAQA/G-5M 19 December 2002
-------
1 A5: Problem ^>
1 Definition/ 1
\_ Background _//\
/ A6: Project/Task \ ^
1 Description and 1
I Schedule ]~
/
[ A9: Documentation |
1 and Records 1
1 B9: Non-Direct \
1 Measurements (data I
V acquisition equirements) /'
_,f
(A7: Quality Objectives ^
and Criteria for Model
Inputs/Output ^/
(B10: Data Management \
and Hardware/Software )
Configuration J
Model Desiqn
X
Theoretical
Development
Mathematical
Formulation
i
\
Identification of
Needed Input
Data and
Parameter
Values
)
f
Science Peer
Review
\
\
\
\
\
*
!
1
}
1
Model Coding
Development of
. Data Management,
Hardware/Software
Configuration
Model
Programming
r
j
i
r
i
J r, r,
[ Peer Review
Model Testing
Eva uate
1 * Uncertainty
i
Determine if model
meets performance
criteria
1
1 ' 1
f
Calibrate the
model
V
!
Peer Review
i r~\
(A7: Qualty Object ves ^\
-i ^ * L M -i i A9: Documentaton \
and Criteria for Model and Records \
Inputs/Output
C1: Assessments and C2: Reports to /
Response Actions Management / \ i
( A7: Quality \
f Objectives J
~""\ and Criteria for Model /
,V Inputs/Output )
A
J A9: Documentation J
"A and Records j
s
[ B7: Model |
V Calibration J
^
''\f C1- Ass
\V Respons
sssments A
d J
e Actions /
[ C2: Reports to j
V. Management J
First Application
Figure 4. Map of the QA Elements to Step 2 Model Design Tasks. Information from the QA
Project Plan Elements is relevant for the indicated modeling tasks in the Model Development
phase of the modeling process.
requirements specified in Element Cl (Assessment and Response Actions) of the project's QA Project
Plan.
3.2.2 Model Coding
The process of converting the science formulations to model code is usually begun after any
science peer review has been completed and any necessary changes (corrective actions) have been
completed. The process begins by developing the data management process and the computer
hardware and software configuration requirements. The data management process consists of tracing
the path, mathematical and computational operations, and quality control of "as-acquired" model input
data and model predictions from their generation to their final use, storage, and documentation.
Computer hardware and software configuration involves devising the data handling equipment and
EPA QA/G-5M
20
Final
December 2002
-------
procedures, including required computer hardware/software and related code performance
requirements (e.g., run time) needed to compile, process, manipulate, evaluate, and assess the input
data and model predictions. It also involves establishing coding and configuration control requirements
to ensure consistency. Modules should be compartmentalized so that they can be tested independently,
and should be tested in sequence to verify consistency and compatibility between the modules. Section
3.2.3 provides more details on model testing. These requirements and criteria should be listed in QA
Project Plan Element BIO (Data Management and Hardware/Software Configuration). During this
phase, the results of the assessments of these requirements should be documented according to the
project-specific requirements specified Element A9 (Documentation and Records) of the project's QA
Project Plan.
Then, model coding, or the transformation of the model equations and related algorithms into
an integrated and efficient computer code, is undertaken. This process involves the work of computer
programmers working closely with the scientists to ensure that the theory is accurately represented in
the code. Decisions might have to be made in cases where the theory cannot exactly be replicated in
the code for technical reasons, and these compromises should be reached by the scientists and
programmers working together.
The next part of model coding involves calibration. Calibration is the process of refining the
model to achieve a desired degree of correspondence between the model output and actual
observations of the environmental system or process the model is intended to represent. The focus is
usually on the estimation and characterization of empirical parameters and parameterizations of the
modeled processes. Element B7 (Calibration) specifies the procedures to be implemented to ensure
that the mathematical representations and parameters used in models are reasonable to achieve the
accuracy requirements specified in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs) of the project's QA Project Plan.
The final task in code development is to test the computer code and have it reviewed by
external peer reviewers (if required) to ensure that the code is error free and achieves all requirements
specified in Elements A7 (Quality Objectives and Criteria for Model Inputs/Outputs) and BIO (Data
Management and Hardware/Software Configuration) of the QA Project Plan. The code assessment
process (i.e., when they are to occur and the possible corrective actions to be taken) is described in
Element Cl (Assessment and Response Actions). Such assessments may result in revisions to the
model code, which may need additional follow-up assessments.
3.2.3 Model Testing
Before formal model testing, uncertainty (i.e., lack of knowledge) should be characterized for
all stages of model design, namely: the conceptual model, the theoretical basis of the model, the
mathematical formulation of the theory, the parameters and parameter values used, and the input data
Final
EPAQA/G-5M 21 December 2002
-------
30
Figure 5. Propagation of Uncertainty and
Variability
and assumptions. The natural variability of
the relevant population(s) should also be
characterized (note natural variability often
limits certainty attainable, better data only
provides better definition of those limits). As
seen in Figure 5, these sources of variability
and uncertainty affect how the model outputs
are distributed and, as a result, the level of
uncertainty in the output values (see EPA
1996 and 1997 for more detail).
Requirements for the acceptable degree of
uncertainty should have been specified in
Element A7 (Quality Objective an Criteria
for Model Inputs/Outputs) in the QA Project
Plan during the planning stage. Element B9
(Non-direct Measurements) defines
acceptable uncertainty associated with the
existing data used in the modeling process.
Finally, the documentation and records of assessments of uncertainty and software performance
requirements are specified in Element A9 (Documentation and Records).
The model design team then evaluates the uncertainty in the modeling output by comparing
the output with available field or lab data. If inadequacies are found [that is, if the uncertainty and
variability exceed the limits specified in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs)], part of the model design process may be revisited and changes made; this assessment
and review phase of model development is frequently iterative in nature. Next, the model design team
assesses whether the final form of the model can achieve the remaining performance criteria
specified in Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs) of the QA Project
Plan. This assessment may be done using statistical model fitting and evaluation methods (e.g., Draper
& Smith, 1981) if appropriate data exist and can be obtained for comparison to model predictions.
Other model-specific statistical evaluations (e.g., American Society for Testing and Materials, 2000)
may also be undertaken. If appropriate data are not available, model outputs can be compared with
those of an accepted model (also known as "benchmarking") or other methods can be used (see Chen
& Beck, 2000 for some examples). The method and results of these assessments are documented
according to specifications given in Element A9 (Documentation and Records). If the model design
team cannot be assured that the performance criteria will be met by the given model, then either the
model's needs and performance criteria need to be reassessed or revisions made to the model such that
the criteria are met.
EPA QA/G-5M
22
Final
December 2002
-------
Once the model design team determines that the model will be able to achieve the overall
performance criteria listed Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs), a
peer review phase is recommended to assess how well the model design specifications were actually
implemented. Although a formal peer review process is preferred, this may not always be possible due
to project constraints. Scheduling of peer reviews, the peer review mechanism to be used, and
documentation of the responses to peer review comments throughout model development is
documented in Element Cl (Assessment and Response Actions). Reports to management resulting
from peer reviews are described in Element C2 (Reports to Management) with management of peer
review records in Element A9 (Documentation and Records).
The quality assurance assessment of the model is documented within the Group D QA Project
Plan elements. Element Dl (Departures from Validation Criteria) details the criteria used to decide if
the model outputs meet the requirements specified in Element A7 (Quality Objectives and Criteria for
Model Inputs/Outputs) of the project's QA Project Plan. Element D2 (Validation Methods) describes
the process for determining whether errors exist in the model output or if model outputs achieve
required project quality levels. Element D3 (Reconciliation with User Requirements) describes the
methods to be used to analyze the model outputs (as well as other site-related information and data) to
determine whether anomalies exist or if assumptions established in the planning phase are not being met.
3.2.4 First Application
The first application of the model is considered to be a part of model development, because the
model cannot be deemed complete until it has been shown that it can be successfully applied in a given
situation. Ideally, the developers will be able to apply the model to a particular scenario during model
testing using data from field and/or lab studies to compare the output. The assessment of quality for
the output is assessed in the same manner that subsequent applications will be. Section 3.3 will
describe the objectives and processes for these steps in more detail.
Elements Cl (Assessment and Response Actions) and C2 (Reports to Management) would
apply to assessments of input data and model parameters as shown in Figure 6 and described in
Sections 3.3.2 to 3.3.5. During the first application, calibrations may be employed to set parameter
values with site-specific data using procedures described in Element B7 (Calibration). The records of
calibration results are identified and their management defined in Element A9 (Documentation and
Records). Element Cl (Assessment and Response Actions) should describe when to assess the
performance of the model through comparisons with field or lab data, and also describes any corrective
actions to be taken. Reports documenting the assessment methods and the results obtained are defined
in Element A9 (Documentation and Records).
Final
EPAQA/G-5M 23 December 2002
-------
3.3 MODEL APPLICATION
Once models have been identified (or developed) for the project or routine application, the
model application process (Step 3) in Figure 2 is executed. Figure 6 shows how the various tasks in
the model application process are linked to specific QA Project Plan elements.
Step 3: Model Application
Select the Most Appropriate
Model
Data Development,
Model Parameterization, and Model
Calibration for the Application Site
Run the Computer Code
Model Output Testing and
Peer Review
1
Summarize Results and
Document
Relevant QA Project Plan Elements
A5: Problem Definition/Background
A6: Project/Task Description and Schedule
A7: Quality Objectives and Criteria for Input/Output Data
A9: Documentation and Records
'
A7: Quality Objectives and Criteria for Input/Output Data
A9: Documentation and Records
B7: Model Calibration
B9: Non-Direct Measurements (data acquisition
requirements)
"*^*^^
^\
___ '
C1 : Assessment and Response Actions
. __ f
C2: Reports to Management
A7: Quality Objectives and Criteria for Input/Output Data
A9: Documentation and Records
C1: Assessment and Response Actions
C2: Reports to Management
D1: Data Review, Evaluation, and Validation
D2: Validation and Evaluation Methods
D3: Reconciliation with User Requirements
A9: Documentation and Records
:r>
Figure 6. Map of QA Project Plan Elements to Modeling Tasks. Information for the QA
Project Plan Elements is relevant for the indicated modeling task in the Model Application phase
of the modeling process.
3.3.1 Select the Most Appropriate Model
When multiple models are relevant to a particular model application (see
www.epa.gov/epahome/models.htm), the most appropriate model is determined by considering the
EPA QA/G-5M
24
Final
December 2002
-------
particular problem that the project is addressing, any important background material that may impact
the type of model used, the project's scope and available resources, the underlying objectives, and any
results of statistical modeling procedures [e.g., as discussed in Draper and Smith (1981) and Berthouex
and Brown (1994)]. The criteria established for selecting an appropriate model are presented in
Elements A5 (Problem Definition/Background), A6 (Project/Task Description and Schedule), and A7
(Qaulity Objectives and Criteria for Model Inputs/Outputs) of the QA Project Plan. The results of the
model selection procedure, in addition to the information going into the selection process, are reported
according to the procedures specified in Element A9 (Documentation and Records).
3.3.2 Consider Site- or Process (Pathway)-Specific Information
The second task is to tailor the model to the specific application by identifying the most current
and appropriate data, parameter values, expert opinion, and assumptions that are consistent with model
requirements. If appropriate, the model is calibrated using data and methods that apply to the site,
process or pathway and application of interest following procedures and methods defined in Element
B7 (Calibration), with the results managed as specified in Element A9 (Documentation and Records).
Statistical methods for fitting calibration curves using appropriate data should be used whenever
possible [see e.g., Berthouex and Brown (1994, pp. 213-220)]. Additional activities include
implementing quality control (QC) for input data and specifying any special data management, computer
hardware/software configuration, or project documentation requirements. The performance criteria for
these activities are provided as appropriate in Elements A7 (Quality Objectives and Criteria for Model
Inputs/Outputs), B7 (Calibration), and B9 (Non-direct Measurements), and the documentation
requirements for these activities are provided as appropriate in Elements C2 (Reports to Management)
and A9 (Documentation and Records) of the QA Project Plan. Element Cl (Assessment and
Response Actions) describes when assessments will be conducted and the corrective actions, if any, to
be taken.
3.3.3 Are Performance Criteria Met?
After the first two tasks, the model inputs and parameter estimates for the given application are
evaluated based on whether they meet the performance and acceptance criteria specified in Element A7
(Quality Objectives and Criteria for Model Inputs/Outputs). If not, previous steps and tasks (e.g.,
overall modeling needs and requirements) need to be revisited, depending on the nature of the
discrepancy. Application of the model using data from one or more field or lab studies not employed
during model development allows assessments to support accepting, rejecting, or qualifying the model
output for its intended use.
Final
EPAQA/G-5M 25 December 2002
-------
3.3.4 Generate and Assess Model Outputs
If model inputs and parameter estimates for the given application meet performance and
acceptance criteria, the model's computer code is executed to obtain initial model outputs (predictions),
and the outputs are assessed against performance criteria specified in QA Project Plan Element A7
(Quality Objectives and Criteria for Model Inputs/Outputs). Initial assessments (i.e., site-specific code
has no errors or mistakes and comparisons with field or lab data) are conducted as indicated in QA
Project Plan Element Cl (Assessment and Response Actions). Assessment findings (e.g., need for
corrective action) are reported as indicated in Element C2 (Reports to Management) and documented
as indicated in Element A9 (Documentation and Records). The methods for final assessments [i.e.,
whether the final results are reasonable, address all project objectives, are technically useable, and meet
performance specifications as specified in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs) are documented in QA Project Plan Elements Dl (Departures from Validation
Criteria) and D2 (Validation Methods). Elements D3 (Reconciliation with User Requirements) and A9
(Documentation and Records) define the documentation needed from the assessments specified in
Elements Dl (Departures from Validation Criteria) and D2 (Validation Methods). Element D2
(Validation Methods) describes the methods to be used to analyze the model outputs (as well as other
site-related information and data) to determine whether anomalies exist or if assumptions established in
the planning phase are not met. Once it is determined that the model provides technically useable
outputs, the full suite of model runs is conducted. Each run would be subject to continued surveillance
of output quality using the methods and criteria specified in the QA Project Plan [Elements A9
(Documentation and Records), Cl (Assessment and Response Actions), C2 (Reports to
Management), Dl (Departures from Validation Criteria), D2 (Validation Methods), and D3
(Reconciliation and User Requirements)].
3.3.5 Summarize Results and Document
The last task in model application is to document the model outputs, the results of assessing
these outputs, and the need to place caveats on the use of those outputs. The project-specific
documentation and reporting requirements are described in Element D3 (Reconciliation and User
Requirements) and A9 (Documentation and Records) of the QA Project Plan.
Final
EPAQA/G-5M 26 December 2002
-------
CHAPTER 4
ELEMENTS OF QUALITY ASSURANCE PROJECT PLANS
FOR MODELING PROJECTS
This chapter provides details on the types of information that are included within each element
of a QA Project Plan for a modeling project. This chapter differs from Chapter 3 in that Chapter 3 is
focused around the general modeling process and provided an overview of how the elements of the QA
Project Plan were linked to various tasks within this process. Chapter 3 follows the typical sequence of
these tasks within a modeling project. In contrast, this chapter describes each QA Project Plan
element in the order in which the elements appear in the QA Project Plan. As such, it is structured
similarly to Chapter 3 of EPA Requirements for Quality Assurance Project Plans (EPA QA/R-5)
(EPA, 2001) and Chapter 5 of EPA Order 5360 Al (EPA 2000a).
As discussed in Section 1.6, EPA quality assurance guidance allows for a graded approach to
preparing QA Project Plans, in which the level of detail is based on the scope and magnitude of the
project and the intended use of the model. This chapter describes QA Project Plan elements for
projects where a new model is developed or an existing model is revised, and this model is being
applied for the first time to a given test application. Various types of modeling projects may need more
or less detail (e.g., subsequent applications) than what is suggested in this chapter based on the scope,
magnitude, and intended use of the model output.
Like the QA Project Plan, this chapter presents each element in a separate section, with the
sections organized by the four element groupings. Each section is preceded by a text box containing a
summary of the element's features as specified within EPA Requirements for Quality Assurance
Project Plans (QA/R-5) (EPA, 2001), with alterations occasionally made to this summary to make it
more specific to a modeling project.
4.1 GROUP A: PROJECT MANAGEMENT
All nine Group A elements presented in EPA Requirements for Quality Assurance Project
Plans (QA/R-5) (EPA, 2001) are relevant and important to address in the QA Project Plan for a
modeling project, and each is addressed in the following Sections 4.1.1 through 4.1.9.
4.1.1 Title and Approval Sheet (Al)
Include title of plan; name of the organization(s) implementing the project; effective date of
the plan; and names, titles, signatures, and approval dates of appropriate approving
officials.
Final
EPAQA/G-5M 27 December 2002
-------
The title and approval sheet documents the following information on the QA Project Plan:
title, document number (if appropriate), revision number (Revision 0 corresponds to the
first version), and effective date of the given revision;
name of each organization involved in implementing the various stages of the modeling
project (e.g., modeling organization, software contractor);
names, titles, and affiliation of the officials of each organization who will approve the
QA Project Plan (e.g., project managers, principal investigators, QA Managers, EPA
project manager/program officer, EPA QA Officer), along with space for their dated
signatures. [These officials are also represented in the project's organizational chart
within Element A4 (Project/Task Organization).]
Examples of title and approval sheets can be found within QA Project Plans that are posted on EPA's
Quality Staff web site (www.epa.gov/quality). The signatures of the key project officials on this page
document their official awareness and approval of the QA Project Plan, and therefore, the project's
goals and procedures. The specified dates of the signatures indicate when the plan was approved,
which constitutes the earliest start date for the model development and/or application efforts addressed
in the plan (i.e., Steps 2 and 3 of the modeling process as documented in Chapter 3.)
4.1.2 Table of Contents and Document Control Format (A2)
Provide a table of contents for the document, including sections, figures, tables, references,
and appendices. Apply a document control format on each page following the Title and
Approval Sheet when required by the EPA Project Manager and QA Manager.
Page of_
The table of contents allows easy access to
information contained in the plan. It includes the titles of all
sections, subsections, tables, and figures, along with the section JNo.
page numbers within the document where each of these can Revision JNo.
be found. When required by the EPA Program Manager
and QA Manager, a document control format such as that in
Figure 7 appears in the upper right-hand corner of each
page of the Q A Project Plan. It includes the Q A Project Figure 7. Example of a Document
Plan's section number, revision number and date (as stated Control Format
on the title and approval sheet), the page number, and the
total number of pages either in the given section or the entire document.
Final
EPAQA/G-5M 28 December 2002
-------
4.1.3 Distribution List (A3)
List the individuals and their organizations who need copies of the approved QA Project
Plan and any subsequent revisions, including all persons responsible for implementation
(e.g., project managers), the QA Managers, and representatives of all groups involved.
Paper copies need not be provided to individuals if equivalent electronic information
systems can be used.
The QA Project Plan's distribution list facilitates the control of the document by helping ensure
that the most current QA Project Plan is in use by all project participants.
4.1.4 Project/Task Organization (A4)
Provide a concise organizational chart showing the relationships and the lines of authority
and communication among all project participants. Include other model results users who
may be outside the group that are developing and applying the model, but for whom the
model outputs are intended. Identify any subcontractor relationships relevant to the
modeling project.
Identify the individuals or organizations participating in the project and discuss their
specific roles and responsibilities. Include the principal users of the model output, the
decision makers, the project QA Manager, and all persons responsible for implementation.
The project QA Manager must be independent of the unit generating the data or software.
Identify the individual responsible for maintaining the official, approved QA Project Plan.
A model development or application project often consists of several different tasks that can
involve different groups and organizations. It is important for the QA Project Plan to officially
document the responsible organizations and key players on the project, lines of authority and reporting,
and the relationships and official lines of communication between and among them.
The organizations involved and their responsibilities on the project should be identified,
followed by text descriptions of key individuals participating in the project and an organizational chart
such as the example in Figure 8. The key individuals on modeling projects may include the project
manager, project QA Manager, managers of the software/hardware configuration, the person
responsible for maintaining the QA Project Plan, principal users of the model output, decision makers
(such as the person responsible for model selection, if a selection is possible), and other persons
responsible for implementing the QA Project Plan and for approving deliverables. The specific roles,
activities, and responsibilities of these positions on the project should be highlighted in their descriptions.
Final
EPAQA/G-5M 29 December 2002
-------
EPA Work Assignment/
Project Manager
*N. Wentworth
202-564-6830
Office of Research and Development
EPA Quality Assurance
Officer
L. Doucet
202-564-1416
Office of Research and Development
Principal Investigator
J. Warren
202-564-6876
Research Institute
Environmental Department
Communication only
Modeling Coordinator and
Assessment
G.Johnson
919-541-7612
Research Institute
Environmental Modeling Department
Project Quality
Assurance Officer
T. Dixon
202-564-6877
Research Institute
QA Department
Model Inputs and
Parameter Estimates
F. Siegelman
202-564-6872
Research Institute
Data Systems Department
Model
Application
D. Sims
202-564-2872
Environmental
Modeling Department
Figure 8. Example of an Organizational Chart for a Modeling Project
The project organization chart shows the lines of authority and communication among key individuals on
the project. If direct contact between project managers and users of model output does not occur
(e.g., between modelers and decision makers), the organization chart should show the path that the
information takes from one group to the other.
4.1.5 Problem Definition/Background (A5)
State the specific problem to be solved, decision to be made, or outcome to be achieved.
Include sufficient background information to provide a historical, scientific, and regulatory
perspective for this particular project.
EPA QA/G-5M
30
Final
December 2002
-------
This element of the QA Project Plan states the problem to be addressed by the modeling
project with detail sufficient to allow a technically-trained reader to understand both the scientific need
for the project and how the project's objectives and activities will address the problem. The types of
decisions to be made are also documented in this QA Project Plan element. "Stating the Problem" and
"Identifying the Decision" are the first two steps of EPA's approach to systematic planning and the
Data Quality Objectives Process, as documented in Guidance for the Data Quality Objectives
Process (QA/G-4) (EPA, 2000b) and highlighted in Section 4.1.7 below. Questions that the QA
Project Plan may address when defining the problem include the following:
What is the specific problem? What are the goals and objectives of this project that
will address this problem?
What population in the environment does the problem specifically target, and what
measure(s) within this population does the problem address?
Why should a modeling approach be used to address the problem? Is there a
regulatory requirement for a modeling analysis?
What specifically will this project produce to address this problem (e.g., a new
predictive tool, modeling results for a new scenario)?
What types of decisions regarding the problem may be made as a result of this project?
Who will be responsible for making these decisions?
Will any aspect of the problem not be addressed in this project?
What other types of problems may this proj ect address?
It is important to place the problem in historical perspective to give readers and users a sense of
the project's purpose and position relative to other project and program phases and initiatives. Such
information also indicates the importance of generating new information and suggests tools that may be
available to do this. Therefore, sufficient background information should be provided in the QA Project
Plan to answer the following types of questions, when applicable:
Why is this project important, and how may it support proposed or existing research,
programs, initiatives, regulations, or other legal directives?
How may this project "fit in" with other on-going, broader efforts?
Final
EPAQA/G-5M 31 December 2002
-------
What types of conflicts or uncertainties currently exist that will be resolved by this
project?
What information, previous work, or previous data may currently exist that this project
can use?
Given that the problem is best solved by a modeling approach, what models currently
exist (if any) that can be used to achieve this project's goals and objectives? If multiple
models exist, how is one determined to be better than the others for this application?
The presentation of background information can include a discussion of initial ideas or
approaches for model development or application that were considered when selecting the approach
described under "Project/Task Description and Schedule" (Element A6).
4.1.6 Project/Task Description and Schedule (A6)
Provide a summary of all work to be performed, products to be produced, and the schedule
for implementation. The discussion need not be lengthy or overly detailed, but should give
an overall picture of how the project will resolve the problem or question described in
Element A5 (Problem Definition/Background).
An overview of the various tasks to be involved in the model development and/or application
effort is provided in this section of the QA Project Plan, along with the general technical approach and
the quality activities and procedures associated with these tasks. The discussion should present the
various tasks in relation to each other, and as a result, should allow the effort to resolve the problem
specified in Element A5 (Problem Definition/Background). This discussion need not be overly lengthy
or burdensome to prepare, as it is primarily meant to provide an overview. For detailed information,
references can be made to other available documents, other sections of the QA Project Plan, or to a
contact person who can distribute the information.
For modeling projects, examples of tasks that can be addressed in the overview (but not
necessarily detailed here) include the following:
how the conceptual model of the problem or site will be developed;
how the structural model and data processing software will be obtained;
how data may be acquired for model development, calibration, and testing;
Final
EPAQA/G-5M 32 December 2002
-------
the criteria used to decide whether probabilistic model output or point estimates are
needed; and
assessments relative to associated project-specific quality requirements.
This element of the QA Project Plan also lists the products, deliverables, and milestones to be
completed in the various stages of the project, along with a schedule of anticipated start and completion
dates for the milestones and deliverables, and the persons responsible for them.
4.1.7 Quality Objectives and Criteria for Model Inputs/Outputs (A7)
Describe the quality objectives for the project and the performance criteria to achieve
those objectives. EPA policy is to use a systematic planning process [see EPA Order
5360.1 A2 (EPA, 2000d)] to define quality objectives and performance criteria.
This element of the QA Project Plan introduces the quality criteria that the expected
components and outcomes of the modeling project need to achieve in order to meet the needs of the
user. These criteria are specified within performance or acceptance criteria that are developed in a
systematic planning process. Systematic planning identifies the expected outcome of the modeling
project, its technical goals, cost and schedule, and the criteria for determining whether the inputs and
outputs of the various intermediate stages of the project, as well as the project's final product, are
acceptable. This is usually an iterative process involving at least modelers and users. The goal is to
ensure that the project will produce the right type, quality, and quantity of data to meet the user's
needs. EPA Order 5360.1 A2 (EPA, 2000d) states that EPA environmental programs should use a
systematic planning process to develop acceptance or performance criteria when collecting, evaluating,
or using environmental data (e.g., information produced from models).
The systematic planning process can be applied to any type of data-generating project. The
seven basic steps of the systematic planning process are illustrated in Figure 9. The first three steps can
be considered preliminary aspects of scoping and defining the modeling effort, while the last four steps
relate closely to the establishment of performance criteria or acceptance criteria that will help ensure the
quality of the model outputs and conclusions. While both are measures of data quality, performance
criteria are used to judge the adequacy of information that is newly-collected or generated on the
project, while acceptance criteria are used to judge the adequacy of existing information that is drawn
from sources that are outside of the current project. Generally, performance criteria are used when
data quality is under the project's control, while acceptance criteria focus on whether data generated
outside of the project are acceptable for their intended use on the project (e.g., as input to a model)
[see Guidance on Systematic Planning for Environmental Data Collection Using Performance
and Acceptance Criteria (QA/G-4A) (EPA, 2002)].
Final
EPAQA/G-5M 33 December 2002
-------
Systematic planning is based on a graded
approach. This means that the extent of systematic
planning and the approach to be taken should match
the general importance of the project and the
intended use of the data. For example, when
modeling is to be used on a project that generates
data to be used either for decision making (i.e.,
hypothesis testing) or to determine compliance with a
standard, EPA recommends that the systematic
planning process take the form of the Data Quality
Objectives (DQO) Process that is explained in detail
within Guidance for the Data Quality Objectives
Process (QA/G-4) (EPA, 2000b).
The primary feature of this QA Project Plan
element is the set of performance or acceptance
criteria associated with data quality within the various
stages of the modeling process, as determined from
applying a systematic planning process. For
decision-making programs in which the systematic
planning process takes the form of the DQO
Process, these criteria are represented within DQOs
1 . State the Problem
t
2. Identify the Study Questions
*
3. Identify Information Needs
t
4. Specify the Characteristics
that Define the Population of
Interest
*
5. Develop the Strategy for
Information Synthesis
*
6. Specify Performance and
Acceptance Criteria
t |
7. Optimize the Design for Obtaining and
Generating Adequate Data or Information
Figure 9. The Systematic Planning Process
(EPA, 2000b) that express data quality relative to achieve tolerable levels of potential decision errors.
What types of issues do the performance or acceptance criteria that result from
systematic planning address in a modeling project? The performance or acceptance criteria
developed by the project planning team (see Section 3.1.1) address the following types of components
for modeling projects:
the particular type of task being addressed and the intended use of the output (e.g.,
predictions) of the modeling project to achieve this task;
the type of output needed to address the specific regulatory decision (if relevant),
including whether probabilistic or point estimates are needed;
the statistical criteria (e.g., limits on decision error) to be used in the model-building
process to identify those variables considered statistically important to the prediction
process and included as input to the model;
EPA QA/G-5M
34
Final
December 2002
-------
desired limits placed on the probability of making a certain type of decision error due to
the uncertainty associated with the model output (if a decision is to be made) and/or
criteria to demonstrate the model performs adequately (e.g., as well or better than a
previously accepted model for a given situation);
how the parameter, input, calibration, and test data necessary for this project are
acquired and evaluated for use in model development and/or in producing output;
requirements associated with the hardware/software configuration (e.g., run time or
processing capabilities) for those studies involving software evaluation.
While DQOs state the user's data needs relative to a given decision, corresponding criteria
need to be placed on the data to determine whether the data have satisfied these needs. For modeling
projects, such quality criteria can be placed on outcomes such as software performance (e.g., run time
or processing capabilities) and model prediction (e.g., acceptable level of uncertainty associated with
model prediction, relative to decision error). This element of the QA Project Plan links the DQOs with
appropriate data quality indicators (DQIs), which measure features of data quality such as precision
(i.e., variability in data under prescribed similar conditions), bias (i.e., systematic error), accuracy,
representativeness, completeness, and comparability. Although the level of rigor with which this is done
and documented within the QA Project Plan can vary widely depending on the particular type of
modeling project, this linkage represents an important advancement in implementing quality assurance.
Evaluation of model data may also include uncertainty and variability assessment which can be
introduced here, with details provided in the Group C and D elements. Requirements for a specified
level of uncertainty and/or variability can also be specified under this element.
Once the performance or acceptance criteria have been established, it is possible to specify
software performance and corresponding parameterization, input, and test requirements. These
specifications are made within the Group B elements of the QA Project Plan.
4.1.8 Special Training Requirements/Certification (A8)
Identify and describe any specialized training or certifications needed by personnel in order
to successfully complete the project or task. Discuss how such training will be provided
and how the necessary skills will be assured and documented.
This element identifies and documents any specialized training requirements and technical
expertise of the project team for the modeling project. Training requirements can include expertise in
certain scientific disciplines (e.g., statistics), in code development or testing in a specific computer
Final
EPAQA/G-5M 35 December 2002
-------
language, in data assessment, and in model development, evaluation or application. Usually the
organizations participating in the project are responsible for conducting training and obtaining any
required certification for their staff.
The issues outlined in this element (when relevant) include:
the specific training, expertise, and certifications required, and the project team
member(s) required to have them;
specific dates or a time frame for training sessions or certifications;
plans for securing the services of qualified staff with the necessary training and/or
certification;
plans for documenting the achievement of training requirements and certifications. (All
certificates or documentation representing completion of specialized training should be
maintained in personnel files.)
4.1.9 Documentation and Records (A9)
Describe the process and responsibilities for ensuring that appropriate project personnel
have the most current approved version of the QA Project Plan, including version control,
updates, distribution, and disposition. Itemize the information and records that must be
included in the given data reporting format and specify the reporting format for hard copy
and any electronic forms. Records can include data from other sources, model input and
output files, and results of model calibration and testing. Identify any other records and
documents applicable to the project that will be produced, such as test or data assessment
reports, interim progress reports, and final reports. Specify or reference all applicable
requirements for the final disposition of records and documents, including location and
length of retention period.
Preparing appropriate documentation for quality assurance purposes is important for all
environmental data operations, but especially so for modeling projects. Note that the title for this
element uses the broader term documentation instead of "documents" as in Requirements for Quality
Assurance Project Plans (QA/R-5) (EPA, 2001) because in modeling items like internal notes in the
code may need to be described. Information on how a model was selected, developed, evaluated, and
applied (as relevant) on a given project needs to be documented so that sufficient information is
available for model testing and assessment, peer review, and future model application.
Final
EPAQA/G-5M 36 December 2002
-------
Like the other QA Project Plan elements, the scope and magnitude of the project drives the
types of records and level of detailed documentation to be kept, and consequently, specified within this
element. For example, organizations conducting basic model research have different reporting
requirements than organizations producing predictions for regulatory decision making, in which
documentation needs to be sufficient to allow the model results to be re-created. Examples of specific
types of issues related to records and documentation for a modeling project that this element may
address include the following:
formats of the different types of data reporting packages (e.g., model parameterization,
model input, model calibration, model output), which need to be consistent with the
requirements and procedures for data assessment described in the remainder of the QA
Project Plan;
document control and distribution procedures;
storage requirements (may be governed by regulatory requirements, organizational
policy, or contractual project requirements);
backup plans for paper and electronic storage;
approvals (e.g., when changes or updates to records or documents are necessary);
retention periods; and
persons responsible for the various facets of record keeping (e.g., storage, tracking,
access, updates, obtaining approvals, distribution, final disposition).
Table 3 contains examples of records and documentation that may be relevant for some
modeling projects. Note that the QA Project Plan itself is addressed here, although information on
document control, distribution, updates (as necessary), and approvals for the QA Project Plan are
specified in earlier elements.
Table 3. Examples of Important Reporting and Documentation on a Modeling Project
Project
Phase/Document
Systematic Planning
Implementation
(Oversight)
Type of Documentation
Quality Assurance Project Plan
Approach/Concept Peer Review Reports
Model User's Guide (including application examples)
Standard Operating Procedures
Final
EPAQA/G-5M 37 December 2002
-------
Table 3. Examples of Important Reporting and Documentation on a Modeling Project
Project
Phase/Document
Type of Documentation
Assessment Reports for Acquired Data
Input and Test Data Sets
Code Verification Report
Hardware/Software Configuration Test Report
Source Code
Calibration Report
Model Assessment Reports (i.e., sensitivity analyses, comparison of output
to field or lab data or output of another model, reasonableness checks,
uncertainty analyses)
Configuration Management and Maintenance Manuals
Results
Assessment
Assessment Reports for Output Data
Output Data Set and Use Constraints Metadata
Report of Peer Review Comments
When peer review is necessary during the modeling process, the materials required for peer
review drive reporting and documentation needs on a modeling project. For example, peer reviewers
need access to model evaluation records in order to address the following types of peer review charge
questions (see EPA, 1994):
What databases were used to provide an adequate test?
What were key assumptions and the model's sensitivity to them?
Is the documentation of model code and verification testing adequate?
How well does the model report variability and uncertainty in its output?
This element of the QA Project Plan should specify the types of documentation that will be necessary
for peer review and how such information will be generated, maintained, and distributed.
If candidate models are assessed as part of a model selection process, certain documentation
(e.g., code verification, testing results, user's guide, application examples) are needed to ensure that the
model selected meets necessary acceptance criteria, such as criteria placed on hardware/software
configuration. Details on documentation of the project's hardware/software configuration are provided
within Section 4.2.3 of this guidance document [QA Project Plan Element BIO (Data Management and
Hardware/Software Configuration)].
EPA QA/G-5M
38
Final
December 2002
-------
The final output data reporting package includes individual records that represent actions taken
to achieve the objectives of the modeling project and the performance of specific QA functions. The
QA Project Plan should discuss how these various components will be assembled to represent a
concise and accurate record of all activities that impact data quality.
Example of EPA policy on documentation. EPA'sRisk Assessment Forum (EPA, 1997) has
established policy requiring sufficient documentation of the approach to a risk assessment before its
conclusions can be considered acceptable (EPA, 1996). Examples of model-related features that
should be documented on risk assessment projects include:
all models used and assumptions made that can impact results,
the rationale used to select the methods to propagate uncertainty and variability through
the model and to conduct sensitivity analyses of the model,
the statistical goodness-of-fit methods and other rationale used to decide which
statistical distributions should be used to characterize the uncertainty or variability of
model input parameters,
how the number of Monte Carlo simulations necessary to achieve sufficiently stable
probabilistic model results was determined,
how the methods used to elicit expert opinion on models and model inputs were
selected and implemented.
4.2 GROUP B: MEASUREMENT AND DATA ACQUISITION
The Group B elements focus on the quality procedures that are implemented when acquiring,
generating, and handling data to develop or apply a predictive tool or model. Chapter 3 showed that
the Group B elements are addressed when determining whether a new model needs to be developed,
when performing various tasks within model design and coding, and when preparing a model for a
particular application. The data acquisition methods are introduced within the project overview
provided in Element A6 (Project/Task Description and Schedule), and the Group B elements provide
detailed information on these methods.
Input data for model development and application efforts are typically collected outside of the
modeling effort or generated by other models or processing software. These data need to be properly
assessed to verify that a model characterized by these data would yield predictions with an acceptable
level of uncertainty. To this end, the Group B elements address various aspects of data acquisition, the
calibration of the model based on these data, management of the data, and the software/hardware
Final
EPAQA/G-5M 39 December 2002
-------
configuration needed for data processing. Of the ten Group B elements presented in EPA
Requirements for Quality Assurance Project Plans (QA/R-5) (EPA, 2001), the following three are
especially relevant for a modeling project:
[Model] Calibration (B7): Documenting the procedures for calibrating the model that
will perform the designated regulatory predictive task.
Non-direct measurements (data acquisition requirements) (B9): Introducing the
types and sources of existing data to be used in building and/or executing the model(s)
to be considered, specifying how these data will be acquired, and documenting the
quality associated with these data and their relevance in addressing project objectives.
Data management and hardware/software configuration (BIO): Documenting the
data management process from data acquisition through transmission and processing,
and to final use; documenting the components of the process to generate model outputs;
and highlighting the QA procedures associated with the configuration of the hardware
and software utilized by the model.
Although other Group B elements may occasionally be relevant for certain modeling projects,
they are generally geared toward projects that collect new data. These elements need to be
represented within the QA Project Plan for modeling projects, but they can be labeled as "not relevant"
for the given project if it is so determined. Sections 4.2.1 through 4.2.3 provide guidance on the above
three QA Project Plan elements for modeling projects.
When addressing the Group B elements, detailed copies of quality assurance methods and/or
standard operating procedures (SOPs) can be either included directly in the discussion, provided as
attachments to the QA Project Plan, or, if easily obtained and readily available to all project
participants [e.g., American Society for Testing and Materials (ASTM) methods], cited within the
discussion and included in the reference list.
4.2.1 [MODEL] Calibration (B7)
Identify all tools affecting quality that must be controlled and, at specified periods,
calibrated to maintain performance within specified limits. Describe or reference how
calibration will be conducted. If no nationally recognized standards exist, document the
basis for the calibration. Identify the certified equipment (approved model) and/or
standards used for calibration. Indicate how records of calibration shall be maintained and
be traceable to the instrument (model).
Final
EPAQA/G-5M 40 December 2002
-------
General QA Project Plan guidance indicates that Element B7 (Calibration) for data collection
addresses calibration of the analytical instruments that will be utilized to generate analytical data. For
modeling projects, by analogy the "instrument" is the predictive tool (the model) that is to be developed
and/or applied. All models, by definition, are a simplification of the environmental processes they are
intended to represent. When formulating the mathematical representations of these processes, there are
empirical relationships and parameters that need to be defined (e.g., the rate of formation or destruction
of a chemical). The estimation of parameters involved in formulating these empirical relationships is
called (model) calibration, and it is most often performed once in the model development phase.
However, some model parameters may need to be estimated for every application of the model, using
site-specific field data. Similar to an analytical instrument, models are calibrated by comparing the
predictions (output) for a given set of assumed conditions to observed data for the same conditions.
This comparison allows the modeler to evaluate whether the model and its parameters reasonably
represent the environment of interest [as determined based on accuracy requirements specified in
Element A7 (Qaulity Objectives and Criteria for Model Inputs/Outputs) of the QA Project Plan].
Statistical methods typically applied when performing model calibrations include regression analyses
(Draper & Smith, 1981) and goodness-of-fit methods (as discussed in EPA, 1999e; Gilbert, 1987; and
D'Agostino & Stephens, 1986). In this portion of the QA Project Plan, the details of the model
calibration procedure, including any statistical analyses that are involved, are documented.
On projects supporting regulatory decision making, the level of detail on model calibration in
the QA Project Plan should be sufficient to allow another modeler to duplicate the calibration method, if
the modeler is given access to the model and to the actual data being used in the calibration process.
For other projects (e.g., some basic research projects), it may be acceptable to provide less detail on
this issue for the QA Project Plan. For example, some projects may use procedures that are somewhat
different from standard calibration techniques, such as "benchmarking" procedures, and although such
procedures may be relevant to document in Element B7 (Calibration), the level of detail may differ from
what is generally portrayed for calibration. Even though some modeling projects may use techniques
other than standard calibration techniques, this section uses the term "calibration" to facilitate the
discussion.
Examples of different features of the model calibration effort that the QA Project Plan may
address include the following:
objectives of model calibration activities, including acceptance criteria;
frequency of model calibration activities;
details on the model calibration procedure;
method of acquiring the input data;
types of output generated by the model calibration;
method of assessing the goodness-of-fit of the model calibration equation to calibration
data;
Final
EPAQA/G-5M 41 December 2002
-------
method of incorporating variability and uncertainty in the model calibration results;
corrective action to be taken if acceptance criteria are not met.
Each of these items is addressed in detail in the paragraphs that follow.
Objectives of Model Calibration Activities. Including Acceptance Criteria - Information related
to objectives and acceptance criteria for calibration activities that generally appear at the beginning of
this QA Project Plan element includes the following:
Objectives of the model calibration., including what the calibration should accomplish
and how the predictive quality of the model might be improved as a result of
implementing the calibration procedures.
Acceptance criteria: The specific limits, standards, goodness-of-fit, or other criteria
on which a model will be judged as being properly calibrated (e.g., the percentage
difference between reference data values from the field or laboratory and predicted
results from the model). This includes a mention of the types of data and other
information that will be necessary to acquire in order to determine that the model is
properly calibrated (e.g., field data, laboratory data, predictions from other accepted
models).
Justifying the calibration approach and acceptance criteria. Each time a model is
calibrated, it is potentially altered. Therefore, it is important that the different
calibrations, the approaches taken (e.g., qualitative versus quantitative), and their
acceptance criteria are properly justified. This justification can refer to the overall
quality of the standards being used as a reference or of the quality of the input data
(e.g., whether data are sufficient for statistical tests to achieve desired levels of
accuracy).
Frequency of Model Calibration Activities - Inputs to the model calibration process can highly
influence the quality of information generated by the model. Therefore, the calibration process may
need to be iterative in nature, repeated whenever some key aspect of the environment changes. Each
iteration would utilize data that accurately portray the changing environment and, therefore, would
provide further necessary refinements to the model. This element of the QA Project Plan should
contain a schedule for addressing model calibration issues, with explanation as needed for the specified
frequency and timing of calibration.
Details on the Model Calibration Procedure - The QA Project Plan documents the procedures
to be used when performing model calibrations during the project, providing information such as the
following:
Final
EPAQA/G-5M 42 December 2002
-------
An overview of each model or model component requiring calibration should be given,
along with the various components of the calibration procedure, some of which may
coincide with the model's components. This could be specified in text format and/or in
a graphic, flow diagram-type figure. This presentation can incorporate how schedule
and other time-dependent factors interplay with the various stages of the calibration
procedure.
Details on specific methods to be used to perform the calibration, for each portion of
the model and at each stage of the procedure. SOPs detailing the methods can be
appended to the QA Project Plan.
Any need to modify the calibration method to accommodate data acquired for
calibration purposes (see below).
The resources necessary to conduct the model calibration, along with the individual
responsible for directing the model calibration efforts [as specified in the organizational
chart in Element A4 (Project/Task Organization)].
The approach to maintaining calibration records and ensuring that the results can be
traced to the appropriate version of the model should be specified [over and above
what has already been discussed in Element A9 (Documentation and Records) on this
subject].
Method of Acquiring the Input Data - Element B9 (Non-direct Methods) provides details on
the types of existing data to be acquired and used as input to model calibration activities. This element
can document some introductory information on these data, such as the following:
The types of data necessary at each stage of the calibration procedure and for each
model component (or each model), along with any need for the data to represent a
specific environmental situation determined by location or some other unique
characteristic;
How the data will be acquired;
How the quality of the data for model calibration will be determined and verified
throughout the calibration procedure. If previous investigations on these data provide
information on the quality of the data, references documenting the level of data quality
should be included in the QA Project Plan. Otherwise, any methods used to verify data
quality in the context of this project should be documented.
Final
EPAQA/G-5M 43 December 2002
-------
Types of Output Generated by the Model Calibration - The important measures and outputs
that are expected to be generated upon implementing the model calibration procedure and that will be
used to assess whether the model is properly calibrated should be documented. In addition, statistical
QC techniques to be used to process the output data for comparison to reference values or other
acceptance criteria should be described. The quality assurance aspects of these analyses should also
be addressed.
Method of Assessing the Goodness-of-fit of the Model Calibration Equation to Calibration
Data - Statistical procedures such as chi-square tests and various regression diagnostic reviews (e.g.,
residual plots, tests for lack of fit) are generally used when comparing the distribution of model output
data that results from calibrating the model to the distribution of data measured within the particular
environment that the model output is to simulate. If such procedures are used on the project, they
should be referenced here along with the criteria to be used in judging the "goodness-of-fit" of the
model-generated distribution with the reference distribution.
Method of Incorporating Variability and Uncertainty in the Model Calibration Results - For a
given environmental condition, uncertainty in the representativeness of the model input data (e.g.,
incompleteness, variability, and unintentional bias) will affect uncertainty in the outcome of model
calibration. Deviations to the input data (reflecting the data's inherent uncertainty) or to the calibration
methods and acceptance criteria can yield different model calibration outcomes. The QA Project Plan
addresses uncertainty in the outcome of model calibration and its potential impact on decisions being
made from this outcome by documenting the following:
The expected sources of uncertainty and variability in the model and their potential
effect on the outcome of model calibration.
The tools to be used to characterize uncertainty and variability in the outcome of model
calibration (e.g., Monte Carlo techniques, sensitivity analysis).
Acceptance criteria to be used to evaluate the level of uncertainty and variability,
relative to whether the resulting uncertainty in the outcome of model calibration falls
within acceptable limits.
Corrective Action to Be Taken If Acceptance Criteria Are Not Met - The QA Project Plan
should document what corrective action would be taken when situations such as the following occur:
when limits, standards, or other criteria that identify whether the model is properly
calibrated are not achieved;
Final
EPAQA/G-5M 44 December 2002
-------
when sensitivity or uncertainty analysis imply that uncertainty in the model calibration
outputs exceeds pre-specified criteria.
Situations in which the model calibration process may need to be repeated after any corrective action is
taken should also be specified.
4.2.2 Non-direct Measurements (Data Acquisition Requirements) (B9)
Identify any types of data needed for project implementation or decision making that are
obtained from non-measurement sources such as computer data bases, programs,
literature files, and historical databases. Describe the intended use of the data. Define the
acceptance criteria for the use of such data in the project and specify any limitations on the
use of the data.
"Non-direct" measurements refer to data and other information that have been previously
collected or generated under some effort outside the specific project being addressed by the QA
Project Plan. Examples include computer databases, literature files, and software processing.
Frequently, using existing data rather than generating new data is sufficient to meet the needs of some
phases of a modeling project. Because the data have already been collected and therefore, the needs
of the project cannot influence how the measurements were generated, these data need special
consideration. Issues regarding how relevant non-direct measurements are identified, acquired, and
used on the project are addressed within this QA Project Plan element. Additional guidance on the
secondary use of data may also be found in other EPA documents currently under development.
Consider an example where the sources of certain model input data are historical databases or
nationally-representative surveys. This element of the QA Project Plan would document how the
approach to collecting these data (as determined from the data's sources) could impact the ability to
meet the modeling project's quality objectives [detailed in Element A7 (Qaulity Objectives and Criteria
for Model Inputs/Outputs)]. The issue is not one of the credibility of the organization which generated
the data, but rather whether this particular data set is relevant for the problem at hand. When used
uncritically, data with characteristics that do not meet the quality objectives of the project may lead to a
higher chance of making decision errors. Therefore, it is important to scrutinize the applicability of any
non-direct measurements to a project, regardless of their sources.
The QA Project Plan should address the following four issues regarding information on how
non-direct measurements are acquired and used on the project:
the need and intended use of each type of data or information to be acquired;
how the data will be identified or acquired, and expected sources of these data;
Final
EPAQA/G-5M 45 December 2002
-------
the method of determining the underlying quality of the data; and
the criteria established for determining whether the level of quality for a given set of data
is acceptable for use on the project.
Each of these items is addressed in detail below.
The Need and Intended Use of Each Type of Data or Information to Be Acquired - This QA
Project Plan element can begin by introducing those components of the modeling process [introduced in
Element A6 (Project/Task Description and Schedule)] in which non-direct measurements are expected
to be utilized as input. For each of these components, the following items should be addressed:
the particular need for non-direct measurements, relative to how the need will allow the
objectives in Element A6 (Project/Task Description and Schedule) to be achieved;
justification for using non-direct measurements, rather than generating new data;
details on the intended use of the data to satisfy the given need.
How the Data Will Be Identified or Acquired, and Expected Sources of These Data - Once
the need for and the intended use of non-direct measurements have been introduced, the approach to
obtaining data to meet these needs can be specified. Sources of non-direct measurements identified for
possible use on the project would be documented here. In addition, the procedure for selecting data
sources would be detailed here, along with the types of information that these sources would be
anticipated to provide. This discussion would also mention the criteria and critical judgment that would
be used to determine whether certain data sources are relevant and how certain data sources would be
selected if competing sources were available. If sources of potentially acceptable data are available but
would likely not be considered for the project, then reasons for not considering these sources should be
given to address the potential for selection bias that this may introduce.
The QA Project Plan should also indicate the extent to which project objectives (or other
criteria, such as limits on decision errors) can be achieved by using specific sources and types of data
and parameters that have been identified through the data acquisition process. One example of how
this may be done is to rank the different data sources relative to how critical their data are to the
modeling process and its output and the extent to which the data have been collected according to
specifications of the modeling project. Data receiving high rankings would have a high likelihood of
being selected (if multiple sources exist or if budgeted resources are limited) and would be placed under
greater scrutiny during data assessment and software testing processes [Elements BIO (Data
Management and Hardware/Software Configuration) and Cl (Assessment and Response Actions)].
Final
EPAQA/G-5M 46 December 2002
-------
The Method of Determining the Underlying Quality of the Data - Once specific types of data
and information and the sources of these data have been identified, it is necessary to express the
acceptance criteria introduced in Element A7 (Qaulity Objectives and Criteria for Model
Inputs/Outputs) relative to each anticipated data source and how these criteria will determine whether
the given data source is an acceptable input to the modeling project. Because these acceptance criteria
are directly tied to the underlying quality of the data, this QA Project Plan element should precede this
discussion by specifying what is known about the quality of the non-direct measurements to be used in
the modeling project.
A first step in gathering information on acceptance criteria for a given source of non-direct
measurement data is to determine whether a QA Project Plan or a sampling and analysis plan exists and
is available for that source. Complete SOPs, other standard procedures, and laboratory-specific
deviations and specifications associated with these procedures may also be important to review and
reference in the QA Project Plan. When analytical data are being considered, it may be sufficient for
the QA Project Plan to simply cite the analytical method if it is a complete SOP. If other analytical
methods were used, a reference to a procedure (e.g., from Test Methods for Evaluating Solid
Waste, SW-846) should be identified, along with other relevant information (e.g., options and
variations used by the lab, detection limits actually achieved, calibration standards, and concentrations
used). If the analytical procedure is unique or an adaption of a "standard" method, the QA Project
Plan should describe or cite the complete analytical and sample preparation procedures.
From these references, the results of data assessment and the quality guidelines associated with
the given type of measurement may be cited in the QA Project Plan for the modeling project, as well as
the extent to which the data abide by these guidelines. In particular, any limitations on the use of the
data that may result from any uncertainty that may exist in the data's quality should be noted.
One approach to determining whether non-direct measurements satisfy acceptance criteria
involves referring to references obtained from the data sources to identify information on data quality
indicators (DQIs) associated with the measurements (Section 4.1.7). The QA Project Plan should
document the relevant findings for the non-direct measurements.
Other information about the projects and procedures that generated the non-direct
measurements may also be important to cite in the QA Project Plan. For example, the extent to which
the measurements reflect the modeling project's target population, the sampling and analytical methods
used, and the parameters of interest to the modeling project would be important to document. The
sampling method can materially affect the representativeness, comparability, bias, and precision of the
final analytical result. Similarly, the analytical methods influence the outcome of the measurements and,
as a result, their ability to achieve decision performance criteria for the model's intended predictive
task. These types of information serve as examples; the information relevant to a given project will
depend on the project's overall scope and objectives.
Final
EPAQA/G-5M 47 December 2002
-------
The Criteria Established for Determining Whether the Level of Quality for a Given Set of Data
is Acceptable for Use on the Project - Documenting how the non-direct measurements achieve pre-
established acceptance criteria is one step toward satisfying the quality objectives for the modeling
project that were specified within Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs) of the QA Project Plan. Thus, as a first step in determining acceptance criteria, the
project's objectives for the quality of the intended output [as determined through the systematic
planning process documented in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs)] are reviewed. During this first step, the quality objectives can be translated into a
statement about the likelihood of making certain decision errors as a result of using the model output in
statistical analysis. This first step corresponds to the first stage of the Data Quality Assessment process
that EPA has established to evaluate data based on scientific and statistical criteria in order to determine
whether the data are of the right type, quality, and quantity to support their intended use. The first stage
of the Data Quality Assessment process is documented in Chapter 1 of Guidance for Data Quality
Assessment (QA/G-9) (EPA, 2000c).
Note that information on project objectives and/or limits placed on decision errors (as
determined within systematic planning) would be used as the basis for ranking the various data types by
their criticality to achieving these objectives, as discussed previously.
Acceptance criteria for individual data values generally address issues such as the following:
Representativeness: Were the data collected from a population sufficiently similar to
the population of interest and the model-specified population boundaries? Were the
sampling and analytical methods used to generate the collected data acceptable to this
project? How will potentially confounding effects in the data (e.g., season, time of day,
location, and scale incompatibilities) be addressed so that these effects do not unduly
impact the model output?
Bias: Would any characteristics of the data set directly impact the model output (e.g.,
unduly high or low process rates)? For example, has bias in analysis results been
documented? Is there sufficient information to estimate and correct bias? If using data
to develop probabilistic distributions, are there adequate data in the upper and lower
extremes of the tails to allow for unbiased probabilistic estimates?
Precision: How is the spread in the results estimated? Is the estimate of variability
sufficiently small to meet the uncertainty objectives of the modeling project as stated in
Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs) (e.g., adequate
to provide a frequency of distribution)?
Final
EPAQA/G-5M 48 December 2002
-------
Qualifiers: Have the data been evaluated in a manner that permits logical decisions on
the data's applicability to the current project? Is the system of qualifying or flagging
data adequately documented to allow data from different sources to be used on the
same project (e.g., distinguish actual measurements from estimated values, note
differences in detection limits)?
Summarization: Is the data summarization process clear and sufficiently consistent
with the goals of this project (e.g., distinguish averages or statistically transformed
values from unaltered measurement values)? Ideally, processing and transformation
equations will be made available so that their underlying assumptions can be evaluated
against the objectives of the current project.
In addition to addressing these questions when establishing acceptance criteria, the QA Project
Plan can document the likely consequences (e.g., incorrect decision-making) for selecting data that do
not satisfy one or more of these areas (e.g., are non-representative, are inaccurate), as well as
procedures in place to minimize the likelihood of selecting such data.
Certain types of data may be needed to meet the project's data needs [Element A7 (Quality
Objectives and Criteria for Model Inputs/Outputs)] but may not need to meet the same level of quality
as other data (e.g., data meant only to provide background information). This should be noted in the
QA Project Plan when relevant.
4.2.3 Data Management and Hardware/Software Configuration (BIO)
Describe the procedures for assuring that applicable Agency information resource manage-
ment requirements (EPA Directive 2100) are satisfied (EPA QA Project Plans only). If
other Agency data management requirements are applicable, such as the data standards,
discuss how these requirements are addressed.
This QA Project Plan element focuses on the procedures in place to manage (i.e., compile,
handle) the data that are expected to be either acquired or generated by the modeling project.
Different phases of the project may consider different types and sources of data, and the use of certain
data may span across multiple project phases. Therefore, the first segment of this QA Project Plan
element (Data Management, labeled "Element BlOa") documents the approaches that the project will
take to manage the various forms and types of data on the project from acquisition or generation -
through their final use. The first subsection (Section 4.2.3.1) provides guidance on documenting the
approach to data management.
Final
EPAQA/G-5M 49 December 2002
-------
Linked with data management is the hardware/software configuration required to input the
acquired data in an existing model and to generate the model output necessary to meet the objectives of
the prediction task. Also, as indicated in Figure 4, the hardware/software configuration will influence
how a model is coded. Therefore, the second segment of this QA Project Plan element
(Hardware/Software Configuration, or Element Bl Ob) documents the software and hardware
configuration needed for data processing within the modeling project. The type of information included
about the hardware/software configuration should be consistent with what is specified in the Software
Quality Assurance Plan addressed by American National Standards Institute/Institute of Electrical and
Electronics Engineers Standard 730 (IEEE, 1999), as integrated into overall project planning. The
second subsection (Section 4.2.3.2) provides guidance on specifying the requirements of and the
approach to developing the hardware/software configuration.
4.2.3.1 Data Management (BlOa)
In general, the QA Project Plan should provide enough documentation on the project's data
management procedures (e.g., data entry, data tracking, data manipulation) to permit the data to be
traced from their source and acquisition through their transmission and processing, to their final use on
the project. Data management procedures are addressed for all types and sources of data, from the
non-direct measurements [introduced in Element B9 (Non-direct Measurements)] used as input to the
model through the model outputs.
Describing how the various data sources can be traced through the life cycle of the project,
from initial source to final use, can be a difficult task. It is often better to portray this process
graphically rather than textually through one or more process flow diagrams. An example of a
process flow diagram is given in Figure 10. The process flow diagram shows the various data inputs
and outputs for each procedure in the modeling process, the interaction of these procedures, and the
reports that the process produces. Therefore, the diagram documents the intermediate and final stages
of the data relative to the flow of the different project phases, as well as the roles that each type of data
play within the modeling process and in generating the final products.
The QA Project Plan also documents the tools to be used to facilitate the data management
process. Examples of such tools include forms, checklists, database documentation, on-line interactive
screens, and program code. Some tools can be included as attachments to the QA Project Plan.
This element of the QA Project Plan should present an overview of the mathematical operations
and data processing operations that will be performed on the data during the pre-processing, model
computation, and post-processing stages of the modeling process, as described below.
Final
EPAQA/G-5M 50 December 2002
-------
(
Point
Source Data
(Land Use ( \ (
Data I ) V^
Stream
Data
( Meteorological f \
\^ Data \^ J
I
Watershed Assessment, Transport,
and Fate/Bioaccumulation Models
I
Contaminant/
Data for
Model
Calibration
C Predicted
Contaminant
Concentrations/
Post Processing of
Model Outputs
Final Report
Figure 10. Example of a Process Flow Diagram for a Modeling Project
In the pre-processing stage, the input data are prepared for use in the modeling stage
by performing procedures such as data formatting, reduction, transformations,
conversions, and subsetting. [Analyses used to verify whether the data satisfy pre-
determined assumptions for statistical analysis are addressed in QA Project Plan
Element D3 (Reconciliation and User Requirements).] As necessary, the discussion
should justify why certain data procedures need to be implemented and address the
impact that the data procedures may have on the ability to achieve the project's quality
objectives documented in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs).
In the model computational stage, the mathematical equations within the model are
derived and applied to the data. [This is different from the mathematical approaches
EPA QA/G-5M
51
Final
December 2002
-------
used to calibrate the model, which are addressed in Element B7 (Calibration).] While a
purpose of the project may be to develop the specific mathematical procedures and
equations that constitute the model computational stage, this element can still highlight
the primary mathematical approaches that are expected to be applied and how these
approaches will ensure that the model's underlying scientific principles will be properly
incorporated.
In the post-processing stage, statistical procedures are applied to analyze the model
output, to generate data summaries and reports introduced in Element A9
(Documentation and Records), and to characterize variability and uncertainty in the
model output.
This element should outline the proposed methodology for data analysis (e.g., summary statistics,
frequency distributions, etc.), verifying that the methodology can be applied successfully to collected or
existing data. If the final report will provide details on the methodology, this should be mentioned in the
QA Project Plan. This element should also address how the final forms of the data will be processed in
support of the project's requirements for the types of information generated from the outputs of the
project (i.e., reporting the synthesis of results in the form of charts, tables, graphs, figures, visualizations)
and how deviations from these requirements would be handled.
Examples of "control mechanisms" associated with data management that can be documented
in the QA Project Plan include the following:
the approach to detecting and correcting errors that may occur during data entry to
forms, reports, or databases, and to preventing loss of data during data reduction and
data reporting;
internal checks to be implemented during data entry or data pre-processing;
the proposed approach to generating an audit trail associated with certain data
management activities such as data reduction and conversion;
guidelines for transmitting data from one person or location to another or copying data
from one form to another;
the approach to characterizing data transmittal error rates;
the approach to minimizing information loss during data transmittal; and
Final
EPAQA/G-5M 52 December 2002
-------
descriptions of the established procedures for tracking the flow of data through the data
processing system portrayed within the process flow diagram.
Relevant control procedures are determined based on the project's overall scope and importance of
the project's objectives.
Examples of what may be addressed in the QA Project Plan regarding documentation of the
data management process include standard record-keeping procedures, audit trails, a system for
controlling and tracing the whereabouts of data and documents, and the approach to storing and
retrieving data records for various storage media types (e.g., hardcopy, electronic).
4.2.3.2 Hardware/Software Configuration (BlOb)
This section of the QA Project Plan provides information on the types of computer equipment,
hardware, and software to be used on the project, including information on how they will be used (e.g.,
for conducting the data management procedures specified earlier in this element of the QA Project
Plan). If different types of equipment, hardware, and software are available, a brief justification should
be provided, as necessary, for selecting certain types over other types. Reasons should also be given if
a certain type of equipment, hardware, or software is required to be used on the project. This type of
information may be best presented in table format.
This element may also address performance requirements (e.g., run times) and other features
that characterize and assess the hardware/software configuration. This discussion should be
incorporated within a general overview of the configuration's QA program. (Assessments that target
the model itself and its ability to make predictions are addressed by the Group C elements within the
QA Project Plan.) The configuration's QA program is jointly planned and implemented by the project
management team and the software developer's independent QA staff, generally as part of systematic
planning [Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs)]. It addresses the
use of standards, test planning and scheduling, level of documentation required, personnel assignments,
and change control. It also ensures that timely corrective action will be taken as necessary. Items
within the systems development life cycle that are relevant to the particular modeling project should also
be considered when establishing the configuration's QA program. Examples of such items, taken from
Chapter 4 of'EPA's Information Resources Management Policy Manual (Directive 2100) (EPA,
1998b) and the Information Technology Architecture Roadmap1, are provided in Table 4.
Published by EPA's Office of Technology Operations and Planning, formerly the Office of Information
Resources Management, Directive 2100 establishes a policy framework for managing information within EPA. It can
be accessed online at www.epa.gov/irmpoli8/pohnan/index.html. The Information Technology Architecture
Roadmap, which contains annual updates of this document, can be found at (Internal EPA website)
http://Basin.rtpnc.epa.gov: 9876/etsd/ITARoadMap.nsf.
Final
EPAQA/G-5M 53 December 2002
-------
Table 4. Typical Activities and Documentation Prepared Within the System Development
Life Cycle of a Modeling Project that Should be Considered When Establishing the QA
Program for the Hardware/Software Configuration
Life Cycle Stage
Needs Assessment
and General
Requirements
Detailed
Requirements
Analysis
Framework Design
Implementation
Controls
Testing,
Verification, and
Evaluation
Installation and
Training
Operations,
Maintenance, and
User Support
Typical Activities
Assessment of needs and requirements
interactions in systematic planning with users
and other experts
Listing of all inputs, outputs, actions,
computations, etc., that the modeling
framework is to perform
Listing of ancillary needs such as security
and user interface requirements
Design team meetings
Translation of requirements into a design to
be implemented
Coding and configuration control
Design/implementation team meetings
Verification that the modeling code, including
algorithms and supporting information
system, meets requirements
Verification that the design has been correctly
implemented
Beta testing (users outside team)
Acceptance testing (for final acceptance of a
contracted product)
Implement necessary corrective actions
Installation of data management system and
training of users
Usage instruction and maintenance resources
for model framework and data bases
Documentation
Needs assessment documentation
(e.g., QA Project Plan)
Requirements document
Detailed requirements document,
including performance, security,
user interface requirements, etc.
System development standards
Design documents) including
technical framework design,
software design (algorithms, etc.)
In-line comments
Change control documentation
Test plan
Test result documentation
Corrective action documentation
Beta test comments
Acceptance test results
Installation documentation
User's guide
User's guide
Maintenance manual or
programmer's manual
It is important that the QA Project Plan specify the particular QA procedures that will be
implemented within the modeling project to ensure that the data generated by the modeling framework
(i.e., the model and its supporting hardware and operating system) are defensible and appropriate for
its planned final use. This section of the QA Project Plan would address QA efforts performed as the
data management and processing systems are being developed. The schedule for these efforts should
also be included in task scheduling [Element A6 (Project/Task Description and Schedule)]. These
efforts may include:
EPA QA/G-5M
54
Final
December 2002
-------
identifying necessary requirements for the hardware/software configuration and
establishing quality criteria that address these requirements within the systematic
planning and needs analysis phase of the project [Element A7 (Quality Objectives and
Criteria for Model Inputs/Outputs)];
implementing an appropriate proj ect management framework to ensure that the
requirements and quality criteria established for the hardware/software configuration are
achieved [as discussed in Elements A4 through A9 and B9 (Non-direct
Measurements)]; and
performing testing and other assessment procedures on the configuration to verify that
the project-specific requirements and quality criteria are being met [details on the
assessment procedures are addressed in Element Cl (Assessment and Response
Actions)].
The magnitude of these QA efforts will depend on the underlying complexity of the modeling effort and
the required hardware/software configuration. Therefore, EPA's graded approach (Chapter 1) will
direct the overall scope on which these QA efforts need to focus.
Elaborating further on the first bullet above, the systematic planning phase of the study [Element
A7 (Quality Objectives and Criteria for Model Inputs/Outputs)] defines requirements and quality
criteria for the data processing system to ensure that the project's end-use needs can be adequately
met. For example, criteria on errors propagated by data processing would be established during
systematic planning to ensure that uncertainty requirements for the model outputs can be met. Such
requirements and criteria, therefore, impact the project's hardware/software configuration.
In systematic planning, questions such as the following may be addressed when defining these
requirements and quality criteria:
What are the required levels of accuracy and uncertainty for numerical approximations?
Are the selected mathematical features of the model (e.g., algorithms, equations,
statistical processes) appropriate for the model's end use?
Are the correct data elements being used in the calculations performed within the
model's algorithms?
What requirements regarding documentation and traceability are necessary for the
model's inputs, interim outputs, and final outputs?
Final
EPAQA/G-5M 55 December 2002
-------
Other items addressed during systematic planning that are likely to impact assessment of the
hardware/software configuration include security, communication, software installation, and system
performance (e.g., response time). These issues are addressed briefly below.
Security issues: Examples of QA efforts that address security issues regarding data handling
and communication (e.g., viruses, hackers, e-mail interception) and system vulnerability include the
following:
ensuring that security needs are included among project requirements and can be found
within the project requirements documentation;
ensuring that security features of the configuration are adequately and thoroughly tested
and the results documented;
planning for audits addressing system vulnerability to be conducted by security
personnel outside the immediate project team.
Other security issues that may be relevant to a particular project (preferably during system planning) are
found in EPA's Information Security Manual (Directive 2195A1) (EPA, 1999a).
Communication issues: The QA program should address communication among the model
developers and users regarding configuration acceptance and control of changes, such as:
ensuring that all user communication interfaces are adequately defined, and that beta
testing is performed to have users test the connections of these various interfaces;
ensuring that proper and complete testing of the communications hardware and
software is done, preferably under "worst-case" (i.e., high-load, adverse) conditions.
Software installation issues: The QA program should specify those QA techniques
associated with installing or uninstalling the software configuration on users' computers. Examples of
such techniques include the following:
performing adequate beta testing to ensure that installation can be performed
successfully on a wide variety of different platforms, including various combinations of
processors, memory sizes, and video controllers;
ensuring that uninstall routines are complete (e.g., removing entries in the initialization
and registry files as well as deleting files);
Final
EPAQA/G-5M 56 December 2002
-------
ensuring that any "auto-install" or "auto-uninstall" routines are thoroughly tested and
debugged before release.
Response time issues: If the QA program provides explicit, testable objectives for response
time performance in various phases of the modeling procedure (e.g., real-time data processing, pre-
processing, post-processing, uncertainty analyses), these should be documented in the QA Project Plan
for all interactive and real-time systems.
When documenting planning and performance components of the hardware/software
configuration, project and QA Managers should tailor the documentation to meet the specific needs of
their project. Examples of different types of documentation that can be generated for various tasks
within the planning phase of the system's life cycle include the following:
Requirements Documentation (IEEE, 1998b): The general requirements document
gives an overview of the functions that the model framework will perform. Detailed
requirements documents define all critical functions that the completed framework
needs to support. Performance goals derived from analysis of the project's quality
objectives [QA Project Plan Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs)] should be included among the requirements. Requirements
documentation should be reviewed by the regulatory program end user, if possible, to
ensure that critical functions and other requirements have not been overlooked.
Design Documentation: Design documents plan and describe the structure of the
computer program. These are particularly important in multi-programmer projects in
which modules written by different individuals interact. Even in small or single-
programmer projects, a formal design document can be useful for communication and
for later reference.
Coding Standards or SOPs: These may apply to a single project or a cumulative
model framework and need to be consistent across the development team. Uniform
standards for code formats, subroutine calling conventions, and in-line documentation
can significantly improve the maintainability of models and processors.
Testing Plans (IEEE, 1986): Testing needs to be planned in advance and address all
requirements and performance goals. Specific procedures for corrective action and
retesting process should be described in the QA Project Plan (similar to IEEE, 1999)
and implemented within the assessments discussed in the QA Project Plan's Group C
elements.
Final
EPAQA/G-5M 57 December 2002
-------
Data Dictionary: A data dictionary can be useful to developers, users, and
maintenance programmers who may need to modify the modeling framework later. The
data dictionary is often developed before code is written as part of the design process.
The dictionary should be updated as necessary when new elements are added to the
data structure. A data dictionary need not be a separately written document. For
example, the record definition files needed for many database systems can serve this
purpose, provided that they are available in a form that is readily accessible to the user
or maintenance programmer.
User's Manual: The user's manual can often borrow heavily from the requirements
document because all the software's functions should be specified there. The scope of
the user's manual should take into account such issues as the level and sophistication of
the intended user and the complexity of the interface. Online help can also be used to
serve this function.
Maintenance Manual: The maintenance manual's purpose is to explain a framework's
software logic and organization for the maintenance programmer. This manual should
also contain crucial references documenting algorithms, numerical methods, and
assumptions. Instructions for rebuilding the system from source code are included. The
maintenance manual will often borrow heavily from the design manual.
Source Code: It is very important to securely store downloadable code and to archive
computer-readable copies of source code according to the policies of the relevant
regulatory program.
Configuration Management Plan (IEEE, 1998a): The configuration management
plan provides procedures to control the software/hardware configuration during
development of the original model version and subsequent revisions. It ensures that
only approved deletions, additions, and changes are introduced in a controlled way to
maintain the integrity of the model. The configuration management process starts when
the unit is successfully tested and placed under project configuration control. Upon
deployment, a documented product baseline configuration enhances the reliability and
maintainability of the software/hardware configuration through the remainder of its
useful life.
Additional information and examples can be found in Chapter 17 of EPA's Information Resources
Management Policy Manual (Directive 2100) (EPA, 1998b). In general, any discussion of
documentation in the QA Project Plan should be coordinated with information presented in Element A9
(Documentation and Records).
Final
EPAQA/G-5M 58 December 2002
-------
The configuration should be designed to comply with applicable EPA information resource
management and data standard policies, which can be found within EP'A's Information Resources
Management Policy Manual (Directive 2100) (EPA, 1998b). Other policies and standards may also
be applicable and should be cited, such as the Federal Information Processing Standards, which govern
the acquisition of U.S. Government information processing systems. This portion of the QA Project
Plan should introduce these standards and discuss how the project will ensure that they will be
addressed.
Sources for determining specific types of standards include the following:
EPA's Information Resources Management Policy Manual (Directive 2100) (EPA,
1998b) includes the EPA hardware and software standards to promote consistency in
use of standard support tools such as computer-aided software engineering tools and
coding languages, as applicable, by contractors and EPA staff in model software
development and maintenance efforts.
Chapter 5 of EPA's Information Resources Management Policy Manual (Directive
2100) (EPA, 1998b) defines applicable EPA data standards.
EPA's Environmental Data Registry (www.epa.gov/edr) promotes data standardization,
which allows for greater ease of information sharing.
The EPA Information Technology Architecture Roadmap (Internal EPA website:
Basin.rtpnc.epa.gov:9876/etsd/ITARoadMap.nsf) provides guidance for the selection
and deployment of computing platforms, networks, systems software, and related
products that interconnect computing platforms and make them operate.
Publications on Federal Information Processing Standards govern the acquisition of
U.S. Government information processing systems.
Directives and standards such as these are frequently revised. Therefore, it is important that these
directives and standards be reviewed frequently to ensure that the latest versions are being utilized. The
QA Project Plan should specify how the acceptability of the configuration will be verified or
demonstrated according to these and other standards.
4.3 GROUP C: ASSESSMENT AND OVERSIGHT
The purpose of the assessment and oversight section of the QA Project Plan for a modeling
project (i.e., the Group C QA Project Plan elements) is to describe the internal and external checks
and activities necessary to assess the effectiveness of the modeling project implementation and
Final
EPAQA/G-5M 59 December 2002
-------
associated QA and QC activities. These checks and activities are needed to detect errors and to
ensure that using the model being developed and applied achieves the modeling task (i.e., meets user
requirements) at the prediction ability and levels of quality specified in QA Project Plan Element A7
(Quality Objectives and Criteria for Model Inputs/Outputs) (Section 4.1.7). Furthermore, the
assessment is needed to ensure that adequate coordination and documentation of all planned internal
and external assessments (e.g., peer reviews) are achieved, and that any needed corrective actions are
implemented and their effectiveness assessed and documented in a timely manner.
All models need to include this type of evaluation in the planning stageseven data-poor or
qualitative models for which traditional model evaluations cannot be done. The QA Project Plan for
these projects should contain information to address the following questions:
Are more studies needed?
How good does my confidence have to be to move forward?
What do I need to know to be able to show this model is sound?
Assessment and Response Actions (Element Cl) includes various model theory and algorithm
evaluation assessments, assessments of the effect of variability within a population and uncertainty of
model structure and parameter values on model outputs, the sensitivity of model outputs to a particular
model and model inputs used, data quality assessments, model verification tests by statistical
comparisons of model outputs with field or laboratory data, internal and external peer reviews, and
hardware/software configuration testing. These assessments may lead to response actions to make
mid-course corrections in project activities. Reports to Management (Element C2) identifies the
frequency and distribution of reports issued to inform management of the project status and ensures
prompt and appropriate corrective actions are taken during model development or application.
Because both Groups C and D of the QA Project Plan address assessment activities,
information presented in these two groups of elements may overlap (as discussed in Section 4.4).
Generally, Group C of the QA Project Plan focuses on interim assessments conducted iteratively
throughout both the model development and application processes to ensure that pre-determined
modeling criteria are being achieved as the model is being developed or implemented. In contrast,
Group D focuses on final assessments performed in the final stages of model development and after the
model has been applied in a given instance to evaluate whether the outcomes meet the project's original
objectives. This distinction between the Group C and Group D elements is consistent with EPA
Requirements for Quality Assurance Project Plans (QA/R-5) (EPA, 2001) for environmental
sampling and data analysis projects, which specifies that the assessments documented in Group C
occur first, focusing on how samples are being collected and data are being generated, while
assessments in Group D focus on ensuring that the newly generated data can be used to achieve the
project's objectives.
Final
EPAQA/G-5M 60 December 2002
-------
4.3.1 Assessment and Response Actions (Cl)
Describe each assessment to be used in the project, including the frequency and type.
Discuss the information expected and the success criteria (i.e., goals, performance
objectives, acceptance criteria and specifications, etc.) for each assessment proposed. List
the approximate schedule of assessment activities.
For any planned self-assessments (utilizing personnel from within the project groups),
identify potential participants and their exact relationship within the project organization.
For independent assessments, identify the organization and person(s) that shall perform
the assessments, if this information is available. Describe how and to whom the results of
each assessment shall be reported. Define the scope of authority of the assessors,
including a description of when assessors are authorized to act.
Discuss how response actions to assessment findings, including corrective actions for
deficiencies and other non-conforming conditions, are to be addressed and by whom.
Include details on how the corrective actions will be verified and documented.
This element of the QA Project Plan documents the types of assessments to be performed
throughout the various stages of both model development and application, the purpose of each
assessment and the specific model features that each assessment is to address, and the expected
periods of time in which the assessments will take place. The specific assessments are based on a clear
understanding and statement of the purpose of the model, the way the model will be used in
management and/or regulatory tasks and decision making, and the accuracy of the model outputs
needed (predictions).
In general, this QA Project Plan element specifies the following types of information:
a description of the assessment/oversight strategies and schedule of assessment
activities, including the order in which the assessments will be conducted and how the
total set of assessments is structured to provide a complete and comprehensive
assessment;
a description of how each assessment will be planned and conducted;
the organizations and individuals that are expected to participate in assessments,
including peer reviews;
the information expected, success criteria, and documentation for each assessment;
Final
EPAQA/G-5M 61 December 2002
-------
a description of how and to whom the results of each assessment will be reported;
a discussion of the limits of authority of assessors to recommend or direct changes
(including corrective actions) in the development or application of models; and
a description of how, when, and to whom response actions to assessment findings will
be addressed and verified as successfully completed and documented.
When relevant, the QA Project Plan addresses the following two classes of project
assessments that are typically used on a modeling project:
Internal assessments are those conducted by the organizations developing or applying
the model. These assessments are usually initiated or overseen by the internal QA
Officer [specified within QA Project Plan Element A4 (Project/Task Organization)].
Such assessments include:
reviews of the model theory, mathematical structure, parameters, and data to
ensure the objectives of the new model or application of an existing model are
being met;
- reviews of the model evaluation and hardware/software configuration testing
conducted to assure the quality requirements for a new application of an
existing model;
- reviews to assess the appropriateness of data being used or considered for use
in a new application of a model.
External assessments are those conducted by organizations that are not directly
involved with the development and application activities. External assessments are
often peer reviews. Such assessments typically address the same issues as listed above
for internal assessments, but the broader perspectives that external reviewers may
provide can lead to new insights into problems not fully recognized or discussed by
internal reviewers.
Since models produce estimates that are averaged in many ways (e.g., in time, over some
volume or space), it is to be expected that the model output will differ from observed field data.
Environmental processes are inherently stochastic, so differences (often quite large) will be seen even
when conditions are apparently identical. Evaluating the effectiveness of model development or model
application projects involves several types of assessments so that all significant sources of uncertainty
(i.e., lack of knowledge) in model output and input parameters and variability in relevant populations
Final
EPAQA/G-5M 62 December 2002
-------
and data can be evaluated. For these reasons, the assessments of model performance will involve both
qualitative and quantitative assessments.
Quantitative Assessments: The uncertainty in some sourcessuch as some model
parameters and some input datacan be estimated through quantitative assessments
involving statistical uncertainty and sensitivity analyses. In addition, comparisons can be
made for the special purpose of quantitatively describing the differences to be expected
between model estimates of current conditions and comparable field observations.
However, model predictions are not what is directly observed, so special care needs to
be taken in any exercise that attempts to make a quantitative comparisons of model
predictions with field data.
Qualitative Assessments: Some of the uncertainty in model predictions may arise
from sources whose uncertainty cannot be quantified. Examples are uncertainties about
the theory underlying the model, the manner in which that theory is mathematically
expressed to represent the environmental components, and theory being modeled. The
subjective evaluations of experts may be needed to determine appropriate values for
model parameters and inputs that cannot be directly observed or measured (e.g., air
emissions estimates). Qualitative assessments are needed for these sources of
uncertainty.
Additional discussion of quantitative and qualitative assessments is provided in various
publications such as Standard Guide for Statistical Evaluation of Atmospheric Dispersion Model
Performance (ASTM D 6589-00), A Guide for Uncertainty Analysis in Dose and Risk
Assessments Related to Environmental Contamination (NCRP, 1996), Uncertainty: A Guide to
Dealing with Uncertainty in Quantitative Risk and Policy Analysis (Morgan & Henri on, 1990),
and Uncertainty Analysis, (Kirchner, 2000), and in recent EPA documents available on the web,
Report of the Workshop on Selecting Input Distributions for Probabilistic Assessments (EPA,
1999d at www.epa.gov/NCEA/input.htm) and Risk Assessment Guidance for Superfund: Volume 3
(Part A): Process for Conducting Probabilistic Risk Assessments (EPA, 1999e at
www.epa.gov/superfund/pubs.htm).
The following two categories of assessment activities are typically addressed when preparing
the QA Project Plan for a model development or application project:
Surveillance Activities: Surveillance consists of the continual or frequent monitoring of
the status and progress of the modeling project and the analysis of records to ensure
that management and technical aspects are being properly implemented according to
the specified schedule and quality requirements developed during systematic planning.
Surveillance activities are conducted to uncover any problems that may adversely
Final
EPAQA/G-5M 63 December 2002
-------
impact the ability to achieve project goals. Surveillance activities include assessing how
project milestones are achieved and documented, corrective actions, budgets,
communications among project staff, peer reviews, data management and timely
acquisition of data, computers and software, and prompt attention to problems that
arise.
Model Performance Evaluations: Model performance evaluations are conducted to
answer the question, "Does the model perform the required task while meeting the
quality objectives?" Examples of such evaluations include the following:
- model theory assessments, which assess whether the theory underlying the
model is appropriate and capable of explaining the processes critical to the
regulatory task or scientific hypotheses being addressed by a new model or a
new application of an existing model;
- model algorithm assessments, which ensure that the mathematical algorithms
(equations) used to obtain quantitative predictions provide an appropriate
translation of the accepted underlying theory into equation form;
- data acquisition assessments, which assess the quality of non-direct
measurements (e.g., computer databases, programs, literature files, and
historical databases) using criteria specified in the QA Project Plan [Element B9
(Non-direct Measurements)];
- model calibration studies, which quantitatively assess the uncertainty and
suitability of data used to develop values of model parameters and the amount
of discrepancy between model predictions and appropriate, measured field or
laboratory data [Element B7 (Calibration)];
- sensitivity analyses, which identify key model inputs (assumptions, expert
opinion, data) and model parameters whose uncertainties account for a large
proportion of the uncertainty of model predictions;
- uncertainty analyses, which assess the range of predicted values from the
model that could occur because of the uncertainty of model components,
assumptions and input data, or the uncertainty over space and time of the value
of some model parameters (if necessary to meet risk assessment guidelines);
- hardware and software configuration testing for programming errors (e.g.,
the coded model equations or modules are correctly inserted, linked, and used
Final
EPAQA/G-5M 64 December 2002
-------
in the software code) and other problems with the development and
implementation of hardware and software needed for obtaining model
predictions (Element BlOb);
- data quality assessments, which use statistical design and analysis tools to
assess whether (1) the model input data have the precision and accuracy
needed and satisfy all assumptions (e.g., lognormality) needed for use in the
model, and (2) the variability and/or uncertainty of initial model outputs achieve
the specified Data Quality Objectives (decision performance criteria) and
model performance criteria specified in Element A7 (Quality Objectives and
Criteria for Model Inputs/Outputs of the QA Project Plan;
- model evaluations (or verification tests), which compare model results with
laboratory tests, field data, analytical solutions, synthetic test data sets, or other
well-accepted models ("benchmarking") to provide documented information
about the model's strengths and weaknesses and permit programmatic
evaluations to identify areas needing corrective action.
- internal and external peer reviews, in which individual experts or a group of
experts not previously involved in the project review all aspects of the model
development, code preparation and use, model application, and results.
Note: This is a list of examples of different types of model performance evaluations that may be
performed on a modeling project and is not meant to imply that all would be required on each project.
For data quality assessments, Guidance for Data Quality Assessment: Practical Methods
for Data Analysis (QA/G-9) (EPA, 2000c) provides nonmandatory statistical guidance for planning,
implementing, and evaluating retrospective assessments of the quality of results from environmental data
collection operations, including methods to characterize and describe data sets, test for outliers, test
distribution assumptions, test for differences between two data sets, and test for deviation from a
population standard. Other resources on data quality assessment include Goodness of Fit Techniques
(D'Agostino& Stephens, 1986), Risk Assessment Guidance for Superfund: Volume 3 (Part A):
Process for Conducting Probabilistic Risk Assessment (EPA, 1999e), and Statistical Methods for
Environmental Pollution Monitoring (Gilbert, 1987), which provide statistical goodness-of-fit tests
of data to hypothesized distribution forms.
If EPA contractual or regulatory requirements specify that peer reviews of certain aspects of
the modeling process are necessary on the project (e.g., on the theoretical basis for the model, the
mathematical model structure, model algorithms and methods of solution of the model code, model
predictions and alternative interpretations, model calibration data, uncertainty and sensitivity analysis
Final
EPAQA/G-5M 65 December 2002
-------
methods, data quality assessment methods, conclusions), the QA Project Plan should specify criteria
that the peer review will meet [e.g., EPA's peer review policy (EPA 2000e) and guidance given in
Guidance for Conducting External Peer Review of Environmental Regulatory Models (EPA,
1994)]. When peer review is to be performed on the project, the QA Project Plan implementation file
should include the names, titles, and positions of the peer reviewers, their report findings, the project
management's documented responses to their findings (or reference to where responses to peer review
comments are located).
Among the model performance evaluations discussed above, hardware and software
configuration testing is conducted to ensure that:
programming errors are minimized;
the coded model equations or modules are correctly inserted, linked, and used in the
software code;
the code is linked to or performs analyses that assess the uncertainty and variability of
predicted results (if necessary to meet risk assessment guidelines); and
the total completed modeling software package meets all user requirements.
While QA Project Plan Element BlOb (Hardware/Software Configuration) introduces the needs and
goals of the hardware/software configuration and presents the configuration's QA program, this element
of the QA Project Plan should address the tests performed to assess the configuration's adherence to
the requirements specified in the QA program. Examples of such tests include the following:
Software code development inspections: Software requirements, software design, or
code are examined by an independent person or groups other than the author(s) to
detect faults, programming errors, violations of development standards, or other
problems. All errors found are recorded at the time of inspection, with later verification
that all errors found have been successfully corrected.
Software code performance testing: Software used to compute model predictions is
tested to assess its performance relative to specific response times, computer
processing usage, run time, convergence to solutions, stability of the solution algorithms,
the absence of terminal failures, and other quantitative aspects of computer operation.
Tests for individual model module: Checks ensure that the computer code for each
module is computing module outputs accurately and within any specific time constraints.
(Modules are different segments or portions of the model linked together to obtain the
final model prediction.)
Final
EPAQA/G-5M 66 December 2002
-------
Model framework testing: The full model framework is tested as the ultimate level of
integration testing to verify that all project-specific requirements have been implemented
as intended.
Integration tests: The computational and transfer interfaces between modules need to
allow an accurate transfer of information from one module to the next, and ensure that
uncertainties in one module are not lost or changed when that information is transferred
to the next module. These tests detect unanticipated interactions between modules and
track down cause(s) of those interactions. (Integration tests should be designed and
applied in a hierarchical way by increasing, as testing proceeds, the number of modules
tested and the subsystem complexity.)
Regression tests: All testing performed on the original version of the module or linked
modules is repeated to detect new "bugs" introduced by changes made in the code to
correct a model.
Stress testing (of complex models): This ensures that the maximum load (e.g., real-
time data acquisition and control systems) does not exceed limits. The stress test
should attempt to simulate the maximum input, output, and computational load expected
during peak usage. The load can be defined quantitatively using criteria such as the
frequency of inputs and outputs or the number of computations or disk accesses per
unit of time.
Acceptance testing: Certain contractually-required testing may be needed before the
new model or model application is accepted by the client. Specific procedures and the
criteria for passing the acceptance test are listed before the testing is conducted. A
stress test and a thorough evaluation of the user interface is a recommended part of the
acceptance test.
Beta testing of the pre-release hardware/software: Persons outside the project
group use the software as they would in normal operation and record any anomalies
encountered or answer questions provided in a testing protocol by the regulatory
program. The users report these observations to the regulatory program or specified
developers, who address the problems before release of the final version.
Reasonableness checks: These checks involve items like order-of-magnitude, unit,
and other checks to ensure that the numbers are in the range of what is expected.
Again, it should be noted that the actual types of tests performed depend on the level of assessment
needed for the program, and not all of these tests would be performed on each project.
Final
EPAQA/G-5M 67 December 2002
-------
Types of information that can be specified in the QA Project Plan for hardware and software
configuration tests include the following:
a list of user needs and code requirements based on the end use of the model;
a description of the types, number, and frequency of configuration tests that will be
conducted;
a structured, orderly, and documented test schedule for testing and audits;
a specification of "correct" or "acceptable" outputs for each configuration test to
compare with test results;
a list of necessary documentation for the planned configuration testing;
a discussion of potential corrective actions that could result from uncovered problems;
and
a description of the internal and external peer reviews that are planned to assess the
hardware/software configuration testing effort and results.
When executed, the assessments discussed above may reveal that the quality measures
associated with the outcomes of the model development or application project do not conform to the
quality specifications defined in the QA Project Plan. The QA Project Plan may specify certain types
of documentation or assessment procedures, such as:
occurrences of specific unsatisfactory conditions upon which the assessors are
authorized to act;
specific response actions needed;
protocol for managing, conducting, evaluating, reporting, and documenting the response
actions and rationale for such actions;
person(s) responsible for seeing that response actions are successfully completed [see
Element C2 (Reports to Management)];
person(s) responsible for overseeing response actions and person(s) to whom the
results of the response action will be reported; and
decision-making hierarchy and the schedule and format for oral and written reports.
Final
EPAQA/G-5M 68 December 2002
-------
4.3.2 Reports to Management (C2)
Identify the frequency and distribution of reports issued to inform management (EPA or
otherwise) of the project status; results of software tests, model performance evaluations,
and other assessments such as output data quality assessments; and significant quality
assurance problems and recommended solutions. Identify the preparer and the recipients
of the reports, and any specific action recipients are expected to take as a result of the
reports.
Reports to management are a critical part of the communication process among all participants
in a model development or application project. Planned reports provide a structure for notifying
management of the following:
adherence to project schedule and budget;
deviations from approved QA Project Plans, as determined from project assessment
and oversight activities (discussed in the previous subsection);
the impact of these deviations on model prediction, application quality, and uncertainty;
the need for and results of response actions to correct the deviations;
potential uncertainties in decisions based on model predictions and data; and
Data Quality Assessment findings regarding model input data and model outputs
(predictions).
Reports to management should provide an understanding of the potential effect that changes made in
one segment of the model input data, the algorithms, or the development and application process may
have on segments of the model algorithms, process, or predictions.
Certain management reports and deliverables mandated by contractual or project requirements
should be specified in the QA Project Plan, along with the necessary routing procedures, due dates,
and distribution lists [if they differ from the distribution list specified in Element A3 (Distribution List)].
Other reports that are less formal in nature (e.g., a report on the outcome of the model selection
process) can be handled less formally in the QA Project Plan. The QA Project Plan specifies the
personnel responsible for preparing reports, evaluating their impact, and implementing response actions.
It also specifies the planned frequency, content, and distribution of relevant reports to ensure that
management may be alerted to data quality problems, necessary changes to the data, viable response
actions, and proposed solutions. The frequency of reports should allow management to properly
anticipate certain events and take corrective actions (e.g., propose viable solutions, procure additional
resources).
Final
EPAQA/G-5M 69 December 2002
-------
Examples of certain types of reports to management that may be relevant to mention for a
modeling project (depending on the scope of the project) include the following:
final version of the QA Proj ect Plan,
monthly progress reports detailing the status of each phase of the project from both a
technical and cost standpoint,
needs assessment report,
modeling framework report,
model documentation,
model calibration report,
model evaluation report,
model outputs uncertainty report,
disposition of peer review comments (for the various stages of peer review on the
project),
summary report of the model application, and
final report of the model application.
4.4 GROUP D: DATA VALIDATION AND USABILITY
The primary purpose of this group of elements is to describe the process to assess the usability
of the model results (whether from the first application of a new or revised model or from application of
an accepted model). Therefore, these elements refer to quality procedures that occur near or at the
end of the modeling project. The QA Project Plan contains the criteria and methods for deciding the
degree to which each component of the modeling process has met its quality specifications as described
in the Group B elements as well as the requirements for the final output [Elements A5 (Problem
Definition/Background) and A7( Quality Objectives and Criteria for Model Inputs/Outputs)].
As mentioned at the beginning of Section 4.3, both Groups C and D of the QA Project Plan
address assessment-type activities regarding the model output (i.e., predictions) generated in a given
application. Therefore, information in these elements may overlap. Generally, Group D elements focus
on final assessments performed in the final stages of model development for a first application or in
subsequent applications to evaluate whether the outcomes meet the project's original objectives.
Group D of the QA Project Plan describes data review, verification, and validation processes
(EPA, 2001). For modeling projects, this is analogous to confirming that the steps of the modeling
process were followed correctly to produce the model outputs and that the results meet project
objectives.
Data (or information) validation and usability activities for modeling projects are represented
within the following three QA Project Plan elements:
Final
EPAQA/G-5M 70 December 2002
-------
Departures from Validation Criteria (Dl): This first element documents the criteria
used to evaluate how deviating from the specifications given in the QA Project Plan
may impact the quality and usability of final results and decisions that are made based
on these results.
Validation Methods (D2): This second element describes the process and methods
for determining whether deviations have occurred within the model components.
Reconciliation with User Requirements (D3): This element combines the information
from the previous two elements to make a final assessment of the usability of the model
results.
Each element is addressed in the following subsections.
4.4.1 Departures from Validation Criteria (Dl)
State the criteria used to review and validatethat is, accept, reject, or qualifythe model
results, in an objective and consistent manner.
Along with Element D2 (Validation Methods), this element of the QA Project Plan elaborates
on the acceptance criteria mentioned in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs), which evaluate the model and its components based on its ability to produce results
that can be used to achieve project objectives. For example, this element would state acceptance
criteria associated with the degree to which each model output item has met its quality specifications.
The possible types of discrepancies that may arise when the acceptance criteria and other QA Project
Plan specifications are not met in their entirety are also addressed, along with the effects that such
discrepancies are likely to have on the outcome of the model development and application processes.
Examples of such acceptance criteria and details about how such criteria may be evaluated in
the various stages of the modeling process are as follows, presented in the context of specific model
applications:
Mathematical basis for the model: Criteria to be used to determine how well the
mathematical theory behind the model represents the processes being modeled are
addressed, in addition to how the criteria were established. In all cases, the possible
ways in which the criteria may not be met are specified, and the effects these conditions
may have on the model output are discussed.
Final
EPAQA/G-5M 71 December 2002
-------
Numerical models: The criteria confirming that the numerical (coded) model
accurately estimates the mathematical theory behind the model are discussed, along
with how the modeling outcome is affected if these criteria are not met. The criteria for
assessing the numerical model (e.g., convergence criteria) and reasons for their
appropriateness are discussed. Also, the expected level of uncertainty from model
structure and parameterization and its effects on the model output are addressed.
Code verification: The final set of computer code is scrutinized to assure that the
equations are programmed correctly and that sources of error, such as rounding, are
minimal. This process is likely to be more extensive for new computer code. For
existing code, the criteria used for previous verification, if known, can be described or
cited. Any additional criteria specific to the modeling project can be specified, along
with how the criteria were established. Possible departures from the criteria are
discussed, along with how the departures can affect the modeling process.
Model evaluations: A model can be evaluated by comparing model predictions of
current conditions with similar field or laboratory data not used in the model calibration
process, or with comparable predictions from accepted models or by other methods
(uncertainty and sensitivity analyses).
Validation of input data: For a first application of the model, where parameter values
are specified and site-specific data are input into the model or subsequent applications,
the input data may need to be validated for their requirements planned in Element A7
(Quality Objectives and Criteria for Model Inputs/Outputs). In addition how the
criteria were established and the possible ways in which the criteria may not be met are
specified, and the effects these conditions may have on the model output are discussed.
Model output: The criteria used to assess the usability of the model output include its
regulatory task requirements, as specified in Element A7 (Quality Objectives and
Criteria for Model Inputs/Outputs) of the QA Project Plan. For model applications in
production mode, model outputs are similarly assessed against program uncertainty and
variability requirements. Comments on the process of choosing these criteria and
objectives should refer to Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs). If the model output fails to meet these criteria, the resulting course of
action is detailed.
Note that many of the assessment approaches used to evaluate these acceptance criteria may have
already been provided in Element Cl (Assessment and Response Actions), and references back to this
section are acceptable.
Final
EPAQA/G-5M 72 December 2002
-------
4.4.2 Validation Methods (D2)
Describe the process to be used for validating the model results, including (if appropriate)
procedures that maintain the integrity of data throughout the life of the project or task.
Discuss how issues shall be resolved and the authorities for resolving such issues.
Describe how the results are conveyed to users. Provide examples of any forms or
checklists to be used. Identify any project-specific calculations required.
The purpose of this element is to describe, in detail, the process for making a final assessment
of whether model components and their outputs satisfy the user requirements specified throughout the
QA Project Plan. The appropriate methods of evaluation are determined by the quality objectives
developed for the project and detailed in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs). The individuals responsible for the evaluation of the various components of the model
together with the lines of authority should be shown on the organizational chart presented in Element A4
(Project/Task Organization). The chart should indicate who is responsible for each activity within the
process.
The review process should be described for each component of the model and its results
discussed in Element Dl (Departures from Validation Criteria). Discussion should address the
difference between processes for newly-developed versus established models and between existing
versus new software code. The review procedure for all model output results should be documented in
the SOPs (e.g., graphical displays or videos).
Examples of issues to be addressed include the following:
Mathematical basis for the models: The process should be described that will be
used to assure that the mathematical basis for the model accurately reflects the process
it represents. For example, the mathematical model can be assessed through peer
review. The method of model evaluation, including responsibilities and schedules for
peer review or other methods of assessment, should be described.
Numerical models: For new or existing models, the process of measuring
characteristics of the numerical model against the set criteria should be addressed,
along with the procedure for confirming that the numerical model is accurately
estimating the mathematical model.
Code verification: The process of specifying how and when code will be checked to
ensure that the program correctly solves the equations representing the mathematical
model should be discussed. The new software code is examined and existing code is
Final
EPAQA/G-5M 73 December 2002
-------
evaluated with respect to the current modeling process. If possible, the verification
status of existing code should be addressed.
Model evaluation: The process of specifying how and when model output will be
compared with independent data to ensure that the modeling results meet project
objectives.
Validation of input data: The process of specifying how and when parameter values
and site-specific data that are input into the model will be validated for their
requirements planned in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs) [orB9 (Non-direct Measurements)].
Model output: The usability of the model output is assessed by comparing it against its
regulatory task requirements and other criteria specified in QA Project Plan Element
A7 (Quality Objectives and Criteria for Model Inputs/Outputs). The procedure that
will be used to assess the model output should be documented.
4.4.3 Reconciliation with User Requirements (D3)
Describe how the results obtained from the project or task will be reconciled with the
requirements defined by the data user or decision maker. Outline the proposed methods to
analyze the output data and determine possible anomalies or departures from assumptions
established in the planning phase of the modeling process. Describe how reconciliation
with user requirements will be documented, how issues will be resolved, and how limitations
on the use of the output data will be reported to decision makers.
The purpose of Element D3 (Reconciliation with User Requirements) is to outline and specify, if
possible, methods for evaluating (relative to user requirements) the model outputs that the project
generates. These methods include scientific evaluations of the model predictions to determine if they
are of the right type, quantity, and quality to support their intended use.
This element discusses the procedures in place to determine whether the final set of model
results meets the requirements for the data quality assessment. This element should also discuss how
departures from the underlying assumptions or output criteria associated with statistical procedures
applied in the data quality assessment will be addressed, the possible effects of departures from
assumptions or specified output criteria on the model results, and what potential modifications will need
to be made to adjust for these departures. Finally, the discussion should specify model limitations that
may impact the usability of the results.
Final
EPAQA/G-5M 74 December 2002
-------
Data quality assessment (DQA) follows the procedures that are documented within the Group
DQA Project Plan elements. Discussed in detail within, Guidance for Data Quality Assessment:
Practical Methods for Data Analysis (QA/G-9) (EPA, 2000c), DQA determines how well the
validated output can support their intended use. In cases where formal DQOs have been established
[Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs)], DQA is a key part of the
assessment phase of the data life cycle. It focuses on evaluating the ability of information to be used for
decision making and provides graphical and statistical tools. If a probabilistic model has been
developed, then the distribution of the model outputs, which is a measure of the uncertainty of the
model predictions, should be compared with the allowable model output uncertainties specified in
Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs Data). The uncertainties of the
various inputs and assumptions used in the model should be examined if that has not been done under
Element Cl (Assessment and Response Actions). Also, sensitivity analyses may be used to gain
information on the particular model inputs that contribute the most uncertainty to model outputs. This
information can be used to provide clues to how reductions in model output uncertainties can be most
cost-effectively reduced to acceptable levels.
Risk Assessment Guidance for Superfund: Volume 3 (Part A): Process for Conducting
Probabilistic Risk Assessment (EPA, 1999e) provides uncertainty and sensitivity analysis methods.
Uncertainty Analysis in Ecological Risk Assessment (Warren-Hicks & Moore, 1998) and
Uncertainty Analysis (Kirchner, 2000) also provide case studies describing current state-of-the-
practice ecological uncertainty analyses. Statistical methods should be used to assess the uncertainty
(e.g., variability and unintended bias) and any abnormalities in the differences (residuals) between model
outputs and comparable data. Graphical and statistical methods as discussed in Applied Regression
Analysis (Draper & Smith, 1981 and in ASTM D6589) are useful for this purpose.
For modeling projects that are data poor (i.e., that have high uncertainty in model inputs and
consequently in model outputs), it is particularly important that procedures are in place to address
whether the high levels of uncertainty that are likely can allow user requirements to be met, and if not,
the actions needed to address that issue
Final
EPAQA/G-5M 75 December 2002
-------
Final
EPAQA/G-5M 76 December 2002
-------
CHAPTER 5
EXAMPLES
This chapter is designed to provide an illustration of the information that could be included in a
QA Project Plan for modeling. The previous chapters discuss the type of content that could be
included, and this chapter is meant to provide examples of this content. This chapter is important for
two reasons: (1) implementation of a new process is always more understandable with examples and,
(2) these examples will provide the reader with some insight into the implementation of the "EPA
graded approach."
In this chapter, three levels of projects are described to illustrate the varying level of detail and
emphasis that may be needed for a QA Project Plan based on the nature of the modeling project and
the intended use of the output data. The first example describes the development of an entire model
made up of many different modules. The second example describes application of a less complex
model. The third describes the planning information needed to test a single module in development as
part of building a larger model system.
In each example, the information provided under each relevant QA Project Plan element is
described to illustrate the application of the QA Project Plan elements to that example. These
examples also discuss the documentation appropriate for each project.
5.1 COMPLEX MODELS - LAKE MICHIGAN MASS BALANCE PROJECT
A QA Project Plan for this complex project describes how planning, implementation, and
assessment activities are coordinated across the extensive model framework (see
www.epa.gov/glnpo/lmmb/planning.htm). Although this would be an overview document, an adequate
level of detail is needed to clearly communicate agreed-upon development, assessment, and reporting
activities and evaluation criteria from project management to the model developers. Referral to
accompanying project documents would be used where possible to keep this QA Project Plan for the
entire framework at a reasonable size.
5.1.1 Part A: Project Management
Section A3: Distribution List. Due to the large scope of this modeling project, this section
contains an extensive list of individuals who will be involved in the review and approval of the QA
Project Plan. This list can include but is not limited to module project managers, QA Managers, data
suppliers, individual module developers, funding managers, and data users.
Final
EPAQA/G-5M 77 December 2002
-------
Section A4: Project/Task Organization. Because coordination among modeling teams and
communication of progress will be a critical part of this QA Project Plan, the management plan for the
project is described. Because this is such a long-term project, management will take a phased
approach to development in order to accommodate the increasing complexity of the system and
changes in resources over time. The stages would be clearly defined, with reviews, evaluations, and
interim decision making clearly identified and described. The planned stages for this project include
using historical data to refine modeling theory and sampling designs, working with data suppliers on
data quality assessment, developing and testing the final models, and documenting the models and
quality assurance results.
Section A5: Problem Definition/Background. The purpose of this model is to develop a
tool to improve the understanding of the sources, transport, fate, and bioaccumulation of toxic
chemicals in Lake Michigan. The information needs driving the model development are twofold: (1) to
determine sources of chemicals in tributary, atmosphere, and sediment categories and, (2) to predict
future in-lake chemical concentrations [e.g., polychlorinated biphenyls (PCBs) in lake trout] under
alternative management scenarios. The target fish species are lake trout and coho salmon.
Section A6: Project/Task Description and Schedule. It is important to have an overall
schedule so that individual module developers take into account testing and evaluating of other modules.
This section lists all the modules being developed, the schedule for testing and evaluation, and the
schedule for testing interfaces between/among modules.
The modeling
framework is shown in
Figure 11. The relation-
ships between the
modules are clearly
specified in the planning
documents, including
input, output, any
processing needed, and
the measurement unit to
be used at each
transition. The method
for moving the data from
one module to another is
specified under Element
BIO (Data Management
and Hardware/Software
Configuration).
Lake Michigan Modeling Framework
Computational
Transport
Mass
Balance
Bioaccumulation
Hydrodynam
Model
(POM/3D)
c
; meteorology
£S_ -
\
G.L.
Wave
Model
-
Advective/
Disoersive
Transport
c
^ 1
sp
-
Collapsed Grid
Transport Model
Eutrophication
Model
Solids/Sorbent
Dynamics
**
/
I Transport & r
^_ Fate Model J
f trophic ^
V^dynamics/
"f^ constituent ^\
V^lpadingjB,^/
1
Food Web
Bioaccumulation
Model
phytoplankton
benthos I
rhpmiral |
exposure & availability
Figure 11. Modeling Framework for Lake Michigan Mass Balance
Project
EPA QA/G-5M
78
Final
December 2002
-------
The model framework QA Project Plan would include QA information on the individual
modules within the framework. Each model team would develop its own individual quality assurance
plan to be included in the QA Project Plan for the framework. These QA plans would include:
background and main features of the model;
source(s) and quality of the data (e.g., used in calibration or testing of the model);
plans for model software code development and maintenance;
information on model parameters (source of the data, other values considered and
reasons for rejection, information supporting the use of these data for this model);
plans for model documentation (e.g., equations, underlying assumptions, boundary
conditions, and numerical methods);
plans for code verification;
plans for code documentation (e.g., code, internal documentation, documentation of
adaptations for the Great Lakes, user's guide);
plans for model calibration, evaluation, and uncertainty;
references, especially to any earlier documentation of the model if it was previously
developed.
Section A7: Quality Objectives and Criteria for Model Inputs/Outputs. This section will
describe several different levels of quality objectives. The Technical Coordinating Committees for this
modeling project developed modeling quality objectives (EPA, 1999b) that were translated into quality
objectives for the individual modules to be used to evaluate the model. The data quality objectives for
the entire framework specified that the model should be capable of calculating pollutant concentrations
in Lake Michigan to within a factor of 2 of observed concentrations in the water column and target fish
species. Furthermore, study managers agreed to accept an uncertainty level for each input to the model
within 20-30% of the mean at the 95% confidence level. The modeling objective was to simulate the
average water quality within ± 2 standard errors of the observed mean (by cruise/segment average) or
to provide 95% confidence that the actual mean falls within this specified range.
Section A9: Documentation and Records. This section includes the plan for long-term
documentation, which is especially important in a multi-year and multi-stage project like this one. The
type of information that will be part of the record is itemized so that modelers are informed beforehand
Final
EPAQA/G-5M 79 December 2002
-------
of the records they should keep. All records, including modeler's notebooks and electronic files, would
be maintained by the laboratory QA officer in a central project archive upon completion of the project.
These records would include documentation of model testing, calibration, and evaluation.
Records would be kept of written rationale for selection of models, record of code verification (hand-
calc checks, comparison to other models), source of historical data, source of new theory, and
documentation of adjustments to parameter values due to calibration. These records will be maintained
by each modeler and transferred to a central project archive. Agency standards and ANSI/IEEE
standards for software QA plans need to be followed.
5.1.2 PartB: Data Acquisition and Management
Section B7: Calibration. In this section, outputs of interest are clearly identified, with special
attention to issues that will need to be coordinated for consistency between different modules. The
analysis procedures and acceptance criteria would be specified in the QA Project Plan for each module
(this might be done under the QA Project Plan for each module under Section A6). If field or other
data will be used for calibration, the QA Project Plan would describe why those data were chosen,
how those data will be acquired, and how the quality of those data will be assessed.
Adjustments due to calibration need to agree with scientific knowledge and with the process
rates within a reasonable range of values found in literature. A list of internal variables used to calibrate
the model outputs would be included in this section.
For this model, simulations will be used to attempt to reproduce the statistical distribution
properties of the data. Parameter optimization will be used to minimize residuals. Model simulation
results will be compared to measurements obtained from the project data collection program.
Section B9: Non-direct Measurements (Data Acquisition Requirements). For this
project, the preference is to use published data that have been through a peer review process overseen
by an editorial board. Unpublished data can be used, and will be reviewed using data quality
assessment (DQA) tools. Instructions for performing DQA are presented in this section. This section
also includes a table listing major data needs and potential and confirmed sources for these data. A list
of the available QA information is included along with identification of the QA information that will be
sought for data sources that are utilized.
For this QA Project Plan, historical data from a variety of sources and data collected for this
modeling project need to be assessed to determine if they meet the objectives of the modeling project.
Information on acquired data would be obtained from the relevant data collection QA Project Plans or
related materials. This information would be used to evaluate indicators such as the comparability of
the data for aggregation and time analyses, its representativeness of lake conditions, bias, and precision.
Final
EPAQA/G-5M 80 December 2002
-------
Assessment of the data by both the Great Lakes National Program Office and the modelers would be
preferred.
Section BIO: Data Management and Hardware/Software Configuration. For this
model, the Revision Control System will be used to coordinate version input and output files. The
compatibility of the input database formats and modeling framework will be worked out along with
hardware/software requirements. Hardware/software configuration would be described in some depth
in accordance with document planning requirements for developers of the individual models, and pre-
and post-processors.
Test planning and results reporting need to be specified and implemented to assure
compatibility of parts of the model framework and final usability. Plans would be needed for
configuration management of revised code throughout the phased development.
5.1.3 Part C: Assessments and Response Actions
Uncertainty will be estimated in a two-step process. The parameter variance-covariance matrix
resulting from calibration will be estimated and applied to generate exceedence levels for model
predictions using Monte Carlo methods. The goal of this process is 95% exceedence limits within a
factor of 2 of the predicted toxic chemical concentrations in water and top predator fish over the
duration of the calibration period (1994-1995).
The ability of the code to correctly represent model theory will be assessed. In addition,
specific tests are planned to ensure operations are verified. Continuity checks, mass conservation
checks, and testing of numerical stability and convergence properties will be conducted.
5.1.4 Part D: Output Assessment and Model Usability
Plans for testing need to be worked into the phases of the project. Model simulations will be
planned to reproduce the statistical distribution properties of the field and laboratory data. Evaluation
will be done by comparing cumulative frequency distribution plots of data to frequency distributions
plots from comparable model predictions. This quantitative evaluation will be integrated with qualitative
assessment. A science review panel independent of the model development process will provide
reviews on a regular basis.
As the individual modules evolve through different scales and levels of uncertainty with
development (e.g., screening models, a medium-resolution segment scheme, and a high level resolution
phase), assessments will be planned to determine if they are acceptable to pass on to the next phase.
Ultimately, the whole framework needs to be assessed, and these individual assessment results will
provide background for planning such an assessment.
Final
EPAQA/G-5M 81 December 2002
-------
5.2 INTERMEDIATE MODELS - PORT RIVER NITROGEN AND PHOSPHATE
TOTAL MAXIMUM DAILY LOAD (TMDL) PROJECT
In this example, a state's Department of the Environment (DOE) wishes to estimate a Total
Maximum Daily Load (TMDL) for selected nutrients (nitrogen and phosphorus) that have impaired the
ability of a given waterway to achieve water quality standards, and estimate how the TMDL is allocated
among different sources of these nutrients. A TMDL is an estimate of the maximum amount of a
pollutant that a waterway can receive and still remain within water quality standards established by the
Section 303 of the Clean Water Act (CWA). The state DOE plans to use EPA's existing Water
Quality Analysis Simulation Program (WASPS), and in particular, its EUTRO5 component on water
quality/eutrophication, as a modeling tool for obtaining the necessary information for establishing the
TMDL estimates. For this project, a QA Project Plan is prepared to describe how planning,
implementation, and assessment activities would be accomplished to achieve defensible outputs from
this modeling process. The following discussion indicates the type of information that would be
included in the QA Project Plan.
5.2.1 Part A: Project Management
Section A4: Project/Task Organization. Because this is largely the effort of a single state's
DOE, this section of the QA Project Plan would focus on roles and responsibilities of internal
departments and managers who will contribute to the modeling effort. For example, each department
that is to provide necessary model input data would be specified here, along with appropriate contact
people. Departments that will contribute to TMDL development but may not be directly involved in the
modeling effort may also be specified here. Individuals will be noted according to their responsibilities
in overseeing various stages of the project. Examples of these project stages include the following:
refining the problem using historical data
working with data suppliers on data quality assessment
reviewing the model for this specific application
calibrating and testing the selected model
performing model fits and identifying appropriate model inputs for each fit
conducting sensitivity analyses
documenting the model application and quality assurance results.
Section A5: Problem Definition/Background. The Port River, a 8.5 mile-long tributary of
the Large River, was included on the CWA Section 303(d) list of impaired waters due to demonstrated
eutrophication and low dissolved oxygen levels that violated water quality standards. Elevated nitrogen
and phosphorus levels in the river have contributed to excessive algae blooms and wide diurnal
fluctuations in dissolved oxygen content which have the potential to cause fish kills. Estimated TMDLs
for total nitrogen and total phosphorus are needed because technology-based controls and other types
Final
EPAQA/G-5M 82 December 2002
-------
of controls did not allow water quality standards to be attained for water contact recreation, aquatic life
protection, and shellfish harvesting. The goals are to keep the concentration of chlorophyll-a (a
surrogate indicator of algae blooms) to below 52|ig/L and dissolved oxygen to no lower than 5mg/L.
Thus, the purpose of this model application is to provide predictions (or projections) on total nitrogen
and total phosphorus to be used in developing TMDLs that will be protective of the river's designated
uses. Information would be generated for both low (base) flow and annual average conditions.
Section A6: Project/Task Description and Schedule. It is important to establish a
schedule that will allow time requirements on approving TMDLs to be met with a defensible product.
This section of the QA Project Plan lists all critical activities for planning criteria; acquiring and assessing
existing data; model evaluation for selection, calibration, and application; and reporting. A process for
tracking this schedule to complete these activities is also needed to assure that deadlines in the overall
TMDL process are met.
Section A7: Quality Objectives and Criteria for Model Inputs/Outputs. This section will
describe the several different levels of quality objectives for the various stages of this modeling project,
as determined from a systematic planning process. The following types of quality objectives are among
those relevant for this effort:
Acceptance criteria for intended use of existing data: Existing data will be used as input
to the modeling effort. This data will be judged acceptable for their intended use if they
meet acceptance criteria that are determined from a systematic planning process and
documented here. The sources of existing data to be used in this effort and their
intended uses are documented in Element B9 (Non-direct Measurements). Examples
of acceptance criteria and the types of data issues addressed by these criteria are as
follows:
- Data reasonableness: Loading data from surveys conducted in 1984 and 1985
will be used as input to the models when predicting baseline conditions. The
reasonableness of these dates needs to be documented here. For example,
these dates occur prior to the implementation of non-point source nutrient
reduction controls placed on this waterway.
Data completeness: Data from at least four surveys need to be available for all
available sampling stations within the tidal portion of the Port River from its
confluence with the Large River to the upper free flowing areas of an unnamed
tributary located above the Orange Sewage Treatment Plant. This criterion
reflects the agreed-upon spatial domain of the model. Overall, the data need to
achieve a 95% minimum criterion on completeness.
Final
EPAQA/G-5M 83 December 2002
-------
Data representativeness: Certain criteria on data representativeness will need
to be achieved, such as time and location for sample collection and
comparability in sampling and sample analysis methods for input data from the
DOE's water quality monitoring and point source databases (e.g., point source
loading data from the most recent year for the four point sources). Other
methods for assessing data representativeness relative to current conditions
(e.g., photograph verification, visual assessment, considering data from
bathymetric and ground truth land use surveys) can be detailed in Element Cl
(Assessment and Response Actions).
Acceptability of model calibration and testing inputs and outputs: Element B7
(Calibration) discusses the approach to be taken to determine whether the EUTRO5
model is properly calibrated. Project managers are defining model calibration in this
setting as how well the model is able to reproduce current observed flow rates and
concentrations of nutrient, dissolved oxygen, and chlorophyll-a (e.g., trends and peak
values), as measured from multiple field surveys and stored in a state-maintained
monitoring database. Because water quality data for multiple surveys exist for each
sampling station in the waterway, each station has multiple measurements for the water
quality parameters that are used as input to the models. Thus, the calibration procedure
is able to divide the total variability of the model predictions into two sources: within-
station variability in the input measurements, and variability and uncertainty associated
with how well the model fits the data (i.e., lack-of-fit). The following criteria, as
determined from stakeholders consulting with study statisticians, have been established
for acceptable model calibration inputs and outputs, respectively:
At least 90% of the input nutrient concentration measurements at a given
sampling station should be within two standard errors of the mean measurement
at that station.
The percentage of total variability and uncertainty that is attributable to lack-of-
fit of the model should not exceed 25% in any of the calibration model fits.
Other goodness-of-fit evaluations may also be considered (EPA, 2000c; Gilbert, 1987)
when determining corrective action associated with these criteria. For example, data
may need to be transformed (e.g., logarithmic) to better achieve these criteria and other
model assumptions, or further investigations into specific data values may be necessary.
Such corrective actions are addressed in Element Cl (Assessment and Response
Actions).
Final
EPAQA/G-5M 84 December 2002
-------
Acceptability of model outputs: Any limitations on uncertainty associated with model
outputs are specified here. For example, managers within the state's Wastewater
Management Program may specify a goal that the uncertainty associated with the
predicted value at a given location will need to be within 25% of the predicted value
with 95% confidence.
Acceptability of flow rate estimates: As discussed in Element B9 (Non-direct
Measurements), multiple regression techniques will be used to predict low and average
flows within various segments of the watershed as a function of land use data. The
regression equations will need to achieve certain criteria on their goodness-of-fit (e.g.,
R2 of at least 0.90).
The above types of data quality criteria tend to be quantitative in nature. However, certain
stages of the model application process may benefit from assessments that are more general and
qualitative. For example, when evaluating the outcome of model calibration, qualitative assessments
may be done by evaluating how well the outputs of the fitted model are able to match the overall trend
in prediction over time or over the entire watershed area. This evaluation would be documented in a
final report by using graphs the axes of these graphs and the data to be used should be documented
in the QA Project Plan. An evaluation of how well these model outputs reflect peaks and valleys in the
predicted water quality values at specified time points or at certain points in the watershed can also be
compared to what has been observed and collected in surveys. Again, these can be documented in the
final report by graphs that should be described in this element.
Section A9: Documentation and Records. This section includes the plan for documentation
which will be important for defending the model predictions used in supporting the TMDLs. The type
of information that will be part of the record would be itemized here so that modelers are aware before
the project begins of the records that should be kept. Examples of appropriate documentation in this
study include: calibration and sensitivity analyses results, records of written rationale for selection of
models or modules, record of code verification (e.g., hand-calculated checks, comparison to other
models), sources of existing data used, and any adjustments to model parameter values that result from
model calibration. All records, including modeler's notebooks and electronic files, should be
maintained by the project manager in a central project archive upon completion of the project.
5.2.2 Part B: Data Acquisition and Management
Section B7: Calibration. Water quality data on such measures as chlorophyll-a,
phosphorus, nitrate, and dissolved oxygen that were collected from earlier surveys on the waterway, in
addition to flow measurements obtained from the United States Geological Survey, will be used in
calibrating the EUTRO5 model to this particular application. Element B9 (Non-direct Measurements)
will document the sources of these data, and Element A7 (Quality Objectives and Criteria for Model
Final
EPAQA/G-5M 85 December 2002
-------
Inputs/Outputs) will specify the criteria for which these data will be judged acceptable for use in model
calibration.
The calibration will judge the extent to which the model is able to predict current water quality
measures that agree with what was actually observed in the surveys. For instance, the extent to which
the model accurately captures observed trends in the water quality data at the various sampling points in
the river and its tributaries, after taking into account the underlying variability in these monitored data,
will be determined and appropriately documented. The discussion should note that in this particular
application, model testing is also occurring during the model calibration process, as the inputs to the
model calibration and the model's corresponding outputs represent a given water quality scenario that
will be used as part of the TMDL development process. The performance criteria upon which the
calibration will be deemed acceptable will be noted in Element A7 (Quality Objectives and Criteria for
Model Inputs/Outputs).
Within the model calibration exercise, model rate coefficients will be adjusted as necessary to
meet the calibration criteria and to reflect current scientific knowledge and various process rates that fall
within a reasonable range of values found in the scientific literature. A list of internal variables used to
calibrate the model outputs should be included in this section, along with any adjustments made to the
model. The rationale for any needed model adjustments based on the results of the calibration process
would be documented according to the procedures specified in Element A9 (Documentation and
Records).
Section B9: Non-direct Measurements (Data Acquisition Requirements). As indicated
previously, different types of data already existing within the state's databases will be used for model
calibration and as input to the model. This section of the QA Project Plan discusses the different
sources of these data, the intended uses of these data on the project, the specific criteria for accepting
an existing data source for use on the project, and any limitations that the use of these data may have on
this project.
The primary types of existing data to be used in the modeling effort are nutrient point source
loads and non-point source loads for the watershed. Loads for the contributing point sources are taken
from discharge monitoring reports that exist in the state's point source database. When average flow
conditions are assumed, non-point source loadings are calculated as a sum (across the different types of
land uses) of land use areas multiplied by their respective land use loading coefficients. Land use areas
originate from the State's Office of Planning. Land use loading coefficients are previously-published
outputs that resulted from fitting a continuous simulation model representing the bay in which the river
deposits. Non-point sources represent such contributors as atmospheric deposition, as well as loads
from septic tanks, urban development, agriculture, and forest land.
Final
EPAQA/G-5M 86 December 2002
-------
The modeling effort will consider four different scenarios that represent either baseline or future
conditions and low versus average flow characteristics. These scenarios simulate seasonality effects.
Therefore, different sets of existing data (i.e., point source loads and non-point source loads) will be
needed as model inputs for each scenario. These data sources should be listed here for each scenario,
along with any limitations that these data may have in terms of predicting the scenarios.
For this project, the water quality and physical field data were collected by the state's field
operations program in earlier surveys according to documented field and laboratory protocols and at
documented monitoring stations. In order to ensure this data is appropriate for use in this model,
survey records will be checked to assure conformance with procedures established for their initial
collection and to assure that the resulting data meet the requirements established in Element A7 (Quality
Objectives and Criteria for Model Inputs/Outputs). Data will be reviewed to be sure that their values
fall within previously-observed and reasonable ranges (e.g., base flow nutrients and groundwater).
As part of the modeling procedure, the watershed will be subdivided into subwatersheds, and
low and average flows will be determined at each subwatershed using multiple regression procedures
that predict flow as a function of land use data (Draper & Smith, 1981). The regression equation
predicting low flow was determined from flow data collected at two sampling stations. Because no
average flow data were available for the basin, the regression equation for average flow was
determined using long-term average daily flow data originating from survey gauges and relating these
data to land use acres within the basins. The equations will need to achieve acceptance criteria
specified in Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs). The source and
relevance of these regression equations will be documented in this element of the QA Project Plan.
In determining whether existing data sources meet acceptance criteria for the project [as
defined in Element A7 (Quality Objectives and Criteria for Model Inputs/Outputs)], information on
acquired field data to be used for assessing acceptance criteria would be obtained from the QA Project
Plans of the operations collecting these data or from related materials from the state's field operations
branch. Any limitations on the existing data that may impact a model's predictive ability for this project
should be discussed here. For example, water quality surveys for which data were used in model
calibration were collected over 15 years ago, and therefore, any changes to the waterway and its
environment since then needs to be taken into consideration.
Section BIO: Data Management and Hardware/Software Configuration. EPA's model
WASP 5 and its sub-model EUTRO5 are being used to estimate the fate and transport of nutrients
because they are capable of studying time-variable or steady state (up to three dimensional), linear or
non-linear, kinetic water quality problems. This element of the QA Project Plan discusses the resource
needs for the modeling system. A process flow diagram would be included to illustrate how various
types of data are input to the system and the different types of model outputs. Details would be
provided on how the model input and output data will be manipulated, managed, and maintained as part
Final
EPAQA/G-5M 87 December 2002
-------
of this project (e.g., all data to be input to the model need to be stored in metric units), along with how
the overall integrity of the data will be ensured through the various stages of the project. For example,
information on how the water quality survey data used in model calibration are averaged across the
different surveys would be discussed here.
The future form of the model needs to be compatible with the input database formats and with
the hardware/software capabilities of the department (e.g., Department PC-compatible, pre- and post-
processors available from distributor). Thus, these types of resource requirements are mentioned in this
element of the QA Project Plan. If available, case histories on prior use of this model will be reviewed
with the test results generated in the model development phase to assure that the model is adequate to
meet the goals of this project.
5.2.3 Part C: Assessments and Response Actions
The different types of assessments and model performance evaluations to be performed in this
project are documented in Element Cl (Assessment and Response Actions) of the QA Project Plan.
Examples are the following:
Testing the Model - The ability of the selected model and modules to correctly
represent modeled conditions will be assessed focusing on changes in eutrophication
due to increases in stream flow. A sensitivity analysis will be performed to determine
the effect of flow in the Port River system focusing on non-point flows and
corresponding loads. The goal of this analysis is to test the sensitivity of the model
during high flows to assure its responses are reasonable. If needed, further verification
will be done by comparing model prediction results with survey data for base
conditions.
Performing Multiple Runs of the Model to Simulate Seasonality Effects - To assess
the extent to which seasonality impacts the model outputs and, ultimately, to incorporate
seasonality into the TMDL development process, the model will be fitted under
different scenarios for nutrient loading and stream flow conditions. Some assessment is
necessary to verify that the different scenarios that are selected represent the critical
conditions for which the TMDL needs to satisfy water quality standards. Results of all
runs will be used in the TMDL development process.
Evaluating Existing Data - As described in Element B9 (Non-direct Measurements),
modeling staff will evaluate data to be used in calibration and as model input according
to criteria discussed in Element A7 (Quality Objectives and Criteria for Model
Inputs/Outputs Data) and will follow-up with the various data sources on any concerns
that may arise.
Final
EPAQA/G-5M 88 December 2002
-------
Calibrating the Model - The model calibration procedure is discussed in Element B7
(Calibration), and criteria for acceptable outcomes are provided in Element A7 (Quality
Objectives and Criteria for Model Inputs/Outputs).
Sensitivity Analyses - The different considerations that are to be addressed in
sensitivity analyses will be documented here. One example is fitting the model after
increasing the assumed non-point source flows and the corresponding loads, in order to
test the sensitivity of the model under high flow situations.
5.2.4 Part D: Output Assessment and Model Usability
This section will describe how the modeling staff will complete and document the following
tasks:
review the set of future model predictions for reasonableness and relevance based upon
observed field data and the literature,
determine the consistency with the requirements documented in Element A7 (Quality
Objectives and Criteria for Model Inputs/Outputs) on acceptable uncertainty,
verify the model predictions relative to the requirements of the TMDL development
process, and
confirm that all steps of the modeling process were followed correctly.
Any problems will be reported to the project manager and corrective actions discussed with the
Wastewater Management Program. The findings, including any limitations associated with their use in
developing TMDLs, will be discussed in the project report.
Note from Element B9 (Non-direct Measurements) above that estimated non-point source load
data for future conditions, representing future conditions after treatment improvement and used as input
to the model, includes a margin of safety factor to account for uncertainty in future projections that are
based on existing data. This margin of safety is generally specified in percentage terms within the QA
Project Plan and is based on previous knowledge of uncertainty in forecasting nutrient levels. It
generally accounts for uncertainty in a manner that is conservative from the standpoint of environmental
protection. For example, it improves the likelihood that the allowable nutrient loads the system can
incorporate without incurring an impairment is not underestimated. In determining whether this margin
of safety was selected appropriately and meets user requirements, some evaluation should be
conducted to assess where the magnitude of this margin of safety is adequate relative to the level of
uncertainty associated with these model inputs. In addition, this margin of safety may be included in the
Final
EPAQA/G-5M 89 December 2002
-------
evaluation of uncertainty in model outputs that is performed within the data quality assessments
documented within Group D of the QA Project Plan.
5.3 SINGLE MODULE - LAND APPLICATION MODULE
This is an example of critical QA Project Plan elements for a single module that is part of a
larger existing model. This QA Project Plan covers the assessment phase of the model development
task. At the beginning of this project, the model has already been developed and will be tested with
this project.
5.3.1 Background
The Land Application Unit (LAU) module is one of five source emission models that are
components of EPA's Multimedia, Multipathway, Multi-receptor Risk Analysis (3MRA) software
system. In a LAU, waste is applied to the land at regular intervals and is tilled into the soil. The LAU
module simulates this process and estimates how much of each waste constituent volatilizes into the air,
leaches into the groundwater, runs off into the stream, or remains bound to the soil. All of these output
concentrations (air emissions, leachate concentration, runoff concentration, and amount bound to the
soil) are used as input to other modules in 3MRA. The LAU module is described in detail in a technical
background document (EPA, 1999c).
The 3MRA model was developed for use in implementing the Hazardous Waste Identification
Rule (HWIR). The 3MRA model establishes risk-based concentration limits that, if exceeded for a
constituent in a waste stream, will identify that waste stream as hazardous, necessitating management of
the waste in a regulated waste management unit designed for hazardous waste. The 3MRA system
schematic diagram is presented in Figure 12. As a 3MRA source module, the LAU Module is the first
module fired by the 3MRA's Multi-Media Simulation Processor.
5.3.2 Part A: Project Management
Section A4: Project Organization. The team will include the following members: client
(Hazardous Waste Program Project Manager), program QA Manager, contractor (Project Manager),
modelers, scientists, information management personnel, and QA personnel. Some of the scientists,
information management personnel, and modelers assigned to the QA project are not assigned to the
LAU model development project and, therefore, are independent in their assessments.
Section A5: Problem Definition/Background. The objective for this project is identified as
follows: to evaluate whether the outputs of the LAU module are acceptable and correspond to the
expectations that the total model places on the LAU module; and to summarize and report to the
project manager and client the results of those assessment tests. Because 3MRA is a probabilistic risk
Final
EPAQA/G-5M 90 December 2002
-------
Source
Media
Food Chain
Exposure/Risk
Figure 12. Module Components of 3MRA.
assessment model, reporting of uncertainty and variability associated with the output of the LAU
module is specified in EPA policy.
Section A6: Project/Task Description and Schedule. It is important to document the
overall schedule because some assessments must be performed sequentially while others may be
conducted independently. This task lists all of the assessments and the time frame in which the tests
should be performed. The schedule for assessment of this module is constrained by the schedule for the
testing of the larger model. Thus, because this module is the first module run for the entire model, it
needs to be assessed first. However, if unforseen problems with the output of this module are
discovered during the assessment of other modules receiving these data outputs, the LAU module must
be revised and re-assessed. This overall project schedule is also constrained by regulatory deadlines.
Section A7: Data Quality Objectives and Criteria for Model Inputs/Outputs.
Because this project involves the assessment of module development, performance and acceptance
criteria are more appropriate than DQOs. Model performance criteria include: (1) evaluation of pre-
simulation calculations and accuracy of calculated input data, (2) evaluation of algorithms for subarea
coupling and water balance, (3) performance of the Generalized Soil Column model for benzene
diffusion and convection and the subarea coupling process, and (4) overall performance of the LAU
module by benchmarking of the LAU module with the widely accepted HELP model. The criteria for
acceptance are specified for each test requirement in Table 5.
The module assessment requirements for the LAU module include testing of algorithms using
hand calculations to ensure that the model calculates as intended, writing another program to ensure that
EPA QA/G-5M
91
Final
December 2002
-------
the results of certain calculations within this model agree within ± 0.1%, or benchmarking the model
using data from an external source for comparison.
Table 5. Description of Modeling Project Tasks
Project Tasks
Read input data
Pre-simulation
calculations
Benzene diffusion
Subarea coupling
process
HELP
benchmarking
Logical flags
Technical
Approach
Echo inputs to
outputs, 10%
check of data
points
Hand
Calculations
Independent
computer
program to check
vertical transport
calculations
Set runoff
parameters to
nonzero and zero
values
Compare 10% of
modeled values
to external
benchmarks from
HELP model
Output logical
flags to external
file and check for
agreement
Quality
Requirements
and Procedures
100% agreement
100% agreement
check.
Complete
agreement with
modeled values
Agreement
between output
within 0.1%
Check of output
is a nonzero value
and zero as
expected
100% agreement
Compare to see
that 100% of
logical flags were
carried forward in
the analysis
Personnel
Requirements
Information
specialists
Scientists
Computer
modeler
Scientists
Scientists
familiar with
external
benchmark data
Information
specialist
Types of
Reports/Records
Input and output
records
Algorithm
spreadsheet and
program
documentation
Output of model
and independent
comparison model
Inputs and outputs
of models,
checking nonzero
and zero values
where appropriate
Model outputs and
corresponding
external
benchmark data
Output of logical
flags from multiple
portions of the
model to assure
correspondence
EPA QA/G-5M
92
Final
December 2002
-------
All data used in the model will be subjected to quality control procedures. All data that are
acquired through electronic transfer from quality-controlled sources will be subjected to 10% check to
assure a valid data transfer has occurred. Any data that are hand entered specifically for this module
will be subjected to a 100% check against primary or secondary data sources.
Section A8: Special Requirements/Certification. This section states that experience with
hazardous waste modeling and information management is needed for this project.
Section A9: Documentation and Records. The documentation of the assessments from
Sections B7 and B9, and Groups C and D will include model input and output files used in the
assessment process, inputs and outputs and model code from any external computer programs
developed to validate the model, and any external data sources used to benchmark the modeled
outputs. In addition, each module assessment step will include a signed QA documentation form for the
model. Any changes made to the model code as a result of QA checks will be documented using the
appropriate change control process.
5.3.3 Part B: Data Generation and Acquisition
Section B7: Calibration. Calibration will not be used for this model due to lack of data [see
Section Cl (Assessment and Response Actions) for benchmark testing].
Section B9: Non-direct Measurements (Data Acquisition Requirements). This section
documents the data used to generate the input data files (site layout data, LAU Module-specific
parameters, chemical properties parameters, and four types of meteorological datadaily, monthly,
annual, and long-term average). This documentation will include the sources of these data, quality
assurance information about these data sources, and an indication of how the data were assessed to
determine if they were appropriate for this model.
They will be assessed based on project requirements (e.g., acceptable uncertainty and
variability). All data used in model testing are provided by EPA and will be evaluated for data quality
for use in the LAU source model (e.g., representativeness, bias, completeness, and comparability to
expected input data). In addition, the data will be checked (100% for data that are hand-entered and
10% for data that are entered by direct electronic transfer) to assure that the correct values have been
recorded in the data files. The sources of data used in the LAU model will be numerous because many
types of data are necessary for this module. For example, meteorologic and climate data are acquired
from the National Climate Center electronically, soil texture data are available from an online database,
and chemical and physical data are available from literature sources (such as hard copies of
handbooks).
Final
EPAQA/G-5M 93 December 2002
-------
Section BIO: Data Management and
Hardware/Software Configuration. In this
section, a flow diagram (like Figure 13) would be
used to show the relationship between the parts of
the module and describe how data flows through
these parts. The QA Project Plan would contain
details about how the data were pre-processed (in
this case, unit conversions were needed) and how
this pre-processing was done. Also in this section,
details would be included about how the data were
entered into the Generalized Soil Column Model
whether it was hand-entered, read from a database
or other file, generated internally, or otherwise
entered into the model.
eteorologic] ILAU-specific
Data / V data
Generalized Soil
Column Model
Model outputs
Figure 13. Data Flow Diagram for the
LAU Module
The hardware and operating system for running this module would be described with enough
detail to determine if they meet the requirements specified for the model (e.g., operating system
compatibility, run time, memory).
5.3.4 Part C: Assessment and Response Actions
These assessments are representative examples of what was planned to be done, and are not a
complete list of model assessments.
Correctly Read Input Data. The LAU Module reads input data from several files, which are
either preexisting (meteorological data) or are generated. Input data will be verified for: (1) site layout
data, (2) LAU Module-specific parameters, (3) chemical properties parameters, and (4) four types of
meteorological datadaily, monthly, annual, and long-term averageby running the simulation in a
mode that documents as output all of the input parameter values to an external file. Ten percent of the
parameter values documented in the output file will be randomly selected and compared with the
corresponding values from the input files.
Results will be reported to the project QA officer in the format provided in Section A9. If
exact agreement is not achieved between the input values and the output values, corrective action will
be taken by the modeling coordinator to assure that the correct files are read appropriately and the test
is repeated to document compliance.
Evaluation of Pre-simulation Calculations. Prior to executing the algorithms in the LAU
Module, each calculation in the spreadsheet will be evaluated for mathematical correctness (checked
against the parent equation in the documentation, if applicable), reviewed for correct coding [e.g.,
EPA QA/G-5M
94
Final
December 2002
-------
ensuring that parentheses were correctly placed in the assignment statement or that the internal
calculation order is correct, (e.g., multiplication before addition)], and reviewed for consistency among
units.
In addition, a number of parameters that are internally generated as a function of the input data
will be assessed to see if they are correct. Single representative calculations of each type will be
selected for testing by the QA officer. A spreadsheet will be created to independently perform the
calculations that generate the parameters of the selected representative LAU Module code for
comparison. To be correct, values can differ by no more than a rounding error (usually less than
±0.1%).
Results will be reported to the project QA officer and the modeling coordinator in the format
provided in Section A9. If incorrect algorithms or deviations from the independent spreadsheet
calculation values are found, corrective action will be taken by the modeling coordinator to assure that
the algorithms are corrected and this test is repeated.
Assessment of Correct Operation of the Generalized Soil Column Model. The
Generalized Soil Column Model is the fundamental chemical fate and transport algorithm driving the
LAU simulation. It simulates the time-variable, vertical distribution of chemical mass in a one-
dimensional soil column as well as losses due to internal sinks, emissions to the atmosphere of volatile
and sorbed chemical, losses from the surficial soil layer due to storm water runoff and erosion, and
losses to the vadose zone from leaching. Computationally, it is a semi-analytical solution to the one-
dimensional, time-variable advective/diffusive differential equation including first-order losses.
The assessment of correct operation of the Generalized Soil Column Model will be conducted
by several test cases, including benzene diffusion and convection, to check the model's response to
several individual key parameters. The planned test cases are described in this section.
Benzene Diffusion This test case will assess the most computationally intensive aspect of the
LAU Module, the diffusive mass transfer algorithm, which simulates the diffusive mass flux emanating
from N individual computational elements constituting the "soil column," combines the effects of each
element on all the other elements, and performs the same operations in a "reflected" soil column to
achieve a zero concentration gradient boundary condition at the soil column's lower boundary.
The diffusion algorithm is impractical to check in a spreadsheet environment because of its
computational complexity. Therefore, an independent computer program will be developed in BASIC
and used to check these calculations. This will be a "clean-room" exercise in which only the
documentation and previous knowledge of the algorithm's theoretical basis will be used to develop it,
rather than by "reverse-engineering" the LAU Module code. The compiled program will be
documented and reported according to requirements specified in Section A9.
Final
EPAQA/G-5M 95 December 2002
-------
An output file of the vertical concentration profile will be produced by running the simulation in
a mode that will also document all of the corresponding input parameter values to an external file. The
simulation will represent the times after the waste application and again after the diffusion algorithm are
applied. Separately, the post-diffusion vertical concentration profile will be simulated by the module
and documented. The results will be reported to the project QA officer and the modeling coordinator
in the format provided in Section A9. If deviations from the independent values are found, corrective
action will be taken by the modeling coordinator to assure that the algorithms are corrected, and this
testing process will be repeated to document compliance.
Benzene Convection. This test case for benzene will be used to check the convective transfer
algorithm of the Generalized Soil Column Model that shifts the vertical concentration gradient
downward by one computational cell when a convection event occurs (i.e., when the chemical mass,
moving at the effective convection velocity, has completely traversed the cell's depth). This test will be
conducted by producing before- and after-convection vertical profiles from a convection event in an
external file mode and comparing them to ensure that the profile has been translated downward by one
cell.
A spreadsheet presenting excerpts from the external file on the convection event will document
the results of the Generalized Soil Column Model convection algorithm assessment, and will be
reported to the project QA officer and the modeling coordinator in the format provided in Section A9.
If the results do not show the downward translation in the profile, corrective action will be taken by the
modeling coordinator to assure that the algorithms are corrected, and this testing process will be
repeated to document compliance.
Assessment of the Correct Operation of the Subarea Coupling Processes. A local watershed is
a land area containing the LAU and over which surface runoff is assumed to occur as sheet flow
(overland flow). It contains up to three subareas: the LAU, a downslope buffer between the LAU and
the water body, and possibly an up-slope subarea lying between the LAU and the local watershed
drainage divide. The local watershed subarea coupling algorithm serves to take chemical (and solids)
mass being transported off the LAU during a runoff event as overland flow and contaminating the
downslope buffer via settling and diffusion. The Generalized Soil Column Model is essentially a one-
dimensional model (vertical), simulating the fate and transport of chemicals in a single subarea. These
subareas are coupled for purposes of simulating fate and transport across the local watershed by code
external to the Generalized Soil Column Model, which is the code of interest for assessing Requirement
4 in Section A7 (Quality Objectives and Criteria for Model Inputs/Outputs).
Subarea Coupling Algorithm Decoupling Test. As noted above, the mechanisms by which a
buffer area downslope of the LAU is contaminated are: (1) settling of eroded solids onto the buffer
and, (2) diffusive exchange of dissolved chemicals between the storm water runoff and the underlying
pore space of the buffer's surficial soil layer. These mechanisms are controlled by the solids' settling
Final
EPAQA/G-5M 96 December 2002
-------
rate in storm water runoff and the diffusion coefficient. A test case will be assessed as a two-run
sensitivity analysis to ensure that cancellation of these two coupling parameters would result in no
contamination of the downslope buffer subarea. Run 1 will use the default values (nonzero) for storm
water runoff and the diffusion coefficient to establish that contamination of the downslope buffer occurs
when these values are nonzero. Run 2 will change these two parameter values to zero to assess
whether contamination will occur.
The buffer concentrations will be reported to the output file, and if the results of Run 1 were not
positive and the results of Run 2 were not zero, corrective action will be taken. The results will be
reported to the project QA officer and the modeling coordinator in the format provided in Section A9.
If the results are not appropriate, corrective action will be taken by the modeling coordinator to assure
that the algorithms are corrected, and this testing process will be repeated.
Benchmark Testing. Benchmark testing will be performed on portions of the module, and
additional model assessment will be performed to test model performance. Benchmark testing is used
in this case because there are no existing data to use in model calibration. This assessment will
compare hydrology outputs of the LAU's water balance algorithm to a widely accepted EPA
hydrology model, the HELP model, which is more complex than the LAU module's water balance
algorithm but uses a similar underlying mathematical construct. Approximate comparability of long-term
average results is expected with the benchmarks. General agreement of the various water balance
outputs with the outputs generated by the HELP model will be acceptable.
This test case will benchmark the water balance algorithm against results of the HELP model
for six sites (e.g., Miami, Dallas, San Diego, Madison, Denver, and Seattle) over five years.
Meteorological files generated for these benchmark sites will be used with edited daily meteorological
files to replace daily precipitation and daily temperature with the precipitation and temperature time
series for the benchmark site being run. The monthly meteorological file will also be edited to replace
the monthly maximum and minimum average temperature time series values with those from the
benchmark site. Each file will end the time series at five years. The appropriate data file will also be
edited to replace the site latitude with the latitude of the benchmark site. In addition, the appropriate
data file will be edited to make any non-LAU sub-areas have an area of 1.0 m2 to minimize water
balance effects of those subareas. Identical values of the following water balance parameters will be
used for the HELP and HWIR water balance algorithm runs.
Results of comparisons of HWIR and HELP will be presented in spreadsheets for infiltration
rates and average run-off In addition, this assessment will also include average annual runoff rates for
the regions of the country pertaining to each of the six sites as taken from the Water Atlas (Geraghty et
al, 1973). The five-year average runoff values will be compared to those of the Water Atlas for each
of the six sites as well as the predictions of the HELP model as a check of reasonableness of the HELP
output. This approach assumes that these empirical Water Atlas data are indeed an appropriate basis
Final
EPAQA/G-5M 97 December 2002
-------
of comparison2. The results will be reported to the project QA officer and the modeling coordinator in
the format provided in Section A9. If the results do not show the appropriate correspondence with the
benchmarks, corrective action will be taken by the modeling coordinator to assure that the algorithms
are corrected, and this testing process will be repeated to document compliance.
5.3.5 Part D: Output Assessment and Model Usability
If the model outputs meet the internal criteria of the module as described in this test plan, they
should meet the requirements for the overall model. Total model planning, development, and testing are
intended to assure that if each module meets its own internal quality standards, the module output will
meet the requirements for the entire model.
Assessment of Correct Operation of System Requirements. Because the LAU is a source
module and is the first module fired by the overall modeling system, system-level interactions with other
modules are relatively minor. However, some system-level functionality is neededfor example,
outputting warning/error messages, output of logical flags, and time series management.
Logical Flags. A set of seven logical outputs will be generated by the LAU Module to check
that the outputs are properly generated. That is, the LAU need only have the potential to output that
variable for the flag to be set at "true." It need not actually have a non-zero output during the run. The
criteria for properly generated flags will be checked via hand calculations.
The data file results with the logical flags will be checked for agreement with the module
documentation, and the agreement will be reported to the project QA officer and the modeling
coordinator in the format provided in Section A9. If the logic flags are incorrectly set, corrective action
will be taken by the modeling coordinator to assure that the logic flags are corrected, and this testing
process will be repeated to confirm compliance.
More benchmark testing would be required for more definitive statements regarding relative accuracy. For
example, only a single curve number was used in the comparisons, and it is unknown to what extent that curve
number might represent average conditions as reflected by the Water Atlas data.
Final
EPAQA/G-5M 98 December 2002
-------
APPENDIX A
BIBLIOGRAPHY
40 CFR 51, Code of Federal Regulations, Guideline on Air Quality Models, Appendix W.
American National Standards Institute/American Society for Quality Control (ANSI/ASQC). (1995).
Specifications and Guidelines for Quality Systems for Environmental Data Collection and
Environmental Technology Programs (E4-1994). American National Standard.
American Society for Testing and Materials. 2000. Standard Guide for Statistical evaluation of
Atmospheric Dispersion Model Performance. Standard D6589-00. West Conshohocken,
Pennsylvania.
Berthouex, P.M. and Brown, L.C. (1994). Statistics for Environmental Engineers. Boca Raton,
FL: Lewis Publishers
Chen, J. and Beck, M.B. (2000). Quality Assurance of Multi-media Model for Predictive
Screening Tasks (EPA/600/R-98/106). U.S. EPA.
Cox, W. M. and Tikvart, J.A., (1990). A Statistical Procedure for Determining the Best Performing
Air Quality Simulation Model. Atmospheric Environment. Vol. 24A: pp. 2387-2395.
D'Agostino, R.B. and Stephens, M.A. (1986). Goodness-of-fit Techniques. New York: Marcel
Dekker.
Draper, NR. and Smith, Jr., H. (1981). Applied Regression Analysis (2nd ed.). New York: Wiley.
Geraghty, J. A., D. W .Miller, F. van der Leeden, and F. L. Troise. (1973). Water Atlas of the
United States. 3rd edition. Water Information Center, Incorporated, Port Washington, NY.
Gilbert, R.O. (1987). Statistical Methods for Environmental Pollution Monitoring. New York:
Wiley.
Institute of Electrical and Electronics Engineers (IEEE). (1986). Standard 1012: IEEE standard for
software verification and validation plans. IEEE Standards Collection: Software
Engineering (Volume 2: Process Standards). Piscataway, NJ.
Final
EPAQA/G-5M A-l December 2002
-------
Institute of Electrical and Electronics Engineers (IEEE). (1998a). Standard 828: IEEE standard for
software configuration management plans. IEEE Standards Collection: Software
Engineering (Volume 2: Process Standards). Piscataway, NJ.
Institute of Electrical and Electronics Engineers (IEEE). (1998b). Standard 830: IEEE recommended
practice for software requirements specifications. IEEE Standards Collection: Software
Engineering (Volume 4: Resource and Technique Standards). Piscataway, NJ.
Institute of Electrical and Electronics Engineers (IEEE). (1999). Standard 730: IEEE standard for
software quality assurance plans. IEEE Standards Collection: Software Engineering
(Volume 2: Process Standards). Piscataway, NJ.
Kirchner, T.B. (2000). Uncertainty analysis. In T.B. Borak (Ed.), Applications of Probability and
Statistics in Health Physics, (pp. 115-161). Madison, WI: Medical Physics Publishing.
Morgan, M.G. andHenrion, M. (1990). Uncertainty: a Guide to Dealing with Uncertainty in
Quantitative Risk and Policy Analysis. New York: Cambridge University.
National Council on Radiation Protection and Measurements (NCRP). (1996). A Guide for
Uncertainty Analysis in Dose and Risk Assessments Related to Environmental
Contamination. (NCRP Commentary No. 14). Bethesda, MD.
National Institute of Standards and Technology (NIST). (1996). NIST Special Publication 500-234:
Reference Information for the Software Verification and Validation Process [On-line].
Available: http:\\Mssa.ncsl.nist.gov\HHRFdata\Artifacts\ITLdoc\234\val-proc.html
Snedecor, G.W. and Cochran, W.G. (1989). Statistical Methods (8th ed.). Ames, IA: Iowa State
University Press.
U.S. Environmental Protection Agency. (1994). Guidance for Conducting External Peer Review of
Environmental Regulatory Models (EPA 100-B-94-001). Washington, DC: Office of the
Administrator.
U.S. Environmental Protection Agency. (1996). Summary Report of Workshop on Monte Carlo
Analysis (EPA/63O/R-96/010). Washington, DC: Office of Research and Development.
U.S. Environmental Protection Agency. (1997). Guiding Principles for Monte Carlo Analysis
(EPA/630/R-97/001). Washington, DC: Office of Research and Development
Final
EPAQA/G-5M A-2 December 2002
-------
U.S. Environmental Protection Agency. (1998a). Guidance for Quality Assurance Project Plans
(QA/G-5) (EPA/600/R-98/018). Washington, DC: Office of Research and Development.
U.S. Environmental Protection Agency. (1998b). Information Resources Management Policy
Manual (Directive 2100). Washington, DC.
U.S. Environmental Protection Agency. (1999a). Information Security Manual (Directive 2195A1).
Washington, DC.
U. S. Environmental Protection Agency. (1999b). Quality Assurance Plan for Mathematical
Modeling. Washington, DC: Office of Research and Development.
U.S. Environmental Protection Agency. (1999c). Quality Assurance of Multi-media Model for
Predictive Screening Tasks (EPA/600/R98/106). Washington, DC: Office of Research and
Development.
U.S. Environmental Protection Agency. (1999d). Report of the Workshop on Selecting Input
Distributions for Probabilistic Assessments (EPA/630/R-98/004) [On-line]. Available:
www.epa.gov/ncea/input.htm
U.S. Environmental Protection Agency. (1999e). Risk Assessment Guidance for Superfund:
Volume 3 (Part A): Process for Conducting Probabilistic Risk Assessments (EPA 000-0-
99-000) (Draft) [On-line]. Available: www.epa.gov/superfund/pubs.htm
U.S. Environmental Protection Agency. (1999f). Source Modules for Nonwastewater Waste
Management Units (Land Application Units, Wastepiles, and Landfills): Background and
Implementation for the Multimedia, Multipathway, and Multireceptor Risk Assessment
(3MRA)for HWIR99. Washington, DC: Office of Solid Waste.
U.S. Environmental Protection Agency. (2000a). EPA Quality Manual for Environmental
Programs (Order 5360 Al). Washington, DC.
U.S. Environmental Protection Agency. (2000b). Guidance for the Data Quality Objectives
Process (QA/G-4) (EPA/600/R-96/055). Washington, DC: Office of Environmental
Information.
U.S. Environmental Protection Agency. (2000c). Guidance for Data Quality Assessment:
Practical Methods for Data Analysis (QA/G-9) (EPA/600/R-96/084, QAOO Update).
Washington, DC: Office of Environmental Information.
Final
EPAQA/G-5M A-3 December 2002
-------
U.S. Environmental Protection Agency. (2000d). Policy and Program Requirements for the
Mandatory Agency-wide Quality System (EPA Order 5360.1 A2). Washington, DC.
U.S. Environmental Protection Agency. (2000e). Science Policy Council Handbook: Peer review
(EPA 100-B-OO-OOl). Washington, DC: Office of Research and Development, Office of
Science Policy.
U.S. Environmental Protection Agency. (2000f). Systems Development Life Cycle Methodology
(Sdlcm) Statement of Policy, Standards and Guidelines. Washington, DC: Office of
Research and Development, Center for Exposure Assessment Modeling.
U.S. Environmental Protection Agency. (2001). EPA Requirements for Quality Assurance Project
Plans (QA/R-5) (EPA/240/B-01/003). Washington, DC: Office of Environmental
Information.
U.S. Environmental Protection Agency. (2002). Guidance on Systematic Planning for
Environmental Data collection Using Performance and Acceptance Criteria (QA/G-4A).
Peer Review Draft.
U.S. General Accounting Office. (1997). Report to the Chairman, Subcommittee on Oversight
and Investigations, Committee on Commerce, House of Representatives: Air Pollution
Limitations of EPA 's Motor Vehicle Emissions Model and Plans to Address Them
(GAO/RCED-97-21).
Warren-Hicks, W. J. and Moore, D.R.J. (1998). Uncertainty Analysis in Ecological Risk
Assessment (SETAC Special Publications Series). Pensacola, FL: Society of Environmental
Toxicology and Chemistry.
Final
EPAQA/G-5M A-4 December 2002
-------
APPENDIX B
TERMS AND DEFINITIONS
assessment - the evaluation process used to measure the performance or effectiveness of a system
and its elements.
audit (quality) - a systematic and independent examination to determine whether quality activities and
related results comply with planned arrangements and whether these arrangements are implemented
effectively and are suitable to achieve objectives.
boundary condition - mathematical expression of a state of the environmental system that constrains
the equations of the mathematical model.
conceptual model - an interpretation or working description of the characteristics of the physical
system.
contractor - any organization or individual that contracts to furnish services or items or perform work;
a supplier in a contractual situation.
data quality assessment - a statistical and scientific evaluation of the data set to determine the validity
and performance of the data collection design and statistical test, and to determine the adequacy of the
data set for its intended use.
design - specifications, drawings, design criteria, and performance requirements. Also the result of
deliberate planning, analysis, mathematical manipulations, and design processes.
environmental data - any measurements or information that describe environmental processes,
location, or conditions; ecological or health effects and consequences; or the performance of
environmental technology. For EPA, environmental data include information collected directly from
measurements, produced from models, and compiled from other sources such as data bases or the
literature.
environmental data operations - work performed to obtain, use, or report information pertaining to
environmental processes and conditions.
environmental processes - manufactured or natural processes that produce discharges to or that
impact the ambient environment.
Final
EPAQA/G-5M B-l December 2002
-------
environmental programs - work or activities involving the environment, including but not limited to
characterization of environmental processes and conditions; environmental monitoring; environmental
research and development; the design, construction, and operation of environmental technologies; and
laboratory operations on environmental samples.
evaluation - a test of the model with known input and output information that is used to assess that the
calibration parameters are accurate without further change and to demonstrate model performance.
graded approach - the process of basing the level of application of managerial controls applied to an
item or work according to the intended use of the results and the degree of confidence needed in the
quality of the results.
inspection - an activity such as measuring, examining, testing, or gauging one or more characteristics of
an entity and comparing the results with specified requirements in order to establish whether
conformance is achieved for each characteristic.
management system - a structured, non-technical system describing the policies, objectives,
principles, organizational authority, responsibilities, accountability, and implementation plan of an
organization for conducting work and producing items and services.
method - a body of procedures and techniques for performing an activity (e.g., sampling, modeling,
chemical analysis, quantification) systematically presented in the order in which they are to be executed.
model calibration - the process of refining the model representation of the environmental framework,
properties, and boundary conditions to achieve a desired degree of correspondence between the model
simulations and observations of the environmental system and processes. The focus is usually on the
estimation and characterization of empirical constants used in parameterizations and mathematical
representations of environmental processes.
participant - when used in the context of environmental programs, an organization, group, or individual
that takes part in the planning and design process and provides special knowledge or skills to enable
the planning and design process to meet its objective.
performance evaluation - a type of audit in which the quantitative data generated in a measurement
system are obtained independently and compared with routinely obtained data to evaluate the
proficiency of an analyst or laboratory.
quality - the totality of features and characteristics of a product or service that bear on its ability to
meet the stated or implied needs and expectations of the user.
Final
EPAQA/G-5M B-2 December 2002
-------
quality assurance - an integrated system of management activities involving planning, implementation,
documentation, assessment, reporting, and quality improvement to ensure that a process, item, or
service is of the type and quality needed and expected by the client.
quality assurance manager - the individual designated as the principal manager within the
organization having management oversight and responsibilities for planning, documenting, coordinating,
and assessing the effectiveness of the quality system for the organization.
quality assurance project plan - a document describing in comprehensive detail the necessary QA,
QC, and other technical activities that must be implemented to ensure that the results of the work
performed will satisfy the stated performance criteria.
quality control - the overall system of technical activities that measures the attributes and performance
of a process, item, or service against defined standards to verify that they meet the stated requirements
established by the customer; operational techniques and activities that are used to fulfill requirements for
quality.
quality system - a structured and documented management system describing the policies, objectives,
principles, organizational authority, responsibilities, accountability, and implementation plan of an
organization for ensuring quality in its work processes, products (items), and services. The quality
system provides the framework for planning, implementing, documenting, and assessing work
performed by the organization and for carrying out required QA and QC activities.
record - a completed document that provides objective evidence of an item or process. Records may
include photographs, drawings, magnetic tape, and other data recording media.
sensitivity - the variation in the value of one or more output variables or quantities calculated from the
output variables due to variability or uncertainty in one or more inputs to a model.
sensitivity analysis - a quantitative evaluation of the impact of variability or uncertainty in model
inputs on the degree of calibration of a model and on its results or conclusions.
simulation - one complete execution of the computer program, including input and output.
source code - the program instructions written in a programming language [TIPS PUB 106 (NIST)
definition].
specification - a document stating requirements and which refers to or includes drawings or other
relevant documents. Specifications should indicate the means and the criteria for determining
conformance.
Final
EPAQA/G-5M B-3 December 2002
-------
supplier - any individual or organization furnishing items or services or performing work according to a
procurement document or financial assistance agreement. This is an all-inclusive term used in place of
any of the following: vendor, seller, contractor, subcontractor, fabricator, or consultant.
surveillance (quality) - continual or frequent monitoring and verification of the status of an entity and
the analysis of records to ensure that specified requirements are being fulfilled.
uncertainty - lack of knowledge about specific factors, parameters, or models.
variability - observed differences attributable to true heterogeneity or diversity in a population or
exposure parameter (variability can be better characterized but not reduced by further measurement or
study).
verification - confirmation by examination and provision of objective evidence that specified
requirements have been fulfilled. In design and development, verification concerns the process of
examining a result of a given activity to determine conformance to the stated requirements for that
activity.
Final
EPAQA/G-5M B-4 December 2002
-------
APPENDIX C
CHECKLIST FOR QA PROJECT PLAN ELEMENTS FOR MODELING
Name of Project this QA Project Plan is for:
GROUP A: PROJECT MANAGEMENT
Al. Title and Approval Sheet
Contents this element may contain:
Title of QA Proj ect Plan
Revision number of Q A Proj ect Plan
Effective date of QA Project Plan revision
Names of all organizations involved in modeling project
Names of all key project officials responsible for the work, with space for dated signatures
A2. Table of Contents and Document Control Format
Contents this element may contain:
Title of all sections, including subsections, tables, figures, references, appendices
Page numbers for each section
Section for each Q A Proj ect Plan element
Document control box
A3. Distribution List
Contents this element may contain:
List of all individuals (and their role on the project) who will be provided copies of the approved
QA Project Plan, including all persons responsible for implementation, including project managers,
QA Managers, and representatives of all groups involved.
A4. Project/Task Organization
Contents this element may contain:
Concise organizational chart showing the relationships and lines of communication among all
project participants, other data users, and any subcontractors relevant to environmental data
operations
Project name and organizations involved, and a description of their respective responsibilities
Final
EPAQA/G-5M C-l December 2002
-------
A5. Problem Definition/Background
Contents this element may contain:
Goals and objectives of this project that will address this problem
Definition of the population the problem targets and what measures within this population the
problem addresses
Reason the project includes a modeling approach to address the problem (is it a new predictive
tool?)
Types of decisions that may be made as a result of this project
Names of those responsible for making these decisions
Any other types of problems that the proj ect may address
Background information on the problem
Reasons the project is important, how it supports other existing research, programs, or regulations
Conflicts or uncertainties that will be resolved by this project
Reasons one model is determined to be better than another for this application
A6. Project/Task Description and Schedule
Contents this element may contain:
Summary of all work to be performed, products to be produced, and the schedule for
implementation
List of products, deliverables, and milestones to be completed in the various stages of the project
Schedule of anticipated start and completion dates for the milestones and deliverables, and
persons responsible for each
A7. Quality Objectives and Criteria for Model Inputs/Outputs
Contents this element may contain:
Project data quality objectives (DQOs), performance criteria, and acceptance criteria
Description of task that needs to be addressed and the intended uses of the output of the
modeling project to achieve the task
List of requirements associated with the hardware/software configuration for those studies
involving software evaluation
A8. Special Training Requirements/Certification
Contents this element may contain:
Types of required training and certification needed by the proj ect team
Plan for obtaining training and/or certification
Documentation of training and/or certification
Final
EPAQA/G-5M C-2 December 2002
-------
A9. Documentation and Records
Contents this element may contain:
Description of information to be included in reports
Proper document control and distribution procedures
Details on document storage
Backup plan for records stored electronically
Description of the change control process (who approves changes, etc.)
Length of retention periods for each record
Data assessment reports, interim project progress reports
Model science formulation report, peer review reports
Model assessment reports, interim project progress reports
Code standards, code auditing and testing reports, interim project progress reports
Model calibration report
Model evaluation records (How well does the model report variability and uncertainty in its
output?)
User's manual
Configuration management (after production version) and code maintenance (e.g., or software
internal documentation of logic and structure) manuals
GROUP B: MEASUREMENT AND DATA ACQUISITION
B7. Calibration
Contents this element may contain:
Objectives of model calibration activities, including acceptance criteria
Frequency of model calibration activities
Details on the model calibration procedure
Method(s) of acquiring input data
Types of output generated by the model calibration
Approach to characterize uncertainty (e.g., sensitivity analysis)
Corrective action to be taken if criteria are not met
Resources and responsibilities for calibrating the model
Analysis of model output relative to acceptance criteria
Final
EPAQA/G-5M C-3 December 2002
-------
B9. Non-direct Measurements (Data Acquisition Requirements)
Contents this element may contain:
Types of data needed for implementing a project that are obtained from non-measurement sources
such as databases, literature files
Need for non-direct measurements, intended use of data
Method(s) of identifying and acquiring data
Method of determining the underlying quality of the data
SOPs and field or lab-specific deviations associated with these procedures
Acceptance criteria for non-direct measurements: such as completeness, representativeness, bias,
precision, qualifying data
BIO. Data Management and Hardware/Software Configuration
Contents this element may contain:
Information on the project data management process (field, office, and lab)
Record-keeping procedures, document control system, audit trails
Control mechanism for detecting and correcting errors, preventing loss of data
Procedures for assuring applicable Agency resource management requirements are satisfied
Required computer hardware/software and any specific performance requirements
Data Management
Any data forms, checklists, on-line interactive screens used in the modeling process
Any graphics developed to document the data management process (process flow diagrams,
modeling flow charts, etc.)
Documentation of internal checks used during data entry
Data calculations and analyses that should to be highlighted in the QA Project Plan
Plans for characterization of uncertainty and variability in the model results (e.g., summary
statistics, frequency distributions, goodness-of-fit tests)
Hardware/Software Configuration
List of equipment, hardware, and software that will be used on the project
Description of performance requirements
Decisions regarding security issues
Decision regarding communication issues
Decisions regarding software installation issues
Decisions regarding response time issues
Plans for requirements documentation
Final
EPAQA/G-5M C-4 December 2002
-------
Coding standards
Testing plans
Plans for data dictionary (may not need to be a separate document)
Plans for a user's manual
Plans for a maintenance manual (explaining software logic and organization)
Plans for source code for the ultimate user of the model or model framework
Configuration management plan (procedures to control software/hardware configuration during
development of the original model version)
GROUP C: ASSESSMENTS AND OVERSIGHT
Cl. Assessment and Response Actions
Contents this element may contain:
Assessment/oversight strategies and schedule of assessment activities, order of events
Organizations and individuals expected to participate in assessments, including peer reviews
Information expected, success criteria
Scope of authority of assessors to recommend or direct changes to the model (corrective actions)
Qualitative and quantitative assessments
Internal assessments (internal QA officer's review of input data, code verification, calibration,
benchmarking) and external assessments (peer review of model theory or mathematical structure)
Surveillance activities (continued monitoring of status and progress of the project, tracking project
milestones and budgets)
Plans for model performance evaluations
Plans for sensitivity analysis
Plans for uncertainty analysis
Plans for data quality assessment
Plans for code testing
Hardware/Software Assessments
Plans for hardware and software configuration testing, if appropriate
Plans for code verification tests
Plans for internal and external peer reviews
Plans for checking for programming errors
Plans for checking for correct insertion of model equations
Plans for checking for code's linkage to analysis of uncertainty
Final
EPAQA/G-5M C-5 December 2002
-------
Hardware/Software Configuration Tests
Plans for software code development inspections
Plans for software code performance testing
Plans for a test of the model framework
Plans for integration tests (check computational and transfer interfaces between modules)
Plans for regression tests
Plans for stress testing of complex models (to ensure that maximum load during peak usage does
not exceed limits of the system)
Plans for acceptance testing (contractually-required testing needed before a new model or model
application is accepted by the customer and final payment is made)
Plans for beta testing of pre-release hardware/software, recording of anomalies
Plans for checking for programming errors
Plans for science and product peer review
Theoretical basis for the model
Mathematical model structure
Model algorithms
Model predictions
Model calibration
Plans for data quality assessment
Plans for peer review of final technical product
C2. Reports to Management
Contents this element may contain: plans for documentation of:
Project reporting schedule
Frequency, content, and distribution of reports
Deviations from approved QA Proj ect Plan
Need for response actions to correct deviations
Potential uncertainties in decisions based on input data and model limitations
Data Quality Assessment findings
Final
EPAQA/G-5M C-6 December 2002
-------
GROUP D: DATA VALIDATION AND USABILITY
Dl. Departures from Validation Criteria
Contents this element may contain:
Criteria used to review and validate (accept, reject, or qualify) model components such as theory,
mathematical procedures, code, and calibration (convergence criteria, etc.)
Criteria used to review and validate input data
Criteria used to test model performance
Criteria used to review or validate model outputs
D2. Validation Methods
Contents this element may contain:
Methods for review of model components such as theory, mathematical procedures, code, and
calibration (peer review, etc.)
Methods for review of input data
Methods for review of model performance tests
Methods for assessment of model output and usability
D3. Reconciliation with User Requirements
Contents this element may contain:
Discussion of project or task results
List of departures from assumptions set in the planning phase of the model
Report on limitations on use of output data for decision makers or users
Final
EPAQA/G-5M C-7 December 2002
------- |