oEPA
United States
Environmental Protection
Agency
EPA/600/R-18/021 January 2018 | www.epa.gov/ord
Temporal-Spatial Ambient
Concentrator Estimator
(T-SpACE): Hierarchical
Bayesian Model Software
Used to Estimate Ambient
Concentrations of NAAQS
Air Pollutants in Support
of Health Studies
Eric S. Hall
Office of Research and Development
National Exposure Research Laboratory
-------
-------
EPA/600/R-18/021
January 24, 2018
Temporal-Spatial Ambient Concentrator Estimator (T-SpACE):
Hierarchical Bayesian Model Software Used to Estimate Ambient Concentrations of
NAAQS Air Pollutants in Support of Health Studies
Eric S. Hall
Systems Exposure Division
National Exposure Research Laboratory
Office of Research and Development
US Environmental Protection Agency
Research Triangle Park, North Carolina 27711
-------
Table of Contents
1.0 Introduction to Temporal-Spatial Ambient Concentrator Estimator (T-SpACE) Software
1.1 EPA's NAAQS and criteria pollutants
1.2 Need for the software
1.2.1 Brief overview of the steps to develop the surfaces
1.2.2 Expected use of the resulting air quality surfaces
2.0 Design and nominal operational behavior of the T-SpACE software
2.1 Design of the software
2.1.1 Transforming monitor data and model-based data to simulation input data
2.1.1.1 Air Quality Monitoring Network Monitor Data
2.1.1.2 Air Quality Model-Based Data
2.1.2 Choosing the parameters for the simulation
2.1.2.1 Choice of the region for simulation
2.1.2.2 T-SpACE parameter choices
2.1.3 Results from the simulation
2.1.4 Summary statistics and validation of the resulting air quality surfaces
2.1.5 Three (consecutive) "year" average of 4th highest concentration surface
(03)
2.2 Operational behavior of the software
2.2.1 System requirements
2.2.2 Typical processing times
2.2.2.1 Preparing the input data for the simulation (Step 2)
2.2.2.2 Full year/Full grid simulations (Step 4)
2.2.2.3 Validation (Step 5)
3.0 Operation of and/or the statistical basis underlying the operation of the T-SpACE
software model
3.1 Grid size
3.2 Timeframe
3.3 Bias Spline
3.4 Control Points
3.5 Priors
3.6 Neighborhood Boundary
3.7 Burn-in/Loop
3.8 Markov Chain Monte Carlo (MCMC) Simulation
4.0 CMAQ and CAMx Air Quality Models Used in T-SpACE
4.1 CAMx model description
4.2 CMAQ model description
5.0
References
-------
Appendix A - Table of Contents
Introduction
A-xxiv
1.1 Software Overview
A-xxiv
1.1.1 Description
A-xxiv
1.1.2 Purpose
A-xxiv
1.1.3 Scope
A-xxv
1.2 Glossary
A-xxv
1.2.1 Definitions
A-xxv
1.2.2 Acronyms
A-xxvi
1.2.3 Abbreviations
A-xxvii
1.3 References
A-xxviii
Design Descriptions
A-xxix
2.1 Design Overview
A-xxix
2.1.1 Requirements
A-xxix
2.1.2 Tradeoffs
A-xxix
2.2 User Interface
A-xxx
2.3 System Design
A-xxxi
2.4 Logical design
A-xxxix
2.5 Data
A-xliii
2.5.1 Permanent Storage
A-xliii
2.5.2 Volatile Storage
A-li
2.6 Communications/Messaging
A-lvi
Operational Design
A-lvii
3.1 Use Cases
A-lvii
3.1.1 Prepare to Create Simulation Input Use Case
A-lvii
3.1.2 Create Simulation Input Use Case
A-lvii
3.1.3 Choose Simulation Parameters Use Case
A-lvii
3.1.4 Run Simulation Use Case
A-lvii
3.1.5 Validate Simulation Use Case
A-lvii
3.1.6 Compare AQS Files Use Case
A-lvii
3.1.7 Compare Airsites Files Use Case
A-lviii
3.1.8 Average 4th Highest Surface Files Use Case
A-lviii
3.1.9 Create States Grid File Use Case
A-lviii
3.2 Typical Processing
A-lviii
3.2.1 Sequence
A-lviii
3.3 Expected System Behavior
A-lix
3.3.1 Approximate Time
A-lix
3.3.2 Memory Usage
A-lxi
Design Issues
A-lxii
4.1 Factors Affecting Design
A-lxii
4.2 Suggested Future Upgrades
A-lxiii
Summary
A-lxiii
-------
Appendix B - Table of Contents
1.0 Background B-l
2.0 Model B-l
3.0 Data B-2
3.1 Models-3/Communitv Multiscale Air Quality (CMAQ) B-2
3.1.1 CMAO B-2
3.1.2 CAMx B-3
3.2 Network Monitors B-6
3.2.1. T-SpACE Model Input Monitor File B-6
3.2.2 Validation B-6
4.0 Input Parameters B-7
5.0 Model validation results B-10
5.1 Explanation of the metrics used in the validation B-ll
5.2 Summary of results B-18
5.2.1. Comparisons for the runs during the "Ozone" season of May 1 through
October 31 B-18
5.2.2. Comparisons for the runs for a "full" year of data (January 1 through
November 29) B-19
5.2.3. Overall summary B-19
6.0 References B-21
Appendix C: Markov Chain Monte Carlo (MCMC) Description (Model 5.1) C-2
Appendix D: SAS Code for Developing the Krige Model Estimates D-10
Appendix E: Simulation.PAR (2001. 12 km. ozone-Run 1) E-1
Appendix F: SAS Code for Generating the Summary Statistics F-l
Appendix G: T-SpACE User's Guide G-l
-------
1.0 Introduction to Temporal-Spatial Ambient Concentrator Estimator (T-SpACE)
Software
1.1 EPA '.s NAA QS and criteria pollutants
To fulfill its mission to protect human health and the environment, EPA has established National
Ambient Air Quality Standards (NAAQS) on six selected air pollutants known as criteria
pollutants: ozone (O3); carbon monoxide (CO); lead (Pb); nitrogen dioxide (NO2); sulfur dioxide
(SO2), and; particulate matter (PM). The states are primarily responsible for maintaining and
improving air quality and complying with the NAAQS. Ozone (O3) and particulate matter (PM)
are two of the criteria pollutants whose levels are regulated by NAAQS. Extensive air pollution
monitoring networks have been set-up across the US to understand the levels (concentrations) of
these air pollutants across the US and over time. Once collected, this information is reported to
EPA and is made available publicly after the data is reviewed and analyzed for quality and
accuracy.
When assessing air quality, the most direct way is to utilize unbiased ground-level air pollution
measurements from the existing surface monitoring network stations located across the US.
However, a good portion of the US, especially rural areas, have very sparse or irregularly spaced
monitoring stations, resulting in large areas of the country with no information available on air
quality. Air quality models estimate the spatial and
temporal gradients of air pollution based on
emissions inventories and meteorological
information. These models, while providing
estimates over large regions at relatively low cost,
have (statistical) bias and have greater error
(variance) than air pollution monitoring networks.
These methods include Model s-3/Community
Multiscale Air Quality (CMAQ) numerical model
and the Comprehensive Air Quality Model with
Extensions (CAMx). Various statistical methods
have been investigated that utilize existing
monitoring data and spatial modeling techniques to
develop a concentration surface that
(retrospectively) estimates the distribution of daily
air pollution levels within a specified region of the
US. Kriging is one such method. Another
technique is Hierarchical Bayesian Modeling
Figure 1. Plot of PM2.5 Measurements on a Specific
Monitoring Day from FRM Monitors in the
NAMS/SLAMS Network Located in the Northeast US
(HBM).
The HBM approach combines air monitoring data from the 2,545 National Air Monitoring
Stations/State and Local Air Monitoring Stations (NAMS/SLAMS) with results from model
estimates, such as from the CMAQ and CAMx air quality models. The approach develops daily
estimated surfaces for both fine particulate matter (PM2.5) and ozone (O3) concentrations.
This report describes the HBM approach developed to model concentration surfaces for O3 and
PM2.5, and the accompanying software that has been developed to produce the surface estimates.
-------
1.2 Need for the software
The Temporal-Spatial Ambient Concentrator Estimator (T-SpACE) software, that was jointly
developed by EPA and Battelle through EPA Contract EP-D-04-068 (Work Assignment #54),
requires data preparation and model parameter input assignments from the model user to run a
simulation to produce an estimated air pollution concentration surface. The T-SpACE software
was designed to streamline model set-up and minimize the amount of work required to manually
implement the data and model preparation steps, including allowing the user to prepare the input
data for the model run, choose the model specifications, run the model, receive output files
containing the estimated surface, and run a set of validation procedures to compare the T-SpACE
output concentrations with a standard kriging method and model-based output concentrations
such as those from CMAQ and CAMx. The T-SpACE software provides a single interface for
running a simulation. The user is asked for particular choices for input data sets, model
parameters, and names for output files and is guided through the simulation. Note: This model
was used in the preparation of two peer-reviewed journal articles: Weber, S. A., Insaf, T. Z.,
Hall, E. S., Talbot, T. O., Huff, A. K., 2016, "Assessing the Impact of Fine Particulate Matter
(PM2.5) Sources on Respiratory-Cardiovascular Chronic Diseases in the New York City
Metropolitan Area using Hierarchical Bayesian Model Estimates", Environmental Research,
Volume 151, November 2016, pp. 399-409, DOI: 10.1016/j.envres.2016.07.012 (Weblink:
http://www.sciencedirect.com/scieiice/aiticle/pii/S0013935116302961), and, "Assessing the
Impact of Fine Particulate Matter (PM2.5) on Respiratory-Cardiovascular Chronic Diseases in the
Baltimore Maryland Metropolitan Area using Hierarchical Bayesian Model Estimates" (in
preparation). The United States Environmental Protection Agency and its Office of Research
and Development did not perform, or provide funding, for data collection and analysis of the
health outcome data described in the two above-mentioned journal articles.
1.2.1 Brief overview of the steps to develop the surfaces
In the software, there are five distinct steps to develop the surfaces and the final step prepares the
visual rendering of the generated air pollution concentration surface.
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Choose Time/Grid
Prepare Model Input Data
Model Specification
Launch Model
Launch Validation
Visualize Surface
Each estimated air pollution concentration surface is calculated for a single calendar year. In
Step 1, the area covered by the air pollution concentration surface is defined by the positional
grid coordinates utilized by the selected air quality model. Step 1 directs the user to choose the
time frame over which to calculate the surfaces, and the positional grid coordinates from the
selected air quality model (CMAQ or CAMx) to provide a boundary for the air pollution
concentration surface estimates. Step 2 transforms the latitude/longitude based monitor data to
the positional grid coordinates of the selected air quality model and develops an analysis data set
to model the air pollution concentration estimates. Step 3 directs the user to specify the air
quality model and the parameters of the simulation. Step 4 launches T-SpACE using a Markov
Chain Monte Carlo (MCMC) simulation. This step produces the file that contains the estimated
-------
air pollution concentration surfaces. Step 5 validates the model results against kriging and the
estimates from one of the air quality models. Step 6 generates a graphical view of the surface.
1.2.2 Expected use of the resulting air quality surfaces
The resulting air quality concentration surfaces are useful to public health researchers,
epidemiologists, air quality assessors, air quality scientists, and those with a basic understanding
of statistical/mathematical modeling.
2.0 Design and nominal operational behavior of the T-SpACE software
The T-SpACE software was designed to easily guide the user through the many steps that were
previously manually implemented utilizing SAS and R
2.1 Design of the software
There were several design trade-offs that were made in the design of this system. The system
was developed from an existing set of SAS programs, an MCMC model executable that required
a particular text structure for the input information and had a specified structure for the output
results, and monitor and model-based concentration data that had a specified structure.
Throughout the software development process, decisions were made to allow the T-SpACE
model to work within the established design constructs, taking into account EPA requirements,
while developing software that was easy for the user.
2.1.1 Transforming monitor data and model-based data to simulation input data
Early in the development of T-SpACE, the following design and operational constructs were
implemented using previous model development work that had been completed:
ฆ Conversion of the Network Common Data Form (NetCDF) model-based
concentration files to the required format for input into the MCMC model executable;
ฆ Conversion of the AQS monitor data to the air quality model-based grid, and
subsequently to the required format for input into the MCMC model executable;
ฆ Adding the latitude and longitude to the MCMC simulation output file.
The decision was made to ensure that the data transformations used for these data types were
implemented according to EPA standards such as utilization of the Lambert projection
transformations and the quality assurance (QA) information from the AQS monitor files.
2.1.1.1 Air Quality Monitoring
Network Monitor Data
For the initial input air quality
monitor data, the user must
supply a text file that is in the
exact same format as EPA's
AQS RD (raw data) monitor
data.
0 Apply Stale or Region to Grid:
Please select a CSV file with State and Region grid information:
C:\Documents and Settings'ฆhartford ' .My Documents -Projects' .EP
-2
Disclaimer:
12 km grid does not cover the entire
Click to see a graphic of 2001 orth
Choose X and Y coordinates of the qrii
v
Virginia
Washington
West Virginia
Wisconsin
Wyoming
North East
NorthCentral
NorthWest
A
a
T 2
V
-------
2.1.1.2 Air Quality Model-Based
Data
For the initial input air quality
model data, the user must supply a
file that is formatted like the
CMAQ NetCDF files. A file
formatted like the CMAQ NetCDF
files is required for model
execution. When the CAMx model
is used in T-SpACE, its format is
converted into the CMAQ NetCDF
file format.
2.1.2 Choosing the parameters for the simulation
The software has been designed to allow the user to choose the parameters for the simulation.
These choices are made in Steps 1 and 3 of the software. The choices the user can make are
described below.
2.1.2.1 Choice of the region for simulation
The software provides users with the ability to identify the appropriate grid coordinates for a
particular region or state.
2.1.2.2 T-SpACE parameter choices
There are several tabs in the software to guide the user through the choices for the model
parameters. By sequentially following the tabs, the user is specifying priors of the model
including the number of control points in the bias spline calculation, the distributional properties
of the mean level of the Conditional Auto Regressive (CAR) process, the vector of coefficient
for the bias, the precision of the measurement error in the monitor observations, the precision of
the measurement error in the computer observations, the precision of the mean process, and the
temporal autocorrelation parameter of the mean process. In addition, the user specifies the
boundaries which define the neighborhood structure of the model, the random seeds for
initializing the simulation, the frequency of sampling during the simulation, the number of initial
model runs (burn-ins) for the simulation, and the total number of simulation loops that will be
completed. Finally, to provide the user information about the simulation, the user can choose to
track the chain of the MCMC simulation.
2.1.3 Results from the simulation
The T-SpACE software stores the results in a tabular (comma-delimited) format. One row exists
in this file for each grid cell within the region and for each day within the time period of interest.
The relevant variables in this file, in the order in which they appear (and are shown in the
column headings), are as follows:
Day
XCoord
YCoord
Longitude
Latitude
PredAvg
PredStd
BiasAvg
BiasStd
CovarAvg
CovarStd
Monitor Data
1/2/2001
41
65
-91.2294
35.24114
3.56199
1.34186
-0.90992
0.940508
-999
-999
-999
1/2/2001
41
66
-91.0884
35.34224
3.55762
1.34159
-0.95024
0.924785
-999
-999
-999
G Hierarchical Bayesian Model - [Step 3: Model Specification] EEIB1
File View Tools
Step 1: Choose Time/Grid | Step 2: Prepare Model Input Data | R1 Step 3: Model Specification | ^ Step 4: Launch Model Step 5: Launch Validation
~ I
| Gnd j Priors Priors Data Boundaries Track Simulation 4th Highest |
I ฆ
1 _
Date of Time Step 1
Month Day Year
| 1| | l| | 20011
Day: Date represented by the data given in this row, in MM/DD/YYYY format.
-------
XCoord: The cell's x-coordinate (east-west) within the grid (where the coordinate value
increases by 1 with each grid cell as you move from west to east).
YCoord: The cell's y-coordinate (north-south) within the grid (where the coordinate
value increases by 1 with each grid cell as you move from south to north).
Longitude: The x-coordinate value transformed to longitude (degrees).
Latitude: The y-coordinate value transformed to latitude (degrees).
PredAvg: The (natural log-transformed) model-predicted value of the mean
concentration for the location given by the specified latitude and longitude.
PredStd: The (natural log-transformed) model-predicted value of the standard error for
the mean predicted concentration.
BiasAvg: Mean of the "bias surface," or the estimate of bias in the CMAQ or CAMx
response surface relative to the monitoring station data (which represents "true"
concentrations).
BiasStd: Standard error of the mean of the bias surface.
CovarAvg: This parameter is obtained from NASA's Moderate Resolution Imaging
Spectroradiometer (MODIS) Aqua and Terra satellites. The number -999 denotes a
missing value. [This is implemented in the Model 6 version of T-SpACE.]
CovarStd: This parameter is obtained from NASA's Moderate Resolution Imaging
Spectroradiometer (MODIS) Aqua and Terra satellites. The number -999 denotes a
missing value. [This is implemented in the Model 6 version of T-SpACE.]
MonitorData: The (natural log-transformed) air pollutant measurement from any air
pollution monitoring station present at the given location (when monitor is present).
For variables with data values of -999, this indicates missing values. Values of -999 for the
variable MonitorData indicate that no monitoring station exists at the location(s) represented in
these (table) rows.
2.1.4 Summary statistics and validation of the resulting air quality surfaces
The user can run validation statistics at the completion of the simulation run. The first sets of
statistics calculated are the distributional summary statistics for each of the model parameters, a
summary of the number of iterations, the mean of the simulated values, and the standard error
observed from the simulated values. In addition, the validation compares the T-SpACE results
with a Krige model that makes predictions solely from monitoring data (using an exponential
variogram approach in which results from daily variograms are combined) and predictions based
solely on the CMAQ (or CAMx) modeling system (i.e., one of the input data sources to the T-
SpACE). The summary statistics provided include the mean square error (MSE) and bias for the
T-SpACE, kriging, and the CMAQ (or CAMx) model surface, the estimated percentage of time
(across all monitoring days and locations) that the calculated prediction intervals (i.e., 95%
credible intervals for the T-SpACE) actually included the observed monitor value, considering
data for those locations in the validation dataset that contain monitors, and the prediction
intervals that represent bounds on the range of measurements that are expected to contain the
(unknown) true value a certain percent of the time (typically, 95%) if the model assumptions are
correct.
2.1.5 Three (consecutive) "year" average of 4th highest concentration surface (Os)
-------
Along with predicting average concentrations within each grid cell on a daily basis, the T-
SpACE has the capability to predict the fourth highest daily maximum 8-hour average ozone
(O3) concentration value that would be expected to occur within the time period of interest (e.g.,
one year) within each grid cell and to average three such surfaces together to create a 3
consecutive "year" average of the 4th highest concentration surface known as the EPA 'design
value' for O3. "Year" is defined as the time frame of interest. The time frame could range from
January 1 to December 31 for a year.
Within a time period of interest (i.e., one year), the T-SpACE software determines the fourth
highest average daily concentration value for ozone (O3) within each grid cell across all
simulated surfaces that the software generates for each day within the time period (as noted in
Section 4.1). The results are averaged and smoothed to generate a single response surface. The
software generates a comma-separated value (CSV) file that contains the results. There are only
four columns included in this CSV file:
XCoord: The cell's x-coordinate (east-west) within the grid (where the coordinate value
increases by 1 with each grid cell as you move from west to east).
YCoord: The cell's y-coordinate (north-south) within the grid (where the coordinate
value increases by 1 with each grid cell as you move from south to north).
PredAvg: The (natural log-transformed) T-SpACE-estimated value of the fourth highest
average daily concentration for the given grid cell.
PredStd: The (natural log-transformed) T-SpACE-estimated value of the standard error
for the fourth highest average daily concentration for the given grid cell.
XCoord YCoord 4HAvg 4HStd
1 1 3.76241 0.173166
1 2 3.8159 0.16131
1 3 3.89272 0.139333
In particular, because the air pollution concentration response surface for the average of the 4th
highest daily maximum 8-hour averaged ozone O3 concentration represents the entire time period
of interest (i.e., one year), a date column is not necessary in this *.CSV file.
2.2 Operational behavior of the software
2.2.1 System requirements
The system requirements for the Hierarchical Bayesian (Air Pollution) Model (Model 5: T-
SpACE) software are as follows:
The target environment is the Windows XP or higher operating system for both the 32
and 64 bit platforms. The software will also work on Windows Vista platform, but was
not developed specifically for this platform.
SAS 9.1.3 or higher is required on the system to run the data validation routines.
For the initial input monitor data, the user must supply a text file that is in the exact same
format as EPA's AQS RD (raw data) monitor data.
-------
For the initial input model based data, the user must supply a file that is formatted like the
CMAQ NetCDF files.
A file formatted like the CMAQ NetCDF files is required for model execution, even if
initial input model based data file already exists.
The user must understand the T-SpACE model to make informed decisions about the
simulation model parameters.
The user will need to disable real-time file scanning anti-virus software. If the user
chooses to run a simulation for a full calendar-year of data and for a full model-based
grid, the output files that are created are large. They are opened and closed several times.
When a file is accessed on a system on which anti-virus scan software is running in real-
time, the anti-virus scan software will lock the T-SpACE simulation, and the T-SpACE
software will not run correctly.
2.2.2 Typical processing times
The computer that was used to determine the operational behavior of T-SpACE was configured
as follows:
Intel Core 2 Duo E6750 Processor, @ 2.66 GHz
3 .48 GB of RAM, @ 2.66 GHz
Windows XP Professional, Service Pack 2
2.2.2.1 Preparing the input data for the simulation (Step 2)
CMAQ Model Input File
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
ozone_2001_12.conc CMAQ input file the time needed was 11 minutes and 16 seconds.
Running a 2001 simulation for PMis using T = (1, 365), X (1, 213), Y (1, 188) and the
pm25_2001_12km.ioapi CMAQ input file the time needed was 11 minutes and 53 seconds.
Monitor Model Input File
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_44201_2001-0.txt monitor input file the time needed was 3 minutes and 8 seconds.
Running a 2001 simulation for PM^s using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_88101_2001-0.txt monitor input file the time needed 1 minute and 51 seconds.
2.2.2.2 Full year/Full grid simulations (Step 4)
Using the ozone files created as described above, the time needed to run the model was as
follows (to the nearest second):
10 burn-in steps and 50 simulation steps:
o Read input files - 35 seconds
o Allocate variables - 1 minute, 15 seconds
o Burn-in steps - 3 minutes, 27 seconds
o Allocate variables - 55 seconds
o T-SvACE model simulation steps 17 minutes, 8 seconds
-------
o Write files - 8 minutes, 35 seconds
o Add latitude and longitude - 7 minutes, 22 seconds
o Write File - 1 minute, 24 seconds
o Create validation chain file - 25 seconds
100 burn-in steps and 2500 simulation steps:
o Read input files - 53 seconds
o Allocate variables - 1 minute, 15 seconds
o Burn-in steps - 3 minutes, 26 seconds
o Allocate variables - 1 minute, 16 seconds
o T-SvACE model simulation steps 14 Hours, 21 minutes, 17 seconds
o Write files - 9 minutes, 52 seconds
o Add latitude and longitude - 7 minutes, 49 seconds
o Write File - 1 minute, 31 seconds
o Create validation chain file - 34 seconds
Using the PJVh.s files created as described above, the time needed to run the model was as
follows (to the nearest second):
10 burn-in steps and 50 simulation steps:
o Read input files - 41 seconds
o Allocate variables - 11 seconds
o Burn-in steps - 3 minutes, 27 seconds
o Allocate variables - 13 seconds
o T-SvACE model simulation steps 17 minutes, 7 seconds
o Write files - 8 minutes, 9 seconds
o Add latitude and longitude - 7 minutes, 45 seconds
o Write File - 29 seconds
o Create validation chain file - 11 seconds
100 burn-in steps and 2500 simulation steps:
o Read input files - 41 seconds
o Allocate variables - 17 seconds
o Burn-in steps - 3 minutes, 25 seconds
o Allocate variables - 24 seconds
o T-SvACE model simulation steps 14 Hours, 16 minutes, 50 seconds
o Write files - 9 minutes, 34 seconds
o Add latitude and longitude - 7 minutes, 52 seconds
o Write File - 32 seconds
o Create validation chain file - 13 seconds
2.2.2.3 Validation (Step 5)
The time needed to validate the model output for the ozone simulation previously described was
as follows (to the nearest second):
10 burn-in steps and 50 simulation steps -14 seconds
100 burn-in steps and 2500 simulation steps - 5 minutes, 44 seconds
The time needed to validate the model output for the PMis simulations previously described was
as follows (to the nearest second):
-------
10 burn-in steps and 50 simulation steps -17 seconds
100 burn-in steps and 2500 simulation steps - 6 minutes, 58 seconds
3.0 Operation of and the statistical basis underlying the operation of the T-SpACE
software model
The Bayesian approach featured by the T-SpACE software model is centered around Bayes'
Theorem, an axiom on statistical probability, first developed by Thomas Bayes and presented
pubically in 1763, but which has received considerable attention in recent years by statisticians
and non-statisticians alike. Specifically, Bayes' Theorem relates how new evidence (data) is
used to update or revise existing knowledge of a given unknown (random) process or model,
thereby improving the ability to make more accurate and precise conclusions using that process
or model, such as predicting the likelihood that something of interest will occur, or obtaining the
best estimate of some unknown parameter such as air pollution levels over a given region. For
example, if A represents such a random process, then existing knowledge of the possible values
that A can hold and their probabilities of occurrence is specified by placing a prior distribution
on A. Note that the selection of the prior distribution of A is subjective. as it is based on existing
knowledge at that time. Now, if B represents a random variable from which values are generated
in an experiment, and the values of B are determined in part by the process or model represented
by A, then Bayes' Theorem indicates that the probability of a specific outcome of A occurring is
proportional to its probability under the prior distribution, multiplied by the likelihood of
observing the particular set of data that was generated from B in the experiment, given that
outcome of A. This updated likelihood is called the posterior distribution for A. The terms
"prior" and "posterior" are introduced because they represent the basic concept behind Bayesian
statistics and the T-SpACE model.
The basic statement of Bayes' Theorem is expanded in order to represent the situation handled
by the T-SpACE model. Specifically, the likelihood that certain events occur can be expressed
as a series of conditional probabilities. For example, if A, B, and C represent three different
types of events, then the probability that A, B, and C occur simultaneously can be expressed as
the product of the following probabilities:
The probability that C occurs.
The probability that B occurs given (or "conditional") that C occurs.
The probability that A occurs given that both B and C occur.
Note that this breakdown of the problem is hierarchical in nature. Therefore, if the joint
probability of {A, B, C} is very difficult or impossible to calculate, such as when it represents a
complex multidimensional process over space and time (as with air pollution), such a
hierarchical breakdown expresses the problem as a series of probabilities that can each be more
easily calculated. While this concept has long been recognized in theoretical statistics, it has
only recently been considered for modeling complicated environmental processes such as that
used by T-SpACE for modeling air pollution levels.
In hierarchical Bayesian modeling, a common setup is to express the primary parameter of
interest ("A" in the example above) as a function of available data possibly from multiple sources
("5"), which in turn can be expressed is a function of some random (often unmeasurable) process
-------
and/or parameters from a model ("C"). (This is a very simplistic and basic expression of a
common setup that can take different forms.) This begins to show why T-SpACE is a
conceptually appealing approach to addressing a wide array of technical problems that are not
easily addressed with standard statistical techniques. T-SpACE allows the user to
Break complex multidimensional problems into simpler, logical, and intuitive
hierarchies;
Combine multiple sources of information, including expert opinion and differing data
sources;
Perform mathematical analyses consistent with the scientific method, and
Implement "learning algorithms" on the unknown entity of interest that are further
refined as more data become available.
In recent years, the rapid decrease in the cost of high performance computer equipment and
significant focused research effort by the statistical community on Markov Chain Monte Carlo
(MCMC) methods for fitting T-SpACE air pollution concentration surfaces (described below)
have changed T-SpACE from a theoretical concept to a practical tool for use in decision sciences
and health exposure assessment.
In this context, the T-SpACE model provides a system to predict air quality data for a specific
time and spatial scale using monitoring and modeling data as input, while minimizing the
limitations either of these methods when applied separately. The primary advantage of the T-
SpACE model is increased flexibility and the ability to predict pollution gradients (and
accompanying uncertainty in the prediction) that might otherwise be unknown using
interpolation results (e.g., from kriging) based solely on relatively sparse monitoring data.
Spatial maps of bias in numerical models are another useful output that will allow modelers to
improve their models to minimize bias. Its major disadvantage is its computational burden.
The T-SpACE model considers the following series of random (unknown) variables and
observed data, each of which is indexed in the range of space and time that is of interest to the
user:
W: (unobservable) log-transformed true concentrations of the pollutant (e.g., PM2.5);
X : (observable) unbiased log-transformed concentration measurements obtained from
monitoring stations;
Y : (observable) biased log-transformed concentration 'measurements' generated by an
air quality (simulation) model;
D : (observable) set of variables that characterize the degree of bias present in the
modeled (simulated) concentration values (relative to their representation of the true
concentrations W); and
C : (observable) covariates associated with the observed concentration measurements.
The first stage of the hierarchical bayesian process in T-SpACE expresses the observed (log-
transformed) concentration measurements X and Y, as (each) originating from normal
distributions with (unknown) mean and variance as follows:
-------
X~N(W,a2x) Y~N(W + Dj8D,a2r)
While measurements originating from monitoring stations are assumed to be unbiased estimates
of the true concentrations at their respective locations, the modeled (simulated) concentrations
are assumed to be biased, as represented by the (unknown) parameters in /fo.
The second stage of T-SpACE modeling expresses the unknown, but true, (log-transformed)
concentrations W, as an unknown overall mean, plus a random process Z, which consists of
multiple components, and has a multivariate normal distribution with a mean of zero, and a
covariance matrix that accounts for the presence of correlation among air pollution
measurements that are collected successively in both space and time. Realizations of the random
process Z for successive points in space and time may be correlated. Therefore, the covariance
matrix for Z is characterized by two (unknown) parameters: o2z and Az. The parameter \). This
abstraction looks simple, however, in reality, the relationship between the components can be
complicated, and the individual components have a large number of dimensions. For example,
the covariate representation (C) is in fact a computational complexity parameter embedded in
the model \fV \ d\. Statistical algorithms to deal with these and other model complexities are
relatively recent, the most notable of them being Markov Chain Monte Carlo (MCMC)
algorithms (e.g., Gilks et. al., 1998; Gelman et. al., 2004).
In the T-SpACE model, the prior distribution (1) is updated through observation of monitoring
data (X) and pollutant data generated by an air quality (simulation) model (7). As noted
above, both are modeled independently given the true pollutant concentrations (W) and some
components of the unknown parameter vector: [X\ W, 6>] and [Y \ W, 6\ Bias present in the
simulated concentration measurements (D) is a complexity factor embedded in the model [Y \ W,
01
The T-SpACE model makes its predictions through characterizing the "posterior distribution"
[W, 6 \X, Y], which captures all available information about the true underlying pollutant
-------
concentrations as well as the parameters governing the relationships among these concentrations
and the observed quantities:
[W,0\X,Y]oz[x,Y\W,0]-[W,0] (2)
3.1 Grid size
Because the air quality model-based data are retrospective 'predictions' (i.e., estimates) across an
entire two-dimensional spatial domain, the set of model predictions at a given point in time
resembles a smooth "surface", when it is graphically displayed, rather than simply a set of points
at various locations. This surface, called a response surface, appears three-dimensional, in that it
covers the two-dimensional spatial region and has a third dimension representing the magnitude
of the predicted concentrations.
The air quality model-based estimates, such as from CMAQ or CAMx, are made relative to a
specified rectangular locational grid that covers the entire region of interest. The grid is
composed of a matrix of rectangular cells. Within each grid cell, the modeling system predicts
average pollution levels at a given point in time. Each cell has equal area which is expressed as
hxv, where h and v denote each cell's horizontal and vertical dimensions, respectively (in
kilometers). Under the CMAQ modeling system, grid cells can either be 36 km x 36 km (which
represents the "parent domain" covering the entire continental US) or 12 km x 12 km (which
represents primarily the eastern and Midwest regions of the US up to 2007, and the entire US
from 2008 and onward). Thus, as the values of h and v decrease for a fixed area, the CMAQ
modeling system generates a greater density of predictions, which increases the resolution of the
area's prediction surface. Currently, EPA recommends that locational grid resolutions not be
smaller than a 12 km x 12 km resolution, as any smaller/finer resolution (e.g., 4 km x 4 km,
which has been used to characterize urban core areas), when applied to a large region, can result
in unacceptably high uncertainties in emissions and meteorological information.
The user is asked to specify the lower x and y and upper x and y coordinates of the grid over
which the simulation will be run. As mentioned earlier, the user is provided guidance on the x,
y-coordinates that cover specific regions of the US.
3.2 Timeframe
T represents the time frame over which the simulation will be run. The study period in this
version of the software is limited to one calendar year. The user cannot choose a time frame that
spans or splits between two different calendar years. The value T represents the number of days
(i.e., Julian [calendar] Days) from January 1 to the value T that is chosen, where January 1 is day
number 1.
3.3 Bias Spline
The bias is evaluated as a linear combination of 2nd order uniform bias spline functions defined
over a 3-dimensional lattice of uniform knots. Using bias splines as basis functions for the bias
allows the user the ability to control the degrees of freedom of the bias through the number of
control points. Furthermore, the piece-wise nature of the bias spline functions respects the
principle of locality, that is, local information does not affect regions far from the region where
-------
the local information is defined. On the numerical side, bias splines allow a tensor factorization
of the bias matrix into three matrixes B. ฆ = h , (z'r), r = 1,2,3 for a total dimensionality of
jVjMj + N2M2 + N3M3, very much less than the total dimension of the full D-matrix, which is
NXMX ฆ N2M2 ฆ N3M3.
Guidance on choosing the bias spline parameters and the number o f control points.
Variable
Bias Spline Min
Bias Spline Max
Number of Control Points
T
First day of the time range over which the bias
will be calculated. By default, this is the first
day of the year (i.e., Jan 1st).
Last day of the time range
over which the bias will be
calculated. By default, this is
the last day of the year (i.e.,
Dec 31st).
This defines the degrees of
freedom in the model for time.
By default, 4 is chosen
representing a control point for
each season of the year.
X
Represents the min x-coordinate of the grid
over which the bias spline will be calculated.
By default, and recommended, the min x-
coordinate is 1, for the 12 km grid. If the user
is using the 36 km grid, then the default should
be 13.
Represents the maximum x-
coordinate of the grid over
which the bias spline will be
calculated. By default, and
recommended, the maximum
possible x-coordinate for the
12 km is given, 213. If the
user is working with the 36 km
grid, then this should be 142.
This is the degrees of freedom
for the x-coordinate in the bias
function. 8 is the default entry.
Please note that if this number
gets too high the performance
of the simulation may be slow.
Y
Represents the min y-coordinate of the grid
over which the bias spline will be calculated.
By default, and recommended, the min x-
coordinate is 1, for the 12 km grid. If the user
is using the 36 km grid, then the default should
be 15.
Represents the maximum y-
coordinate of the grid over
which the bias spline will be
calculated. By default, and
recommended, the maximum
possible x-coordinate for the
12 km is given, 188. If the
user is working with the 36 km
grid, then this should be 94.
This is the degrees of freedom
for the x-coordinate in the bias
function. 7 is the default entry.
Please note that if this number
gets too high the performance
of the simulation may be slow.
3.4 Control Points
The following should be considered when choosing the number of control points:
o Since the number of control points is roughly equal to the number of degrees of freedom
of the simulation, the user wants to be conscious of not over- or under-fitting the model
by choosing too many or too few control points
o For T, time, the number of control points has been chosen to represent the four seasons.
If the user believes there to be a monthly bias in the concentrations, then they may choose
this value to be 12.
o For X and Y, the number of control points has been chosen to represent an estimated rate
on which bias changes in each direction. A metric that may be used to choose the
number of control points for X and Y could be the length of the grid over which the
simulation is to run divided by the average scale of the spatial bias (every 25 grid cells).
It is recommended that, in general, there should be less than 100 control points in the x-y
coordinate system.
o Finally, it is recommended that a user of the system run sensitivity analyses to understand
how sensitive the surface predictions are to the number of control points.
3.5 Priors
-------
In Bayesian modeling, unknown model parameters are assumed to represent random quantities
that originate from some underlying distribution. In the absence of any new data, the modeler
selects a "prior distribution" for each parameter based on existing knowledge or their 'best
judgment'. In the T-SpACE model, prior distributions are assigned as follows (to the model
parameters introduced earlier):
For the unknown variance parameters g\, al, and o\, their reciprocals are known as
rl, and t\ respectively (where ris the Greek letter "tau"), and are each assigned a
gamma distribution (with shape and scale parameters specified by the user).
The parameter D and the unknown "slope" parameters fiu are each assigned a normal
distribution (with mean and variance specified by the user)
The temporal correlation parameter Az is assigned a uniform distribution (with lower and
upper bound specified by the user).
3.6 Neighborhood Boundary
The neighborhood boundary defines the
neighborhood structure of the model. There are four
possible neighborhood structures that are available
for the simulation, first through fourth order
neighborhoods. As seen in the figure, as the order
increases, the number of neighbors used in the
calculation increases. The default in the software is a
first order neighborhood.
3.7 Burn-in/Loop
Users of the T-SpACE software encounter the
reference to a "burn-in" period for the MCMC. This
signifies the number of initial model runs, in the iterative process, whose results are discarded
from consideration in the model computations, as they are assumed to represent estimates that
deviate considerably from the final posterior distribution. The ideal situation is to construct an
MCMC algorithm that has a short burn-in period and yields stable 'draws' (selections) from the
posterior distribution.
3.8 Markov Chain Monte Carlo (MCMC) Simulation
Because it is not possible to obtain a closed form of a posterior distribution on the air pollutant
concentrations (i.e., fitting the T-SpACE model) using standard statistical modeling techniques,
an alternative, computationally-intensive procedure is needed to obtain a good approximation of
the posterior distribution. This procedure involves performing a computer simulation centered
on a Markov Chain process in an effort to converge numerically to the proper distribution. A
series of random variables {Xi, X2, ..., Xn} representing some ordered process (e.g., values over
time) is known as & Markov Chain if the probability that one of the variables (Xi) holds a specific
value depends only on the value of the variable that immediately precedes it (X-i) and is
independent of the values of all other prior variables (X, ..., X-2). For example, if you are about
to roll a die to determine how many spaces to advance in a board game, the probability that you
will land on a particular space as a result of that roll is dependent only on where you currently
First order neighborhood
Second order neighborhood
Third order neighborhood
Fourth order neighborhood
-------
are on the board, and not on the outcome of any prior rolls that determined how you got to your
current position. This is a simple example of a Markov Chain process.
In the MCMC process used in T-SpACE, the parameters of the posterior distribution [W, 9 | X,
Y] are random variables that must be estimated. They are assigned initial estimates (based on
the assigned 'prior' distribution), which are then repeatedly updated in a linear, stage-wise
Markov Chain process, where the estimates in one stage of the process depend only on the
estimates in the prior stage. The stage-wise process allows for the estimates to slowly improve
until a state of "stationarity" is achieved, that is when the estimates have converged to a final set
of values, and no further significant improvement can be made. In statistics, the term "Monte
Carlo" refers to the use of random simulation techniques to characterize some unknown physical
or mathematical system. The Monte Carlo component of the MCMC is the series of simulation
runs in which one moves in a probabilistic fashion from one stage to the next in the Markov
Chain process to get to the final posterior distribution. The most widely used simulation
technique in MCMC is known as the Gibbs sampler (e.g., Gelfand and Smith, 1990). To
perform MCMC, the T-SpACE model uses an algorithm based on the Gibbs Sampler. In this
algorithm, the posterior distribution [W, 9 | X, Y] is sampled by simulating successively from the
steps:
[w\e,x,r]
[e\ w,x,y]
Each step is conditioned on the latest values obtained from the previous step. These distributions
are referred to as the 'full conditional' distributions. When these full conditional distributions
are not in a recognizable form and can only be calculated up to a normalizing constant, the
algorithm incorporates a,Metropolis-type step (Robert and Casella, 2004). Software such as
WinBugs is available to perform MCMC techniques, although specially-prepared software has
been developed to handle the MCMC necessary to fit the T-SpACE model.
4.0 CMAQ and CAMx Air Quality Models Used in T-SpACE
4.1 CAMx model description
The Comprehensive Air Quality Model with Extensions (CAMx) is an Eulerian (gridded)
photochemical dispersion model that allows for integrated "one-atmosphere" assessments of
tropospheric air pollution (e.g., ozone, particulates, air toxics) over spatial scales ranging from
neighborhoods (local) to continents (global). It is a modern, open-source system that is
computationally efficient, flexible, and pubically available. CAMx's Fortran source code is
modular and well-documented. The Fortran binary input/output file formats are based on the
Urban Airshed Model (UAM) convention, and are compatible with many existing pre- and
post-processing tools. Meteorological input fields are supplied to CAMx from separate weather
prediction models. CAMx specifically supports the Weather Research and Forecasting (WRF)
model, the Mesoscale Model 5 (MM5) and the Regional Atmospheric Modeling System
(RAMS). All emission inputs are supplied from external pre-processing systems (e.g., the
Sparse Matrix Operator Kernel (SMOKE) system and the Emissions Processor System Version 3
(EPS3).
-------
CAMx simulates the emission, dispersion, chemical reaction, and removal of pollutants by
'marching' the Eulerian continuity equation forward in time (t) for each chemical species (1) on a
system of nested three-dimensional grids. The continuity equation specifically describes the
time dependency of volume-average species concentration within each grid cell as a sum of all
physical and chemical processes operating on that volume. CAMx can perform simulations on
four types of Cartesian map projections: Lambert Conic Conformal, Polar Stereographic,
Mercator, and Universal Transverse Mercator (UTM). CAMx also offers the option of operating
on a geodetic latitude/longitude grid system. The vertical grid structure is defined externally, so
layer interface heights may be specified as any arbitrary function of space and/or time. This
flexibility in defining the horizontal and vertical grid structures allows CAMx to be configured to
match the grid of any meteorological model that is used to provide environmental input fields.
Detailed information on the CAMx system can be found at: http://www.camx.com.
4.2 CMAQ model description
The Models-3/Community Multiscale Air Quality (CMAQ) modeling system goals are to
improve: 1) the environmental management community's ability to evaluate the impact of air
quality management practices for multiple pollutants at multiple scales, and; 2) the scientist's
ability to better probe, understand, and simulate chemical and physical interactions in the
atmosphere. Traditionally, the CMAQ modeling system has been used to predict air quality
across an entire regional or national domain, and then to simulate the effects of various changes
in emission levels for policy-making purposes. However, for health studies, it is frequently
applied to provide supplemental information on predicted air quality in areas where few or no
monitors exist. It also accounts for meteorological conditions to better evaluate health outcomes.
Detailed information on the CMAQ modeling system is available at https://www.epa.gov/cmaq
and at https://www.cmascenter.org.
Because the CMAQ modeling system makes predictions across an entire two-dimensional
domain, the set of model predictions at a given point in time resembles a smooth "surface" when
it is graphically displayed, rather than simply a set of points at various locations. This surface,
called a response surface, appears three-dimensional in that it covers the two-dimensional region
and has a third dimension representing the magnitude of the predicted concentrations. The
CMAQ model predictions are made relative to a specified rectangular grid that covers the entire
region of interest. The grid is composed of a matrix of rectangular cells. Within each cell, the
CMAQ modeling system predicts average pollution levels at a given point in time. Each cell has
equal area which is expressed as h x v, where h and v denote each cell's horizontal and vertical
dimensions, respectively (in kilometers). Under the CMAQ modeling system, grid cells can
either be 36 km x 36 km (which represents the "parent domain" covering the entire continental
US) or 12 km x 12 km (which represents primarily the eastern and Midwest regions of the US up
to 2007, and the entire US from 2008 and onward). Thus, as the values of h and v decrease for a
fixed area, the CMAQ modeling system generates a greater density of predictions, which
increases the resolution of the area's prediction surface.
-------
5.0 References
[1] T.W. Tesche, Ralph Morris, Gail Tonnesen, Dennis McNally, James Boylan, Patricia
Brewer, "CMAQ/CAMx annual 2002 performance evaluation over the eastern US",
Atmospheric Environment 40 (2006) 4906-4919.
[2] Jinyou Liang, Philip T. Martien, Su-Tzai Soong, and Saffet Tanrikulu, A Photochemical
Model Comparison Study: CAMx and CMAQ Performance in Central California,
Proceedings of the Thirteenth Conference on the Application of Air Pollution
Meteorology with the Air and Waste Management Association, Vancouver Canada;
American Meteorological Society, Boston (2004)
[3] McMillan, N.J., Coutant, B.W., Morara, M., Young, G.S., and Zhou, J. (2005).
"Combining Monitored and Modeled PM2.5 and Satellite Aerosol Optical Depth
Information Across Space and Time Using Bayesian Modeling" Technical Memorandum
from Battelle to US EPA, Office of Air Quality Planning and Standards, Contract No.
68-D-98-030, Work Assignment 3-06.
[4] User's Guide: Hierarchical Bayesian Space-Time Modeling of Air Pollution Data. April
2009. Version 5_4h.0.1.21.
[5] US EPA Final Report: Overview of EPA's Hierarchical Bayesian Model For Predicting
Air Quality Patterns In The United States Over Space And Time, For Use With Public
Health Tracking Data. Contract No. EP-D-04-068, Work Assignment 54, Task 1, April
15, 2009.
[6] Draft Final Report: Hierarchical Bayesian Model Software Architecture/Design
Document. April 30, 2009. Contract No. EP-D-04-068, Work Assignment 54, Task 4.
[7] USEPA, Draft Report Hierarchical Bayesian Model Evaluation For Ozone Data, April
30, 2009, Contract No. EP-D-04-068, Work Assignment 54.
[8] User's Guide: Comprehensive Air Quality Model With Extensions Version 6.40,
Ramboll Environ, December 2016, http://www.camx.com/files/camxusersguide v6-
40.pdf.
-------
Appendix A
-------
TEMPORAL-SPATIAL AMBIENT CONCENTRATOR
ESTIMATOR (T-SPACE) SOFTWARE ARCHITECTURE/DESIGN
1. INTRODUCTION
1.1 Software Overview
1.1.1 Description
EPA has developed a method to combine air pollution monitoring network data and simulation
model output to maximize the advantages offered by both data sources while minimizing the
disadvantages of each when predicting fine particulate matter (PM2.5) and ozone (O3)
concentrations. Air quality (simulation) models estimate the spatial and temporal gradients of air
pollution based on emissions inventories and meteorological information. These models, while
providing estimates over large regions at relatively low cost, have been found to have statistical
bias and have greater errors than air pollution monitoring networks. Air pollution monitoring
networks generally provide relatively accurate, unbiased results. However, because monitoring
networks are sparsely and irregularly spaced over large spatial domains, they are not able to
produce large-scale predictions.
EPA developed the T-SpACE model that combines monitoring data from the 2,545 National Air
Monitoring Stations/State and Local Air Monitoring Stations (NAMS/SLAMS) with estimated
air pollution concentrations from air quality models such as the Models-3/Community Multiscale
Air Quality (CMAQ) and the Comprehensive Air Quality Model with Extensions (CAMx),
which maximizes the advantages offered by both data sources, while minimizing the
disadvantages of each.
1.1.2 Purpose
The T-SpACE model requires data preparation and model parameter input from the user to run a
simulation to produce an estimated air pollution concentration surface. To streamline the model
set-up, the user is allowed to prepare the input data for the model run, choose the model
specifications, run the model, receive output files containing the estimated air pollution
concentration surface, and run a set of validation procedures to compare the T-SpACE model run
with a standard kriging method, and the model-based air pollution output concentrations from
CMAQ or CAMx. To minimize the amount of work for the user, a graphical user interface (GUI)
"wrapper" was developed that provides a single interface for configuring and running a
simulation. The user is asked for particular choices for input data sets, model parameters, and
names for the output files, and is guided through the simulation.
-------
1.1.3 Scope
This document was generated to provide a detailed description of the architecture of the
Hierarchical Bayesian Space-Time Model for Air Pollution, Version 5-4h.0.1.21 (T-SpACE). In
particular, this document provides information on the physical structure of the source code, the
logical structure of the classes, the input data, the output data, and the expected behavior of the
application.
1.2 Glossary
1.2.1 Definitions
Abstract Class An object-oriented programming class that provides basic
methods (functions/procedures) and data attributes
(variables/parameters) which are specifically defined in
lower-level classes through the mechanism of inheritance.
This is a 'template' for building a functional class.
Application Programming Interface A set of protocols for using a programming library.
Assembly A compiled low-level machine code (library).
Base Class A functional object-oriented programming class that is
derived from an abstract object-oriented programming
class. It provides defined methods (functions/procedures)
and data attributes (variables/parameters) which can be
inherited in lower level (child[ren]) classes. This is a
'template' for building an 'object' (object = and
'instance'/working example of the base class).
Bitmap Image file format for storing digital images.
C# An object-oriented programming language developed by
Microsoft.
Console Application A program that has no user interface. It is run from the
computer command line.
DOS Batch File A Disk Operating System (DOS) file that contains a series
of command line instructions to be run in sequence.
DOS Short File Name
A file name compatible with the DOS operating system,
allowing up to 8 characters in the file name, 3 characters in
the extension and a limited allowable character set.
-------
Dynamic-Linking Library
A shared library for use on the Windows platform. Files in
the Dynamic-Linking Library are designated by the '.dll'
extension.
Kriging
A geostatistical method that uses interpolation to develop a
response (air pollution concentration) surface from the
nearest set of monitors.
.NET Framework
Singleton Class
Static Method
Visual Basic
Visual Studio
Windows Console Project
Windows Deployment Project
Windows Forms Project
Windows Library Project
Windows Long File Name
A software framework for the Windows platform.
A object-oriented programming class that generates a
single instance (object) of itself at any given time.
A method that is shared by all instances of a class. It can
be invoked without creating an instance of the class.
An object-oriented programming language.
An integrated development environment used for
programming in C# and Visual Basic.
A Visual Studio project for creating a console application.
A Visual Studio project for creating an installation package
for an application.
A Visual Studio project for creating a Windows application
with a GUI (Graphical User Interface).
A Visual Studio project for creating a DLL.
A file name that permits up to 255 characters with an
expanded allowable character set.
1.2.2 Acronyms
API Application Programming Interface
AQS Air Quality System
CAMx Comprehensive Air Quality Model with Extensions
CASTNET Clean Air Status and Trends Network
CMAQ Community Multiscale Air Quality
-------
CSV
The file extension for a Comma Separated Value file
DLL
The file extension for a Dynamic-Link Library file
DOS
Disk Operating System
GUI
Graphical User Interface
HBM
Hierarchical Bayesian Model
IOAPI
Input/Output Application Programming Interface
MCMC
Markov Chain Monte Carlo
MSI
The file extension for a Microsoft Installer file
NAMS
National Air Monitoring Stations
NetCDF
Network Common Data Format
OOP
Object-Oriented Programming
POC
Parameter Object Code
PNG
The file extension for a Portable Network Graphic file
RAM
Random Access Memory
SAS
Statistical Analysis System
SLAMS
State and Local Air Monitoring Stations
TauX
The precision of the measurement error in the air quality monitor observations
TauY
The precision of the measurement error in the air quality model 'observations'
TauZ
The precision of the mean process
T-SpACE
Temporal-Spatial Ambient Concentrator Estimator
TSV
The file extension for a Tab-Separated Value file
UC
User Controls
1.2.3
Abbreviations
-------
BMP The file extension for a Bitmap file
EXE The file extension for an Executable file
GHz Gigahertz
ICO The file extension for an Icon file
K Kilobyte. IK = 1,024 bytes
KM Kilometer. 1KM is approximately 0.625 miles
MB Megabyte. 1MB = 1,024K = 1,048,576 bytes
O3 Ozone (pollutant)
03 Ozone (file-name)
PM2.5 Fine Particulate Matter less than 2.5 microns in (aerodynamic) diameter
TXT The file extension for a Text file
1.3 References
[1] McMillan, N.J., Holland, D.M., Morara, M., and Feng, J., Combining different sources of
particulate data using Bayesian space-time modeling, Environmetrics, 2010, Volume 21,
pp 48 - 65, DOI: 10.1002/env.984
[2] USEPA, Draft Report Overview of EPA's Hierarchical Bayesian Model For Predicting
Air Quality Patterns In The United States Over Space And Time, For Use With Public
Health Tracking Data, March 13, 2009, Contract No. EP-D-04-068, Work
Assignment 54.
[3] User's Guide: Hierarchical Bayesian Space-Time Modeling of Air Pollution Data, March
2009, Version 5-4h.0.1.21.
-------
2. DESIGN DESCRIPTIONS
This chapter discusses the design of the T-SpACE Model software. Software design is the
process of problem-solving and planning for a software solution.
2.1 Design Overview
This section provides an overview of the design of the T-SpACE Model software. Each
subsection details an aspect of the design.
2.1.1 Requirements
The requirements for the T-SpACE Model software are as follows:
The target environment is the Windows XP, or higher operating system for both the 32-
and 64-bit platforms. The software will work on the Windows Vista platform, but was
not developed specifically for this platform.
SAS 9.1.3 or higher is required on the system to run the data validation routines.
For the initial input air quality monitor data, the user must supply a text file that is in the
exact same format as EPA's AQS RD (raw data) monitor data.
For the initial input air quality model data, the user must supply a file that is formatted
like the CMAQ NetCDF files.
A file formatted like the CMAQ NetCDF files is required for model execution. When the
CAMx model is used in T-SpACE, its format is converted into the CMAQ NetCDF file
format by T-SpACE.
The user must understand the T-SpACE model to make informed decisions about the
simulation model parameters. There are three documents that can help with this
understanding: [1], [2], [3],
The user will need to disable real-time file scanning/anti-virus software. If the user
chooses to run a simulation for a full calendar-year of data for a full model-based grid,
the output files that are created are large. They are opened and closed several times.
When a file is accessed on a system on which file scanning/anti-virus software is running
in real-time, the file scanning/anti-virus software will 'lock' the T-SpACE simulation
file, and the T-SpACE model software will not run correctly.
2.1.2 Tradeoffs
There were several design trade-offs that were made in the design of this system. The system
was developed from an existing set of SAS programs, a Markov Chain Monte Carlo (MCMC)
model executable that required a particular structure for the input information and had a specified
structure for the output results, and monitor and model-based concentration data that had a
-------
specified structure. Throughout the development process, decisions were made to allow the T-
SpACE model to work within the established design constructs, and implement EPA
requirements. The system design trade-offs included the following:
~~~ A decision was made to have EPA scientists develop the algorithms for:
ฆ Conversion of the NetCDF model-based concentration files to the required format for
input into the MCMC model executable;
ฆ Conversion of the AQS monitor data to the air quality model-based grid, and
subsequently to the required format for input into the MCMC model executable;
ฆ Adding the latitude and longitude to the MCMC simulation output file.
The decision was made to ensure that the data transformations used for these data types
were implemented according to EPA standards. The code that performs the data
transformations had been previously designed by EPA, and was easily updated. When
changes to the format of the input or output data was made during the design process,
additional routines were written to open up the files, make the requested changes, and
close the files. This led to the issue with the real-time, virus scanning software
mentioned above.
The designed model processes are pollutant-specific, limiting the simulations to PM2.5
and ozone. For instance, there are two separate executables to process ozone and PM2.5
CMAQ or CAMx data. Additional air pollutants will require additional simulation
algorithms and associated components.
~> Utilization of SAS and the previously developed SAS validation programs to run the
validation routine.
ฆ This decision requires that any computer system that will be using the validation
portion of the software must have SAS installed on that system.
This decision was made because the statistical functions in SAS have been validated, and
the previously developed SAS validation programs for T-SpACE have been verified.
2.2 User Interface
Each subsection provides a brief description of the forms that were developed as part of the T-
SpACE user interface. The subsection headings represent the name of the form, visible in the
form's title bar.
2.2.1 Hierarchical Bayesian Melding
This form acts as a container for the forms needed to prepare, run and validate simulations. This
form also contains the buttons used to walk through the modeling process.
2.2.2 Step 1: Choose Time/Grid
This form is used for Step 1 of the process, where the user specifies the temporal and spatial
grids for the simulation, as well as the folder that will hold the files created in Step 2.
-------
2.2.3 Step 2: Prepare Model Input Data
This form is used for Step 2 of the process, where the user generates the air quality model-based
and air pollution monitor files that will be input to the simulation. The raw air quality model and
air pollution monitor data files are transformed into the required format in this step.
2.2.4 Step 3: Model Specification
This form is used for Step 3 of the process, where the user chooses the simulation parameters.
2.2.5 Step 4: Launch Model
This step does not have, nor does it require, a configurable GUI/'form' setup process for the user
to select/modify simulation parameters. All of the necessary configuration setup steps for the
model run (implemented in Step 4) have been performed in the previous 3 Steps (and other
associated pre-Step 4 preparation/activities). When the Step 4 button is pushed, a DOS window
opens and at this point, the T-SpACE model simulation run begins.
2.2.6 Step 5: Launch Validation
This form is used for Step 5 of the process, where the user specifies the kriging and network files
used to validate the simulation output.
2.2.7 Average 4th Highest Surface Files
This form is where the user specifies three consecutive 4th highest (daily maximum 8-hour
average ozone O3 concentration) surface files to average into a single ozone concentration
surface file.
2.2.8 File Compare
This form is where the user specifies 2 AQS monitor files, or 2 air sites (airsite.tsv) files, to
compare in order to determine differences in monitor locations.
2.2.9 Apply States to Grid
This form is where the user can create a state grid file by applying state and region definitions to
cells in a (locational) grid file.
2.2.10 Step 6: Visualize Surface
Step 6 of the process is where the user generates a three-dimensional (3-D: latitude, longitude,
concentration) graphical representation of the air pollution concentration surface generated in T-
SpACE.
2.3 System Design
The system design section provides a detailed listing of the physical structure of the application's
source code modules and support files. There are two components: Projects and Other Files.
The projects listed below are each a set of source code modules that compile into a single
executable or DLL.
-------
The other files are those files developed elsewhere that are required to run the software.
2.3.1 Projects
The subsection heading is the project name as found in the source code. Each module in the
project is listed with a brief description.
2.3.1.1 BayesianMelding
This is a Windows Forms project written in Visual Basic 2005. The output of this project,
GUI 2.1.exe, is referenced by the main application. BayesianMelding is used to provide the user
with a friendly interface with which to maintain the cryptic parameter files used by the model.
Modules:
frmParams.vb - The form as well as all logic needed to verify input, load parameter
files and save parameter files.
StringTokenizer.vb - Class used to parse (split) strings into individual tokens based on a
specified delimiter.
2.3 .1.2 BayesianMonitorCheck
This is a Windows Console project written in C# 2005. The output of this project,
BayesianMonitorCheck.exe, is located in the application folder. This application is used to
post-process the monitor PM2.5 file (containing PM2.5 monitor concentration values) created in
Step 2 so that it can be used in the model. This post-processing consists of:
Removing any data record that has a negative value for row and/or column
Replaces "m" in the logpm column with an empty string (this represents a missing value)
Modules:
Program.cs - Contains the program's entry point and all logic needed to complete the
processing
2.3.1.3 CMAQColumnRenamer
This is a Windows Console project written in C# 2005. The output of this project,
CMAQColumnRenamer.exe, is located in the application folder. This application is used to
post-process the CMAQ or CAMx (Ozone and PM2.5) files created in Step 2 so that the column
names are consistent. These column names are:
Day
Column
Row
CMAQ log (ozone)/CMAQ log (pm) - This is a generic designation that incorporates:
CAMx log (ozone)/CAMx log (pm)
Modules:
Program.cs - Contains the program's entry point and all logic needed to complete the
processing.
-------
2.3.1.4 CMA QPostProcessor
This is a Windows Console project written in C# 2005. The output of this project,
CMAQPostProcessor.exe, is located in the application folder. This application is used to post-
process the CMAQ PM2.5 file created in Step 2 so that it can be used in the model. This post-
processing consists of:
Replaces the value in the average column with the natural log of the average.
Modules:
Program.cs - Contains the program's entry point and all logic needed to complete the
processing.
2.3.1.5 CreateValidationChainFile
This is a Windows Console project written in C# 2005. The output of this project,
ValidationChainFileCreator.exe, is located in the application folder. This application is used
to post-process the chain file created in Step 4 so that it can be used in the validation step. This
post-processing consists of:
Removes all columns except for TauX, TauY, and TauZ
Modules:
Program.cs - Contains the program's entry point and all logic needed to complete the
processing.
2.3.1.6 EPAWA41 State
This is a Windows Library project written in C# 2005. The output of this project,
EP AW A41 State.dll, is a reference of the main application. This class serves as a repository for
values that are needed throughout program execution. These values are accessible as properties
of the WA41_StateTracker object.
Modules:
WA41_StateTracker.cs - Contains the StateTracker's (WA41_StateTracker object)
properties as well as the methods/procedures that access these properties.
DosLongToShortPathConverter.es - Contains static methods that:
o Convert a file's Windows long file name to its DOS short file name
o Convert a file's DOS short file name to its Windows long file name
2.3 .1.7 MonitorColumnRenamer
This is a Windows Console project written in C# 2005. The output of this project,
MonitorColumnRenamer.exe, is located in the application folder. This application is used to
post-process the Monitor (Ozone and PM2.5) files created in Step 2 so that the column names are
consistent. These column names are:
Day
Column
Row
monitor log (ozone) / monitor log (pm)
-------
Modules:
Program.cs - Contains the program's entry point and all logic needed to complete the
processing.
-------
2.3.1.8 Bayesian Air Pollution Model
This is a Windows Deployment project written in Visual Studio 2005. The output of this project,
Bayesian Air Pollution Model.msi and setup.exe, are used to install the application on the
target computer.
It has the following detected dependencies (i.e., files that must be already generated/present and
contain the requisite data required to initiate the process):
.NET Framework 2.0
EPAWA41State.dll
GUI 2.1.exe
2.3.1.9 EPAWA41
This is a Windows Forms project written in C# 2005. The output of this project,
EPAWA41.exe, is the main application. The EPAWA41 application is used to provide the user
with a single interface to execute the steps needed to run a simulation. Since this project has
several modules, multiple subfolders have been created for organization. The module list is
provided by subfolder.
Modules:
Application folder
o program.cs - Contains the program's entry point and launches the main form.
BatchFileProcessors subfolder
o BatchFileCreator.es - Class for creating DOS batch files needed in Step 2.
o BatchFileCreatorAddLL.es - Class for creating DOS batch files needed to
execute ADDLL.exe (Add Latitude and Longitude executable file).
o BatchRunner.es - Class for executing DOS batch files.
CSYFileProcessors subfolder
o CSVProducer.es - Abstract class used as a basis for the batch file creator classes.
o CSVProducerNetCDF.es - Class that contains the interoperability functions that
allows the application to use the NetCDF API library.
o IOAPILatLonInserter.es - Class that inserts latitude and longitude information
from CMAQ/CAMx 10API files into the CMAQ/CAMx and monitor files created
in Step 2.
o RowColToLatLongConverter.es - Class that inserts latitude and longitude
information from CMAQ/CAMx 10API files into the surface file created by the
model.
o TextLatLonInserter.es - Class that inserts latitude and longitude information
from a cells text file (cells*.txt) into the CMAQ/CAMx and monitor files created
in Step 2.
Forms subfolder
o Average4thHighest - Form that allows the user to select three consecutive,
yearly 4th Highest Surface files and average them into a single concentration
surface file.
o FileCompare - Form that allows the user to compare two AQS monitor files or
two site TSV files to see which monitor site locations are missing from one of the
files.
-------
o Main - This is the application's main form, the one that opens when the
application starts.
o Step20zone - Form that allows the user to create the CMAQ/CAMx and/or
monitor file for ozone. This is a Step 2 element,
o Step2PM25 - Form that allows the user to create the CMAQ/CAMx and/or
monitor file for PM2.5. This is a Step 2 element,
o Validation - Form that allows the user to run a validation SAS program against
the model output. This is a Step 5 element.
Welcomelntro - Form that allows the user to specify the time period and
geographic area of interest for the simulation. This is a Step 1 element.
Resources subfolder
o CSVFile3.bmp - Bitmap representing a CSV file,
o Earth.bmp - Bitmap representing the Earth,
o Earth.ico - Application's icon.
o GreenArrow.bmp - Bitmap representing a green arrow,
o GreenGoAirowBitmap.bmp - Bitmap representing a large green arrow,
o GridToEarthModify.bmp - Bitmap representing the application of a grid to the
Earth.
o GridToEarthModifyPNG.png - PNG representing the application of a grid to
the Earth.
o Home2.bmp - Bitmap representing a home,
o HomeInPS.bmp - Bitmap representing a home,
o OpenFile.bmp - Bitmap representing the opening of a file,
o OpenFolder.png - Bitmap representing the opening of a folder,
o PM25.bmp - Bitmap representing the PM2.5 air pollutant,
o PM25Red.bmp - Bitmap representing the PM2.5 air pollutant,
o SmallDisk.bmp - Bitmap representing a diskette.
Tools subfolder
o AQSFileComparer.es - Class that compares two AQS monitor files,
o FileComparer.es - Base class that provides functionality for all file comparers,
o MonitorSite.es - Class that represents a monitor site from an AQS monitor file or
an air pollution monitor site TSV file,
o SurfaeeCell.es - Class that represents a cell from a surface file,
o SurfaeeFileAverager.es - Class that averages 4th highest surface files,
o TSVFileComparer.es - Class that compares two sites TSV files.
UserControls subfolder
o CalendarRangeSelector - Contains the two calendar controls used in Step 1.
o TabPageGoBetween - Container for the two tabs in the Step 2 form,
o UC-CMAQ - The CMAQ/CAMx interface for Step 2.
o UC-IntroScreen - The interface for Step 1.
o UC-Monitor - The monitor interface for Step 2.
Utility subfolder
o AppSettings.es - Class that saves application settings between program sessions,
o TextFileCreator.es - Class for saving batch file output to text files,
o ValidationAppSettings.es - Class that saves validation application settings
between program sessions.
-------
2.3.2 Other Files
This section lists the files that are written to the computer when T-SpACE is installed. These
files are different from the output of the project files listed previously. This list is presented by
folder. A description of each file is provided. The following files are placed in the application
folder upon installation of the T-SpACE software. The application folder uses 45.8 MB of disk
space.
Application Folder:
addll .exe - Add latitude and longitude (addll) executable provided by the EPA. It inserts
latitude and longitude information in the air pollution concentration surface file created
by the T-SpACE software. This is automatically executed when the model completes.
airsites.tsv - Tab-delimited file provided by the EPA. It contains the locations of the
PM2.5 monitors. This is used by monpm25.exe.
AqVis4.exe - Surface visualization software.
cmaqpm25.exe - Executable provided by the EPA. It creates the CMAQ file for PM2.5
in Step 2.
cmaqpm2503.exe - Executable provided by the EPA. It creates the CMAQ file for
ozone in Step 2.
Model5-4H.0.1.19_Mc2_Win32.exe - Model executable for the 32-bit platform. This is
executed in Step 4.
Model5-4H.0.1.19_Mc2_Win64.exe - Model executable for the 64-bit platform. This is
executed in Step 4.
mon03.exe - Executable provided by the EPA. It creates the monitor file for ozone in
Step 2.
mon03airsites.tsv - Tab-delimited file provided by the EPA. It contains the locations of
the ozone monitors. This is used by mon03.exe
monpm25.exe - Executable provided by the EPA. It creates the monitor file for PM2.5 in
Step 2.
netcdf.dll - API for accessing NetCDF format data files.
tz.csv - Comma-delimited file provided by the EPA. It contains time zone information.
It is used by cmaqpm25.exe and cmaqpm2503.exe.
ValidationTemplate.sas - This is a template of a SAS program that is used to validate
the model output files. The template is completed with user selected files and then
executed in Step 5.
CellsFiles subfolder:
cells_12km_2001.txt - Cells text file for the year 2001, 12 km grid.
cells_12km_2002.txt - Cells text file for the year 2002, 12 km grid.
cells_36km_2001.txt - Cells text file for the year 2001, 36 km grid.
cells_36km_2002.txt - Cells text file for the year 2002, 36 km grid.
Help subfolder:
2001CMAQ12km.PNG - Graphic depicting the 12 km grid for the year 2001.
2002CMAQ12km.PNG - Graphic depicting the 12 km grid for the year 2002.
HBMUsersGuide.doc - The User's Guide.
OtherDataFiles subfolder:
-------
aod_fake.csv - An empty file representing Aerosol Optical Depth (AOD) data. This is a
remnant of the manual, multi-step process.
StateLatLon.csv - This is a CSV representation of the States.txt file, including only the
data needed to create a StateGrid.csv file.
States.txt - Contains states with their latitude and longitude ranges. The source of this
information is http://www.netstate.com.
StateFiles subfolder:
StateGrid2001-12km .csv - Contains a list of states and regions, with their associated X,
Y ranges according to the 2001 cells text file for the 12 km grid.
StateGrid2001-36km.csv - Contains a list of states and regions, with their associated X,
Y ranges according to the 2001 cells text file for the 36 km grid.
StateGrid2002-12km .csv - Contains a list of states and regions, with their associated X,
Y ranges according to the 2002 cells text file for the 12 km grid.
StateGrid2002-36km.csv - Contains a list of states and regions, with their associated X,
Y ranges according to the 2002 cells text file for the 36 km grid.
ValidationDataFiles subfolder:
castnet_grid_o3_2001_12km.sas7bdat - SAS dataset containing CASTNET ozone data
for the 12 km grid for the year 2001. This file can be used in validation (Step 5).
castnet_grid_o3_2001_36km.sas7bdat - SAS dataset containing CASTNET ozone data
for the 36 km grid for the year 2001. This file can be used in validation (Step 5).
castnet_grid_o3_2002_12km.sas7bdat - SAS dataset containing CASTNET ozone data
for the 12 km grid for the year 2002. This file can be used in validation (Step 5).
castnet_grid_o3_2002_36km.sas7bdat - SAS dataset containing CASTNET ozone data
for the 36 km grid for the year 2002. This file can be used in validation (Step 5).
krigepredozoneerrorl22001.sas7bdat- SAS dataset containing kriged ozone data for
the 12 km grid for the year 2001. This file can be used in validation (Step 5).
krigepredozoneerrorl22002.sas7bdat- SAS dataset containing kriged ozone data for
the 12 km grid for the year 2002. This file can be used in validation (Step 5).
krigepredozoneerror362001.sas7bdat- SAS dataset containing kriged ozone data for
the 36 km grid for the year 2001. This file can be used in validation (Step 5).
krigepredozoneerror362002.sas7bdat- SAS dataset containing kriged ozone data for
the 36 km grid for the year 2002. This file can be used in validation (Step 5).
pm25_improve_validation_2001_36.sas7bdat - SAS dataset containing IMPROVE
PM2.5 data for the 36 km grid for the year 2001. This file can be used in validation (Step
5).
pm25_krige_predict_eiror_2001_36.sas7bdat - SAS dataset containing kriged PM2.5
data for the 36 km grid for the year 2001. This file can be used in validation (Step 5).
pm25_stn_validation_2001_36.sas7bdat - SAS dataset containing STN PM2.5 data for
the 36 km grid for the year 2001. This file can be used in validation (Step 5).
validation_pm25_2001_36.sas7bdat - SAS dataset containing IMPROVE and STN
PM2.5 data for the 36 km grid for the year 2001. This file can be used in validation (Step
5).
-------
2.4 Logical design
Each subsection lists the object-oriented programming (OOP) classes defined in each project. A
description of each is provided, as well as the objects that use the class. See Figures la and lb
for a graphical representation of the logical structure of the software.
tura3Eev3mdUOnliding.frmPararm
EPAWA41.Main
EPAWA41 ,UC-Wonitor
EPAWA41 -UC-CMAQ
| EPA WA41, RowCoJT oi.atLonConverter
>.WA41raSlateTraeker
| EPAWA41 .UC-lrrtroScreen
| EPAWA41 JOAPfLationinserter
EPAWA41 .ToxtLatLonlnsertor
i EPAWA41 .BatehiRunneif1
EPAWA41 .BatchFileCreatorADDLL
EPAWA41 .TextFiieCfeator
EPAWA41 .FiteCompare
EPAWA41 CSVProduterNetCDF
! EPAWA41 >MonStorSf I
EPAWA41.SurfaceFiieAvcr^ger
EPAWA41 .Surf aceCeti
Figure 1a. Class Usage Diagram (Class A uses> Class B).
-------
Validation
WA41 SlafcTracterl
DwLengTeShortPaftCmrwater
A A A
BafchFtteCreatarftDDLL
1 if* i
u o h rttroacreen
TextLaltomlnserter
A A
Figure 1b. Class Usage Diagram (Class A uses> Class B) (Continued).
2.4.1 BayesianMelding Classes
frmParams - The form as well as all logic needed to verify input, load parameter files
and save parameter files.
o Used by EPAWA41 .Main
StringTokenizer - Parses (splits) strings into individual tokens based on a specified
delimiter.
o Used by BayesianMelding.frmParams
2.4.2 BayesianMonitorCheck Classes
Checker - Performs the post processing on the monitor PM2.5 file created in Step 2
o Used by BayesianMonitorCheck.Main
2.4.3 CMAQColumnRenamer Classes
-------
CMAQColumnRenamer - Performs the post processing of the CMAQ/CAMx files
created in Step 2.
o Used by CMAQColumnRenamer.Main
2.4.4 CMAQPostProcessor Classes
CMAQPostProcessor - Performs the post processing of the CMAQ/CAMx PM2.5 file
created in Step 2.
o Used by CMAQPostProcessor.Main
2.4.5 CreateValidationChainFile Classes
ValidationChainFileCreator - Performs the creation of the validation chain file.
o Used by ValidationChainFileCreator.Main
2.4.6 EPAWA41 State Classes:
WA41_StateTracker - This is the repository for values that are needed throughout
program execution.
o Used by:
BayesianMelding.frmParams
EPAWA41.BatchFileProcessors.BatchFileCreator
EPAWA41.BatchFileProcessors.BatchRunner
EPAWA41.CSVFileProcessors.IOAPILatLonInserter
EPAWA41.CSVFileProcessors.RowColToLatLongConverter
EPAWA41.CSVFileProcessors.TextLatLonInserter
EPAWA41.Forms.Main
EPAWA41.Forms.Validation
EPAWA41.UserControls.UC-CMAQ
EPAWA41.UserControls.UC-IntroScreen
EPAWA41.UserControls.UC-Monitor
DosLongToShortPathConverter - Contains two static methods allowing the application
to obtain either the DOS short name (8.3, no spaces) or the Windows long name of a file
from the other one.
o Used by:
EPAWA41.BatchFileProcessors.BatchFileCreator
EPAWA41.BatchFileProcessors.BatchFileCreatorAddLL
EPAWA41.CSVFileProcessors.IOAPILatLonInserter
EPAWA41.CSVFileProcessors.TextLatLonInserter
EPAWA41.Forms.Main
EPAWA41.Forms.Validation
EPAWA41.UserControls.UC-IntroScreen
WA41 State. WA41 StateTracker
2.4.7 MonitorColumnRenamer
-------
MonitorColumnRenamer - Performs the post processing of the monitor files created in
Step 2.
o Used by MonitorColumnRenamer.Main
2.4.8 Bayesian Air Pollution Model Classes
None.
2.4.9 EPAWA41
BatchFileProcessors
o BatchFileCreator - Creates batch files used in Step 2.
ฆ Used by:
EPAWA41.UserControls.UC-CMAQ
EPAWA41.UserControls.UC-Monitor
o BatchFileCreatorAddLL - Creates batch file to run addll.exe.
ฆ Used by:
EPAWA41.CSVFileProcessors.IOAPILatLonInserter
EPAWA41.CSVFileProcessors.TextLatLonInserter
o BatchRunner - Executes batch files created by EPAWA41.
BatchFileProcessors.BatchFileCreator.
ฆ Used by:
EPAWA41.CSVFileProcessors.IOAPILatLonInserter
EPAWA41.CSVFileProcessors.TextLatLonInserter
EPAWA41.UserControls.UC-CMAQ
EPAWA41.UserControls.UC-Monitor
CSYFileProcessors
o CSVProducer - Base class to provide functionality to other classes.
ฆ Inherited by:
EPAWA41.BatchFileProcessors.BatchFileCreator
EPAWA41.BatchFileProcessors.BatchFileCreatorAddLL
o CSVProducerNetCDF - Provides an interface to the NetCDF library.
ฆ Used by:
EPAWA41.CSVFileProcessors.IOAPILatLonInserter
o IOAPILatLonlnserter - Inserts latitude and longitude into the CMAQ/CAMx
and monitor files created in Step 2.
ฆ Used by:
EPAWA41.UserControls.UC-CMAQ
EPAWA41.UserControls.UC-Monitor
o RowColToLatLongConverter - Inserts latitude and longitude values into the
surface file created by the model
ฆ Used by: EPAWA41.Forms.Main
o TextLatLonlnserter - Inserts latitude and longitude values into the
CMAQ/CAMx and monitor files created in Step 2.
ฆ Used by:
EPAWA41.UserControls.UC-CMAQ
-------
EPAWA41.UserControls.UC-Monitor
Tools
o AQSFileComparer - Compares two AQS monitor files
ฆ Used by: EPAWA41.Forms.FileCompare
o FileComparer - Base class to provide functionality to other classes.
ฆ Inherited by:
EPAWA41.Tools.AQSFileComparer
EPAWA41.Tools.TSVFileComparer
o MonitorSite - Object representing a monitor site
ฆ Used by:
EPAWA41.Tools.AQSFileComparer
EPAWA41.Tools.FileComparer
EPAWA41.Tools.TSVFileComparer
o SurfaceCell - Object representing a cell from surface file
ฆ Used by: EPAWA41.Tools.SurfaceFileAverager
o SurfaceFileAverager - Averages 3 consecutive (annual) 4th Highest surface files
ฆ Used by: EPAWA41.Forms.Average4thHighest
o TSVFileComparer - Compares two air pollution monitor sites tsv files
ฆ Used by: EPAWA41.Forms.FileCompare
Utility
o AppSettings - Saves application settings for use in future program sessions
ฆ Used by: EPAWA41.UserControls.UC-IntroScreen
o TextFileCreator - Creates text files that capture execution messages from batch
files
ฆ Used by:
EPAWA41.BatchFileProcessors.BatchRunner
EPAWA41.Forms.Main
o ValidationAppSettings - Saves validation settings for use in future program
sessions
ฆ Used by: EPAWA41.Forms.Validation
2.5 Data
Each subsection lists a category of data used by the application. A description of each data file is
provided, as well as the listing of the files that fall into each category.
2.5.1 Permanent Storage
Each subsection lists a category of permanent data files. These are the data files that persist
when the application is closed.
2.5.1.1 Input Files
Each subsection lists a data file that must exist before a simulation is started. These do not
include the previously described data files that are installed to the application folder. These data
files are not part of the application's installation but are obtained otherwise.
A description of each data file is provided, as well as the names and the representative data files.
-------
2.5.1.1.1 CMAQ/CAMx File
These files are in NetCDF format, also referred to as 10API files. These files are specific to each
air pollutant, year, and grid size. The air pollutant (ozone/03 or PM/PM25) and grid size
(12/12km or 36/36km) are indicated by the file name. These files are generated and supplied by
EPA. These files are input for the CMAQ/CAMx (i.e., air quality model) portion of Step 2. This
file is also necessary for Step 4 to obtain latitude and longitude information for the air pollution
concentration surface file created by the model. This file is also required in the Monitor portion
of Step 2 if the user decides to place 10API latitude and longitude information into the monitor
file created there. Table 1 lists a few sample CMAQ files.
Table 1. Sample CMAQ Files.
CMAQ File Name
Type of Data
Size
ozone_2001_12.cone
Ozone, 2001, 12 km2
1338.22 MB
2002ac_v4.61_L3b_EUS_12km. combine. hourly.03.total365
Ozone, 2002, 12 km2
2237.66 MB
2002ac_v4.61_L3b_us36b.combine.hourly.O3.total365
Ozone, 2002, 36 km2
553.99 MB
pm25_2001_12km.ioapi
PM2.5, 2001, 12 km2
1338.22 MB
2002ac_v4.61_l3b_eus_12km.dailyavg.pm25
PM2.5, 2002, 12 km2
93.25 MB
2.5.1.1.2 Monitor File
These files are text files, delimited by the pipe character ('|'). These files are specific to air
pollutant and year. A portion of a file is provided to understand the required format of the file.
These files are available for download from the AQS download page at
https://www3.epa.gov/ttn/airs/airsaqs/detaildata/downloadaqsdata.htm. These files are input for
the Monitor portion of Step 2. Table 2 lists the names of several sample monitor files.
-------
Table 2.
Sample Monitor Files.
Monitor File Name
Pollutant
Size
RD_501_44201_2001-0.txt
Ozone
521.26 MB
RD_501_44201_2002-0.txt
Ozone
531.41 MB
RD_501_88101_2001-0.txt
PM25
12.19 MB
rd 501 pm fine spec 2001-0.txt
PM25
36.89 MB
RD_501_88101_2002-0.txt
PM25
12.30 MB
rd 501 pm fine spec 2002-0.txt
PM25
76.83 MB
2.5.1.1.3 Cells File
These files are text files, delimited by spaces. These files are specific to a year and locational
grid, providing latitude and longitude values for each cell of the grid. These files are provided
by EPA and are required in the CMAQ/CAMx and Monitor portions of Step 2, if the user
decides to place the latitude and longitude information for each cell in the created files. Table 3
lists the names of several sample cells files.
Table 3. Sample Cell Files.
Cell File Name
Size
cells_12km_2001 .txt
4.58 MB
cells_12km_2002.txt
7.66 MB
cells_36km_2001 .txt
1.90 MB
cells_36km_2002.txt
1.90 MB
2.5.1.2 Output Files
Each subsection lists a file that is created during a simulation. These files are not deleted when
the simulation is complete. They are available at the completion of the simulation. A
description of each file is provided, as well as file sizes for a typical simulation.
2.5.1.2.1 CMAQ Model Input file
This is a CSV file created in the CMAQ/CAMx portion of Step 2, and is an input into the T-
SpACE model. This file has 4 columns: Day, Column, Row, and CMAQ log (i.e., the log-
transformed value of air quality model [CMAQ or CAMx] estimated concentration). The
number of rows, and total size, is dependent upon the grid selections (T, X, Y) made in Step 1.
T = time of simulation in calendar days (start, end)
X = rows of grid cells (start, end)
Y = columns of grid cells (start, end)
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
ozone_2001_12.conc CMAQ/CAMx input file; the resultant CMAQ/CAMx model input file is
245.00 MB in size.
-------
Running a 2001 simulation for PM2.5 using T = (1, 365), X (1, 213), Y (1, 188) and the
pm25_2001_12km.ioapi CMAQ/CAMx input file; the resultant CMAQ/CAMx model input file
is 399.65 MB in size.
2.5.1.2.2 Monitor Model Input File
This is a CSV file created in the monitor portion of Step 2, and is an input to the T-SpACE
model. This file has 4 columns: Day, Column, Row, and monitor log (i.e., the log-transformed
air pollution monitor concentration value). The number of rows, and total size, is dependent upon
the grid selections (T, X, Y) made in Step 1.
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_44201_2001-0.txt monitor input file; the resultant monitor model input file is 3.66 MB
in size.
Running a 2001 simulation for PM2.5 using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_88101_2001-0.txt monitor input file; the resultant monitor model input file is 5.35 MB
in size.
2.5.1.2.3 CMAQ Model Input file with Latitude and Longitude
This is a CSV file that is optionally created in the CMAQ/CAMx portion of Step 2. This file is
identical to the CMAQ Model Input File previously described, except that it has 2 additional
columns: Longitude and Latitude.
Adding the latitude and longitude values to the ozone file example previously described
(ozone_2001_12.conc) resulted in a file that is 509.12 MB in size.
Adding the latitude and longitude values to the PM2.5 file example previously described
(pm25_2001_12km.ioapi) resulted in a file that is 660.81 MB in size.
2.5.1.2.4 Monitor Model Input file with Latitude and Longitude
This is a CSV file that is optionally created in the monitor portion of Step 2. This file is identical
to the Monitor Model Input File previously described, except that it has 2 additional columns:
Longitude and Latitude.
Adding the latitude and longitude to the ozone file example previously described
(RD_501_44201_2001-0.txt) resulted in a file that is 7.34 MB in size.
Adding the latitude and longitude to the PM2.5 file example previously described
(RD_501_88101_2001-0.txt) resulted in a file that is 11.82 MB in size.
2.5.1.2.5 Chain File
This is a CSV file that is created by the T-SpACE model in Step 4. This file contains the values
for TauX, TauY, TauZ, RhoZ (i.e., the temporal autocorrelation parameter of the mean process)
and each Mu (the mean level of the CAR [conditional autoregression] process) for each
simulation step.
-------
The chain file is helpful in understanding if the choices made by the user for the number of burn-
in steps and number of simulations steps is appropriate. By plotting the Tau parameters, by each
iteration, the user can see where the Tau parameter values are increasing or decreasing,
indicating that those iterations should be included in the (number of) burn-in steps. If the user
sees that the Tau parameters jump between a range (e.g., bouncing back and forth), this would
indicate the point at which the simulation steps should begin, and where the number of burn-in
steps should end. See Tables 4, 5, 6 and 7 for typical chain file sizes.
2.5.1.2.6 Global File
This is a CSV file that is created by the T-SpACE model in Step 4. This file contains a brief
summary of the number of burn-in steps, simulation loop steps, and 'thinning' choices made for
the simulation. Thinning is the process of saving a sample every given number of steps (n). It is
used to reduce the self-correlation in the chain file. See Tables 4, 5, 6 and 7 for typical Global
file sizes.
2.5.1.2.7 Simulation Data File
This is a machine readable file that is created by the T-SpACE model in Step 4. This file is for
Aerosol Optical Depth (AOD) satellite input, which is implemented in the Model 6 version of T-
SpACE. See Tables 4, 5, 6 and 7 for typical simulation data file sizes.
2.5.1.2.8 Summary File
This is a CSV file that is created by the T-SpACE model in Step 4. This file contains summary
statistics for all calculated parameters of the model including the bias surface. Appendix G, the
T-SpACE User's Guide, provides a detailed description of the contents of this file. See Tables 4,
5, 6 and 7 for typical summary file sizes.
2.5.1.2.9 Surface File
This is a CSV file that is created by the T-SpACE model in Step 4. This file is used by the
validation process (Step 5). This file contains the estimated surface from the simulation. The T-
SpACE User's Guide provides a detailed summary of the contents of this file. See Tables 4, 5, 6
and 7 for typical air pollution concentration surface file sizes.
Table 4. Sizes of Model Output Files Using Ozone Model Input Files for 10 Burn-in
Steps and 50 Simulation Steps.
Model Output File
Size
Chain.csv
146 K
Globals.csv
< 1 K
Simulation.dat
1,334.50 MB
Summary.csv
1,642 MB
Surface.csv
1,479.30 MB
-------
Table 5. Sizes of Model Output Files Using Ozone Model Input Files 100 Burn-in
Steps and 2500 Simulation Steps.
Model Output File
Size
Chain.csv
6.97 MB
Globals.csv
< 1 K
Simulation.dat
1,334.50 MB
Summary.csv
1,664.34 MB
Surface.csv
1,106.42 MB
Table 6. Sizes of Model Output Files Using PM2.5 Model Input Files Previously
Described, 10 Burn-in Steps and 50 Simulation Steps.
Model Output File
Size
Chain.csv
146 K
Globals.csv
< 1 K
Simulation.dat
1,334.50 MB
Summary.csv
1,659.52 MB
Surface.csv
1,531.19 MB
Table 7. Sizes of Model Output Files Using PM2.5 Model Input Files Previously
Described, 100 Burn-in Steps and 2500 Simulation Steps.
Model Output File
Size
Chain.csv
6.97 MB
Globals.csv
< 1 K
Simulation.dat
1,334.50 MB
Summary.csv
1,681.85 MB
Surface, csv
1,145.24 MB
2.5.1.2.10 Validation Chain File
This is a CSV file that is created by a process that runs immediately after completion of the T-
SpACE model run in Step 4. This file is used by the validation process (Step 5). This file is a
subset of the chain file previously described, containing only the values for TauX, TauY, and
TauZ. The size of a validation chain file is typically 10% of the size of the parent chain file from
which it was created.
2.5.1.2.11 Validation Report File
This is a text file that is created by running the validation process (Step 5). This file contains the
reports generated by the SAS validation program. The size of this file is approximately 20 K.
2.5.1.2.12 Validation Log File
This is a text file that is created by running the validation process (Step 5). This file contains the
logs generated by the SAS validation program. The size of this file is approximately 90 K.
2.5.1.2.13 Validation SAS Program
-------
This is a text file that is created by running the validation process (Step 5). This file contains the
SAS source code generated and executed by the T-SpACE model. The size of this file is 16 K.
2.5.1.3 Intermediate Files
Each subsection lists a file that is created during a T-SpACE simulation run. These files are
deleted automatically by the system when no longer needed. A description of each file is
provided, as well as file sizes for a typical process.
2.5.1.3.1 CMAQ Ozone Column Rename Temp File
This CSV file is a duplicate of the CMAQ Ozone Model Input File created in Step 2. It is used
by the process that runs automatically in Step 2 that changes the columns names in the CMAQ
Ozone Model Input File. This temp file, which is the same size as the CMAQ Ozone Model
Input File, is deleted when the file with the desired column names is generated.
2.5.1.3.2 Monitor Ozone Column Rename Temp File
This CSV file is a duplicate of the Monitor Ozone Model Input File created in Step 2. It is used
by the process that runs automatically in Step 2 that changes the column names in the Monitor
Ozone Model Input File. This temp file, which is the same size as the Monitor Ozone Model
Input File, is deleted when the file with the desired column names is generated.
2.5.1.3.3 CMAQ PM2.5 Column Post-Process Temp File
This CSV file is a duplicate of the CMAQ PM2.5 Model Input File created in Step 2. It is used by
the process that runs automatically in Step 2 that performs post-processing on the CMAQ PM2.5
Model Input File. This temp file, which is the same size as the CMAQ PM2.5 Model Input File,
is deleted when the post-processed file is written.
2.5.1.3.4 CMAQ PM2.5 Column Rename Temp File
This CSV file is a duplicate of the CMAQ PM2.5 Model Input File created in Step 2. It is used by
the process that runs automatically in Step 2 that changes the columns names in the CMAQ
PM2.5 Model Input File. This temp file, which is the same size as the CMAQ PM2.5 Model Input
File, is deleted when the file with the desired column names is generated.
2.5.1.3.5 Monitor PM2.5 Column Post-Process Temp File
This CSV file is a duplicate of the Monitor PM2.5 Model Input File created in Step 2. It is used
by the process that runs automatically in Step 2 that performs post-processing on the Monitor
PM2.5 Model Input File. This temp file, which is the same size as the Monitor PM2.5 Model Input
File, is deleted when the post-processed file is generated.
2.5.1.3.6 Monitor PM2.5 Column Rename Temp File
This CSV file is a duplicate of the Monitor PM2.5 Model Input File created in Step 2. It is used
by the process that runs automatically in Step 2 that changes the columns names in the Monitor
PM2.5 Model Input File. This temp file, which is the same size as the Monitor PM2.5 Model Input
File, is deleted when the file with the desired column names is generated.
2.5.1.3.7 CMAQ Reformatted File for ADDLL Temp File
-------
This CSV file is created when, in Step 2, the user chooses to have latitude and longitude
information added to a CMAQ Model Input File. This file contains all of the data from CMAQ
Model Input File created in Step 2, reformatted to match what ADDLL.exe expects as input.
This file is deleted once ADDLL.exe completes execution. The file is generated as an
intermediate data file, which allows air quality model concentration surfaces to be generated
for individual states and regions in the US. With ozone, the size of this file is approximately
150% of the size of the CMAQ Model Input File from which it is created. With PM2.5, the size of
this file is approximately 200% of the size of the CMAQ Model Input File from which it is
created.
2.5.1.3.8 CMAQ Reformatted File with Lat/Lon Temp File
This CSV file is created when, in Step 2, the user chooses to have latitude and longitude
information added to a CMAQ Model Input File. This file is created by ADDLL.exe and
contains all of the data from the temporary file previously described, plus latitude and longitude
for each record. Another process reformats this file into the CMAQ Model Input file with
Latitude and Longitude previously described. The file is generated as an input data file, which
allows air quality model concentration surfaces to be generated for individual states and regions
in the US. Once the file reformatting is complete, the file is deleted. With ozone, the size of this
file is approximately 175% of the size of the temporary file from which it is created. With PM2.5,
the size of this file is approximately 150% of the size of the temporary file from which it is
created.
2.5.1.3.9 Monitor Reformatted File for ADDLL Temp File
This CSV file is created when, in Step 2, the user chooses to have latitude and longitude
information added to a Monitor Model Input File. This file contains all of the data from Monitor
Model Input File created in Step 2, reformatted to match what ADDLL.exe expects as input.
This file is deleted once ADDLL.exe completes execution. The file is generated as an
intermediate data file, which allows air quality monitor concentration surfaces to be generated
for individual states and regions in the US. With ozone, the size of this file is approximately
150% of the size of the Monitor Model Input File from which it is created. With PM2.5, the size
of this file is approximately 150% of the size of the Monitor Model Input File from which it is
created.
2.5.1.3.10 Monitor Reformatted File with Lat/Lon Temp File
This CSV file is created when, in Step 2, the user chooses to have latitude and longitude
information added to a Monitor Model Input File. This file is created by ADDLL.exe and
contains all of the data from the temporary file previously described, plus latitude and longitude
for each record. Another process reformats this file into the Monitor Model Input file with
Latitude and Longitude previously described. Once that reformat is complete, this file is deleted.
The file is generated as an input data file, which allows air quality monitor concentration
surfaces to be generated for individual states and regions in the US. With ozone, the size of this
file is approximately 200%) of the size of the temporary file from which it is created. With PM2.5,
the size of this file is approximately 225% of the size of the temporary file from which it is
created.
2.5.1.3.11 Raw Surface Temp File
-------
This CSV file is created by the T-SpACE model in Step 4. It is used by the process that runs
automatically when the T-SpACE model simulation completes which creates the final air
pollution concentration surface file. This process takes the raw surface temp file, adds latitude
and longitude to each record, and converts the date on each record to a recognizable format. This
file is deleted once the post-processing successfully completes. The file allows final (monitor
plus model) concentration surfaces to be generated for individual states and regions in the US.
The size of this file is approximately 75% of the size of the final surface file based on this
temporary file.
2.5.2 Volatile Storage
This subsection lists the properties contained in the volatile storage (i.e., computer memory) used
during a simulation. This storage disappears when the application is closed. A description of
each property is provided, as well as the classes that initialize the property and read the property.
Volatile storage is provided by the state tracker (Singleton) class, which acts as a repository for
values that are needed throughout the application. When the T-SpACE application starts, a
single object of this class {state tracker) is instantiated (created/generated), and a reference
('pointer') to this object is passed to objects that need to initialize or retrieve values stored within
them. No computer hard disk input/output (I/O) is utilized with the state tracker.
The state tracker properties are:
ฆ ApplicationLocation - Application executable's folder.
o Initialized by:
ฆ EPAWA41.Forms.Main
o Read by:
ฆ EPAWA41.BatchFileProcessors.BatchFileCreator
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.RowColToLatLonConv
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ฆ EPAWA41.Forms.Validation
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
ฆ BayesianMeldingStartDate - Start date in format needed by Bayesian Melding.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ BayesianDay - Day of BayesianMeldingStartDate.
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
ฆ BayesianMelding.frmParams
ฆ BayesianMonth - Month of BayesianMeldingStartDate.
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
-------
ฆ BayesianMelding.frmParams
BayesianYear - Year of BayesianMeldingStartDate.
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
ฆ BayesianMelding.frmParams
CMAQPath - Location of the CMAQ/CAMx file created in Step 2.
o Initialized by:
ฆ EPAWA41.UserControls.UC-CMAQ
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.BatchFileProcessors.BatchFileCreator
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
CSVFileType - Denotes if the current file in Step 2 is CMAQ/CAMx or monitor.
o Initialized by:
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
o Read by:
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
EndDateBatchFormat - End date in format needed by Step 2 batch files.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
EndTime - Ending T value, in date format, as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.Forms.Validation
FullPathOfSurfaceCSV - Location of air pollution concentration surface file created by
the T-SpACE model,
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
ฆ EPAWA41.CSVFileProcessors.RowColToLatLonConv
ฆ EPAWA41.Forms.Validation
InputDirectory - Directory specified by user in Step 1, where the files created in Step 2
will be saved
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
-------
ฆ EPAWA41.BatchFileProcessors.BatchFileCreator
ฆ EPAWA41.BatchFileProcessors.BatchRunner
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.RowColToLatLonConv
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ฆ EPAWA41.Forms.Main
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
IOAPIFullFilePath - Location of the CMAQ input file specified in Step 2.
o Initialized by:
ฆ EPAWA41.UserControls.UC-CMAQ
o Read by:
ฆ EPAWA41.BatchFileProcessors.BatchFileCreatorAddLL
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.RowColToLatLonConv
ฆ EPAWA41.Forms.Main
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
Is64BitVersion - Boolean used to determine if application is running on a 64-bit
machine.
o Initialized by:
ฆ EPAWA41.Forms.Main
o Read by:
ฆ EPAWA41.Forms.Main
LatLonCMAQPath - Location of CMAQ/CAMx 10API file used as source of latitude
and longitude information for CMAQ/CAMx CSV file created in Step 2.
o Initialized by:
ฆ EPAWA41.UserControls.UC-CMAQ
o Read by:
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
LatLonMonitorPath - Location of CMAQ/CAMx 10API file used as source of latitude
and longitude information for monitor CSV file created in Step 2.
o Initialized by:
ฆ EPAWA41.UserControls.UC-Monitor
o Read by:
ฆ EPAWA41.UserControls.UC-Monitor
MonitorPath - Location of the monitor file created in Step 2.
o Initialized by:
ฆ EPAWA41.UserControls.UC-Monitor
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.BatchFileProcessors.BatchFileCreator
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ParFilePath - Location of model parameter file created in Step 3.
-------
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
ฆ EPAWA41.Forms.Main
PartialPathOfSurfaceCSVDirectory - Folder of surface file created by the T-SpACE
model.
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
ฆ EPAWA41.Forms.Validation
PartialPathOfSurfaceCSVFile - Name of surface file created by the T-SpACE model.
o Initialized by:
ฆ BayesianMelding.frmParams
o Read by:
ฆ EPAWA41.Forms.Validation
Pollutant - Pollutant for the current T-SpACE simulation run.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
ฆ EPAWA41.Forms.Main
o Read by:
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
ฆ EPAWA41.Forms.Validation
StartDateBatchFormat - Start date in format needed by Step 2 batch files.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ EPAWA41.UserControls.UC-CMAQ
ฆ EPAWA41.UserControls.UC-Monitor
StartTime - Beginning T value, in the appropriate date format, as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.Forms.Validation
TBegin - Beginning T value as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
TEnd - Ending T value as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ValidationChainFile - Chain file specified in Step 5.
-------
o Initialized by:
ฆ EPAWA41.Forms.Validation
o Read by:
ฆ EPAWA41.Forms.Validation
ValidationKrigeFile - Krige file specified in Step 5.
o Initialized by:
ฆ EPAWA41.Forms.Validation
o Read by:
ฆ EPAWA41.Forms.Validation
ValidationMonitorFile - Monitor file specified in Step 5.
o Initialized by:
ฆ EPAWA41.Forms.Validation
o Read by:
ฆ EPAWA41.Forms.Validation
ValidationReportFile - Report file specified in Step 5.
o Initialized by:
ฆ EPAWA41.Forms.Validation
o Read by:
ฆ EPAWA41.Forms.Validation
ValidationReportTitle - Report title specified in Step 5.
o Initialized by:
ฆ EPAWA41.Forms.Validation
o Read by:
ฆ EPAWA41.Forms.Validation
ValidationSurfaceFile - Surface file specified in Step 5.
o Initialized by:
ฆ EPAWA41.Forms.Validation
o Read by:
ฆ EPAWA41.Forms.Validation
XBegin - Beginning X value as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ฆ EPAWA41.Forms.Validation
XEnd - Ending X value as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ฆ EPAWA41.Forms.Validation
YBegin - Beginning Y value as specified in Step 1.
-------
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ฆ EPAWA41.Forms.Validation
ฆ YEnd - Ending Y value as specified in Step 1.
o Initialized by:
ฆ EPAWA41.UserControls.UC-IntoScreen
o Read by:
ฆ BayesianMelding.frmParams
ฆ EPAWA41.CSVFileProcessors.IOAPILatLonInserter
ฆ EPAWA41.CSVFileProcessors.TextLatLonInserter
ฆ EPAWA41.Forms.Validation
2.6 Communications/Messaging
Messaging within the T-SpACE application is minimal, as most of the components reside within
the same logical structure or process. There are several instances where the application must
spawn (generate) a process so that it can run a user interface console application. This is done
using the Process class of the Microsoft.NET framework. Communication to this class is done
using 2 methods of that class, Start and WaitForExit. The software spawns a process by
invoking the start method. Invoking the 'WaitForExit' method causes the software to pause
until the process has completed.
The Process object is used by:
ฆ EPAWA41.BatchFileProcessors.BatchFileCreator
o Run the batch file to create the CMAQ/CAMx and monitor files (ozone and
PM2.5) in Step 2.
o Run the batch file to add latitude and longitude to CMAQ/CAMx and monitor
files (ozone and PM2.5) created in Step 2.
o Run the batch file to add latitude and longitude to the air pollution concentration
surface file created by the model.
ฆ EPAWA41.Forms.Main
o Run the model
o Run the process to create the validation chain file
-------
3. OPERATIONAL DESIGN
Each subsection details a facet of the T-SpACE operational design. These facets include Use
Cases and Typical Processing.
A use case is a description of a system's behavior as it responds to a request that originates from
outside of that system.
Typical processing describes the time and memory constraints of the software during a typical
process.
3.1 Use Cases
The subsection heading is the name of the use case. A brief description of each use case is
provided.
3.1.1 Prepare to Create Simulation Input Use Case
Preparation for creating the T-SpACE simulation input is complete when the user selects the
temporal and spatial grids for the simulation, as well as the destination folder for the simulation
input.
3.1.2 Create Simulation Input Use Case
Simulation input is created when the user selects a pollutant and creates the CMAQ/CAMx and
monitor simulation input files.
3.1.3 Choose Simulation Parameters Use Case
Choosing simulation parameters is complete when the user saves the simulation parameter
(*.par) file.
3.1.4 Run Simulation Use Case
Running the simulation is completed when the user launches the T-SpACE model using a
simulation parameter file.
3.1.5 Validate Simulation Use Case
Validating a completed simulation is complete when the user launches validation using the T-
SpACE simulation output files, monitor data, and kriging data.
3.1.6 Compare AQS Files Use Case
Differences in monitor locations between two AQS data files can be obtained when the user
utilizes the AQS File Compare tool.
-------
3.1.7
Compare Airsites Files Use Case
Differences in monitor locations between two air sites data files can be obtained when the user
utilizes the Airsites File Compare tool.
3.1.8 Average 4th Highest Surface Files Use Case
A file containing the average of three consecutive annual 4th Highest Surface Files is created
when the user launches the Average 4th Highest Surface Files tool.
3.1.9 Create States Grid File Use Case
A file containing the states and regions, along with their corresponding grid cells, is created
when the user launches the Apply States to Grid tool.
3.2 Typical Processing
Each subsection details one aspect of a typical processing for the application. These aspects are
Sequence and Expected System Behavior.
Sequence lists the steps taken during a typical processing.
Expected system behavior details the software's time and memory constraints of a typical
processing.
3.2.1 Sequence
Once the T-SpACE is launched, the sequence for running a complete T-SpACE simulation is as
follows:
1. Using the calendars, choose the beginning and end dates for the simulation.
2. Select a folder where the CSV files (created in Step 2) will be saved.
3. Select the X and Y ranges for the simulation. These can either be entered manually, or
a state/region can be selected from a list.
4. Select a pollutant for the simulation by clicking "Ozone" or "PM 2.5". The application
automatically advances to Step 2.
5. Select a CMAQ/CAMx input file.
6. Optionally change the default output file name.
7. Optionally create an additional version of the output file that also contains latitude and
longitude. A source of this information must be provided, either the file selected in
sequence 5 above, or from a cells.txt file.
8. Click "Execute"; wait for "Complete" message.
9. Still in Step 2, click the "Monitor" tab.
10. Select a Monitor input file.
11. Optionally change the default output file name.
-------
12. Optionally create an additional version of the output file that also contains latitude and
longitude. A source of this information must be provided, either the file selected in
sequence 5 above, or a cells file.
13. Click "Execute"; wait for "Complete" message.
14. Click "Step 3: Model Specification" on the tool bar above the Step 2 form.
15. Optionally open an existing parameter file by clicking the open button (the one with
folder opening) and selecting the file
16. Review the values on the "Grid" tab, change as desired.
17. Click the first "Priors" tab and review/change the values.
18. Click the second "Priors" tab and review/change the values.
19. Click the "Data" tab and review/change the values.
20. Click the "Boundaries" tab and review/change the values.
21. Click the "Track" tab and review/change the values.
22. Click the "Simulation" tab and review/change the values.
23. Click the "4th Highest" tab and review/change the values.
24. Click the save button to save the parameter (*.par) file to the computer hard disk.
25. Click "Step 4: Launch Model" on the tool bar above the Step 3 form; wait for
"Complete" message.
26. Click "Step 5: Launch Validation" on the tool bar above the Step 3 form
27. Review/change the values
28. Click "Run Validation"; wait for "Complete" message.
3.3 Expected System Behavior
The computer that was used to determine the expected behavior of T-SpACE was configured as
follows:
Intel Core 2 Duo E6750 Processor, @ 2.66 GHz
3 .48 GB of RAM, @ 2.66 GHz
Windows XP Professional, Service Pack 2
Each subsection details one of the software's constraints during a typical processing. These
constraints are time and memory.
The simulation parameter files used for these runs are provided in Appendix E.
3.3.1 Approximate Time
The subsection name is the process being timed. A description of the process is provided, as
well as a breakdown of the time needed to complete the process.
3.3.1.1 Creating the CMAQ Model Input File (Step 2)
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
ozone_2001_12.conc CMAQ/CAMx input file the time needed was as follows (to the nearest
second):
Creating the CMAQ file - 2 minutes 44 seconds
Renaming the columns - 12 seconds
-------
Adding latitude and longitude (optional) - 8 minutes, 20 seconds
Running a 2001 simulation for PM2.5 using T = (1, 365), X (1, 213), Y (1, 188) and the
pm25_2001_12km.ioapi CMAQ/CAMx input file the time needed was as follows (to the nearest
second):
Creating the CMAQ file - 2 minutes 49 seconds
Post-processing - 27 seconds
Renaming the columns - 14 seconds
Adding latitude and longitude (optional) - 8 minutes, 23 seconds
3.3.1.2 Creating the Monitor Model Input File (Step 2)
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_44201_2001-0.txt monitor input file the time needed was as follows (to the nearest
second):
Creating the Monitor file - 2 minutes 44 seconds
Renaming the columns - 1 seconds
Adding latitude and longitude (optional) - 23 seconds
Running a 2001 simulation for PM2.5 using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_88101_2001-0.txt monitor input file the time needed was as follows (to the nearest
second):
Creating the Monitor file - 1 minutes 41 seconds
Post-processing - 1 second
Renaming the columns - 1 second
Adding latitude and longitude (optional) - 8 seconds
3.3.1.3 Running the Model (Step 4)
Using the ozone files created as described above, the time needed to run the T-SpACE model
was as follows (to the nearest second):
10 burn-in steps and 50 simulation steps:
o Read input files - 35 seconds
o Allocate variables - 1 minute, 15 seconds
o Burn-in steps - 3 minutes, 27 seconds
o Allocate variables - 55 seconds
o Running T-SpACE simulation steps - 17 minutes, 8 seconds
o Write files - 8 minutes, 35 seconds
o Add latitude and longitude - 7 minutes, 22 seconds
o Write File - 1 minute, 24 seconds
o Create validation chain file - 25 seconds
100 burn-in steps and 2500 simulation steps:
o Read input files - 53 seconds
o Allocate variables - 1 minute, 15 seconds
o Burn-in steps - 3 minutes, 26 seconds
o Allocate variables - 1 minute, 16 seconds
o Running T-SpACE simulation steps - 14 Hours, 21 minutes, 17 seconds
o Write files - 9 minutes, 52 seconds
-------
o Add latitude and longitude - 7 minutes, 49 seconds
o Write File - 1 minute, 31 seconds
o Create validation chain file - 34 seconds
Using the PM2.5 files created as described above, the time needed to run the T-SpACE model
was as follows (to the nearest second):
10 burn-in steps and 50 simulation steps:
o Read input files - 41 seconds
o Allocate variables - 11 seconds
o Burn-in steps - 3 minutes, 27 seconds
o Allocate variables - 13 seconds
o Running T-SpACE simulation steps - 17 minutes, 7 seconds
o Write files - 8 minutes, 9 seconds
o Add latitude and longitude - 7 minutes, 45 seconds
o Write File - 29 seconds
o Create validation chain file - 11 seconds
100 burn-in steps and 2500 simulation steps:
o Read input files - 41 seconds
o Allocate variables - 17 seconds
o Burn-in steps - 3 minutes, 25 seconds
o Allocate variables - 24 seconds
o Running T-SpACE simulation steps - 14 Hours, 16 minutes, 50 seconds
o Write files - 9 minutes, 34 seconds
o Add latitude and longitude - 7 minutes, 52 seconds
o Write File - 32 seconds
o Create validation chain file - 13 seconds
3.3.1.4 Validation (Step 5)
The time needed to validate the model output for the ozone simulation previously described was
as follows (to the nearest second):
10 burn-in steps and 50 simulation steps - 14 seconds
100 burn-in steps and 2500 simulation steps - 5 minutes, 44 seconds
The time needed to validate the model output for the PM2.5 simulation previously described was
as follows (to the nearest second):
10 burn-in steps and 50 simulation steps - 17 seconds
100 burn-in steps and 2500 simulation steps - 6 minutes, 58 seconds
3.3.2 Memory Usage
The subsection name is the process being evaluated for memory usage. A description of the
process is provided, as well as the maximum memory used during the process. When initially
loaded, the T-SpACE application uses approximately 19,000 K (kilobytes) of Random Access
Memory (RAM, i.e., computer memory). This subsection only details processes which cause
this consumption to increase.
-------
3.3.2.1 Creating the CMAQ Model Input File (Step 2)
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
ozone_2001_12.conc CMAQ/CAMx input file used 22,700 K of RAM.
Running a 2001 simulation for PM2.5 using T = (1, 365), X (1, 213), Y (1, 188) and the
pm25_2001_12km.ioapi CMAQ/CAMx input file used 27,900 K of RAM.
3.3.2.2 Creating the Monitor Model Input File (Step 2)
Running a 2001 simulation for ozone using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_44201_2001-0.txt monitor input file used 23,900 K of RAM.
Running a 2001 simulation for PM2.5 using T = (1, 365), X (1, 213), Y (1, 188) and the
RD_501_88101_2001-0.txt monitor input file used 27,900 K of RAM.
3.3.2.3 Running the Model (Step 4)
When running the simulation steps, the T-SpACE model used approximately 1,828,600 K (1.828
MB - megabytes) of RAM.
4. DESIGN ISSUES
Each subsection details a facet of T-SpACE's design issues.
4.1 Factors Affecting Design
Each subsection describes a factor that had an impact on the T-SpACE design.
4.1.1 Using existing applications for Steps 2 and 3.
The existing EPA-developed console programs that were used by the T-SpACE model
application were run by batch files, because a large number of environment variables required
initialization. Processes to create and run these batch files, as well as to wait for their
completion, had to be developed.
4.1.2 Batch files usedfor Step 2 require short file names.
The batch files needed to execute the programs for Step 2 require short DOS names for input
files, output files, auxiliary files and application name. Routines to translate between long and
short names were researched, developed, and implemented so that users had no restrictions on
file naming.
4.1.3 Not having the source code for Step 2 programs.
The files created in Step 2 had to be altered, either by necessity (i.e., the model expects input
values to be natural logarithm transformations of the air pollution concentration values) or by
request (i.e., need different column names). Since the source code for the programs that created
these files was not available, new routines were developed to read in the files and create new,
slightly different, versions of them. These additional steps add to the execution time and
significantly increase the amount of disk input/output.
-------
4.2 Suggested Future Upgrades
Each subsection describes a recommended future upgrade.
4.2.1 Remove States/Regions outside of grid.
All states are currently available for selection on the initial screen. When the states are defined
by the locational grid, the closest grid cell to the lower left and upper right "corners" of the state
are determined. There is always a closest cell, so every state has a corresponding grid region.
Not all of the grids cover the continental United States so it is possible for a state's grid region to
not actually contain any portion of the state.
4.2.2 Creation of validation network and kriging files.
Currently the network and kriging files used in the data validation process must be created
outside of the application.
5. SUMMARY
Several different programming languages were used to create the source code for the T-SpACE
model. Each computer programming language used in the development of T-SpACE has
strengths and unique capabilities, which were leveraged to build a well-engineered software
solution.
The current T-SpACE code base is as follows:
The programs that create the four model input files in Step 2 (CMAQ/Monitor files for
Ozone/PM2.5) are standalone command-line applications written in FORTRAN (*.f files).
The program that adds latitude and longitude information is a standalone command-line
application written in FORTRAN (*.f files).
The model is a standalone command-line application written in C++ (*.cpp).
The Step 2 post-processing programs are standalone command-line applications written
in C# (*.cs files).
The program that creates the chain file for use in data validation is a standalone
command-line applications written in C# (*.cs files).
The program that allows for maintenance to the model simulation files (Step 3) is a
standalone Windows-form application written in Visual Basic (*.vb).
The state tracker object is written in C# (*.cs files).
The validation program (Step 5) is written in SAS (*.SAS files).
All source code listed above, except for the FORTRAN, C++ and SAS programs, are accessible
through the EPAWA41.sln file, which opens with Microsoft Visual Studio. Visual Studio
allows for the editing, building and debugging of the source code.
-------
Appendix B
-------
Temporal-Spatial Ambient Concentration Estimator: Evaluation For Ozone Data
1.0 Background
To accomplish its mission to protect human health and the environment, EPA has established
National Ambient Air Quality Standards (NAAQS) on six selected air pollutants known as
criteria pollutants: ozone (O3); carbon monoxide (CO); lead (Pb); nitrogen dioxide (NO2); sulfur
dioxide (SO2), and; particulate matter (PM). The states are primarily responsible for maintaining
and improving air quality and complying with the NAAQS. Ozone (O3) and particulate matter
(PM) are two of the criteria pollutants whose levels are regulated by NAAQS. Extensive air
pollution monitoring networks have been set-up across the US to understand the levels
(concentrations) of these air pollutants across the US and over time. Once collected, this
information is reported to EPA and is made available publicly after the data is reviewed and
analyzed for quality and accuracy.
When assessing air quality, the most direct way is to utilize unbiased ground-level air pollution
measurements from the existing surface monitoring network stations located across the US.
However, a good portion of the US, especially rural areas, have very sparse or irregularly spaced
monitoring stations, resulting in large areas of the country with no information available on air
quality. Air quality models estimate the spatial and temporal gradients of air pollution based on
emissions inventories and meteorological information. These models, while providing estimates
over large regions at relatively low cost, have (statistical) bias and have greater error (variance)
than air pollution monitoring networks. These methods include Models-3/Community
Multiscale Air Quality (CMAQ) numerical model and the Comprehensive Air Quality Model
with Extensions (CAMx). Various statistical methods have been investigated that utilize existing
monitoring data and spatial modeling techniques to develop a concentration surface that
(retrospectively) estimates the distribution of daily air pollution levels within a specified region
of the US. Kriging is one such method. Another technique is the T-SpACE model. The focus of
this analysis is how T-SpACE compares to kriging and the CMAQ/CAMx model results for
ozone data in the years of 2001 and 2002.
2.0 Model
The T-SpACE model, which uses the ozone monitoring data from EPA's Federal Reference
Method (FRM) monitors in the NAMS/SLAMS network and CMAQ/CAMx model output, is
used to estimate (retrospectively predict) pollutant levels at daily time scales. T-SpACE assumes
that both monitoring data and CMAQ/CAMx data provide good information about the same
underlying pollutant surface, but with different measurement error structures. It gives more
weight to accurate monitoring data in areas where monitoring data exists, and relies on bias-
adjusted CMAQ/CAMx output data in non-monitored areas. The problem is divided into
hierarchical components, and each level in the hierarchy is modeled conditionally on its
preceding levels. Appendix C provides a detailed description of the hierarchical Bayesian
structure of the T-SpACE model.
-------
3.0 Data
Table 1 lists the datasets utilized in the T-SpACE simulation and subsequent data validation.
Table 1. Ozone concentrations (ppm) - Daily 8-hour maximums.
Source
Dates
Resolution
Input file name
File
Format
Community
Multiscale Air
Quality
(CMAQ) model
January
1, 2001 -
November
29, 2001
12 km
ozone.cone
NetCDF
36 km
The appropriate format of this file was not available for
analysis.
January
1, 2002 -
November
30, 2002
12 km
2002ac_v4.61_L3b_EUS_12km.combine.hourly.03.total365
NetCDF
36 km
2002ac_v4.61_L3b_us36b.combine.hourly.03.total365
NetCDF
EPA/AQS
(FRM
NAMS/SLAMS)
January
1, 2001 -
December
31,2001
NA
R D_501 _44201 _2001 -0 .txt
Text
January
1, 2002 -
December
31,2002
NA
R D_501 _44201 -2002-0.txt
Text
3.1 Models-3/Communitv Multiscale Air Quality (CMAQ) and Comprehensive Air
Quality Model with Extensions (CAMx)
3.1.1 CMAQ
The Models-3/Community Multiscale Air Quality (CMAQ) modeling system goals are to
improve: 1) the environmental management community's ability to evaluate the impact of air
quality management practices for multiple pollutants at multiple scales, and; 2) the scientist's
ability to better probe, understand, and simulate chemical and physical interactions in the
atmosphere. Traditionally, the CMAQ modeling system has been used to predict air quality
across an entire regional or national domain, and then to simulate the effects of various changes
in emission levels for policy-making purposes. However, for health studies, it is frequently
applied to provide supplemental information on predicted air quality in areas where few or no
monitors exist. It also accounts for meteorological conditions to better evaluate health outcomes.
Detailed information on the CMAQ modeling system is available at https://www.epa.gov/cmaq
and at https://www.cmascenter.org.
Because the CMAQ modeling system makes predictions across an entire two-dimensional
domain, the set of model predictions at a given point in time resembles a smooth "surface" when
it is graphically displayed, rather than simply a set of points at various locations. This surface,
called a response surface, appears three-dimensional in that it covers the two-dimensional region
and has a third dimension representing the magnitude of the predicted concentrations. The
CMAQ model predictions are made relative to a specified rectangular grid that covers the entire
-------
region of interest. The grid is composed of a matrix of rectangular cells. Within each cell, the
CMAQ modeling system predicts average pollution levels at a given point in time. Each cell has
equal area which is expressed as h x v, where h and v denote each cell's horizontal and vertical
dimensions, respectively (in kilometers). Under the CMAQ modeling system, grid cells can
either be 36 km x 36 km (which represents the "parent domain" covering the entire continental
US) or 12 km x 12 km (which represents primarily the eastern and Midwest regions of the US up
to 2007, and the entire US from 2008 and onward). Thus, as the values of h and v decrease for a
fixed area, the CMAQ modeling system generates a greater density of predictions, which
increases the resolution of the area's prediction surface.
3.1.2 CAMx
The Comprehensive Air Quality Model with Extensions (CAMx) is an Eulerian (gridded)
photochemical dispersion model that allows for integrated "one-atmosphere" assessments of
tropospheric air pollution (e.g., ozone, particulates, air toxics) over spatial scales ranging from
neighborhoods (local) to continents (global). It is a modern, open-source system that is
computationally efficient, flexible, and pubically available. CAMx's Fortran source code is
modular and well-documented. The Fortran binary input/output file formats are based on the
Urban Airshed Model (UAM) convention, and are compatible with many existing pre- and
post-processing tools. Meteorological input fields are supplied to CAMx from separate weather
prediction models. CAMx specifically supports the Weather Research and Forecasting (WRF)
model, the Mesoscale Model 5 (MM5) and the Regional Atmospheric Modeling System
(RAMS). All emission inputs are supplied from external pre-processing systems (e.g., the
Sparse Matrix Operator Kernel (SMOKE) system and the Emissions Processor System Version 3
(EPS3).
CAMx simulates the emission, dispersion, chemical reaction, and removal of pollutants by
'marching' the Eulerian continuity equation forward in time (t) for each chemical species (1) on a
system of nested three-dimensional grids. The continuity equation specifically describes the
time dependency of volume-average species concentration within each grid cell as a sum of all
physical and chemical processes operating on that volume. CAMx can perform simulations on
four types of Cartesian map projections: Lambert Conic Conformal, Polar Stereographic,
Mercator, and Universal Transverse Mercator (UTM). CAMx also offers the option of operating
on a geodetic latitude/longitude grid system. The vertical grid structure is defined externally, so
layer interface heights may be specified as any arbitrary function of space and/or time. This
flexibility in defining the horizontal and vertical grid structures allows CAMx to be configured to
match the grid of any meteorological model that is used to provide environmental input fields.
Detailed information on the CAMx system can be found at: http://www.camx.com.
-------
Figure 1. The maximum area of the United States covered by the 2001 CMAQ
12 km x 12 km Grid.
Figure 2. The maximum area of the United States covered by the 2001 CMAQ
36 km x 36 km grid.
-------
;12km Eastern (large) CMAQ Domain
x,y: -1008000, -1620000
J col: 279 row: 240
36km Domain X V \, k
Lower Left Corner: -2736000, -208800I
Number of Cols, Rows: 148x112 "V
Lambert Projection: ^
1st std parallel: 33 \ \
2nd std parallel: 45
central meridian: -97
latitude of projection's origin: 40
Figure 3. The maximum area of the United States covered by the 2002 CMAQ data for
the 12 km x 12 km and 36 km x 36 km grids.
Table 2. The coordinates used in the simulation runs.
Grid
Year
Coverage Used in Simulation Runs
Full Grid Coverage Available
Column (x)
(Start, End)
Row (y)
(Start, End)
Column (x)
(Start, End)
Row (y)
(Start, End)
12 km
2001
(40, 213)
(30, 188)
(1, 213)
(1, 188)
2002
(100, 279)
(100, 240)
(1, 279)
(1, 240)
36 km
2001
(30, 170)
(30, 154)
(1, 200)
(1, 184)
2002
(28, 120)
(30, 82)
(1, 148)
(1,112)
-------
3.2 Network Monitors
Two sets of network monitor data are used in the T-SpACE model simulation. The EPA-
validated Federal Reference Method (FRM) monitors that are part of the NAMS/SLAMS
network are used to develop the estimated (retrospectively predicted) air pollution concentration
surface. For the ozone simulation, concentrations from monitors in the Clean Air Status and
Trends Network (CASTNET) are used for comparing predictions from the three types of
modeling processes used: T-SpACE, Kriging, and CMAQ/CAMx.
3.2.1. T-SpA CE Model Input Monitor File
EPA has a repository of ambient air quality data. This repository is called the Air Quality
System (AQS), and is located at: https://www.epa.gov/aqs. The AQS contains measurements of
concentrations for the six (6) criteria pollutants (i.e., ozone [O3]; carbon monoxide [CO]; lead
[Pb]; nitrogen dioxide [NO2]; sulfur dioxide [SO2], and; particulate matter [PM]) from EPA-
developed and funded monitoring networks within the United States. These AQS monitors are
FRM monitors. For this evaluation, ozone data available for 2001 and 2002 were extracted from
AQS. The exact names of the files are listed in Table 1. Ozone concentrations were collected
every day. Daily 8-hour maximum concentrations were used in the analysis. The number of
monitors included in the analysis varied with the CMAQ grid that was used in the analysis.
There are upwards of 1,200 ozone monitors with data available in AQS. Table 1 lists the original
data files downloaded by EPA from EPA's AQS and utilized in the analysis.
The CMAQ files had missing data from November 30, 2001 through December 31, 2001 and
December 1, 2002 through December 31, 2002. The data recorded on November 30, 2002 were
ten times higher than any daily value in that year and were not included in the analysis.
Therefore, the analysis conducted on both the 2001 and 2002 data ran from January 1 through
November 29.
3.2.2 Validation
In order to evaluate the predictive ability of T-SpACE, and to compare it to the predictive ability
of the Krige model, data from the Clean Air Status and Trends Network (CASTNET) was
utilized in the data validation. This network is similar to the FRM network, where measurements
are taken on a daily basis. Data from 2001 and 2002 were available. Those monitors that were
located on the CMAQ grid were included in the data validation analysis. The red stars in Figure
4 indicate that 47 2001 CASTNET monitors were located within the CMAQ grid boundaries,
and were used for the analysis. Four of these 47 monitors were co-located with FRM monitors.
-------
Figure 4. Location of the 2001 CASTNET monitors relative to the FRM monitors
included in the analysis.
4.0 Input Parameters
The T-SpACE model requires input from the user. A file with a *.PAR (i.e., parameter)
extension is where the user's choices for the model simulation are stored. Appendix E provides
an example of the *.PAR file used for Run 1 of the analysis of the 2001, 12 km ozone data.
Table 3 provides a brief description of the file and parameter choices. USEPA (2007), and
McMillan et al., (2010) provides more details about the model.
Table 2 lists the choices that were made for the x and y grid sizes for each year, and
CMAQ/CAMx coordinate system. Note that a simulation was not run for the 36 km x 36 km
grid for 2001. The CMAQ data were not available in the appropriate format (NetCDF).
Two time frames for both years were chosen for the 2001 simulations, May 1 (Julian Day 121) to
October 31, the ozone season, and January 1 through November 29, a "full" year. Note that the
CMAQ data was missing for November 30 through December 31 in 2001 and 2002, and this is
why the full year was defined to be January 1 through November 29.
The burn-in steps and the simulation steps were reduced to 10 and 50 for the simulations, to run
the simulations relatively quickly. Table 4 illustrates the estimated time to run various scenarios
using the T-SpACE software.
-------
Table 3. Summary of the T-SpACE Model Input Parameters.
Parameter name
Parameter
Values
Description
Model version
5-4 H
This is the version of the model utilized in the analysis. Currently
the MCMC version is 5.1, but 5-4h is used as the current version
of T-SpACE. PAR file to designate the MCMC model 5.1 and the
addition of the calculation of the 4th highest average concentration
(4h).
Grid size
t
start, end
Time frame within a calendar year in days from January 1 (or the
date specified below).
X
start, end
The range for the x-coordinates of the CMAQ grid illustrated in
Figures 1, 2, and 3. The choice is dependent on the year and the
grid size being used for the analysis. See Table 2 for the ranges
used in the analysis and the full range available.
y
start, end
The range for the y-coordinates of the CMAQ grid illustrated in
Figures 1, 2, and 3. The choice is dependent on the year and the
grid size being used for the analysis. See Table 2 for the ranges
used in the analysis and the full range available.
Year, month, day
of time step 1
year, month, day
This is a reference point for the model. It is best to leave this as
January 1. The values in "Grid size:t" above, are dependent upon
this value. Changing it to anything other than January 1 may
result in incorrect information later on.
Bias Spline
t
start, end, number of
basis points
Start and end are the same as above. The basis points are used
to define the way the model calculates the splines. 4, 8, and 7 are
the recommended basis points for t, x, and y, respectively, when
the analysis is similar in size to the grid utilized for this validation.
These values can be changed depending on the size of the grid
chosen for the analysis.
X
start, end, number of
basis points
y
start, end, number of
basis points
Prior Parameters
Mu
Mean, variance
The mean for the underlying space-time log transformed ozone
process. (Natural log of the concentrations is used in the
analysis.) This is initialized to a non-informative normal prior
(recommended).
BetaD
Mean, variance
The covariates for the CMAQ bias structure. This is initialized to a
non-informative normal prior (recommended).
TauX
Mean, precision
The precision of the measurement error in the monitor (TauX) and
the CMAQ (TauY) observations, respectively. This is assigned a
Gamma distribution. The mean and precision are chosen based
on existing knowledge about measurement errors associated with
FRM monitoring data and CMAQ results. Various scenarios for
the relationship between TauX and TauY are examined in the
validation process.
TauY
Mean, precision
TauZ
Mean, precision
The precision of the underlying space-time log transformed ozone
process. A non-informative Gamma distribution with mean =
0.001 and precision = 0.001 is assigned. This is recommended.
RhoZ
Mean, precision
Temporal autocorrelation parameter of the underlying space-time
log transformed ozone process. A uniform prior distribution
between 0 and 1 is defined for RhoZ. This is recommended.
Directory where
the input data are
stored
Directory
Both the monitor and the CMAQ input data files must be located in
the same directory.
Monitor file
.CSV file
The input monitor data must be a .CSV file of a specified format,
which is generated through EPA software.
CMAQ file
.CSV file
The input CMAQ data must be a .CSV file of a specified format
that is generated through software developed for EPA. -999 is
used to indicate a missing value. Missing value data must be
excluded from the analysis by the user choosing the appropriate
days and grid cells to include in the analysis.
Order of
neighborhood
numeric value
Defines the neighborhood structure of the model. This value can
range from first order (1) to fourth order (4). Default is a first order
neighborhood. Figure 4 illustrates the neighborhood definitions.
-------
Table 3. (Continued).
Parameter name
Parameter
Values
Description
Track chain flag
numeric value
0 or 1 to not write or write, respectively, the chain of the simulation
to the screen. By default, the value is initialized to 0 and the chain
is not written to a file. This speeds up the simulation which can be
laborious as the grid and time frames become large.
Enable calculation
of 4th highest
surface
numeric value
0 or 1 to have the simulation calculate an average 4th highest
concentration surface for the grid and time frame chosen. The
default is 0 (no calculation). A file with _4h.csv as the suffix is
developed if this is initialized to 1. (This surface is not evaluated.)
Track chain flag
for 4th highest
surface
numeric value
0 or 1 to not write or write, respectively, the chain of the simulation
for the calculation of the average 4th highest concentration surface.
By default, the value is initialized to 0 and the chain is not written
to a file. (This surface is not evaluated.)
Directory for
where the output
data from the
model run will be
stored
directory
Location to store the series of output files from the model run.
Simulation output
file name
.CSV file
Name to be used to store the predicted ozone surface and
associated statistics.
Random seeds
1
numeric value
Two seeds are needed for the simulation. By default, these are
initialized to 1234 and 5678. These can be changed.
2
numeric value
Sampling Period
numeric value
This is how often sampling occurs. The default is 1 (i.e., for each
model iteration).
Prompting Period
numeric value
The user can see the simulation progress on the screen. The
default is to see each iteration (1). Higher numbers will reduce the
amount of information written to the screen as the simulation
occurs, thus potentially speeding up the process.
Simulation Flag
numeric value
This indicates whether the simulation results will overwrite a file if
a file already exists with the same name.
Number of burn-in
steps
numeric value
Number of burn-in steps for the simulation. The default is 1000.
For the validation simulations, 10 were used.
Number of
simulation steps
numeric value
Number of simulation loops. The default is 5000. For the
validation simulations, 50 were used.
-------
First order neighborhood
Second order neighborhood
ooooo ooooo
o
o
o
Third order neighborhood Fourth order neighborhood
Figure 4. Neighborhood Definitions.
Table 4. Estimated times to run differing number of burn-ins, iterations, and
timeframes.
Scenario
Grid size
Estimated Time
10 burn-in, 50 iterations, ozone season
12km x 12 km
9 minutes
36 km x 36 km
2 minutes
50 burn-in, 100 iterations, "full" year
12km x 12 km
50 minutes
36 km x 36 km
9 minutes
The only other parameters that were changed from the default values were the mean and
precision for TauX (Tx) and TauY (Ty). The choices for the values were provided by EPA and
were also based on previous research. Some of the previous research indicated a coefficient of
variation (CV) around 20% for tx and a CV of 80% for xy may be appropriate starting points for
choosing the mean and precision (USEPA, 2007). Note that there is an overwhelming amount of
CMAQ data available as compared to the air pollution monitor data. It takes a strong prior on xy
for the prior to have much effect on the posterior. The actual choices for their parameters are
illustrated in Tables 5, 6, and 7.
5.0 Model validation results
Using the data and parameter files described above, data validation runs were performed to
assess the stability of the T-SpACE model when parameters are changed, and to compare the
predictive abilities of T-SpACE with other models, in particular, a Krige model and the
CMAQ/CAMx model results.
-------
Table 5 provides a summary of the statistics that are used in the validation analysis.
Table 5. Description of the Validation Summary Statistics.
Statistic
Description
Run number
The number of the run. The choices for the Tx and Ty are the same for each run number across the year
and grid choices.
Prior Assumptions
Parameter
These are the t parameters which vary across the runs.
Mean
The assumed prior mean for the Gamma distribution.
Precision
The assumed prior precision for the Gamma distribution.
Marginal Posterior
Means
These are the average of the estimated posterior mean for the t's that were produced during the simulation.
Predicted Standard
Error (SE)
This is the average of the model-predicted standard error for the mean predicted concentration.
95% Prediction
Interval
Using the predicted standard error of the mean predicted concentration from T-SpACE, this represents the
% of the validation monitor concentration that fall within the 95% confidence intervals for T-SpACE.
Krige Prediction
Interval
Using the predicted standard error of the mean predicted concentration from the Krige model, this
represents the % of the validation monitor concentration that fall within the 95% confidence intervals for the
Krige model.
Overall Bias
Average of the absolute difference between the monitor concentration and T-SpACE, Krige, and
CMAQ/CAMx predicted concentrations, respectively.
Overall MSE
Average of the square of the bias (described above). The average is over location and time.
% improvement by
day
For each prediction the comparison represents the % of days which the Krige model and the CMAQ/CAMx
model have MSEs that are greater than the MSE for T-SpACE.
% improvement by
time
For each prediction the comparison represents the % of time which the Krige model and the CMAQ/CAMx
model have MSEs that are greater than the MSE for T-SpACE.
Tables 6, 7, and 8 provide the results of the validation runs using data from May 1 through
October 31, the ozone season, while Tables 9, 10, and 11 provide the results for a "full" year of
data, January 1 through November 29.
For each time frame, there were five validation runs per year (2001, 2002) and CMAQ grid (12
km x 12 km, 36 km x 36 km). Runs for the 2001, 36 km x 36 km was not possible, since the
CMAQ data were not available in the appropriate format (NetCDF), so there are three sets of
data validation runs for the ozone data.
5.1 Explanation of the metrics used in the validation
Table 5 describes the information provided in Tables 6 through 11. For reference, the SAS code
that generated the summary statistics for the 2001, 12 km, Run 4, is provided in Appendix F
(pages F-2 to F-8). The SAS code for developing the Krige model used in the analysis of the
2001, 12 km, Run 4, is provided in Appendix D (pages D-2 to D-5). Similar SAS code was
utilized for the other year/grid combinations.
-------
Table 6. Summary Statistics for Model Runs (Model 5_4h) for ozone, 36 kilometers, 2001 (May 1 - October 31).
Run
numbe
r
Prior Assumptions*
Marginal Posteriors
Mean
Predicted SE
95% Prediction
Interval
Krige
Prediction
Interval
Surface
Overall Bias
Overall MSE
%
Improvement
by Day
%
Improvement
by Site
Parameter
mean
precision
Tx
Ty
Tz
1
Tx
75.0E+6
1.0E+6
74.9
0.10
4.84
0.32
94.8
66.5
T-SpACE
0.06
0.04
Ty
0.1 E+8
1.0E+8
Krige
0.02
0.03
5.43
25.5
Tz
1 .OE-3
1 .OE-3
CMAQ
0.09
0.06
78.3
78.4
2
Tx
1.0E+7
1.0E+6
10.0
0.93
7.87
0.29
94.9
66.5
T-SpACE
0.04
0.04
Ty
0.5E+6
1.0E+6
Krige
0.02
0.03
6.52
23.5
Tz
1 .OE-3
1 .OE-3
CMAQ
0.09
0.06
85.9
82.4
3
Tx
8.0E+6
1.0E+6
8.04
0.93
7.80
0.30
94.9
66.5
T-SpACE
0.04
0.04
Ty
0.5E+6
1.0E+6
Krige
0.02
0.03
5.43
17.7
Tz
1 .OE-3
1 .OE-3
CMAQ
0.09
0.06
83.2
82.4
4
Tx
1.6E+7
1.0E+6
16.0
0.93
7.98
0.28
94.8
66.5
T-SpACE
0.03
0.04
Ty
0.5E+6
1.0E+6
Krige
0.02
0.03
8.70
25.5
Tz
1 .OE-3
1 .OE-3
CMAQ
0.09
0.06
90.2
84.3
5
Tx
1.6E+7
1.0E+6
16.0
1.42
9.56
0.26
94.8
66.5
T-SpACE
0.03
0.04
Ty
1.0E+6
1.0E+6
Krige
0.02
0.03
9.24
27.45
Tz
1 .OE-3
1 .OE-3
T-SpACE
0.09
0.06
90.76
84.31
The other parameters are assumed to be non-informative.
-------
Table 7. Summary Statistics for Model Runs (Model 5_4h) for ozone, 12 kilometers, 2001 (May 1- October 31).
Run
numbe
r
Prior Assumptions*
Marginal Posteriors
Mean
Predicted SE
95% Prediction
Interval
Krige
Prediction
Interval
Surface
Overall Bias
Overall MSE
%
Improvement
by Day
%
Improvement
by Site
Parameter
mean
precision
Tx
Ty
Tz
1
Tx
75.0E+6
1.0E+6
75.0
0.12
3.62
0.44
93.2
84.1
T-SpACE
0.08
0.07
Ty
0.1 E+8
1.0E+8
Krige
0.02
0.02
0
8.51
Tz
1 .OE-3
1 .OE-3
CMAQ
0.14
0.08
52.7
51.1
2
Tx
1.0E+7
1.0E+6
10.0
1.96
10.7
0.26
93.1
84.1
T-SpACE
0.02
0.06
Ty
0.5E+6
1.0E+6
Krige
0.02
0.02
0.5
0
Tz
1 .OE-3
1 .OE-3
CMAQ
0.14
0.08
76.6
78.7
3
Tx
8.0E+6
1.0E+6
8.03
1.96
10.7
0.27
93.1
84.1
T-SpACE
0.02
0.06
Ty
0.5E+6
1.0E+6
Krige
0.02
0.02
0.5
0
Tz
1 .OE-3
1 .OE-3
CMAQ
0.14
0.08
76.1
78.7
4
Tx
1.6E+7
1.0E+6
16.0
1.96
10.8
0.26
93.0
84.1
T-SpACE
0.01
0.05
Ty
0.5E+6
1.0E+6
Krige
0.02
0.02
1.1
0
Tz
1 .OE-3
1 .OE-3
CMAQ
0.14
0.08
80.4
80.9
5
Tx
1.6E+7
1.0E+6
16.0
2.32
35.3
0.15
92.1
84.1
T-SpACE
-0.03
0.05
Ty
1.0E+6
1.0E+6
Krige
0.02
0.02
2.72
0
Tz
1 .OE-3
1 .OE-3
CMAQ
0.14
0.08
74.5
76.6
* The other parameters are assumed to be non-informative.
-------
Table 8. Summary Statistics for Model Runs (Model 5_4h) for ozone, 12 kilometers, 2002 (May 1 - October 31).
Run
number
Prior Assumptions*
Marginal Posteriors
Mean
Predicted SE
95% Prediction
Interval
Krige
Prediction
Interval
Surface
Overall Bias
Overall MSE
%
Improvement
by Day
%
Improvement
by Site
Parameter
mean
precision
Tx
Ty
Tz
1
Tx
75.0E+6
1.0E+6
75.0
0.12
3.50
0.43
93.8
80.1
T-SpACE
0.07
0.07
Ty
0.1 E+8
1.0E+8
Krige
0.01
0.02
1.09
2.13
Tz
1.0E-3
1 .OE-3
CMAQ
0.11
0.08
52.2
42.6
2
Tx
1.0E+7
1.0E+6
10.0
1.83
10.2
0.27
93.7
80.1
T-SpACE
0.01
0.05
Ty
0.5E+6
1.0E+6
Krige
0.01
0.02
2.17
0
Tz
1.0E-3
1 .OE-3
CMAQ
0.11
0.08
76.1
78.7
3
Tx
8.0E+6
1.0E+6
8.03
1.83
10.2
0.27
93.7
80.1
T-SpACE
0.01
0.06
Ty
0.5E+6
1.0E+6
Krige
0.01
0.02
1.09
2.13
Tz
1.0E-3
1 .OE-3
CMAQ
0.11
0.08
72.3
78.7
4
Tx
1.6E+7
1.0E+6
16.0
1.83
10.2
0.27
93.5
80.1
T-SpACE
0.00
0.05
Ty
0.5E+6
1.0E+6
Krige
0.01
0.02
3.26
0
Tz
1.0E-3
1 .OE-3
CMAQ
0.11
0.08
77.7
78.7
5
Tx
1.6E+7
1.0E+6
16.0
2.17
10.6
0.26
93.5
80.1
T-SpACE
0.00
0.05
Ty
1.0E+6
1.0E+6
Krige
0.01
0.02
3.26
0
Tz
1.0E-3
1 .OE-3
T-SpACE
0.11
0.08
77.7
78.7
The other parameters are assumed to be non-informative.
-------
Table 9. Summary Statistics for Model Runs (Model 5_4h) for ozone, 12 kilometers, 2001 (01/01/2001-11/29/2001).
Run
number
Prior Assumptions*
Marginal Posteriors
Mean
Predicted SE
95% Prediction
Interval
Krige
Prediction
Interval
Surface
Overall Bias
Overall MSE
%
Improvement
by Day
%
Improvement
by Site
Parameter
mean
precision
Tx
Ty
Tz
1
Tx
7.5E+01
1.3E+04
75.0
0.15
6.76
0.34
92.1
81.6
T-SpACE
0.03
0.06
Ty
1.0E-01
1.0E+09
Krige
0.08
0.04
12.31
21.28
Tz
1.0E+00
1.0E-03
CMAQ
0.13
0.10
81.68
76.60
2
Tx
1.0E+01
1.0E+05
10.0
4.73
38.53
0.14
91.2
81.6
T-SpACE
-0.02
0.07
Ty
5.0E-01
2.0E+06
Krige
0.08
0.04
11.41
17.02
Tz
1.0E+00
1.0E-03
CMAQ
0.13
0.10
66.97
76.60
3
Tx
8.0E+00
1.3E+05
8.03
4.73
38.62
0.14
91.6
81.6
T-SpACE
-0.02
0.07
Ty
5.0E-01
2.0E+06
Krige
0.08
0.04
11.41
17.02
Tz
1.0E+00
1.0E-03
CMAQ
0.13
0.10
67.27
76.60
4
Tx
1.6E+01
6.3E+04
16.0
4.73
38.18
0.14
90.0
81.6
T-SpACE
-0.03
0.07
Ty
5.0E-01
2.0E+06
Krige
0.08
0.04
12.01
17.02
Tz
1.0E+00
1.0E-03
CMAQ
0.13
0.10
70.27
78.72
5
Tx
1.6E+01
6.3E+04
16.0
5.21
39.7
0.14
90.0
81.6
T-SpACE
-0.03
0.07
Ty
1.0E+00
1.0E+06
Krige
0.08
0.04
12.01
17.02
Tz
1.0E+00
1.0E-03
CMAQ
0.13
0.10
69.97
78.72
-------
Table 10. Summary Statistics for Model Runs (Model 5_4h) for ozone, 12 kilometers, 2002 (01/01/02-11/29/02).
Run
number
Prior Assumptions*
Marginal Posteriors
Mean
Predicted SE
95% Prediction
Interval
Krige
Prediction
Interval
Surface
Overall Bias
Overall MSE
%
Improvement
by Day
%
Improvement
by Site
Parameter
mean
precision
Tx
Ty
Tz
1
Tx
7.5E+01
1.3E+04
75.0
0.14
6.48
0.34
93.0
79.4
T-SpACE
0.03
0.06
Ty
1.0E-01
1.0E+09
Krige
0.05
0.03
7.21
12.77
Tz
1.0E+00
1.0E-03
CMAQ
0.09
0.08
65.8
70.21
2
Tx
1.0E+01
1.0E+05
10.0
4.42
38.32
0.14
92.0
79.4
T-SpACE
-0.01
0.07
Ty
5.0E-01
2.0E+06
Krige
0.05
0.03
6.61
10.64
Tz
1.0E+00
1.0E-03
CMAQ
0.09
0.08
48.35
63.83
3
Tx
8.0E+00
1.3E+05
8.03
4.42
38.48
0.14
92.4
79.4
T-SpACE
-0.01
0.07
Ty
5.0E-01
2.0E+06
Krige
0.05
0.03
4.8
10.64
Tz
1.0E+00
1.0E-03
CMAQ
0.09
0.08
46.85
63.83
4
Tx
1.6E+01
6.3E+04
16.0
4.42
37.77
0.14
91.1
79.4
T-SpACE
-0.01
0.06
Ty
5.0E-01
2.0E+06
Krige
0.05
0.03
6.91
8.51
Tz
1.0E+00
1.0E-03
CMAQ
0.09
0.08
53.75
65.96
5
Tx
1.6E+01
6.3E+04
16.0
4.90
39.51
0.14
90.9
79.4
T-SpACE
-0.01
0.06
Ty
1.0E+00
1.0E+06
Krige
0.05
0.03
6.91
8.51
Tz
1.0E+00
1.0E-03
CMAQ
0.09
0.08
53.15
65.96
The other parameters are assumed to be non-informative.
-------
Table 11. Summary Statistics for Model Runs (Model 5_4h) for ozone, 36 kilometers, 2002 (01/01/02-11/29/02).
Prior Assumptions*
Marginal Posteriors
Mean
LU
W
C
o
c
Surface
in
ra
LU
W
c
(D
"c
O
2
ra
O
%
Improvem
by Day
%
Improvem
by Site
Tx
7.5E+01
1.3E+04
T-SpACE
0.02
0.04
1
Ty
1.0E-01
1.0E+09
74.99
0.11
7.20
0.27
99.7
69.6
Krige
0.05
0.03
26.13
34.62
Tz
1.0E+00
1.0E-03
CMAQ
0.08
0.06
78.08
84.62
Tx
1.0E+01
1.0E+05
T-SpACE
0.01
0.05
2
Ty
5.0E-01
2.0E+06
10.05
1.28
14.14
0.23
99.7
69.6
Krige
0.05
0.03
9.01
30.77
Tz
1.0E+00
1.0E-03
CMAQ
0.08
0.06
58.56
80.77
Tx
8.0E+00
1.3E+05
T-SpACE
0.01
0.05
3
Ty
5.0E-01
2.0E+06
8.05
1.28
14.28
0.23
99.8
69.6
Krige
0.05
0.03
7.21
19.23
Tz
1.0E+00
1.0E-03
CMAQ
0.08
0.06
54.35
80.77
Tx
1.6E+01
6.3E+04
T-SpACE
0.01
0.04
4
Ty
5.0E-01
2.0E+06
16.04
1.28
13.75
0.23
99.6
69.6
Krige
0.05
0.03
13.21
34.62
Tz
1.0E+00
1.0E-03
CMAQ
0.08
0.06
69.67
84.62
Tx
1.6E+01
6.3E+04
T-SpACE
0.01
0.05
5
Ty
1.0E+00
1.0E+06
16.04
1.77
15.55
0.21
99.5
69.6
Krige
0.05
0.03
10.81
34.62
Tz
1.0E+00
1.0E-03
CMAQ
0.08
0.06
66.37
80.77
The other parameters are assumed to be non-informative.
-------
5.2 Summary of results
As previously discussed, there were two sets of data validation runs that were performed.
Surfaces were developed for the ozone season, defined as May 1 through October 31 and for a
"full" year of data. As mentioned earlier, a full year was January 1 through November 29,
because data were missing for the CMAQ surface from November 30 through December 31 in
both 2001 and 2002. Below are a series of observations from the validation runs. We first
compare within the time frame assessed, and then provide overall observations.
5.2.1. Comparisons for the runs during the "Ozone" season of May 1 through October 31
In general, across years and grid size, Runs 2 through 5 produce very similar results. Run
1 produces similar results across years and grid size, but produces different results than
runs 2 through 5.
o The biggest difference between Runs 2-5 and Run 1 is the choice of the prior
mean for tx andxy. The Run 1 TauX mean prior values are large, relative to the
marginal posterior mean, and small relative to the precision priors chosen for the
other runs.
o More variability appears to have been introduced into the predictions in Run 1,
yielding a higher Predicted SE, Overall Bias, and Overall MSE. In addition, T-
SpACE did not perform as well against the CMAQ model as it did in Runs 2-5.
Focusing on Runs 2-5,
o The marginal means are relatively closer to each other, and appear to produce
relatively stable results even though the means are changed across the runs.
o The 95% Prediction Interval for the T-SpACE model in general produces results
where nearly 95% of the validation monitors are within the prediction intervals.
The Krige Prediction Interval, in general, captures less than 95% of the validation
monitors, ranging from 67% for the 36 km, 2001 runs, to 84% for the 12 km,
2001 runs.
o When comparing the three models, the T-SpACE and Krige models tend to have
similar overall bias and overall MSE. The CMAQ model overall bias and overall
MSE are relatively larger than both the T-SpACE bias and the krige bias.
o T-SpACE in general shows a smaller MSE, by day and by site, than the CMAQ
data. This indicates that by utilizing the monitor data, the CMAQ variability is
"smoothed" out in the T-SpACE model, which produces better predictions.
o The Krige model in general shows a smaller MSE, by day and by site, as
compared to T-SpACE. During the time frame utilized for the simulation runs,
the ozone monitors have relatively little variation across time and across the US.
Figure 5 provides a view of the data for September 17, 2001. The monitor values
are the circles. There is very little variation in the values observed. Kriging will
perform very well in situations such as this. Also notice the CMAQ surface in
Figure 7 (the yellow surface that is below the orange circles). This indicates that
-------
there might be occasions when the CMAQ results are introducing unnecessary
variability into the T-SpACE model.
5.2.2. Comparisons for the runs for a "full" year of data (January 1 through November 29)
In general, the results for the "full" year of data do not differ significantly from the results
observed for the ozone season results.
We do observe that the Krige model has a higher bias than T-SpACE when a full year of
data is used. This is consistent across all runs.
The difference between the MSE for T-SpACE and Krige models is very similar to that
observed for the simulations using the ozone season. The Krige model has a slightly
smaller MSE than the T-SpACE model across all runs.
We do see that the T-SpACE percent (%) improvement, by day and by site, is higher for
the full year of data. This may indicate that outside the ozone season, the ozone
concentrations are more variable, and T-SpACE is able to capture this.
LLI
T3
ฃ
re
3
O"
<0
ฃ
re
0)
~
Model Runs (Model 5) for ozone, 12 kilometer, 2002
Time Frame:
05/01/02-10/31/02
AAA 01/01/02-11/29/02
A
A
A
Run
Figure 5. Illustration of the MSE for the ozone season versus the "full" year.
5.2.3. Overall summary
5.2.3.1 Comparison of Ozone Season results to the Full Year results
-------
As seen in Figure 5, the T-SpACE MSE for Runs 2 through 5 is slightly higher for the
full year than the ozone season. This is consistent across the Runs. This would seem to be
expected as the ozone concentrations observed outside of the ozone season are more varied than
those observed during the ozone season. For Run 1, the prior Gamma mean for xy is significantly
larger (0.1E+8 as compared to 0.5E+6) than those chosen for the other runs. Though the MSE
difference is not significantly large, it does produce different results from the other runs.
In general, the T-SpACE and Krige models appear to have their strengths, regardless of
the time frame over which the simulation is run. There is an indication with the bias, the %
improvement statistics, and prediction interval statistics that the T-SpACE may be able to better
handle the variability that may occur outside the ozone season when the concentrations are
relatively stable. Though, it should be noted that the indications are slight and it is difficult to
make a definitive decision.
5.2.3.2 The T-SpACE model does a better job of capturing the variability as compared to the
Krige prediction model
The 95% Prediction Interval summary statistic records the rate at which the prediction
intervals included the monitor value (across all days and sites). Ideally the coverage should be
close to 95%. As seen across all years and the two time frames, the T-SpACE model is close to
or achieves the 95% rate. The Kriging method achieves rates that range from 67% to 84%.
A lower rate, such as the kriging summary statistics, could indicate that the estimated
uncertainty is too small or that there is a systematic positive or negative pattern of error (a
positive or negative bias) in the estimate of the true value. A rate higher than 95% could indicate
that the estimated uncertainty is too large - meaning that the estimated interval should be
narrower.
In the case of kriging, performance may be impacted by the choice of covariance
structure and the estimates of its parameters. Also, kriging assumes a stationary random process,
i.e., the response characteristics such as mean, variance and covariance do not depend on
location. Deviations from these assumptions may also impact performance.
T-SpACE can derive posterior intervals using the posterior estimate of response variance,
or numerically by taking the sample standard deviation of the MCMC chain outcomes of the
MCMC process after the burn-in period. The T-SpACE method assumes that the true response is
characterized by a conditional autoregressive process for which the mean values at FRM and
CMAQ/CAMx locations are indicated by a combination of the FRM and CMAQ/CAMx values.
-------
5.3.2.3 Uncertainty of the krige prediction model is less than T-SpACE
The bias is the difference between the monitor's reading and the predicted response. The
bias is not an absolute value and equidistant high and low values can create a bias that is low
when in fact the variability is high. The MSE is the square of these differences. The MSE
penalizes large differences more than small differences, thus highlighting the larger differences
that may occur between the predicted value and the monitor reading.
The T-SpACE model tends to have a lower bias than the Krige model while the T-
SpACE MSE is larger than the Krige model. Figure 6 provides a comparison of the MSE for the
three "models", T-SpACE, Kriging, and CMAQ for the 12 km, 2002, full year simulation runs.
In general, the differences between these statistics is not very large, indicating that the
performance as measured by these statistics is very similar.
Model Runs (Model 5) for ozone, 12 kilometer, 2002
0.12 -
0.10 -
0.08 -
0.06 -
0.04 -
0.02 -
0.00 -
Surface: ~~~ HBM
AAA Krige
ฆ ฆฆ CMAQ
c
(Q 0.04-I
Run
Figure 6. Comparison of the MSE across the three models for the "full" year for 12 km, 2002.
Note: HBM = T-SpACE
-------
I vis_krige_ozone_12_2001 .csv - AqVis4.1 - Combined Data for Space Time Models
HPS
File View Help
Date 9/17/2001
-j
Play | Rewind | Previous | Next
Predicted
Pre. Error
Monitors
CMAQ
Min 25% Median
2.03 3. GO 3.73
0.14 0.24 0.27
2.22
-0.42
3.70
3.45
LN Concentration ug/m'"3
-G1.09 -42.49 -23.99 -5.29 13.32 31.93
Monitor and CMAQ LN Concentration
Values in diagram are shown on a natural log scale.
Batteiie
TL v.,
Figure 7. Illustration of the relatively little variation in monitor concentrations during the
time frame for the validation runs.
In trying to understand why the krige results had a consistently smaller MSE across time and
space than the T-SpACE, an interesting predictive estimate was located over a week in July of
2001. Figures 8 through 14 illustrate the predictions that start off below expected levels on July
3, 2001 and gradually become more pronounced over time with the largest prediction occurring
on July 7, 2001. Figures 15 and 16 provide another view of the CMAQ surface and the monitor
results. These figures illustrate a "hole" in the CMAQ surface that remains the same on both
July 3, 2001 and July 7, 2001 (the brownish surface with the "nick" at the top of the grid to the
top, right-side of the grid. The predicted surface (the yellow surface) displays a "hole" a little
larger than the "nick" on July 3, 2001, but shows an even larger "hole" on July 7, 2001 where the
"nick" in the CMAQ surface is still the same size.
This indicates that the completeness of the CMAQ grid must be examined prior to the
simulation, otherwise, there could potentially be a pronounced effect on the T-SpACE predicted
results. In our validation analysis, the impact is minimal.
Overall, the Krige model appears to do slightly better than T-SpACE in consistently estimating
the concentrations of the individual monitors used in the model validation process, due to the
relatively little variability within the monitors. But, the T-SpACE model appears to do a
relatively better job of capturing the variability in the model validation with ozone
concentrations, allowing the T-SpACE the ability to capture nearly 95% of the validation
monitor concentrations within its 95% prediction interval. This is compared to 67% to 84% for
the Krige model.
-------
.*] vte_fcrlge_ซzorw_12_20O1.ซv - AqVis4.1 - Comblrad Data for Space Time Models
| Die View Help
Dai- 7/3/2001
'3Bi
Hay | Hnwnl
Mm 25* MnWm KX M.w
ttwfctod U,W an 3.83 3.93 4.33
Prr Fiim 012 0?4 076 029 0S6
Monioo 2. CO 3.70 3.30 403 *40
LMAll 211 358 a/1 382 4.13
IM Vjvx*rtt.-tYr\
ฆ
Jsl.US -42.43 -2383 -528 1132 31.83
Mcriw ซnd OHAQIN CcocemiWw
VaLim in (fcigmn .*r sfowi cn a mIiiaI log vsir
Batiene
SSI
Figure 8. July 3, 2001-Anomolous Prediction.
9 vJs_krlfje_ozorป_12_2001.csv - AqVM.1 - Combined Data for Space Time Models
I ฃile Vซw Help
T-JIsM
Dale 7/4/2001
?>S
Makv
ft*
M*
KrKktod
2.83
365
3.80
334
4.44
Pte Flint
ni3
0?4
036
029
?ป
Morion
J.03
374
i TO
303
IMAU
214
354
an
383
4.12
IN ConMrtMikn ug/a"3
ฆ61.03 -42.43 -2389 -528 1332 31.33
Morfef 4rปj CMAQ LN Corcentiafon
VMjm in
-------
9 \rts_kr1ge_ozone_ 13_2001.csv - AqV1s4.1 - Combined Rata for Space Time Models
ฃile yiBw Ltalp
3(sIS
Dale //V2IWI
a | Ned |
in:a ass m iai 4.49
Pw run 013 024 0% 039 C,F4
Moniซi 2.79 3.70 J06 4.24
IXAU 21S 346 i.&) 382 4.27
I tJ C.rtvfftr.ive\ ug^ซ"3
ฆ61.09 -42.ซ -2389 -6.28 1132 31.SB
Morrtx *vj CWAQ LN Corcenbaticn
VaLra in Aiginin am ifaun m a r.vm>il log avtn
Batreiie
ปฆ i
Figure 10. July 5, 2001-Anomolous Prediction.
B vls_krtge_070ซe_12_W01 .ซv AqVh4.1 - Combined Data for 5p*cป T1
:~Q
Date 7/V/200I
Ptop ] famid [ ftgviflui | | Hat |
LN CMWnMbori u$An~3
Baltene
033
Figure 11. July 6, 2001-Anomolous Prediction.
-------
H vte_krige_ozoi>e_12_20O1.ซv - AqVM.1 - Combined Data for Space Time Models
I ฃi(fl View Help
Date t/irim
J
Ha/ j HmwnJ | Pluvious | | Had ||
Mm 2S* Mmfcm KS Mm
Rwfctocl -til US 3/8 388 338 4 -10
Pw Fiim 012 023 0 76 029 3193
Morrtot] 2.94 393 4.0ฎ 4 19 *40
LMiVJ 2.'jU 363 3Jl 3,87 4.13
IM ConcmbuHon ug/ซ"3
ฆ i m
A
i
1.09 -42.49 -2389 -!*28 1332 31.83
I
Morix* ซnd OHAQ LN Cwcemistioo
Vatra in rfngmm .nr shnun m & iuIii/iI Ing srvir
1
Batreiie
Figure 12. July 7, 2001-Anomolous Prediction.
E9 vis_krige_ozone_12_2001 .csv - AqVis4.1 - Combined Data for Space Time Models BtUBI
File View Help
Date 7/8/2001
Play | Rewind | Previous | |" Next
Predicted -25.37
Pre. Error 0.15
Monitors
CMAQ
LN Concentration ug/m"3
61.09 -42.49 -23.89 -5.28 13.32 31.93
Monitor and CMAQ LN Concentration
Values in diagram are shown on a natural log scale.
Batreiie
Figure 13. July 8, 2001-Anomolous Prediction.
-------
S3 vis kri^e ozone 12 2001.csv - AqVis4.1 - Combined Data for Space Time Models
~ "UJlnJli
File View Help
Date 7/9/2001
Play | Rewind | Previous | I Next ;J
Min 25% Median 75X Max
Predicted -9.58 3.80 3.92 4.04 4.48
Pre. Error 0.14 0.24 0.26 0.29 4.77
Monitors 2.39 3.91 4.05 4.19 4.57
CMAQ 2.37 3.68 3.79 3.93 4.41
LN Concentration ug/nT3
ฆ
1
-61.09 -42.49 -23.89 -5.28 13.32 31.93
Monitor and CMAQ LN Concentration
Values in diagram are shown on a natural log scale.
Batreiie
Figure 14. July 9, 2001-Anomolous Prediction.
Batfetie
Figure 15. July 5, 2001-Anomolous Prediction - View of CMAQ "hole"
-------
Date 7/7/2001
Rewind Previous I Next
25% Median
3.78 3.88
0.23 0.2G
Predicted
Pre. Error
Monitors
CMAQ
LN Concentration yg/m"3
-61.09 -42.49 -23.89 -5.28 13.32 31.93
Monitor and CMAQ LN Concentration
Values in diagram are shown on a natural log scale.
Batreiie
Figure 16. July 7, 2001-Aiioinolous Prediction-View of CMAQ "hole".
tH vis_krige_ozone_12_2001 .csv - AqVis4.1 - Combined Data for Space Time Models
-------
6.0 References
[1] McMillan, N.J., Holland, D.M., Morara, M., and Feng, J., Combining different sources of
particulate data using Bayesian space-time modeling, Environmetrics, 2010, Volume 21,
pp 48 - 65, DOI: 10.1002/env.984
[2] USEPA (2002) AOS basics: What data are in AQS? Presentation by David Lutz at the
2002 AQS Conference. Office of Air Quality Planning and Standards, Office of Air and
Radiation, US Environmental Protection Agency.
[3 ] U SEP A (2007) Draft Report: Overview of EPA 's Hierarchical Bayesian Model for
Predicting Air Quality Patterns in the United States over Space and Time, for Use with
Public Health Tracking Data, Contract No. EPA-D-04-068, Work Assignment 44,
September 28, 2007.
-------
Appendix C
Appendix C: Markov Chain Monte Carlo (MCMC) Description (Model 5.1)
-------
Monte Carlo Markov Chain (MCMC) Description (Model 5.1)
Data:
NT Number of time points
Np Number of space points
N = NT xNp Number of events (space-time points)
NX
x e R ' i = \,...,N Monitoring data at event i
Xi e R i = I,..., N Sum of the monitoring data at event / (the number of monitoring
data for each event i can be 0, 1 or more than 1).
,N CMAQ data at event i (the number of CMAQ data for each event i
is always and only 1).
Di} e R i = I,...,N 7 = 1,...,ND Bias jth basis function evaluated at event /
The bias is evaluated as a linear combination of 2nd order uniform B-spline functions defined
over a 3-dimensional lattice of uniform knots. The coefficients of the linear combination
represent unidimensional control points, that is,
Nฐ
Bias = ^Dvfif.
j=i
If we indicate with N1,N2,N3 the dimensions of the CMAQ grid (that is, h\ = N1 ,
N2 ฆ N3 = N1' and N1-N2-N3 = N), and with Ml ,M2 ,M3 the dimensions of the control-points
grid (this defines the degrees of freedom of the bias, that is, Mx -M2 -M3 = N"), and we
decompose the indexes as: i = ix + N} (i2 +N2i3), / = /, + M] (j2 +M2j3), then the bias matrix is
then defined as
1) /' (/ )/' (/.)/',(/.)
where bk(u) is the 2nd order M1 B-spline basis function evaluated at the parameter point u.
Setting [a2,Zป2]x [t/3,h-, ] the domain over which the bias is defined (such domain must be
bigger than or equal to the CMAQ grid domain), the uniform knots vectors over which the B-
spline basis functions are defined are respectively
Ur = frr,ar,ar,ar+sr,ar+2sr...,ar+(Mr-3)sr,br,br,br} r = 1,2,3
-------
where sr = r = 1,2,3 .
r Mr- 2
Using B-splines as basis functions for the bias allows to control the degrees of freedom of the
bias through the number of control points. Furthermore, the piece-wise nature of the B-spline
functions respects the principle of locality, that is, local information does not affect regions far
from the region where the local information is defined. On the numerical side, B-spline allow a
tensor factorization of the bias matrix into three matrixes B" , = h , (z'r), r = 1,2,3 for a total
dimension of NlMl + N2M2 + N3M3, very much less than the total dimension of the full D-
matrix, which is NXMX ฆ N2M2 ฆ N3M3.
Parameters
Zi i = 1,..., N CAR process at event i.
Iut t = 1,..., NT Mean level of the CAR process.
pD g R"" Vector of coefficient for the bias.
tx Precision of the measurement error in the monitor observations.
ty Precision of the measurement error in the computer observations.
tz Precision of the mean process.
pT Temporal autocorrelation parameter of the mean process.
pF Spatial autocorrelation parameter of the mean process.
Likelihood:
[% I ft(,).Z,.rV]= n(wtx),
\y,\Hm,PD,Ziy]=N(w,+l3DDiy),
where
wt =^ + 2^
Priors:
[z I t\p\p'}= 4>,rz(AV)ฎ aV)))
where AT (p) is the precision matrix corresponding to a time autoregressive model with
parameter p and Ap is the precision matrix corresponding to a space autoregressive model of
order r with a zero boundary condition.
Writing
-------
1 t = lN
J
T
nt ='
np = number of spatial neighbors
and
IX
n Hrf W*,i
T P
z n.tii\zi>vt(i)>0Dy]x
oc
2 = 1
-------
Explicitly:
fl[z,iz,_y,Pr,pF]x
2 = 1
D Y Z
? ,T ,T ,P
,Pf\X,y\
oc
N*
N ( X \2
nfe,
N f tY V
nfe
TX N N,
eXPl"V^^^
Z i=1 ifc=l
exp] - vS [>' - U(i) + Z< + P"l).)]2
i =1
(t* #|Ar (/ )ฎ K(pFj exp|- ^-(Zj [a7 (p*)ฎ K(pFM X
Full Conditional Distributions
Variable: Z, e R
where
\z,\-]=n{a->b,a->)
A = TXNX +Tr +TZ
B -TX [xt - N?{lt^]+ Tr [)', - + /?"/} )]+
Variable: //(eR
where
[^( I"]= A'1)
A = zxYJNf +rYNT +tm
iGlt
B=TX^J[X,-Nfz\+
z'g/j
?'Yh,-(z,+PDD$+T'ef
iฃlt
-------
Variable: f3D g R
\pd\-]=n(a1b,a1)
where
N
a, = ''Zn"d^d,>+sซ'd
i=1
B, = rIf1Dll[>>,-(fWl + Z,)]+TDffD
i=1
Variable: zx
[zx \-\=G(A,B)
where
A = \t"?+r*
^ i=1
54ilk-(A(0 + Z,)f+^
Z z=l Ar=l
Variable: zY
\z'\-\ = G(A,B)
where
2
B = -(fl,i,l + Z,+PDD$+S<
^ i=1
Variable: rz
[tz\-\=G(A,B)
where
2
-------
1 N t \
2 T i=i
Variable: p1
\p'\-hx(sslpTi-(pT)f-''''{'''~')^-\p;(pry)zl(z,-ttV.p'))}
This full conditional is not a recognized form, so it has to be sampled using a Metropolis-
Hastings step.
Jump:
P' ~ Ap' ~ pt )
Acceptance criteria:
mini
exp|- \ S T> (?'> p )z< (z< - ^ (p'-> p ))J
hip'f)
TY\-2N^NT-1'
<=xp \-\tT-(p\ppHz,-rf{p\ppi
Variable: p'
Let
A = i cos
2
f
71
v^+l'y
^ ( IT ^
7t
+ cos /,
iV +1 J
( \ N
f(pp)=n
V y J,
1-/A
Np f i ^
T
y
k I -]ซ ^p\-lp;(pT,pFHz,-k(pt,p-)\
This full conditional is not a recognized form, so it has to be sampled using a Metropolis-
Hastings step.
Jump:
p' ~ Ap' ~pF)
Acceptance criteria:
-------
mini
X(sy)(p')D(p')~i"T exp|-|2!7'(pr./"')z,(z, - A'(pr,/0'))|
Dip' f'1'' expj- l-ฑr;(pT,pr)z,(z, - Kip'.p-i
-------
Appendix D
Appendix D: SAS Code for Developing the Krige Model Estimates
-------
Program Name: Krige Ozone Surface_2001_12.sas
* Note -
* Getting a useful empirical variogram is an iterative process
* Be sure to plot tempvariogramestiamte predicted vs. distance to verify that
* the distance bins reach a distance at which the variogram is at the sill (where it levels off)
* and with distance bin sizes small enough to adequately represent the variogram
* but not so small that there are too few distance pairs in any particular bin
* for example:
* lagdistance used for PM2.5 2001 emp. variogram = 5;* lagdistance used for 03 2002 emp.
variogram =10;
* and each year used 20 distance bins
?
%let cooperpath=C:\OAQPS\Data;
%let otherpath=C:\OAQPS\Data;
%let pathl=&cooperpath.;
%let path2=&cooperpath.;
%let monitor=Monitor_03 2001_12. csv;
%let castnet=castnet_grid_o3_2001_12km;
%let krigeerror=krigepredozoneerror122001;
%let krigedata=krige_ozone_l 2_2001;
Libname sd "&pathl";
libname sd2 "&otherpath";
*
* Importing FRM Monitoring data
*
9
* AQS Ozone Data;
data ozone;
infile "&path2.\&monitor" missover delimiter=',';
input time col row logo3;
if time = . then delete;
proc sort data=ozone(rename=(logo3=value)) out=ozone; * was "Log(ozone)";
by time col row;
run;
*
* Construct variogram
*
-------
%macro Calcvario(ds,day);
proc datasets;
delete tempoz tvario tempmean tempoz2;
* Get single days data;
data tempoz;
set &ds;
if time=&day.;
* Calculate empirical variogram values;
proc variogram data = tempoz outvar = t vario;
compute lagdistance =10 maxlags = 20;
coordinates xcoord = col ycoord = row;
var value;
* Count number of observations for the day;
data tempoz;
set tempoz;
unify = _n_;
proc means data=tempoz noprint;
var unify;
output out=tempmean;
data tempmean;
set tempmean;
if (_stat_ = "MAX");
data _NULL_;
set tempmean;
call symput('nobsa', trim(left(unify)));
* Repeat dataset nobsa times;
data tempoz2;
do unify = 1 to &nobsa.;
output;
end;
data tempoz2;
set tempoz;
do i=l to max;
set tempoz2 nobs = max point = i;
output;
end;
proc sort data=tempoz2;
by unify;
* Add squared differences of values within same cell to variogram dataset;
data tempoz2;
set tempoz2;
-------
rename col = x;
rename row = y;
rename value = val;
data tempoz;
merge tempoz tempoz2;
by unify;
if ((x = col) & (y = row) & (val A= value));
sqdiff = (val - value)* *2;
proc means data=tempoz noprint;
var sqdiff;
output out=tempmean mean=sqdiff;
data tempmean;
set tempmean(rename=(_FREQ_=count));
variog=sqdiff/2;
distance=0;
keep distance variog count;
data tvario;
set tempmean t vario;
if variogA=;
keep distance variog count;
* Append this days empirical variogram to rest;
data allvario;
set allvario t vario;
run;
%mend Calcvario;
%macro Loopvario(ds,DatasetIn);
data allvario;
delete;
run;
%let DatasetID = %sysfunc(Open(&DatasetIn.));
%syscall set(DatasetlD);
%let ReturnVal = %sysfunc(fetch(&DatasetID));
%do %while (&ReturnVal = 0);
%CalcVario(&ds.,&time.);
%let ReturnVal = %sysfunc(fetch(&DatasetID));
%end;
%let ReturnVal = %sysfunc(Close(&DatasetID));
%mend Loopvario;
* Construct a variogram, but subdivide by day;
data datefile;
do time = 1 to 365 by 1;
output;
end;
-------
run;
quit;
%Loopvari o(ozone, datefil e);
* Fit an Exponential Semivariogram Model with nugget;
ods output ParameterEstimates = tempParameterEstimates;
ods exclude ANOVA ConvergenceStatus CorrB EstSummary IterHistory MissingValues
ParameterEstimates;
proc nlin data=allvario Maxlter= 10000;
parms Nugget = I, Scale = 200, Range = 2;
bounds Nugget > 0, Scale > 0, Range > 0;
model variog = Nugget + (Scale - Nugget) * (1 - exp(-(Distance)/(Range)));
output Out=tempVariogramEstimate Predicted=PredictedValue;
run;
proc transpose data=tempParameterEstimates out=tempParameterEstimates;
id Parameter;
var Estimate;
run;
data _NULL_;
set tempParameterEstimates;
call symput('time',trim(left(" 126")));
if NuggetA=. then call symput('NuggetIn',trim(left(Nugget)));
else call symput('NuggetIn',0);
call symput('RangeIn',trim(left(Range)));
call symput(' Scaleln',trim(left(Seale)));
run;
*
* Krige macro
*
9
%macro krige(ds,day);
data forkrige;
set &ds;
where time =&day.;
proc krige2d data=forkrige outest=nugget;
predict var=value;
model Form=Exponential Nugget=&NuggetIn. Range=&RangeIn. S cal e=& Seal eln.;
* model Form=Gaussian Nugget=&NuggetIn. Range=&RangeIn. S cal e=& Seal eln.;
coord xc=col yc=row;
grid griddata=predlocs xc=col yc=row;
data merged;
set nugget(rename=(estimate=krige_predict));
time =&day.;
keep gxc gyc krige_predict stderr time;
-------
data krige_predict;
set krige_predict merged;
ran;
%mend krige;
%macro Loopkrige(ds,DatasetIn);
data krige_predict;
delete;
ran;
%let DatasetID = %sysfunc(Open(&DatasetIn.));
%syscall set(DatasetlD);
%let ReturnVal = %sysfunc(fetch(&DatasetID));
%do %while (&ReturnVal = 0);
%put Loopkrige time = &time.;
%krige(&ds.,&time.);
%let ReturnVal = %sysfunc(fetch(&DatasetID));
%end;
%let ReturnVal = %sysfunc(Close(&DatasetID));
%mend Loopkrige;
* Bring in the CASTNET data for ozone
data ozone validation;
set sd2.&castnet;
proc sort data=ozone_validation out=unique_site nodupkey;
by col row;
data predlocs;
set uniquesite;
keep row col;
run;
%Loopkrige(ozone,datefile);
data sd2.&krigeerror;
set krige_predict;
rename gxc=col gyc=row;
run;
proc export data=krige_predict outfile="&otherpath.\&krigedata..csv" replace; run;
-------
Appendix E
Appendix E: Simulation.PAR (2001,12 km, Ozone-Run 1)
-------
# Simulation parameter file
5-4H # Model version
1, 333 # Grid Size t
40, 213 # Grid Size x
30, 180 # Grid Size y
2001, 01, 01 # Year, month, day of time step 1
1, 333, 4 # Bias Spline in t
40, 213, 8 # Bias Spline in x
30, 188, 7 # Bias Spline in y
0, 1.0E-3 # Prior Parameters for Mu ~ Normal(mean, prec)
0, 1.0E-3 # Prior Parameters for BetaD ~ Normal(mean, prec)
7.5E+7, 1.0E+6 # Prior Parameters for TauX ~ Gamma(mean, prec)
1.0E+7, 1.0E+8 # Prior Parameters for TauY ~ Gamma(mean, prec)
1.0E-3, 1.0E-3 # Prior Parameters for TauZ ~ Gamma(mean, prec)
0, 1 # Prior Parameters for RhoZ ~ Gamma (shape, rate)
C:\ # Directory where data are stored
aqs 12km o3 2001.csv # Pile name where monitor data X are stored
cctm 12km 2001.csv # Pile name where CMAQ csv is stored
1 # Order of neighborhood for Z (between 1 and 4)
0 # Track chain flag for Z
1 # Enable calculation of 4th highest
0 # Track chain flag for 4th highest
C:\ # Directory where the simulation outputs are
written
Surface 03 12 2001.csv # Pile name where mean, covariate, bias,
monitoring, and computer average surfaces are written
Surface 03 12 2001 4h.csv # Pile name where 4th highest surface is written
1234 # Random seed 1
5678 # Random seed 2
1 # Sampling period for thinning
1 # Prompting period for diagnostic
1 # New/Load simulation flag. 1 = New, 0 = Load
50 # Number of burn-in steps
100 # Number of simulation steps
-------
Appendix F
Appendix F: SAS Code for Generating the Summary Statistics
-------
* *
Program Name: Summarize model parameters-0zone_12_2001 .SAS
*====> Please change the following to match your system;
%macro CallVal(RunNum, SimFile, ValMonFile, KrigePredErrFile);
%1 et LocModel OutFil e=C: \0 AQP S\Ozone_2001_ 12\&RunNum;
%1 et LocMonitor V alDat=C: \0 AQP S\Data;
* HBM data*;
%let inchainfile=Chain.csv;
%let modelsurfacefile=&SimFile..csv;
* Edit the date specifying the day before the start date of the calendar year
* 'afterDate' is the day before the start date;
%let afterDate = '31dec2000'd;
%let ttl=Ozone 2001 12 km - &RunNum;
%let minx=40;
%let maxx=213;
%let miny=30;
%let maxy=188;
* Kriged data*;
%let krigedsurface=&KrigePredErrFile;
* Monitor data: improve and stn for PM2.5 and CASTNet for 03*;
%let castnet=&ValMonFile;
%let monrdg = o3_8hrmax;
%let logrdg = lno3_8hrmax;
%let o3Valcols= &monrdg. &logrdg.;
%let Valcols = &o3Valcols.;
libname sd "&LocModelOutFile";
libname sd2 "&LocMonitorValDat";
*
* Calculate the summary statistics on the HBM model parameters
*
?
proc import out = chain
datafile = "&LocModelOutFile.\&inchainfile"
dbms = csv replace;
* tau is the precision parameter in HBM
x denotes FRM monitor precision
y denotes CMAQ 'precision'
z denotes precision for Conditional autoregressive (CAR) process
These are estimated from the characteristics of the HBM Markov chain
-------
setaux is se_taux; ;
ods rtf file= "&LocModelOutFile.\PerformanceOutput_&runnum..rtf';
proc univariate data = chain plot;
var taux tauy tauz;
output out = sumparms
mean = mntaux mntauy mntauz
stderr = setaux setauy setauz
n = ntaux ntauy ntauz;
proc print data = sumparms;
title "Summary Statistics for &ttl";
* posterior variance to be computed from x and z precision ******;
data _null_;
set sumparms;
call symput("sigmax", 1/mntaux);
call symput("sigmaz",l/mntauz);
run;
%put &sigmax;
%put &sigmaz;
* Comparison summaries *;
* Read in HBM predicted surface, referred to as modelsurface;
proc import out = hbm
datafile = "&LocModelOutFile.\&modelsurfacefile"
dbms = csv replace;
data monhbm;
set hbm;
where monitordata > 0;
run; quit;
data hbm_surfacepreds(keep = time xcoord ycoord predavg predstd computerdata test
monitordata
rename = (xcoord=col ycoord=row predAvg=predsurface
ComputerData=CMAQ Surface));
set hbm;
* Remove the data where a prediction was not made *;
time=time-&afterDate.;
* time=time-14975-365;
if predAvg A= -999;
test = predstd*predstd;
* This step exponentiates the results back to normal units *;
array vars [4] predAvg biasAvg monitorData computerData;
do i =1 to 4;
if vars[i] ne -999 then vars[i]=exp(vars[i]);
-------
end;
* The bias is transformed *;
if vars[2] ne -999 then vars[2]=(vars[2]-l)*vars[l];
drop i;
* For the validation results, any monitor result greater than 600 truncated to 600 *;
if vars[3]>600 then vars[3]=-999;
if index(varl2,'0D'x)=0 then covardata=input(varl2,8.);
else covardata=input(substr(varl2,l,index(varl2,'0D'x)-l),8.);
run; quit;
proc sort data = hbmsurfacepreds;
by time col row;
data check;
set hbm surfacepreds;
diffpredCMAQ = predsurface-cmaqsurface;
if monitordata > 0 then diffpredmon = predsurface-monitordata;
else diffpred mon = .;
proc sort data = check;
by time;
proc means data = check noprint;
by time;
var diffpred CMAQ diffpred mon;
output out = sumt(drop = type freq)
mean = mnCMAQdiff mnMondiff
stderr = seCMAQdiff seMondiff
n = nCMAQdiff nMondiff;
run; quit;
proc sort data = check;
by col row;
proc means data = check noprint;
by col row;
var diffpred CMAQ diffpred mon;
output out = sumcr(drop = type freq_)
mean = mnCMAQdiff mnMondiff
stderr = seCMAQdiff seMondiff
n = nCMAQdiff nMondiff;
data moncr;
set sumcr;
where mnmondiff > 0;
run; quit;
* Kriging data for comparison *;
data krige_predict;
set sd2.&krigedsurface.;
krigepred = exp(krige_predict);
-------
run; quit;
proc sort data=krige_predict;
by time col row;
run;
* IMPROVE and STN, or CASTNet, data for comparison *;
data surfvalidation (keep = time col row site &Valcols.);
set sd2.&castnet. (rename=(siteID= site));
where (col ge &minx.) and (row ge &miny.) and (row le &maxx.);
run; quit;
proc sort data = surf validation;
by time col row;
run;
* Combine the model prediction, Kriging results, and the IMPROVE/STN data
* NOTE: The bias is calculated as bias = validation value - predicted value
data combined;
merge krige_predict
surf validation (in = invalidation)
hbmsurfacepreds (in = inpred);
by time col row;
* Data to keep in the file *;
if invalidation;
if predsurfaceA=;
if CMAQsurface = -999 then CMAQsurface = .;
if predsurface = -999 then predsurface = .;
* Calculate the summary stats *;
* regular units *;
msemodel = (predsurface-&monrdg.)**2;
mse_krige = (krigepred-&monrdg.)**2;
mse_CMAQ = (CMAQsurface-&monrdg.)**2;
msediff = msekrige-msemodel;
mse_diff_2 = mse_CMAQ-mse_model;
biasmodel = &monrdg. -predsurface;
biaskrige = &monrdg.-krigepred;
bias_CMAQ = &monrdg.-CMAQsurface;
-------
* log units *;
mse_lmodel = (log(predsurface)-&logrdg.)**2;
mse_lkrige = (krige_predict-&logrdg.)**2;
mse_lCMAQ = (log(CMAQsurface)-&logrdg.)**2;
* compare MSE of Krige to HBM and of CMAQ to HBM;
mseldiffKrige = mselkrige - mselmodel;
mse ldiffCMAQ = mselcmaq - mse lmodel;
biaslmodel = &logrdg. - log(predsurface);
biaslkrige = &logrdg. - krige_predict;
bias_lCMAQ = &logrdg. - log(CMAQsurface);
* - Summarize performance by time *;
proc sort data = combined out=combined_oz;
by time;
run;
* Average daily MSE for HBM, Kriging, and CMAQ *;
* avg mseldiff... > 0 implies HBM error is smaller on average;
* semse... provides measure of variation in the magnitude of square errors;
proc means data = combinedoz noprint;
by time;
var mse lmodel mse lkrige mse lcmaq mseldiffkrige mseldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
output out = mseoz (drop = type freq)
mean = mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ
stderr = semselmodel se mse lkrige semselcmaq
se mse ldiffkrige se mse ldiffcmaq
se bias lmodel se bias lkrige se bias lCMAQ
n = n mse lmodel n mse lkrige n mse lcmaq
n mse ldiffkrige n mse ldiffcmaq
n bias lmodel n bias lkrige n bias lCMAQ;
proc print data = mse oz;
var mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ
se mse lmodel se mse lkrige se mse lcmaq
se bias lmodel se bias lkrige se bias lCMAQ
nmselmodel;
format mse lmodel mse lkrige mse lcmaq
bias lmodel bias lkrige bias lCMAQ
-------
semselmodel se mse lkrige semselcmaq
sebiaslmodel sebiaslkrige se bias lCMAQ
6.4;
title 1 'Average Daily MSE, Bias, and number of observations';
run; quit;
* Annual MSE for HBM, Kriging, and CMAQ. Bias summary for HBM, kriging, and CMAQ
___ *
9
proc means data=combined_oz noprint;
var mselmodel mselkrige mselcmaq mseldiffkrige mseldiffcmaq
biaslmodel biaslkrige bias lCMAQ;
output out = overallmseoz (drop = type freq_)
mean = mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
proc print data = overall mse oz label;
var mse lmodel mse lkrige mse lcmaq
bias lmodel bias lkrige bias lCMAQ;
label mse lmodel = 'Overall MSE HBM'
mse lkrige = 'Overall MSE Krige'
mse lcmaq = 'Overall MSE CMAQ'
bias lmodel = 'Overall Bias HBM'
bias lkrige = 'Overall Bias Krige'
biaslcmaq = 'Overall Bias CMAQ';
format mse lmodel mse lkrige mse lcmaq
bias_lmodel bias_lkrige bias_lCMAQ 5.2;
run; quit;
* Summarize performance by site *;
proc sort data = combinedoz;
by site;
proc means data = combined oz noprint;
by site;
var mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
output out = mse_oz2
mean = mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
proc print data = mse_oz2;
var mse lmodel mse lkrige mse lcmaq
bias lmodel bias lkrige bias lCMAQ;
title 1 'Average MSE by site for 03';
* Comparing HBM to CMAQ and Krige based on MSE *;
-------
* mse_oz is daily statistics (by 'time');
data mseozl;
set mseoz;
retain suml sum_2;
if mseldiffkrige > 0 then do;
krige_larger=l;
sum_l+l;
end;
if mseldiffcmaq > 0 then do;
cmaq_larger=l;
sum_2+l;
end;
proc means data = mse ozl noprint;
var sum_l sum_2 time;
output out = Improve_Time(drop = type freq_ tl t2)
max = krigeworse CMAQworse
n = tl t2 ndays;
data Improvetime;
set Improve time;
Pct_Time_Krige_worse = (krigeworse / ndays) * 100;
Pct_Time_CMAQ_worse = (cmaqworse / ndays) * 100;
proc print data = improve time;
title 1 "&ttl: % Improvement over Kriging and CMAQ";
var ndays pct_time_krige_worse krigeworse pct_time_CMAQ_worse cmaqworse;
format pct_time_krige_worse pct_time_CMAQ_worse 7.2;
run; quit;
* mse_oz2 is statistics by site;
data mse_oz3;
set mse_oz2;
retain sum l sum_2;
if mse ldiffkrige > 0 then do;
krige_larger=l;
sum_l+l;
end;
if mse_ldiffcmaq>0 then do;
cmaq_larger=l;
sum_2+l;
end;
run; quit;
proc means data = mse_oz3 noprint;
var sum_l sum_2 _freq_;
output out = Improve_site(drop = type freq_ tl t2)
max = krigeworse CMAQworse
-------
n = tl t2 nsites;
data Improvesite;
set Improve site;
Pct_site_Krige_worse = (krigeworse / nsites) * 100;
Pct_site_CMAQ_worse = (cmaqworse / nsites) * 100;
proc print data = improve site;
title 1 "&ttl: % Improvement over Kriging and CMAQ";
var nsites pct_site_krige_worse krigeworse pct_site_CMAQ_worse cmaqworse;
format pct_site_krige_worse pct_site_CMAQ_worse 7.2;
run; quit;
*
* 95% Prediction Interval: this interval is calculated using modelstd=sqrt(sigmaz+sigmax)
* MCMC Prediction Interval: this interval is calculated using
modelstd=sqrt(predstd*predstd+sigmax)
* here begin to calculate the prediction interval
*
?
* pmsurfacepreds is HBM predicted surface;
* sigmax is posterior variance of FRM data;
* sigmaz is posterior variance of CAR process;
data compare;
merge krige_predict(keep=time col row krige_predict stderr)
hbm_surfacepreds(keep=time col row predsurface cmaqsurface predstd)
surfval i dati on(i n=Val i dati on);
by time col row;
if Validation;
if predsurface A=;
si gmax=& sigmax;
sigmaZ=& sigmaz;
* modelstd=sqrt(predstd*predstd+sigmax);
modelstd=sqrt(sigmaz+sigmax);
* keep track of confidence interval coverage of prediction vs. monitor measurement;
if ((&logrdg.>log(predsurface)-2*modelstd) and (&logrdg.krige_predict-2*stderr) and (&logrdg.
-------
*proc datasets library=work kill;
run; quit;
%mend CallVal;
*%CallVal(Runl, simulation_o3_2001_12_runl, castnet_grid_o3_2001_12km,
krigepredozoneerror122001);
*%CallVal(Run2, simulation_o3_2001_12_run2, castnet_grid_o3_2001_12km,
krigepredozoneerror122001);
*%CallVal(Run3, simulation_o3_2001_12_run3, castnet_grid_o3_2001_12km,
krigepredozoneerror122001);
*%CallVal(Run4, simulation_o3_2001_12_run4, castnet_grid_o3_2001_12km,
krigepredozoneerror122001);
%CallVal(Run5, simulation_o3_2001_12_run5, castnet_grid_o3_2001_12km,
krigepredozoneerror122001);
-------
Appendix G
-------
T-SpACE User's Guide
Hierarchical Bayesian
Space-Time Modeling
of Air Pollution Data
-------
January, 2018
Version 5 4h.0.1.21
-------
-------
Table of Contents
1.0 Introduction 4
2.0 Model Data 4
2.1 Format of initial monitor data from AQS 9
2.2 Format of the initial simulated surface data (CMAQ) 9
3.0 SIMULATION Model 11
4.0 Software to Run the Model 11
4.1 Installation 11
4.2 Menu Bar 14
4.2.1 File 15
4.2.2 View 15
4.2.3 Tools 15
4.2.3.1 Compare A QS Data Files 16
4.2.3.2 Compare Air sites Files 17
4.2.3.3 Average 4th-Highest Surface Files 18
4.2.4 Choosing a Region or State for the Simulation 19
4.3 Step 1: Choose Time/Grid 20
4.3.1 Please select a date range for the study period 20
4.3.2 Please select a folder to save all CSV files 21
4.3.3 Choose X and Y coordinates of the grid 21
4.3.4 Choose Air Pollutant 22
4.4 Step 2: Prepare Model Input Data 22
4.4.1 CMAQ: Choose Input CMAQ data file 23
4.4.2 CMAQ: Output file name 24
4.4.3 CMAQ: Create a second output file with latitude and longitude 24
4.4.4 CMAQ: Execute 24
4.4.5 Monitor 28
4.5 Step 3: Model Specification 29
4.5.1 New, Open, Save 29
4.5.2 Grid 30
4.5.3 Priors-First Tab 31
4.5.4 Priors-Second Tab 32
4.5.5 Data 33
4.5.6 Boundaries 34
4.5.7 Track 36
4.5.8 Simulation 36
4.5.9 4th Highest 38
4.5.10 *.PAR file 39
4.6 Step 4: Launch Model 40
4.7 Step 5: Launch Validation 45
4.7.1 Select Surface File 45
4.7.2 Select Chain File 45
4.7.3 Select Krige Prediction File 46
4.7.4 Select Validation Monitors File 47
4.7.5 Select Report File 48
T-SpACE User's Guide Page 2 Version 5_4h.0.1.21
January 2018
-------
4.7.6 Report Title 48
4.7.7 Run Validation 48
5.0 References 51
Section A: Detailed Description Of The Hierarchical Bayesian Space-Time Model For Modeling
Air Pollution Data 55
Section B: Validation SAS Program Example 60
Section C: Validation Report Example 71
Section D: Cells.txt 85
Section E: Summary.CSV file description 87
Section F: 4th Highest concentration surface 89
Section G: Diagram of Preparation of Validation Files 91
Section H: Visualization of T-SpACE Three-Dimensional Coordinate System 85
Section I: T-SpACE Utilities 89
T-SpACE User's Guide Page 3 Version 5_4h.0.1.21
January 2018
-------
1.0 INTRODUCTION
EPA has developed a method to combine air pollution monitoring data with air quality model
output to maximize the advantages offered by both data sources, while minimizing the
disadvantages of each, when predicting fine particulate matter (PM2.5) and ozone (O3)
concentrations. Air quality models estimate the spatial and temporal gradients of air pollution
based on emissions inventories and meteorological information. These models, while providing
estimates over large regions at relatively low cost, have been found to be biased and to have
greater error than air pollution monitoring networks. Air pollution monitoring networks
generally provide relatively accurate, unbiased results. However, because air pollution
monitoring networks are sparsely and irregularly spaced over large spatial domains, they are not
able to produce large-scale predictions of air pollutant concentrations.
EPA has developed a Temporal-Spatial Ambient Concentrator Estimator (T-SpACE) model that
combines air pollution monitoring data from the 2,545 National Air Monitoring Stations/State
and Local Air Monitoring Stations (NAMS/SLAMS) with estimates from the Models-
3/Community Multiscale Air Quality (CMAQ) and the Comprehensive Air Quality Model with
Extensions (CAMx) air quality models to maximize the advantages offered by both data sources,
while minimizing the disadvantages of each. T-SpACE requires data preparation and model
parameter input from the user to run a simulation to produce an estimated air pollution
concentration surface. To streamline the set-up of the T-SpACE model, EPA developed the
software and the associated guidance described in this User's Guide to allow the user to prepare
the input data for the model run, choose the model specifications, run the model, receive output
files containing the estimated air pollution concentration surface, and run a set of validation
procedures to compare the T-SpACE run with a standard kriging method, and the CMAQ/CAMx
results.
2.0 MODEL DATA
Currently, the T-SpACE model requires two input data sources, air pollution estimates from the
Models-3/Community Multiscale Air Quality (CMAQ) or the Comprehensive Air Quality Model
with Extensions (CAMx) air quality models, and air pollution monitoring data from the
NAMS/SLAMS air pollution monitoring network. Each of these two data sources offers unique
advantages and disadvantages. One strong advantage of air pollution monitoring data is that it
provides information at specific spatial locations. Another advantage of air pollution monitoring
data is that the data tend to be more accurate and exhibit less bias than comparable
CMAQ/CAMx output. However, air pollution monitoring data do not provide information about
every location within a spatial domain, since it is impossible to know exactly what an air
pollution monitor would have recorded in a currently unmonitored location. In contrast to air
pollution monitoring data, CMAQ/CAMx data does provide information across an entire domain
of interest, but CMAQ/CAMx data can provide estimates of air pollutant concentrations.
However, CMAQ/CAMx data are averaged over square grid cells, so the spatial information is
not as detailed as that obtained with air pollution monitors. Also, CMAQ/CAMx data can be
biased in some cases, due to the fact that the accuracy of the results can be strongly influenced
by the input emissions inventories and meteorological data.
-------
Under the CMAQ modeling system, grid cells can either be 36 km x 36 km (which represents the
"parent domain" covering the entire continental US) or 12 km x 12 km (which represents
primarily the eastern and Midwest regions of the US up to 2007, and the entire US from 2008
and onward). The 2001 and 2002 PM2.5 and O3 CMAQ data have been assessed with the CMAQ
model. Figures 1, 2, and 3 illustrate the area of coverage for the 2001 12 km CMAQ data, 2001
36 km CMAQ data, and the 2002 12 km and 36 km CMAQ data, respectively.
T-SpACE User's Guide
Page 5
Version 5_4h.0.1.21
January 2018
-------
Figure 1. The maximum area of the United States covered by the 2001 CMAQ data 12 km
grid.
Figure 2. The maximum area of the United States covered by the 2001 CMAQ data 36 km
grid.
T-SpACE User's Guide
Page 6
Version 5_4h.0.1.21
January 2018
-------
Figure 3. The maximum area of the United States covered by the 2002 CMAQ data for
both the 12 km and 36 km grids.
j12km Eastern (large) CM AO Domain
x,y: -1008000, -1620000
J col: 279 row: 240
36km Domain V V k
Lower Left Corner: -2736000, -208800I
Number of Cols, Rows: 148x112 "V
Lambert Projection: >
1st std parallel: 33 \ \
2nd std parallel: 45
central meridian: -97
latitude of projection's origin: 40
The CMAQ computer model uses emissions information and meteorological data as inputs and
calculates, via dispersion modeling, predicted 24-hour integrated concentrations. The grid cells
are approximately 36 km M 36 km for the 36 km file and 12 km x 12 km grids for the 12 km file.
Detailed information on the CMAQ modeling system is available at https://www.epa.gov/cmaq
and at https://www.cmascenter.org.
The monitor data consists of daily concentrations that are collected from Federal Reference
Method (FRM) samplers within the National Air Monitoring Stations/State and Local Air
Monitoring Stations (NAMS/SLAMS) networks. Only those concentrations with a sample
duration of one day (24-hour integrated sample) are included in the analysis. The data was
extracted from the EPA Air Quality System (AQS). Detailed information AQS can be found on
the AQS web page at https://www3.epa.gov/ttn/airs/airsaqs/detaildata/downloadaqsdata.htm.
Table 1 lists the data sets that have been analyzed using the T-SpACE software.
T-SpACE User's Guide
Page 7
Version 5_4h.0.1.21
January 2018
-------
Table 1. Input data to the T-SpACE.
Data Source
Air
Pollutant
(units)
Dates
Resol
ution
Input file name
File
Format
Available?
*
Community
Multiscale Air
Quality (CMAQ)
model data
January 1,
2001 -
December
31, 2001
12 km
pm25_12km.ioapi
NetCDF
Yes
pm25
(|jg/M3)
36 km
cmaq_36km.ioapi
NetCDF
Yes
January 1,
2002-
12 km
2002ac_v4.61_l3b_eus_12km.dailyavg.pm25
NetCDF
Yes
December
31,2002
36 km
2002ac_v4.61_l3b_us36b.dailyavg.pm25
NetCDF
Yes
January 1,
2001 -
December
31, 2001
12 km
ozone.cone
NetCDF
Yes
36 km
Not available
No
03 (ppm)
January 1,
2002-
12 km
2002ac v4.61 L3b EUS 12km.combine.hourly.
03.total365
NetCDF
Yes
December
31,2002
36 km
2002ac v4.61 L3b us36b.combine.hourly.03.to
tal365
NetCDF
Yes
Daily
concentrations
from the Federal
Reference
Method (FRM)
samplers of the
National Air
Monitoring
Stations/State
and Local Air
Monitoring
Stations
(NAMS/SLAMS)
networks
24-hour
integrated
PM2.5
concentration
(|jg/M3)
January 1,
2001 -
December
31, 2001
NA
AQSPROD_373180-1.txt
Text
Yes
January 1,
2002-
December
31,2002
NA
AQSPROD_38771-1.txt
Text
Yes
03 (ppm)
January 1,
2001 -
December
31, 2001
NA
R D_501 _44201 _2001 -0 .txt
Text
Yes
January 1,
2002-
December
31,2002
NA
R D_501 _44201 -2002-0.txt
Text
Yes
* The data is available from EPA directly. The AQS monitor data is directly available from
https://www3.epa. gov/ttn/airs/airsaqs/detaildata/downloadaqsdata.htm.
T-SpACE User's Guide
Page 8
Version 5_4h.0.1.21
January 2018
-------
2.1 Format of initial monitor data from AQS
Currently the input file must be the same format as the AQS *.txt files that are available on
EPA's web site. The two lines that describe the contents of each data row are reproduced here
and Figure 4 illustrates the required structure of the data.
# RD | Action Code | State Code | County Code | Site ID | Parameter | POC | Sample Duration |
Unit | Method | Date | Start Time | Sample Value | Null Data Code | Sampling Frequency j
Monitor Protocol (MP) ID j Qualifier - 1 j Qualifier - 2 j Qualifier - 3 | Qualifier - 4 j Qualifier - 5
Qualifier - 6 | Qualifier - 7 | Qualifier - 8 | Qualifier - 9 | Qualifier - 10 |
Alternate Method Detectable Limit | Uncertainty
# RC j Action Code | State Code j County Code | Site ID j Parameter | POC | Unit | Method | Year
Period j Number of Samples | Composite Type | Sample Value | Monitor Protocol (MP) ID |
Qualifier - 1 | Qualifier - 2 | Qualifier - 3 | Qualifier - 4 | Qualifier - 5 | Qualifier - 6 | Qualifier - 7
| Qualifier - 8 | Qualifier - 9 j Qualifier - 10 | Alternate Method Detectable Limit | Uncertainty
Figure 4. Screen shot of the required AQS Monitor data file structure.
RD_LCDaily_88101_01"l.tKt - Notepad
File Edit Format View Help
iTsTT]
of
# RD
Acti on
Code
state code | County code | site id [Parameter
# RC
Action
Code
St ate
code|county code|site id| parameter
RD
i
01
003
0010
88101
1
7
105
120
20010101
00:00
9.1|
3 |
RD
i
01
003
0010
88101
1
7
105
120
20010104
00:00
5.5|
31
RD
i
01
003
0010
88101
1
7
105
120
20010107
00:00
12.4
13
RD
i
01
003
0010
88101
1
7
105
120
20010110
00:00
12. 8
13
RD
i
01
003
0010
88101
1
7
105
120
20010113
00:00
IavI:
ฆ i I
RD
i
01
003
0010
88101
1
7
105
120
20010116
00:00
1AV13 11
RD
i
01
003
0010
88101
1
7
105
120
20010119
00:00
4.1|
3!
RD
i
01
003
0010
88101
1
7
105
120
2001012 2
00:00
11.4
| 3
RD
i
01
003
0010
88101
1
7
105
120
2001012 5
00:00
15.3
[3
RD
i
01
003
0010
38101
1
7
105
120
2001012 8
00:00
14.1
13
RD
i
01
003
0010
88101
1
7
105
120
20010131
00:00
6.1|
31
RD I I
01
003
0010
88101
1
7
105
120
20010203
00:00
7.6|
3|
RD
i
01
003
0010
88101
1
7
105
120
20010206
00:00
14. 2
| 3
RD
i
01
003
0010
88101
1
7
105
120
20010209
00:00
7.4|
31
RD
i
01
003
0010
88101
1
7
105
120
20010212
00:00
14. 3
13
RD
i
01
003
0010
88101
1
7
105
120
20010215
00:00
4.7|
3!
RD
i
01
003
0010
88101
1
7
105
120
20010218
00:00
14. 8
| 3
RD
i
01
003
0010
88101
1
7
105
120
20010221
00:00
5.1|
3 |
RD
i
01
003
0010
88101
1
7
105
120
20010224
00:00
16. 6
13
RD
i
01
003
0010
88101
1
7
105
120
20010227
00:00
10. 2
13
RD
i
01
003
0010
88101
1
7
105
120
20010302
00:00
8.7|
3 I
RD
i
01
003
0010
88101
1
7
105
120
20010305
00:00
11.4
13
RD
i
01
003
0010
88101
1
7
105
120
20010308
00:00
16.4
I 3
RD
i
01
003
0010
88101
1
7
105
120
20010311
00:00
19. 3
13
RD
i
01
003
0010
88101
1
7
105
120
20010314
00:00
7.5|
3 !
-------
The CMAQ data files are NetCDF format data files with the following variables. Figure 5
illustrates the structure of the data file. Note that the illustration is specific to PM2.5. Ozone
would replace PM2.5 for the ozone data.
netcdf file: pm25.conc
variables:
TFLAG (Integer)
PM25 (Float)
LambertConformal (Char)
X (Double)
Y (Double)
Z (Double)
Time (Integer)
Timestep, format is YYYYDDD
In ppbV
Lambert Conformal Conic
Synthesized x coordinate,
Synthesized y coordinate,
Synthesized z coordinate,
(where DDD is 1-36
or HHMMSS
in km
in km
in km
Seconds since 2001-01-01 01:00:00 UTC
Figure 5. Structure of the simulated surface input data file.
^ pm25.conc.cdf (Table View)
JnjjcJ
Dimensions
Name
Description
rrSTEP
length = S760; (index)
DATE-TIME
length = 2; (index)
LAY
length = 1; (index)
VAR
length = 1; (index)
ROW
length = 1 88; (index)
COL
length = 213; (index)
Name
Value
IOAPI VERSION
$ld: @(#) ioapi library version 3.0beta OpenMP enabled $
EXEC ID
combineV3.0
FTYPE
1
CDATE
2005294
CTIME
131246
WDATE
2005294
WTIME
131246
SDATE
2001001
STIME
10000
TSTEP
10000
NTHIK
1
NCOLS
213
NROWS
188
NLA.YS
1
NVARS
1
GDTYP
2
Description
TFL^G; ()
USTE P, VAR, DATE-TIM E]_
PM25; (ppbV
USTE P, LAY, R OW, COL]
T-SpACE User's Guide
Page 10
Version 5_4h.0.1.21
January 2018
-------
3.0 SIMULATION MODEL
A component of the T-SpACE software contains the model executable software which processes
the input data. In this version of the T-SpACE software, the model executable software
component is Model5-4H.exe. This executable file is transparent to the user and is only listed
here for documentation purposes only. The executable file included in this software has been
developed for a 32-bit, Windows XP or higher machine. A 64-bit, Windows version of the
executable is also available. The executable file implements a custom-designed Markov Chain
Monte Carlo (MCMC) algorithm and can calculate an average, three consecutive year (annual)
4th highest concentration surface for PM2.5. Section A provides a detailed description of the
MCMC algorithm that has been developed for T-SpACE.
4.0 SOFTWARE TO RUN THE MODEL
The software that runs the T-SpACE model, using the data described above, can be grouped
under five major components (steps). The last step visualizes the resultant air pollutant
concentration surface. The steps are as follows:
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Choose Time/Grid
Prepare Model Input Data
Model Specification
Launch Model
Launch Validation
Visualize Surface
Each of these steps is described below. A user must start with Step 1 and follow each step
sequentially. Even if the user has run the T-SpACE software previously and has an existing
parameter (*.PAR) file that they would like to use to run a simulation, the user must run Step 1
first, proceed to Step 2 to specify the location of the CMAQ/CAMx surface data file that
contains the grid information, and then proceed to Step 3 to locate the user-generated *.PAR file.
Please note that all *. CSV files generated by the Temporal-Spatial Ambient Concentrator
Estimator (T-SpACE) are accessed (opened and closed) multiple times during the simulation
session. During development and testing of the Temporal-Spatial Ambient Concentrator
Estimator, it was discovered that real-time virus scanning software interfered with writing to the
* CSV files, especially when large files are generated. When real-time virus scanning software
operates in the background, whenever a request is made to open a file, the real-time virus
scanning software scans the file, thus locking it from use by any other program. If the file is
large, this can take a relatively large amount of time. Because the T-SpACE software must open
and close the *. CSV files several times, when the files are locked by real-time virus scanning
software, this interferes with the ability of T-SpACE to write the results of the T-SpACE
simulation to the appropriate files. Because of this interaction between virus scanning software
and T-SpACE, it is recommended that the model be run on a computer without virus scanning
software, or on a computer where the virus scanning software has been disabled.
4.1 Installation
When the T-SpACE installation CD is inserted in the CD drive of the computer, the installation
process should begin immediately as shown in Figure 6. The user must click the "Next" button to
continue the installation.
T-SpACE User's Guide
Page 11
Version 5_4h.0.1.21
January 2018
-------
Figure 6. First screen during the installation process.
Bayesian Air Pollution Model
s m
Welcome to the Bayesian Air Pollution Model Setup
Wizard
The installer will guide you through the steps required to install Bayesian Air Pollution Model on your
computer.
WARNING: This computer program is protected by copyright law and international treaties.
Unauthorized duplication or distribution of this program, or any portion of it may result in severe civil
or criminal penalties, and will be prosecuted to the maximum extent possible under the law.
< Back
I
Figure 7. Location for the Hierarchical Bayesian Air Pollution Model Software (T-SpACE)
js' Bayesian Air Pollution Model
Select Installation Folder
The installer will install Bayesian Air Pollution Model to the following folder.
To install in this folder, click "Next". To install to a different folder, enter it below or click "Browse".
Folder:
|C:\Program Files\
Install Bayesian Air Pollution Model for yourself, or for anyone who uses this computer:
C Everyone
< Just me
T-SpACE User's Guide
Page 12
Version 5_4h.0.1.21
January 2018
-------
Figure 7 prompts the user to select the directory/folder where the T-SpACE software is installed.
(Please note that this is not where results are stored. Only system files are stored here.) The user
must specify a directory/folder other than the default folder (C:\Program Files) to store results.
The user must also select whether T-SpACE will be available only to the assigned user/owner of
the computer or available to anyone logging into the computer where T-SpACE is installed.
Figure 8. Screen Indicating that Installation of the Software has completed.
ig) Bayesian Air Pollution Model
EJ n 0
Confirm Installation
The installer is ready to install Bayesian Air Pollution Model on your computer.
Click "Next" to start the installation.
User must choose "Next" to continue the installation (Figure 8).
Figure 9. Screen indicating that installation of the software has completed.
Bayesian Air Pollution Model
~ a
Installation Complete
Bayesian Air Pollution Model has been successfully installed.
Click "Close" to exit.
Please use Windows Update to check for any critical updates to the .NET Framework.
T-SpACE User's Guide
Page 13
Version 5_4h.0.1.21
January 2018
-------
The user must click "Close" to complete the installation (Figure 9). Please note that T-SpACE
requires the Microsoft .NET 2.0 framework. During the installation .NET 2.0 has been installed
on your computer. If you encounter problems during the installation it may be that you do not
have privileges for installing software on your machine. If problems do arise, please see your
system administrator to assist you in the installation.
Figure 10. Opening screen for the T-SpACE (Hierarchical Bayesian Model) software.
G Hierarchical Bayesian Model
~H
File View Xools Help
Step 1: Choose Time/Grid
Step 2: Prepare Model Input Data
IB Step 3: Model Specification J* Step 4: Launch Model jj] Step 5: Launch Validation
Step 1: Choose Time/Grid
Please select a date range for the stud1/ period
Start date of study period
January. 2001
S3
31 1
2
3
4
c
6
7 8
3
10
11
12
13
14 15
16
17
18
19
20
21 22
23
24
25
26
27
28 29
30
31
1
2
3
E
January, 2001
> ]
1
2
3
4
c
6
y
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
5
30
31
Please select a folder to save all CSV files to. This folder will
be remembered throughout the process.
I I .Apply State or Region to Grid:
Disclaimer
12 km grid does not cover the entire United States.
Click to see a graphic of 2QD1 or the 2QQ2 12 km grids.
Choose X and Y coordinates of the grid
T 2 I |2
X 1
Y 1
213
180
Choose Air Pollutant
Ozone
PM 2.5
4.2 Menu Bar
There is a menu bar which is available to users throughout the simulation process. The menu bar
is at the top of the screen as illustrated in Figure 10. The menu bar includes the following:
T-SpACE User's Guide Page 14 Version 5_4h.0.1.21
January 2018
-------
File
View
Tools
Step 1: Choose Time/Grid
Step 2: Prepare Model Input Data
Step 3: Model Specification
Step 4: Launch Model
Step 5: Launch Validation
File, view, and tools will be described here, while the following sections will provide details for
each of the steps.
4.2.1 File
Figure 11 illustrates what is available when the user clicks on File.
Figure 11. Main Menu: File.
ฉ Hierarchical Bayesia... - )j ~ H3!i
File
View Too
s
L t- - * i W
J.
Exit
f *
The only option available to a user is to Exit the T-SpACE software. Clicking on exit will close
the software.
4.2.2 View
Figure 12 illustrates that there are two options for the user under 'View'. The user can choose to
display the T-SpACE steps (Step 1 through Step 5) on the toolbar by checking
Figure 12. Main Menu: View.
G Hierarchical Bayesia... ~(30
File
View
Tools
| ^ j Toolbar
j Status Bar
Status
"ii
'Toolbar', or choose not display the model steps on the toolbar, by un-checking 'Toolbar'.
4.2.3 Tools
T-SpACE User's Guide Page 15 Version 5_4h.0.1.21
January 2018
-------
Seven tools have been developed for the user. Three are shown below (and the entire list will be
provided in Section I):
Compare AQS Data Files
Compare Airsites Files
Average 4th-Highest Surface Files
Figure 13. Main Menu: Tools.
G Hierarchical Bayesia...
-
~ X
File View
Tools
-ซ ni d e!_
Status
Compare AQS Data Files
Compare Airsites Files
Average 4th-Highest Surface Files
Figure 13 illustrates the three options available under Tools.
4.2.3.1 Compare AOS Data Files
Figure 14. AQS File Compare.
J AQS File Compare
Select 2 AQS Data Files to Compare
file 1
File 2
Compare
BBS
[ Browse ..
Save
Results
T-SpACE User's Guide
Page 16
Version 5_4h.0.1.21
January 2018
-------
This tool provides the user with the ability to compare *.CSV model input monitor data files to
understand the differences between the two files. The *.CSV file is the file developed in Step 2
(discussed later). Figure 14 shows that the user chooses two files to compare, clicks on the
"Compare" button, and the results of the comparison are provided in the window below the
"Compare" button. Note that this is a line-by-line file comparison.
This tool is useful for users who want to know if the input files they have created in the past are
the same or not.
4.2.3.2 Compare AirsUes Files
Airsites is a file that is used to provide the latitude and longitude for each of the air pollution
monitoring sites. This tool allows the user to compare two airsites files to determine if they are
the same or if there are differences. If there are differences, the differences are listed. Figure 15
illustrates the structure of the Airsites file, while Figure 16 illustrates the comparison screen.
Figure 15. Airsites.TSV file.
T
airsites.tsv -
Notepad
(300
File Edit Format
View Help
010010001
-86.472778
+32.437222
6
A
010010002
-86.443611
+32.428333
6
H
010010003
-86.791667
+32.332222
6
010030002
-87.706944
+30.552222
6
010030003
-87.713611
+30.555278
6
010030010
-87.881389
+30.497778
6
010050001
-85.135278
+31.891111
6
010050002
-85.606234
+31.664137
6
010070001
-87.233056
+32.903889
6
010090001
-86.464444
+33.956944
6
010090002
-86.473056
+33.949444
6
010150001
-85.824444
+33.656667
6
010150002
-85.837778
+33.685556
6
010210001
-86.637778
+32.839444
6
010210002
-86.628889
+32.839444
6
0102 30001
-88.051944
+32.209167
6
010230002
-88.053611
+32.205556
6
0102 50001
-87.781944
+31.966389
6
010250002
-87.773333
+31.694167
6
010270001
-85.802222
+33.281111
6
010310002
-87.259167
+31.056667
6
010330001
-87.886667
+34.736944
6
010330007
-87.665000
+34.760000
6
010330008
-87.643056
+34.760833
6
0
T-SpACE User's Guide
Page 17
Version 5_4h.0.1.21
January 2018
-------
Figure 16. Airsites.TSV file comparison screen.
AQS File Compare
Select 2 Arsites Data Files to Compare
File 1 |
File 2
Compare
Close
4.2.3.3 Average 4th-Highest Surface Files
This tool allows the user to calculate a three consecutive year average 4th highest surface file.
The calculation of a single 4th highest surface file will be discussed in Section 4. See [2] for an
explanation of the calculation of the three-year average.
In order to calculate the three-year average, a user must specify three 4th highest surface files. It
is recommended that these files be calculated on similar time frames and with a similar number
of burn-in and simulation steps. Figure 17 illustrates the input screen.
T-SpACE User's Guide
Page 18
Version 5_4h.0.1.21
January 2018
-------
Figure 17. Average 4,h highest surface calculation input screen.
ฆt? Average 4th Highest Surface Files
. D|X|
Select 3 Fourth-Highest Surface Files to Average
File 1
I
Browse ...
File 2
Browse ...
File 3
Browse ...
Average File {Output)
Browse .
Average
Close
4.2.4 Choosing a Region or State for the Simulation
0 Apply State or Region to Grid:
Please select a CSV file with State and Region grid information:
4?
Lj
Disclaimer:
12 km grid does not cover the entire United Siates.
Click to see a araohic of 2001 or the 2002 12kmarids.
T-SpACE has a feature
which allows users the
ability to identify the
appropriate grid coordinates
for a particular region or
state. On the screen for
Step 1, the user checks the
box next to "Apply State or
Region to Grid", The user
is then provided four default CSV files to choose the appropriate coordinate system. The default
files are found in the directory
\StateFiles where the software was
loaded and are named:
StateGrid2G01 -12kni.csv
StateGrid200 l-36km.csv
StateGrid2002-12km.csv
StateGrid2002-36km.csv
The user must first load this file into the
system by choosing the file. Once the
T-SpACE User's Guide
0 Apply State or Region to Grid:
Please select a CSV file with State and Region grid information:
C:\Documents and Settings'''ฆฆhartfondXMy Documents-.Projects'ฃP
Disclaimer:
12 km grid does not cover the entire
Click to see a graphic of 2D01 orth
Choose X and Y coordinates of the grii
T 2
Virginia
Washington
West Virginia
Wisconsin
Wyoming
North East
NorthCentral
NorthWest
Page 19
Version 5_4h.0.1.21
January 2018
-------
user has chosen the file, a
dropdown menu will be provided
to the user. This dropdown menu
shows the states that are available
and provides six regions. Note the
disclaimer (The 12 km2 grid does
not cover the entire United States),
which is true for all years before
2008. Two hyperlinks are
provided to allow the user the
opportunity to see the 12 km2 state
coverage for the years 2001 and
2002.
Once the user makes their choice,
the grid coordinates are placed in
the X and Y rows on the screen.
Using the latitude and longitude of
the lowest left corner and highest right corner, the state is transformed to grid coordinates. The
latitude and longitude of the centroid of the grid cell is used to define whether the grid appears in
the state or not. Note that this essentially defines each state as a rectangle, therefore portions of
surrounding states may be 'connected' to the chosen state.
4.3 Step 1: Choose Time/Grid
Step 1 consists of the following actions:
Choosing the time frame over which the simulation will be run
Establishing the location of all output files
Choosing the area of the United States over which the simulation will be run
As shown above, Figure 10 illustrates the input screen for collecting this information.
4.3.1 Please select a date range for the study period
The study period is limited to at most one calendar year of data. The user cannot choose a time
frame that spans two calendar years. For instance, the software will not allow a user to choose a
date in one year and choose a second date in another year, i.e., December 15, 2006 to January 5,
2007 would not be allowed. Please note that all testing to date has utilized only data from the
years 2001 and 2002.
In order to choose the time frame, the user can click on the month. They will see a drop down
box that will allow them to choose the month of interest. Or, the user can use the arrows at the
top of each calendar to navigate between months. The date within the month is chosen by
clicking on the number on the calendar.
0 Apply State or Region to Grid:
Please select a CSV file with State and Region gnd information:
C:\Documents and Settings hartfond .My Documents .Projects'.EP
Delaware
Disclaimer:
12 km grid does not cover the entire United States.
Click to see a graphic of 2DD1 orthe 2002 12 km grids.
Choose X and Y coordinates of the grid
T 2 2
174
176
111
125
Choose .Air Pollutant
Ozone
PM 2.5
T-SpACE User's Guide
Page 20
Version 5_4h.0.1.21
January 2018
-------
As the user chooses the start and end date for the period, they will see the T: row under "Choose
X and Y coordinates of the grid" change. The value in T represents the number of days from
January 1 for the value that is chosen, where January 1 is day 1. As can be seen in Figure 10,
since January 2 is both the start and end date, the value of T is 2, 2.
4.3.2 Please select a folder to save all CSV files
Next, the user is asked to choose a location where all files created during the simulation will be
written and stored. (Note that this is different from the system file location discussed earlier.)
Input data files are created in Step 2, the model parameter specification file (*.PAR) is created in
Step 3, and the model simulation output files are created in Step 4. All these files will be placed
in the specified folder. If no folder is specified, the default location of the files created with this
software will be "C:\", the root folder.
4.3.3 Choose X and Y coordinates of the grid
As mentioned in 4.3.1, the T coordinates are automatically updated as the user navigates the start
and end date calendars. The user cannot make changes to the T row manually. All changes must
be made using the calendar described above. The X and Y coordinates that are referred to in this
area of the input screen, are the coordinates of the square grid over which the user would like to
run T-SpACE model. As shown in Figure 18, the coordinates are read as follows:
Figure 18. Assigning grid coordinates.
Choose XandI Y coordinates of the grid
I CN
2
1 LLx
250 URx
1 LLv
25C URy
(LLx, LLy) -> (URx, URy)
where
LLx Lower Left x-coordinate of the square grid. This is the first column of the input
screen.
LLy Lower Left y-coordinate of the square grid. This is the first column of the input
screen.
URx Upper Right x-coordinate of the square grid. This is the second column of the
input screen.
URy Upper Right y-coordinate of the square grid. This is the second column of the
input screen.
If the user would like to use the CMAQ grid, as illustrated in Section 2, there are the 12
kilometer CMAQ grid or the 36 kilometer CMAQ grid. For the CMAQ data, Figures 1, 2, and 3
illustrate the maximum coordinate range for each grid size and year and the area of coverage for
T-SpACE User's Guide
Page 21
Version 5_4h.0.1.21
January 2018
-------
the respective grids. Note that the software does not provide default ranges, but after each T-
SpACE model run, the coordinates from the previous run are listed.
Figures 1, 2, and 3 illustrate the maximum ranges available. Recommendations for CMAQ grid
sizes, based on areas of study conducted by EPA, are as follows:
2001, 12 km (41, 49) -> (213, 180) (Eastern half of the US coverage and illustrated in
Figure 19)
2001, 36 km (13, 15) -> (142, 94) (US only coverage)
4.3.4 Choose Air Pollutant
The user must choose the air pollutant for which the T-SpACE model simulation will be run.
Currently, the two choices are PM2.5 or ozone.
If the user does not choose an air pollutant and attempts to run another step in the simulation,
they will see the following error screen.
Missing Information
M
You have not selected a Pollutant: Select "Ye!
5 for PM2.5, S
elect No for Ozone or Select Cancel'to return to step 1.
Yes.
No
Cancel
4.4 Step 2: Prepare Model Input Data
As discussed in Section 2, there are two sets of input data that are required to run the simulation.
The data are the CMAQ/CAMx estimated air quality model concentrations, and the actual
NAMS/SLAMS air pollutant concentrations obtained at monitoring stations. This step takes the
information from Step 1, gathers additional information including the name of the raw data file
and the name the user would like to call the analysis data set created in this step, and performs
the appropriate transformations to create .CSV files that are used as input to the simulation run.
Figure 20 illustrates the initial screen when Step 2 is chosen. There are two tabs on this screen
T-SpACE User's Guide
Page 22
Version 5_4h.0.1.21
January 2018
-------
Figure 19. The area of the United States over which 2001,12 Km CMAQ published
results of the model have been run. (The NAMS/SLAMS network sites are red
circles).
Figure 20. Step 2: Prepare Model Input Data - CMAQ (or CAMx).
G Hierarchical Bayesiari Model
EMM
File View Tools
ฆ ^ Step 1: Choose Time/Grid Step 2: Prepare Model Input Data
Step 2: Prepare Model Input Data
CMAQ i Monitor
-CMAQ
Choose Input CMAQ data file [NetCDFformat)
The input file is transformed to a .csv file. Please choose an output file name:
Output file name:
CMAQ 03
Q Create second output file with Latitude and Longitude
Obtain Latitude and Longitude from CMAQ data file
Obtain Latitude and Longitude from cells text file
Execute
File creation status
Status
T-SpACE User's Guide
Page 23
Version 5_4h.0.1.21
January 2018
-------
4.4.1 CMAQ: Choose Input CMAQ data file
In Step 1, the user has already made a decision on the area of coverage. What the user is doing
here is choosing the appropriate raw data file to be processed. The user can enter the full path,
ending with the name of the file, or the user can hit the folder button and browse their computer
to locate the file. If a user utilizes the browse function, when they have selected a file, the full
pathname and filename will be entered in this box. A post-processing step requires this file, so it
is absolutely necessary that this file be specified, even if the development of the input data files
is going to be skipped. Note that if an attempt to run Step 4 is made without a CMAQ/CAMx
data file, the user will be asked to supply one. The format of this file was discussed earlier in
Section 2. This is a file that is in NetCDF format and has the structure listed in Section 2.
4.4.2 CMAQ: Output file name
The user can specify any file name in this area of the input screen. The created file will
automatically have a .CSV extension. The user should NOT specify a pathname here, since the
pathname has already been entered in Step 1: Please select a folder for saving .CSV files. If
the user does specify a pathname, an error will occur, and the analysis data set will not be
created.
4.4.3 CMAQ: Create a second output file with latitude and longitude
This is a utility that was developed to provide the user with a file that looks like the input
CMAQ/CAMx file, but also includes the latitude and longitude for the center of the
CMAQ/CAMx locational grid cell. The user can choose to use the CMAQ/CAMx NetCDF file,
which contains the latitude and longitude for each grid cell, or may input a file that is called
Cells.txt. Section D illustrates the structure of Cells.txt. This file is available from EPA, and is
year-specific, with a specified grid size.
4.4.4 CMAQ: Execute
Once the input raw data file is generated, and the name of the analysis data set is created, and
when the decision to create a second file with latitude and longitude has been made, and the file
created, and all the information in Step 1 is the way that the user wishes, the user should click on
the 'Execute' button.
The completion of the T-SpACE model execution is indicated by a box
that pops up with "Done". There are three files created during the model
execution. The CMAQ/CAMx input file, a copy of the CMAQ/CAMx
input file with latitude and longitude (not to be used as input), and a log
file containing the steps that were completed. Figure 21 shows the
CMAQ/CAMx input file compared to the CMAQ/CAMx input file with latitude and longitude.
The file labeled "CMAQ_03_2001_12km.csv" is the CMAQ/CAMx input file created during
the example run. The file has a series of information in the header of the file. Row 18 is where
the actual data begins. The header information is as follows:
Finished Processing
Done
OK |
T-SpACE User's Guide
Page 24
Version 5_4h.0.1.21
January 2018
-------
Figure 21. CMAQ simulation input file created in Step 2 compared to the
CMAQ input file with latitude and longitude.
l^l Microsoft Excel
I ฆ File
Edit Vie
w Insert
Format Tools Data
Window Help Adobe PDF
Type a question for help ~ II
I! J d I l&J z -J.100% * =
j Arial
* 10 "
B I U
I m m 4
iaiป
0/ ซ *-.(
/o * .oc
.00 I i
"*"0 I =r-
p: | \ฑ\ ^ <3* t ,A T - I
I .J
ijji
Syfi |
Reply with Changes...
Review..
..
:^L
2r*
I7
f*
CM/
i0_0 3_2001 _ 12. csv
A
B
C
D
E
F
G
H I I
J
K
L
M
1
Data read from file C :\DOCUME~
1,Jiartford,dVIYDOCLM\Proiects' IJ -:'
11 Ready
NUM
T-SpACE User's Guide
Page 25
Version 5_4h.0.1.21
January 2018
-------
Relevant CMAQ/CAMx header information:
Line 1: Provides the full pathname and the filename of the CMAQ/CAMx NetCDF file
used to create the file.
Line 3: Lists the species. In this case, it is ozone (O3).
Line 4: Indicates if daylight saving shift was employed in the appropriate time frame.
Line 5: Indicates the type of value the concentration is. In this case, it is the log of the
daily average concentration.
Line 7: This provides the projection (Lambert) used to transform the grid to the contour
of the earth.
Line 10: Indicates the number of columns of the grid available in the NetCDF file.
Line 11: Indicates the number of rows of the grid available in the NetCDF file.
Line 17: These are the names of the columns.
Day is the number of the day (Julian Day) from January 1
Column is the column number of the grid
Row is the row number of the grid
CMAQ/CAMx log (ozone) is the log ozone concentration.
The file named 'CMAQ 03 2001 12-wLatLon .csv" has the exact same structure as
"CMAQ 03 2001 12km .csv", except there are two more columns of data (starting at line 18).
Longitude is the longitude of the center of the indicated CMAQ/CAMx grid cell
Latitude is the latitude of the center of the indicated CMAQ/CAMx grid cell
Comparing the two files in Figure 21 shows the following:
1. The structure of the two files is exactly the same except there are two additional
columns of data in the "wLatLon" file.
2. The input CMAQ/CAMx data file includes all available data from the NetCDF file,
while the "wLatLon" file only includes the CMAQ/CAMx grid specified in Step 1.
Finally, to ensure that the conversion from the NetCDF to the *.CSV input file has been
performed correctly, the user is provided a log file of the conversion. The log file is a text file
that is written to the directory specified in Step 1. The name of the log file is
RecordedOutput_{MM-DD-YYYY} {H-M}_{ AM/PM} S.txt
where
MM
= Month in which file was created
DD
= Day on which file was created
YYYY
= Year in which file was created
H
= Hour of creation
M
= Minute of creation
AM/PM
= AM or PM creation time
S
= Seconds of the minute in which the file was created
Figure 22 presents an example log file for creating a CMAQ/CAMx analysis input data file.
T-SpACE User's Guide
Page 26
Version 5_4h.0.1.21
January 2018
-------
Figure 22. Log file for the conversion of the CMAQ NetCDF file to the *.CSV format.
Recorded0utput_3-16-200911-41_AM216.txt- Notepad
File Edit Format View Help
value for ioapi_no_stdout not defined;returning default: true
This program uses the EPA-AREAL/MCNC-Environmental Programs
Models-3 I/O Applications Programming interface, which is
built on top of the netCDF I/O library (copyright 1993, 1996
University Corporation for Atmospheric Research/Unidata
Program). Portions copyright (C) 1992-2002 mcnc and
carlie 3. coats, Jr., and 2003 Baron Advanced Meteorological
systems, LLC. See URL
"http://www. erac. mcnc. org/products/ioapi/notices, html " for
conditions of use.
Library release tag: 2003280
$ld: @(#) ioapi library version 2.2 $
netCDF version version of Jan 30 2001 08:38:56 $
Value for EXECUTION_ID: 'cmaqpm2 503.exe'
value for infile:
"C:\DOCUME~l\hartford\MYDOCU~l\Projects\EPA\HEEMR\HBMTOO~l\Data\CMAQ\03\OZONE_~l.CON"
Value for IOAPI_CHECK_HEADERS not defined;returning default: FALSE
"infile" opened as old:read-only
File name
"C:\DOCUME~l\hartford\MYDOCU~l\Projects\EPA\HEEMR\HBMTOO~l\Data\CMAQ\03\OZONE_~l.CON"
File type GRDDED3
Execution id "????????????????"
Grid name "m_12_oaqps"
Dimensions: 188 rows, 213 cols, 1 lays, 1 vbles
NetCDF ID: 3 opened as READONLY
starting date and time 2001001:010000 (1:00:00 Jan. 1, 2001)
Timestep 010000 (1:00:00 hh:mm:ss)
Maximum current record number 8760
Value for CALTYPE: 'AVG'
value for species: '03'
value for islocal: t returning TRUE
value for FACTOR not defined; returning default: 1.000000
value for USELOG: T returning TRUE
value for outfile:
'C:\DOCUME~l\hartford\MYD0CU~l\Projects\EPA\HEEMR\WA54~l\TASK5l\CMAQ_O3_2001_12. CSV'
value for START_DATE: 2001001
Value for END_DATE: 2001031
value for ioapi_isph: '19'
initspheres: input sphere Normal sphere
value for tzfile:
'C:\DOCUME~l\hartford\MYDOCU~l\ProjeCts\EPA\HEEMR\WA54~l\SOFTWA-l\SOFTWA~l\tZ. CSV'
Time step error reading file:" infile
Requested date & time: 2001001:000000
File starting date & time: 2001001:010000
File time step: 010000
ป ป warning in subroutine READ3
Time step error reading file: infile
M3WARN: DTBUF 0:00:00 Jan. 1, 2001
ป ป WARNING in subroutine READ3
Time step not available for file: infile
M3WARN: DTBUF 0:00:00 Jan. 1, 2001
End of run
CMAQ column renamer started at 3/16/2009 11:42:01 AM
column Renaming complete in 00:00:oi.5000288 g
The log file indicates if any problems occurred during the conversion process. Figure 22
illustrates a conversion that executed properly.
T-SpACE User's Guide Page 27 Version 5_4h.0.1.21
January 2018
-------
Figure 23. Step 2: Prepare Model Input Data - CMAQ/CAMx.
G Hierarchical Bayesiari Model
~dฎ
File View Tools
'
. * Step 1: Choose Time/Grid | Step 2: Prepare Model Input Data
Step 2: Prepare Model Input Data
CMAQ
Monitor
Monitor
Choose input Monitor file {NAMS/S LAMS) data text file):
The input file is transformed to a .csv file. Please choose an output file name:
Output file name: fv1onitor_03
I I Create second output file with Latitude and Longitude
Obtain Latitude and Longitude from CMAQ data file sleeted on CMAQ tab
Obtain Latitude and Longitude from cells .text file
Execute
File creation status
Status
4.4.5 Monitor
The steps for creating the Monitor input data file are exactly the same as the steps for creating
the input CMAQ/CAMx data file.
To create an input monitor data file, the CMAQ/CAMx NetCDF file MUST be specified
on the CMAQ tab, even if you are not developing a CMAQ/CAMx input file.
The input monitor file must be a NAMS/SLAMS data file. This file is available through
AQS in the appropriate text format. The exact structure of this file is presented and
discussed in Section 2.
T-SpACE User's Guide
Page 28
Version 5_4h.0.1.21
January 2018
-------
4.5 Step 3: Model Specification
Step 3 is where the user specifies the parameters for the T-SpACE simulation.
Figure 24 illustrates the opening screen for Step 3. The purpose of Step 3 is to specify the
parameters of the model for the T-SpACE simulation, identify the input data, name the output
data, and determine whether the average 4th highest surface will be calculated. Once the user has
made all these decisions, a *.PAR (parameter) file that stores all the choices made, is saved and
is then available for later use. Below is a description of the choices available to the user.
Figure 24. Step 3: Opening Screen.
-ฆ
G Hierarchical Rayesian Model - [Step 3: Model Specification]
Eile View Tools _ & X
Step 1: Choose Time/Grid J3? Step 2: Prepare Model Input Data IB Step 3: Model Specification | ^ Step 4: Launch Model Step 5: Launch Validation
~ & H
Grid I Prions Priors Data Boundaries Track Simulation 4th Highest
T:
X: 1 25D
Y: 1| | 250|
Date of Time Step 1
Month Day Year
I 11 I 11 I 2DD1
Status
4.5.1 New, Open, Save
T , , , , , D H
Just above the tabs, the user will see three icons.
"Open", and "Save" for the *.PAR (parameter) file.
D
These icons represent "New",
New Clicking on this icon clears all information that has been entered in the various
choices available to the user. No files will be deleted when this button is selected.
Only the choices made on-screen during this session in Step 3 will be changed.
Note that the T, X, and Y values chosen in Step 1 will be changed to reflect any
changes made here.
Open If a user would like to utilize a previously developed *.PAR file, the user can
click on this icon. Note that prior to opening a *.PAR file, the user must have
specified the CMAQ/CAMx NetCDF file in Step 2.
T-SpACE User's Guide
Page 29
Version 5_4h.0.1.21
January 2018
-------
o
Save When a user completes their choices, clicking this icon allows the user to save
their choices to a *.PAR file.
4.5.2 Grid
On this tab, the user is specifying the grid coordinates over which the T-SpACE simulation will
be run. Below is a description of each of the inputs available to the user.
T Represents the range of time in days (see "Date of Time Step 1" described below). The
user cannot change the time frame on this tab and can only make changes in Step 1. The
changes made in Step 1 will be reflected here.
X Represents the x-coordinates of the locational grid over which the T-SpACE simulation
will be run. By default, these are the x-coordinates chosen in Step 1. The user may make
changes here to the x-coordinates if another grid is desired for the analysis. The user
should ensure that the input data files are specified.
Y Represents the Y-coordinates of the grid over which the T-SpACE simulation will be run.
By default, these are the y-coordinates chosen in Step 1.
Date of Time Step 1
The first day from which to start counting the day range. The default is January 1,
YYYY and this is the recommended date. If a user chooses, they can enter any date here.
For instance, if a user chooses to put in May 1, 2001, and for example, the T range is 2 to
4, then T=2 is May 2, 2001 (and not January 2, 2001).
Figure 25. Priors-First tab.
G Hierarchical Bayesiari Model - [Step 3: Model Specification] [JnM
a9 File View Xools _ fi1 X
i Step 1: Choose Time/Grid | 'JjStep 2: Prepare Model Input Data IB Step 3: Model Specification Step 4: Launch Model ji Step 5: Launch Validation
~ & y
Grid 1 [Priors...]| Priors | Data Boundaries Track | Simulation || 4th Highest
Bias Spline
Bias Spline Min
T I
Bias Spline Max of Control Points
365 1 4
X I
213|| >|
YI
1 ml 7|
Mu: Normal jjj I 1 .DE-3
BetaD: Normal q1 I 1 .OE-3
Status
T-SpACE User's Guide
Page 30
Version 5_4h.0.1.21
January 2018
-------
4.5.3. Priors-First Tab
When clicking on the first 'Priors' tab, the user sees the window illustrated in Figure 25. The
purpose of this tab is to allow the user to specify the information required to calculate the Bias
Splines. Note that the selection of prior information is up to the user. For the T-SpACE model,
the priors are intentionally initialized to 'non-informative' values, to ensure that they have no
effect on the posterior distribution. If the user has some understanding of the statistical
distribution of the data, then the user can change the priors to reflect their knowledge.
A detailed explanation of the simulation model calculations is provided in Section A. Below, in
Table 2, is a description of the input a user should provide.
Table 2. Definition of Prior Parameters - Prior Screen 1.
Variable
Bias Spline Min
Bias Spline Max
number of Control
Points
T
First day of the time range over which the bias
will be calculated. By default, this is the first
day of the year.
Last day of the time range
over which the bias will be
calculated. By default, this is
the last day of the year.
This defines the degrees of
freedom in the model for time.
By default, 4 is chosen
representing a control point
for each season of the year.
X
Represents the min x-coordinate of the grid
over which the bias spline will be calculated.
By default, and recommended, the min x-
coordinate is 1, for the 12 km grid. If the user
is using the 36 km grid, then the default should
be 13.
Represents the maximum x-
coordinate of the grid over
which the bias spline will be
calculated. By default, and
recommended, the maximum
possible x-coordinate for the
12 km is given, 213. If the
user is working with the 36 km
grid, then this should be 142.
This is the degrees of freedom
for the x-coordinate in the bias
function. 8 is the default
entry. Please note that if this
number gets too high the
performance of the simulation
may be slow.
Y
Represents the min y-coordinate of the grid
over which the bias spline will be calculated.
By default, and recommended, the min x-
coordinate is 1, for the 12 km grid. If the user
is using the 36 km grid, then the default should
be 15.
Represents the maximum y-
coordinate of the grid over
which the bias spline will be
calculated. By default, and
recommended, the maximum
possible x-coordinate for the
12 km is given, 188. If the
user is working with the 36 km
grid, then this should be 94.
This is the degrees of freedom
for the x-coordinate in the bias
function. 7 is the default
entry. Please note that if this
number gets too high the
performance of the simulation
may be slow.
Consideration of the following should occur when choosing the number of control points:
o Since the number of control points is roughly equal to the number of degrees of freedom
of the T-SpACE simulation, the user wants to be conscious of not over- or under-fitting
the model by choosing too many or too few control points
o For T, time, the number of control points has been chosen to represent season. If the user
believes there to be a monthly bias in the concentrations, then they may choose this value
to be 12.
o For X and Y, the number of control points has been chosen to represent an estimated rate
on which bias changes in each direction. A metric that may be used to choose the
number of control points for X and Y could be the length of the grid over which the
simulation is to run divided by the average scale of the spatial bias (every 25 grid cells).
T-SpACE User's Guide
Page 31
Version 5_4h.0.1.21
January 2018
-------
It is recommended that there should be less than 100 control points in the x-y coordinate
system.
o It is also recommended that a user of the system run sensitivity analyses to understand
how sensitive the surface predictions are to the number of control points.
In addition, there are two normal regression priors which are specified on this screen.
//, t = 1,..., Nt Mean level of the Conditional Auto Regressive (CAR) process
(Mu)
P" e R Vector of coefficient for the bias (BetaD)
By default, each of these is defined by a non-informative normal distribution with |i equal to 0
and the variance equal to 1.0E-3 (0.001). If the user has information about the distribution of Mu
or BetaD, the user can enter the information here.
4.5.4. Priors-Second Tab
The priors being entered on this second 'Priors' tab are used to specify prior (statistical)
information about the unknown, but true, variances of the air pollution monitor observations, the
air quality model simulated air pollution concentration surface, and random process included in
the simulation to account for presence of correlation among air pollution measurements collected
successively in both space and time. In addition, the user can specify prior information about the
correlation of the random process on this screen. Figure 26 illustrates the input screen for these
priors. Note that if the user is interested in the details of these parameters, they should see [1]
and [2],
Specifically, the t's represent the reciprocal of the unknown variance of the air pollution monitor
observations (measured concentrations), the air quality model generated air pollution
concentration surface, and the random process that is a part of T-SpACE. The parameters that
the user is specifying on this screen are
rx Precision of the measurement error in the monitor observations (TauX)
Tr Precision of the measurement error in the computer observations (TauY)
tz Precision of the mean process (TauZ)
pz Temporal autocorrelation parameter of the mean process (RhoZ)
The t's of the model are defined by a gamma distribution. For all three t's, the first column
represents the shape of the gamma distribution and the second column represents the rate for the
distribution. By default, the t's are uninformative (shape = 0.001, and rate = 0.001) because each
prior needs to be assigned by the user with prior knowledge of the data and the model. These
default choices squeeze the gamma distribution to zero. This is a non-informative choice since
the sampling equation for the t's in the posterior distributions is equal to sample size of data set
plus the shape of the prior. If shape is small (i.e., the non-informative prior), adding that value
does not affect the posterior. Thus, without prior knowledge of the data and the model, the non-
informative priors have no effect on the posterior distribution.
T-SpACE User's Guide
Page 32
Version 5_4h.0.1.21
January 2018
-------
Figure 26. Priors-Second Tab.
G Hierarchical Bayesian Model - [Step 3: Model Specification]
~ฉ0
a9 File View Tools _ B X
ฆ Step 1: Choose Time/Grid | Step 2: Prepare Model Input Data J.B Step 3: Model Specification Step 4: Launch Model Step 5: Launch Validation
Grid || Priors 11 Pira j | Data || Boundaries || Track || Simulation J 4th Highest |
TauX:
Gamma
1.DE-3
1.DE-3
TauY:
Gamma
1.0E-31
1.0E-3
TauZ:
Gamma
1.DE-31
1.DE-3
RhoZ:
Gamma
0
1
p (Rho) is the temporal autocorrelation parameter that is modeled by a uniform distribution to
constrain the correlation between the minimum, 0, and the maximum, 1. By default, the
constraints are initialized to non-informative with the first column representing the minimum
range for the uniform distribution and the second column representing the maximum range. If a
user has an understanding of the possible range for the correlation, they can change the default
values to numbers between 0 and 1.
4.5.5 Data
As illustrated in Figure 27, the data tab is where the user specifies the input air pollution monitor
and simulated air pollution concentration surface files for the T-SpACE simulation. By default,
the input directory is specified in Step 1, and the air pollution monitor data, and the
CMAQ/CAMx data are the .CSV files created in Step 2. If the user does not want to use the
default files, they can make the changes here.
Please note that if the user chooses to change the inputs, they should be as follows:
Input Directory The path name for the location of the monitor and CMAQ/CAMx
data. Please note that both sets of data need to reside in the same
directory.
Monitor Data This needs to be a .CSV file in the format that is created in Step 2.
If the file is not of this format, then the user will receive an error.
T-SpACE User's Guide
Page 33
Version 5_4h.0.1.21
January 2018
-------
CMAQ Data This needs to be either a 12 km or 36 km CMAQ/CAMx .CSV file
in the format created in Step 2. If the file is not of this format, then
the user will receive an error.
Figure 27. Input Data.
G Hierarchical Bayesiari Model - [Step 3: Model Specification] DM
File View I00'5 _ & X
i (J Step 1: Choose Time/Grid | Step 2: Prepare Model Input Data JE Step 3: Model Specification Ji Step 4: Launch Model Step 5: Launch Validation
Q 0
Grid 1 Priors || Priors | Data | Boundaries | Track || Simulation II 4th Highest I
Input directory C:\DOCUME~1\MYDOCU~l\Prajects\EPA\HEEMR\
Monitor Data Monitor 03
CMAQ CMAQ 03
Status
4.5.6. Boundaries
As illustrated in Figure 28, the boundaries tab is where the user defines the neighborhood
structure of the model. Figure 29 illustrates the four possible choices for neighborhoods in the
model. By default, the simulation uses a first order neighborhood. Detailed information about
the boundaries is provided in Section A.
T-SpACE User's Guide Page 34 Version 5_4h.0.1.21
January 2018
-------
Figure 28. Boundaries.
O Hierarchical Bayesian Model - [Step J; Model Specification] - O'EJ
fc-
Fie Tools. _ S
X
Stto lฃ Cnaese jwtjand j Ijfssea 2: Rrtove Mas* Irout Cซป f Sitsi: Hope $sea ฃUKXfi Hoot J SSe s uu*h > ปซ;<*>
d a w
&-(3 Pikki Pr,ซ
-------
4.5.7 Track
Figure 30 illustrates where the user can track the chain flag for z during the simulation. This
chain is written to a file located in the output directory specified on the Simulation screen. By
default, this remains unchecked as it slows down the processing of the simulation. For
diagnostic purposes, the user may want to turn this on, by marking a 'check' the 'Track Z' box.
Figure 30. Track.
ฉ Hierarchical Bayesian Model - [Step 3: Model Specification]
~ฉ0
File View Tools
_ &
X
$ Step 1: Choose Time/Grid | Step 2: Prepare Model Input Data | ffr] Step 3: Model Specification | ^ Step 4: Launch Model Step 5: Launch Validation
~
IjcI
Grid Priors Priors Data Boundaries |i Track jl Simulation 4h Highest
~ Track Z
Status
4.5.8 Simulation
This tab is where the user specifies how the simulation will be run and where the simulation
results will be saved. As shown in Figure 31, there are several inputs the user can specify.
Output Directory
By default, this is the location provided in Step 1. This directory
where the user would like the generated simulation files placed.
Est. Surface File
This is the name of the file that will contain the estimated air
pollution concentration surface. By default, this is named
'surface.csv'. The user can name the file any name they want
provided the extension is .CSV.
Seed
Two random seeds are need to seed the simulation. By default, the
two seeds utilized are 1234 and 5678. The user can specify the
seeds they would like if they prefer something other than the
default values.
Sample every During the simulation, this is how often sampling occurs. By
default, this is initialized to 1 (i.e., sample each model iteration).
T-SpACE User's Guide
Page 36
Version 5_4h.0.1.21
January 2018
-------
Print every
During the simulation, the user can see the progress on screen.
This determines at what step the results are printed to the screen.
By default, this is 1 (i.e., print each step on the screen). Please
note that choosing a lower number may slow down the processing.
New Simulation
Burn-in
Choosing this indicates whether to overwrite current files or not.
Number of burn-ins loops for the simulation. The default is
initialized to 1000.
Loop
Number of simulation loops to complete. The default is initialized
to 5000.
For "Burn-in" and "Loop", the default values were chosen heuristically based on the size of the
default grid. The user can utilize the chain file to help understand if their choice for the number
of burn-ins and simulations loops is appropriate. In particular, TauX, TauY, and TauZ are in the
chain file. Plot these parameters by the iteration. If the parameter value keeps increasing or
decreasing, then those runs should be in burn-in iterations. If the parameter jumps between a
range (i.e., bouncing back and forth), we consider this to be the point at which the simulation
loop should begin. In terms of the total number of iterative loops, there is not a magic number. It
is best if the user plots the parameter values and determines where the parameters appear to
converge.
Figure 31. Simulation.
G Hierarchical Bayesiari Model - [Step 3: Model Specification]
a9 File View Tools _ & X
^ Step 1: Choose Time/Grid | ^|Step 2: Prepare Model Input Data JJg Step 3: Model Specification | ^ Step 4: Launch Model Step 5: Launch Validation
d ^ y
| Grid || Priors || Priors || Data || Boundaries || Track |. | 4th Highest|
Output directory
C:\
[W]
Est. Surface File
Surface .csv
Seed
1234 |
56781
Sample every
11
Print every
11
0 New Simulation
Bum-in
1000
Loop
5000
Status
um
T-SpACE User's Guide
Page 37
Version 5_4h.0.1.21
January 2018
-------
4.5.9 4th Highest
As discussed earlier in this section, under "Tools" on the top menu bar, the user can choose to
calculate the average 4lh-highest
concentration surface using three
consecutive annual (individual) 4th
highest air pollution concentration
surfaces. This tab is where the user can
choose to create an individual 4th-highest
concentration surface. As shown in
Figure 32, by default, the 4th highest
concentration surface is not calculated. In order to calculate the surface during the simulation,
the user must check "Enable 4th Highest Concentration" and provide the name of the file for the
surface. It is recommended that the user use the same name as the surface file, but add "_4h" to
the filename so that the user knows which surface run the 4th highest concentration surface is
related to. The note on this input screen reminds users that if they are looking to create a 3-year
average 4th highest concentration surface, each of the individual runs should utilize the same time
frame, the same grid size, and roughly, the same order of magnitude of burn-in and loop sizes.
-
O Hierarchical Bayesia... - [:~
File View
Tools
I
Status
Compare AQS Data Files
Compare Airsites Files
Average 4th-Highest Surface Files
Figure 32. 4th highest concentration surface.
G Hierarchical Bayesian Model - [Step 3: Model Specification]
Hu B'e View Tools _ & X
3 Step 1: Choose Time/Grid 2: Prepare Model Input Data IE Step 3: Model Specification 4: Launch Model Step 5: Launch Validation
D 12^ 121
I I Enable 4th Highest Concentration
4th Highest Surface File
Note: Number of simulation runs to develop the 4th highest surface
should be similar in magnitude if those files are to be averaged
T-SpACE User's Guide
Page 38
Version 5_4h.0.1.21
January 2018
-------
4.5.10 *.PAR file
Prior to moving to Step 4, the user will want to save their choices to a *.PAR (parameter) file.
This file is excellent for archiving the model runs. The *.PAR file is a text file with a specific
format that is needed to perform the simulation. As shown in Figure 33, the third line of the file
indicates the version of the file. Currently, we are using version "5-4H". If you have an earlier
* PAR file, you may enter your parameter choices manually in Step 3 and create a new * PAR
file, or you may compare the previous *.PAR file to the current version and add the necessary
lines.
Figure 33. Simulation.PAR file.
^ Example-Run 1-12km-ozone-2001.par - Notepad
Qmm]
File Edit Format View Help
# simulation parameter file
5-4H # Model version
1, 333
40, 213
30, 180
# Grid size t
# Grid size x
# Grid size y
2001, 01, 01 # Year, month, day of time step 1
1,
40,
30.
333,
213,
188,
# Bias spline in t
# Bias spline in x
# Bias spline in y
0, 1.0E-3 #
0, 1.0E-3 #
7.5E+7, 1.0E+6
1.0E+7, 1.0E+8
1.0E-3, 1.0E-3
Prior Parameters for Mu ~
Prior Parameters for BetaD
Normal(mean, prec)
Normal(mean, prec)
# Prior Parameters for Taux
# Prior Parameters for TauY
# Prior Parameters for Tauz
Gamma(mean, prec)
Gamma(mean, prec)
Gamma(mean, prec)
0, 1
# Prior Parameters for Rhoz ~ Gamma (shape, rate)
c:\
aqs_12km_o3_2001.csv
cctm_12km_2001.csv
# Directory where data are stored
# File name where monitor data X are stored
# File name where CMAQ csv is stored
# order of neighborhood for z (between 1 and 4)
# Track chain flag for z
# Enable calculation of 4th highest statistics
# Track chain flag for 4th highest statistics
c:\ # Directory where the simulation outputs are written
surface_03_12_2001.csv # File name where mean, covariate, bias, monitoring, and computer
average surfaces are written
surface_03_12_2001_4h.csv # File name where 4th highest surface is written
1234
5678
1
1
1
50
100
I
# Random seed 1
# Random seed 2
# sampling period for thinning
# Prompting period for diagnostic
# New/Load simulation flag. 1 = New,
# Number of burn-in steps
# Number of simulation steps
Load
T-SpACE User's Guide
Page 39
Version 5_4h.0,1.21
January 2018
-------
4.6 Step 4: Launch Model
Prior to launching Step 4, the user must specify the simulated air pollution surface file in Step 2
(NetCDF format) and must open a *.PAR file or create a *.PAR file in Step 3. Once the user has
specified the model information and has created a simulation.PAR file, the user should choose
the "Step 4: Launch Model" tab. Hitting this tab once will initiate the T-SpACE model run.
It may take a few minutes for a DOS screen to appear. This screen will show the progression of
the T-SpACE model run, including reading the data into the model, executing burn-in loop steps,
and executing the simulation loop. Figure 34 illustrates an example of the output that is written
to the DOS screen.
Figure 34. Step 4: DOS Screen output - progression of the simulation.
m C:\D0CUME~1\hartford\L0CALS~1\Apps\2.0\YWMY8G15.XNM\DQX5J4ZD.6D8\EPAWTI~2.0... - [~[ x|
Step
60 TauX
0.171 TauV
0.999
TauZ
9.75
RhoZ
0.5
Step
70 TauX
0 TauV
1
TauZ
10.4
RhoZ
0.5
Step
80 TauX
0 TauV
0.999
TauZ
12.3
RhoZ
0.5
Step
90 TauX
0 TauV
1
TauZ
11.2
RhoZ
0.5
Step
100
TauX
0
TauV
TauZ
9.33
RhoZ
0.5
Step
110
TauX
0
TauV
TauZ
16.2
RhoZ
0.5
Step
120
TauX
0
TauV
TauZ
18
RhoZ
0.5
Step
130
TauX
0
TauV
TauZ
21.7
RhoZ
0.5
Step
140
TauX
0
TauV
0.999
TauZ
19.6
RhoZ
0.5
Step
150
TauX
0
TauV
0.999
TauZ
18.7
RhoZ
0.5
Step
160
TauX
0
TauV
TauZ
18.2
RhoZ
0.5
Step
170
TauX
0
TauV
TauZ
19.2
RhoZ
0.5
Step
180
TauX
0
TauV
0.997
TauZ
23.1
RhoZ
0.5
Step
190
TauX
0
TauV
TauZ
19.7
RhoZ
0.5
Step
200
TauX
0
TauV
TauZ
34.5
RhoZ
0.5
Step
210
TauX
0
TauV
TauZ
18.5
RhoZ
0.5
Step
220
TauX
0
TauV
TauZ
18
RhoZ
0.5
Step
230
TauX
0
TauV
TauZ
16.3
RhoZ
0.5
Step
240
TauX
0
TauV
0.999
TauZ
23.1
RhoZ
0.5
Step
250
TauX
0
TauV
TauZ
30.3
RhoZ
0.5
Step
260
TauX
0
TauV
TauZ
33.3
RhoZ
0.5
Step
270
TauX
0
TauV
TauZ
39.2
RhoZ
0.5
Step
280
TauX
0
TauV
0.999
TauZ
34.9
RhoZ
0.5
Step
290
TauX
0
TauV
0.999
TauZ
35.8
RhoZ
0.5
At the completion of the T-SpACE model run, the DOS window will close. Figure 35 illustrates
the files that are created (and stored) in the directory selected by the user for the output files.
This example shows the files when the 4th highest air pollution concentration surface is also
generated.
The following are the simulation results files available to the user:
Chain.CSV Contains output that was printed to the DOS screen for each parameter in
the model.
Summary.CSV Contains summary statistics for all calculated parameters in the model
including the bias surface. Section E provides a detailed summary of the
information in this file.
T-SpACE User's Guide
Page 40
Version 5_4h.0.1.21
January 2018
-------
Figure 35. Files saved from the simulation run.
l-) C:\Documents and Settings\hartford'vMy Document... ~taa
File Edit View Favorites Tools Help
Back "
: Search
Folders
Q
Address |Q EMR^WA 54\Task 5-Users Guide'ฃxSimulation_2009031S [vj H Go
Narme
งJjCMAQ_03_200 l_12.csv
CMAQ_03_200 l_12-wLatLon. csv
^Monitor_03_200 l_12km,csv
^ Monitor_0 3_2001_ 12km-wLatLon. csv
0 RecordedOutput_3-18-200912-39_PM436.txt
Ifl RecordedOutput_3-18-2009 12-40_PM573.txt
ฎ RecordedOutput_3-18-200912-41_PM94.txt
||] RecordedOutput_3-13-2O0912-44_PM973.txt
0 RecordedOutput_3-18-200912-47_PM633.txt
1 ExamoieSimulation ozone 2001 12km.par
<^5urface_ozone_200 l_12km.csv
ง]Globais,csv
^ Summary, cs v
Chain, csv
& Simulatjon.dat
Surface ozone 2001 12km 4h.csv
Output Files
Size
7,590 KB
341KB
50 KB
5 KB
3 KB
3 KB
3 KB
3 KB
3 KB
2KB
4KB
1KB
7 KB
2 KB
I KB
1KB
<
>
Globals.csv
Simulation.dat
Surf ace*, csv
Day
Time
XCoord
YCoord
Longitude
Latitude
Contains a brief summary of the number of burn-in loop and simulation
loop steps and thinning choices made for the simulation.
This file is used for another application. It is in a machine readable
format.
Contains the estimated air pollution concentration surface from the T-
SpACE simulation. Figure 36 provides a snapshot of the variables seen in
the file. A brief description of each variable in the Surface.csv file is
listed below.
Date for the estimated surface in a MM/DD/YYYY format.
This is the date in Julian format (i.e., number of Julian Day)
The grid x-coordinate
The grid y-coordinate
The transformed grid x-coordinate to a Longitude (degrees)
The transformed grid y-coordinate to a Latitude (degrees)
T-SpACE User's Guide
Page 41
Version 5_4h.0.1.21
January 2018
-------
The natural log of the mean predicted air pollution concentration for each
day and grid cell. (Values of -999 denote a missing value.) The predicted
daily concentration represents a daily average for PM2.5, and an eight-hour
maximum concentration for ozone. Section A contains a detailed
description of the calculation of this value.
The natural log of the standard error for the mean predicted concentration
for each day and grid cell. (Values of -999 denote a missing value.)
Section A contains a detailed description of the calculation of this value.
Mean of the bias surface. (Values of -999 denote a missing value)
Standard error of the mean of the bias surface. (Values of -999 denote a
missing value.)
This parameter is obtained from NASA's Moderate Resolution Imaging
Spectroradiometer (MODIS) Aqua and Terra satellites. The number -999
denotes a missing value. [Implemented in Model 6 version of T-SpACE. ]
This parameter is obtained from NASA's Moderate Resolution Imaging
Spectroradiometer (MODIS) Aqua and Terra satellites. The number -999
denotes a missing value. [Implemented in Model 6 version of T-SpACE.]
MonitorData The natural log-transformed air pollution measurement from any air
pollution monitor present at the given location.
ComputerDataThe natural log-transformed simulated concentration (CMAQ) for the day
and grid cell.
CovarData This parameter is obtained from NASA's Moderate Resolution Imaging
Spectroradiometer (MODIS) Aqua and Terra satellites. The number -999
denotes a missing value. [Implemented in Model 6 version of T-SpACE.]
Figure 36. Screen shot of the Surface.csv file.
i*l Microsoft Excel - Surface
ozone
2001_
12km. csv
:aj
File Edit
View Insert
Format
lools
Data Window Help Adobe PDF
Type a question for help . S x
I J
jdi.jaia
ป l
ฆฃ
I
100% - try Z I Aria
- 10
B I
U I IE
mmm
$ % '
ฆ*.0 .00 1
ฑ= ;= | r
j-A-A* -
ihJ
ii u iiJ
UaLftifcal.
d
j feisr k
A1
fit Day
A
B |
C
D
E
F
G
H
I
j
K
L
M
N
0 I IT1
1
Day
Time XCoord
YCoord
Longitude
Latitude
PredAvg
PredStd
BiasAvg
BiasStd
CovarAvg
CovarStd
MonitorDal Computed CovarData ~
2
1/1/2001
149761
40
50
93 2974
33 70528
3 03319
0 031737
0 350338
0 005622
-999
-999
-999
3.412
-999
3
1/1/2001
149761
40
51
-93 1622
33 80876
3 03614
0 059832
0.350287
0.004851
-999
-999
-999
3 392
-999
4
1/1/2001
14976
40
52
-93 0266
33 91209
3 01409
0 048485
0.349624
0 004183
-999
-999
-999
3.377
-999:
5
1/1/2001
14976
40
53
92.8906
34 01527
3 02995
0 066245
0 348348
0 003637
-999
-999
-999
3 385
-999"
6
1/1/2001
14976[
40
54
-92.7542
34 11829
3 05291
0 029719
0 346458
0 003233
-999
-999
-999
3405
-999
7
1/1/2001
149761
40
55
-92.6175
34.22116
3.0597
0 059276
0.343956
0 00299
-999
-999
-999
3 423
-999
8
1/1/2001
14976
40
56
-924804
34 32388
3 06442
0 048682
0 34084
0 002908
-999
-999
-999
3.422
-999
9
1/1/2001
14976
40
57
-92 3429
34 42644
3 08116
0 058002
0 337112
0 002967
-999
-999
-999
3.414
-999
10
1/1/2001
14976
40
58
-92 205
34 52884
3 03842
0 062
0 332771
0 003128
-999
-999
-999
3.399
-999
11
1/1/2001
14976
40
59
-92.0668
34 63108
3.01995
0 043449
0.327816
0 003351
-999
-999
-999
3.372
-999
12
1/1/2001
14976
40
60
-91 9281
34 73316
2 98439
0 054586
0.322249
0 003605
-999
-999
-999
3 328
-999
13
1/1/2001
14976
40
61
-91 7392
34 83508
2 92579
0 063693
0 316069
0 003866
-999
-999
-999
3 253
-999
14
1/1/2001
14976
40
62
-91 6493
34 93684
2 91706
0 063015
0 309275
0 004119
-999
-999
-999
3.243
-999
15
1/1/2001
14976
40
63
-91.51
35.03844
2 91819
0 035311
0 301869
0 004354
-999
-999
-999
3.234
-999
16
1/1/2001
14976
40
64
-91.3699
35 13987
2 9097
0 06018
0.29385
0 004565
-999
-999
-999
3.225
-999
17
1/1/2001
14976
40
65
-91 2294
35 24114
2 91497
0 054396
0 285217
0 004749
-999
-999
-999
3 239
-999
18
1/1/2001
14976
40
66
-91 0884
35 34224
2 97243
0 03857
0.275972
0 004906
-999
-999
-999
3 269
-999
19
1/1/2001
14976
40
67
-90.9471
35 44318
2.97051
0 038224
0.266114
0 005035
-999
-999
-999
3 254
-999
20
1/1/2001
14976
40
68
-90.8055
35 54394
2 91857
0 059165
0.255642
0 005138
-999
-999
-999
3 178
-999
21
1/1/2001
14976
40
69
-90 6634
35 64454
2 89655
0 043664
0 244558
0 005218
-999
-999
-999
3 158
-999
22
1/1/2001 14976 40
~ w \Surface_ozone_2001_
2km /
-90.521 35 74496 2 8Z5fil.
JLQ4S43fi_JL2a2Ml
0 005279
l 1
Ready
NUM
PredAvg
PredStd
BiasAvg
BiasStd
CovarAvg
CovarStd
T-SpACE User's Guide
Page 42
Version 5_4h.0.1.21
January 2018
-------
Figure 37. Screen shot of the surface_4h.csv file.
[3] Microsoft Excel - Surface_03_12_2001_4h.csv \ [[njfesj
: tSjj File Edit View Insert Format Tools Data Window Help Adobe PDF
_ B X
i A ? iArial - 10 j
i _j -j
1 ^ J3L ft .X' Reply with Changes... "
IS sru
A1 fx XCoord
CD
<
o
o
rn
F
G |
Ready NUM
Surface 4h.csv
XCoord
YCoord
PredAvg
PredStd
Chain.csv
Contains the results of the calculation of the 4th highest air pollution
concentration surface for the specified grid and time frame. The result is a
single air pollution concentration surface. Below is a description of the
variables included in this file. Section F provides a summary of the
calculations for results shown.
The cell's x-coordinate (east-west) within the grid (where the coordinate
value increases by I with each grid cell as you move from west to east).
The cell's y-coordinate (north-south) within the grid (where the coordinate
value increases by 1 with each grid cell as you move from south to north).
The (natural log-transformed) T-SpACE-predicted value of the fourth
highest concentration for the given grid cell.
The (natural log-transformed) T-SpACE-predicted value of the standard
error for the fourth highest concentration for the given grid cell.
As seen in Figure 38, this file contains the values of Z for each T-SpACE
simulation loop and each cell in the grid. This file is needed for Step 5.
T-SpACE User's Guide
Page 43
Version 5_4h.0.1.21
January 2018
-------
Figure 38. Chain.csv.
LMl Microsoft Excel - Chain.csv
LJloJU
File Edit
View Insert Format
Jools Data Window Help Adobe PDF
Type a question for help
X
= d
" : Arial
- 10 -
| B I D|t#
*
S % ป
*-.0 .00 I +=
.00 +.Q ] ="
I 't * * _A t
i _J
LA 'La -J ^ J_
1 3 Mil (P 1
r^egl^ti
Changes...
End Review^^M
m
A1
~
fx Step
A
B
c
D
E
F
G
H
I
J
1
Step
Z[0]
Z[1]
Z[2]
Z[3]
Z[4]
Z[5J
Z[6]
Z[7]
Z[8]
2
1
0 700317
0 770636
0 710751
0 661592
0.735422
0 674667
0 739773
0 820402
0.62631
3
2
0 68949
0 658159
0 714905
0.759022
0 708368
0774725
0.742268
0 743323
0.75007
4
3
0 694499
0.776128
0.631446
0 704516
0 769079
0 806871
0 720507
0.739452
0 66921
5
4
0 69389
0 707193
0 701781
0650142
0 731961
0.774079
0 732833
0.804482
0.7038
6
5
0 744461
0 73102
0.68394
0.762956
0 740933
0.737042
0.772245
0.778464
0 74270
7
6
0 74093
0.7358191 0.729278
0 688976
0.758334
0.754458
0 713486
0.76148
0 73357
8
7
0.706675
0 7257
0667901
0.735419
0702833
0.748386
0.769604
0 301226
0 74902 v
H 4
~ n|\ Chain/
l<
D
l>l
Ready
NUM
Figure 39. Step 5: Launch Validation.
G Hierarchical Bayesian Model
QbJQ
File View Xools
ฆ tj Step 1: Choose Time/Grid J ^JjStep 2: Prepare Model Input Data
m
Step 5: Launch Validation
Select Surface File
ulation_20C!SC'318 .Surface_ozone_2C'C 1 _12km .csv
Select
I
Select Chain File
C:\Documents and Settings''hartford\My Documents
Select
Select Krige Prediction File
Select
Select Validation Monitors File
Select
Select Report File
Select
Report Title
Validation
( Run Validation |
Status
T-SpACE User's Guide
Page 44
Version 5_4h.0.1.21
January 2018
-------
4.7 Step 5: Launch Validation
This step allows the user to run several validation procedures for the results from the T-SpACE
simulation. The validation procedures used to analyze the output from T-SpACE are written in
SAS and requires SAS to be installed on the computer used to validate the model. The file that is
used to run the validation is called ValidationTemplate.sas. The instance of the file is loaded
along with the T-SpACE software at installation time. It can be found in the main directory of
the T-SpACE software. The user should not make any changes to this file, because Step 5 needs
to have it in this format to appropriately run the validation process. A listing of the file is
provided in Section B.
If a user has gone from Step 1 to Step 5 sequentially, then all information needed for the
validation will have been generated. If a user is choosing to run the validation separately from a
model run, then the following MUST be specified in Step 1 in order for the SAS program to be
configured correctly:
The appropriate dates associated with the T-SpACE model output file need to be
specified here.
The grid that was used in the T-SpACE model run needs to be specified here.
Figure 39 illustrates the initial screen when Step 5 is chosen.
4.7.1 Select Surface File
This is the file where the predicted air pollution surface concentration(s) is written. By default, if
Step 3 was completed during this session, then the name of output .CSV file specified on the
Simulation tab will be listed. Otherwise, this entry will list the last file used for a validation and
the user will have to supply the name of an existing surface file that contains the predicted air
pollution surface concentrations. Note that this file must have the same format as the file
illustrated in Figure 36.
Please note that if real-time scanning software is installed on the computer and is running, then
choosing a surface file can take several minutes if the file is large, such as a surface file for a full
year of data (e.g., a full 12 km x 12 km grid).
4.7.2 Select Chain File
The validation process requires that a Chain file be available for the analysis. If Step 3
(Simulation) was completed and "Track Z" was checked on the Track tab, then a file named
"Chain.CSV" will be in the same folder as the surface file specified in Step 3 on the Simulation
tab. Otherwise, this entry will list the last file used for a validation, and the user will need to
supply an existing chain file. This is a required file to run the T-SpACE model validation.
T-SpACE User's Guide
Page 45
Version 5_4h.0.1.21
January 2018
-------
Figure 40. Krige Prediction File.
SAS System Viewer - krigepredozoneerrorl22001.5as7b.(T1 ~ 1^
File Edit View Window Help
a & -~ I & a #4 v il I m fir in e
Q krig
epredozone
error1220C
1 ,sas7bdat
y
~ g
col
row
krige_pr&
diet
STDERR
time
a
-
1
47
186
3.3322
0
1
2
51
54
3.4569180
0.2501500
1
3
63
1 57
3.4582450
0.3049588
1
4
69
130
3.3846446
0.2446051
1
5
74
99
3.2805756
0.2244247
1
6
77
54
3.3343528
0,2685807
1
7
82
111
3.1939931
0.2276211
1
8
89
81
3.1686774
0.2479028
1
9
90
99
3.1444332
0.2638963
1
10
96
151
3.4069039
0.3063113
1
11
100
120
3.1126613
0.2806400
1
12
106
76
3,1186594
0.2416826
1
13
106
60
3.2320541
0.2511275
1
14
108
92
3.1075800
0,2682689
1
1 5
108
109
3.1191540
0.2932817
1
16
110
136
3.2743663
0,3005892
1
17
112
148
3,3828412
0.3121669
1
18
118
22
3.5428243
0.2395353
1
| Ready |Hdn cols:0 Obs
1 /a
4.7.3 Select Krige Prediction File
The krige prediction file is a SAS data set that has been developed using a series of programs
detailed in Section G. Figure 40 illustrates the format of one of these files for ozone, 2001, 12
km x 12 km.
Col
Row
Krige_predict
StdErr
Time
Is the x-coordinate on the grid where a monitor was available for kriging.
Is the y-coordinate on the grid where a monitor was available for kriging.
Is the predicted natural log transformed concentration using a typical kriging
method that is detailed in [3],
Is the log of the standard error of the predicted concentration,
number of days (Julian Days) since January 1.
Four files that have been prepared provided to EPA are as follows:
krigepredozoneerrorl22001. sas 7bdat
2001, 12 km predicted error file for kriging the
CASTNET ozone data.
T-SpACE User's Guide
Page 46
Version 5_4h.0.1.21
January 2018
-------
krigepredozoneerrorl22002. sas 7bilat
krigepredozoneerrorS62001. sas 7hdat
2002, 12 km predicted error file for kriging the
CASTNET ozone data.
2001, 36 km predicted error file for kriging the
CASTNET ozone data.
krigepredozoneerrorS62002.sas 7bdat 2002, 36 km predicted error file for kriging the
CASTNET ozone data.
4.7.4 Select Validation Monitors File
The validation monitor file is a SAS data set that has been developed for EPA using the program
and data detailed in Section G. Figure 41 illustrates the format of one of these files for ozone,
2001, 12 km x 12 km.
Figure 41. Validation file for ozone - uses CASTNET data.
SAS System Viewer - castnet_grid_o3_2001_12km.sas7bdat - |[~ ]
File Edit View Window Help
a & m ฅ si in jiit hi *
2 cast
net_grid_o
3_200t_12
-------
then placed on the appropriate CMAQ/CAMx grid for comparison to the AQS monitors. The
four ozone data files created are listed below.
castnet grid o3 2001 12km, sas 7bdat
2001,
12
km
CASTNET ozone data.
castnet grid o3 2002 12km. sas 7bdat
2002,
12
km
CASTNET ozone data.
castnet grid o3 2001 36km sas 7bdat
2001,
36
km
CASTNET ozone data.
castnet grid o3 2002 3 6km sas 7bdat
2002,
36
km
CASTNET ozone data.
A similar set of PM2.5 files using the IMPROVE/STN are also available. All of these files are
available to the user.
4.7.5 Select Report File
The user is asked to choose a file name for the file where the SAS program will write its output.
This will be a plain text file. The SAS log will be written to the same folder as the report file.
This is a required entry.
4.7.6 Report Title
The user is asked to supply a title for the SAS reports that are generated during this session. This
is a required entry.
4.7.7 Run Validation
Clicking this button will launch the SAS program. The cursor will change to an hourglass while
the validation program is executing and will return to the default shape when it completes. Note
that if you have a large simulation, this can take up to 20 minutes to run.
Immediately, upon initiation, the SAS program is written to the directory as
{Report File name}-SAS Code.sas
This provides the user an archive of the code that was run for the validation and allows the user
to run the code through SAS separately if they desire. Once complete, the specified report file
and the SAS log file will be available in the directory that the user specified.
Section C provides a listing of an example report for a validation performed on a 12 km, 2001,
ozone model run.
The output provided in the validation report is as follows:
1. Distributional summary statistics for each of the model parameters tx , rf , rz is listed
below:
T-SpACE User's Guide
Page 48
Version 5_4h.0.1.21
January 2018
-------
The UNIVARIATE Procedure
Variable: TauX
N
Mean
Std Deviation
Skewness
Uncorrected SS
Coeff Variation
50
75.002386
0.00890517
-0.1340276
281267.899
0.01187318
Sum Weights
Sum Observations
Variance
Kurtosis
Corrected SS
Std Error Mean
50
3750.1193
0.0000793
-0.8349036
0.0038858
0.00125938
Location
Basic Statistical Measures
Variability
Mean 75.00239 Std Deviation 0.00891
Median 75.00090 Variance 0.0000793
Mode 74.99030 Range 0.03710
Interguartile Range 0.01490
NOTE: The mode displayed is the smallest of 3 modes with a count of 2.
Tests for Location: Mu0=0
Test -Statistic- p Value-
Student ' s t
Sign
Signed Rank
59554 . 95
25
637.5
Pr > ItI
Pr >= IMI
Pr >= ISI
<.0001
<.0001
<.0001
Quantiles (Definition 5)
Quantile
100% Max
99%
9b%
90%
15% Q3
501 Median
251 Q1
10%
5%
1%
0% Min
Estimate
75.0184
75.0184
75.0152
75.0138
75.0105
75.0009
74 .9956
74 .9905
74 .9881
74 .9813
74 .9813
Variable: TauX
Extreme Observations
Lowest
Value Obs
Highest
Value Obs
74 .9813
74 .9872
74 .9881
74 .9903
74 .9903
1
23
34
33
75.0141
75.0146
75.0152
75.0154
75.0184
36
28
37
Stem Leaf
75018 4
75016
75014 1624
75012 604
75010 0455688
75008 3558
75006 2
75004
75002 0101
75000 153
74998 0347908
74996 689
74994 22456
74992 8
74990 3376
74988 1
74986 2
74984
74982
74980 3
Boxplot
+ +
Multiply Stem.Leaf by 10*
T-SpACE User's Guide
Page 49
Version 5_4h.0.1.21
January 2018
-------
The UNIVARIATE Procedure
Variable: TauX
Normal Probability Plot
74 . 981 + + + + *
+ + + + + + + + + + +
-2 -1 0 +1 +2
2. A summary of all three parameters, tx , r Y, rz, number of iterations, the mean of the
simulated values, and the standard error observed from the simulated values.
n n n
Obs TauX TauY TauZ mnTauX mnTauY mnTauZ seTauX seTauY seTauZ
1 50 50 50 75.0024 0.16231 3.70759 .001259381 .000035409 0.11092
3. A summary of the performance of the Mean Square Error (MSE) and bias for the T-
SpACE, kriging, and the CMAQ model surface.
Overall Overall Overall
Overall MSE Overall Overall Bias Bias
Obs MSE HBM Krige MSE CMAQ Bias HBM Krige CMAQ
1 0.10 0.04 0.06 0.03 0.06 -0.09
4. A summary of the estimated percentage of time (across all monitoring days and
locations) that the calculated prediction intervals (i.e., 95% credible intervals for the
T-SpACE model) actually included the observed monitor value, using data for those
locations in the validation dataset that contain monitors.
Pct_Time_
Krige Pet Time
Obs ndays worse krigeworse CMAQ worse CMAQworse
1 333 3.60 12 11.71 39
Pet site
Krige Pet site
Obs nsites worse krigeworse CMAQ worse CMAQworse
1 41 9.76 4 31.71 13
5. A summary of the prediction intervals that represent bounds on the range of
measurements that are expected to contain the (unknown) true value a certain percent
of the time (typically, 95%) if the model assumptions are correct.
T-SpACE User's Guide Page 50 Version 5_4h.0.1.21
January 2018
-------
Obs
1
modelCI
0.97616
krigeCI
0.84095
predStd
0. 4232558534
A more detailed discussion of these summary statistics can be found in [2],
5.0 References
[1] McMillan, N.J., Holland, D.M., Morara, M., and Feng, J., Combining different sources of
particulate data using Bayesian space-time modeling, Environmetrics, 2010, Volume 21,
pp 48 - 65, DOI: 10.1002/env.984
[2] USEPA, Final Report Overview of EPA's Hierarchical Bayesian Model For Predicting
Air Quality Patterns In The United States Over Space And Time, For Use With Public
Health Tracking Data, March 13, 2009, Contract No. EP-D-04-068, Work Assignment
54.
[3] USEPA, Draft Report Hierarchical Bayesian Model Evaluation For Ozone Data, March
13, 2009, Contract No. EP-D-04-068, Work Assignment 54.
T-SpACE User's Guide
Page 51
Version 5_4h.0.1.21
January 2018
-------
SECTION A:
Detailed Description Of The
Markov Chain Monte Carlo (MCMC) Model
-------
Covariate Space-Time Auto-Regressive (STAR) {MCMC} Model
Model 5.1
Data:
Nt Number of time points
Np Number of space points
N = NT xNp Number of events (space-time points)
NX
xt g R ' i = \,...,N Monitoring data at event /
Xi g R i = \,...,N Sum of the monitoring data at event i (the number of monitoring
data for each event i can be 0, 1 or more than 1).
/ = 1,... ,N CMAQ data at event i (the number of CMAQ data for each event i
is always and only 1).
Di} g R i = 1,..., N 7 = 1,..., ND Bias jth basis function evaluated at event i
The bias is evaluated as a linear combination of 2nd order uniform B-spline functions defined
over a 3-dimensional lattice of uniform knots. The coefficients of the linear combination
represent unidimensional control points, that is,
nd
Bias = ^D^f.
}=i
If we indicate with N1,N2,N3 the dimensions of the CMAQ grid (that is, h\ = N1 ,
N2-N3= Np and N1-N2-N3 = N), and with Ml,M2,M3 the dimensions of the control-points
grid (this defines the degrees of freedom of the bias, that is, Ml -M2 -M3 = N"), and we
decompose the indexes as: i = ix + N} (i2 +N2i3), / = /, + M] ( j2 +M2j3), then the bias matrix is
then defined as
where bk(u) is the 2nd order kx\\ B-spline basis function evaluated at the parameter point u.
Setting \a], A, ]x [a2,^2]x [c/3,h:, ] the domain over which the bias is defined (such domain must be
bigger than or equal to the CMAQ grid domain), the uniform knots vectors over which the B-
spline basis functions are defined are respectively
-------
Ur = {ar,ar,ar,ar + sr,ar + 2sr.. .,ar + (Mr - 3)sr,br,br,br} r = 1,2,3
where sr = r = 1,2,3 .
r Mr- 2
Using B-splines as basis functions for the bias allows control of the degrees of freedom of the
bias through the number of control points. Furthermore, the piece-wise nature of the B-spline
functions respects the principle of locality, that is, local information does not affect regions far
from the region where the local information is defined. On the numerical side, B-splines allow a
tensor factorization of the bias matrix into three matrixes B' . = h , (z'r), r = 1,2,3 for a total
dimension of NXMX + N2M2 + N3M3, very much less than the total dimension of the full D-
matrix, which is NlMl ฆ N2M2 ฆ N3M3.
Parameters
Zi i = 1,..., N CAR process at event i.
jj.t t = 1,..., NT Mean level of the CAR process.
pD g Vector of coefficient for the bias.
tx Precision of the measurement error in the monitor observations.
ty Precision of the measurement error in the computer observations.
tz Precision of the mean process.
pz Temporal autocorrelation parameter of the mean process.
Likelihood:
k- lft(,).Z,,rv]=A'(M<,,rr),
|'y,\nl[l,,pD,ZT']=N(wl+fiDDT'),
where
wi =n + Zt.
Priors:
[z|rz,^]=Ar(o,rz(AV)ฎAP))
where AT (p) is the precision matrix corresponding to a time autoregressive model with
parameter p and A'' is the precision matrix corresponding to a space autoregressive model of
order r with a zero boundary condition.
Writing
-------
1 t = l,N
T i
T
nt =
np = number of spatial neighbors
and
nt(i) Jฃdti n jGdrpi nt(i)n jฃdtixdrpi
T P
Tz = TZ _ฃ}
1 + (pzf 1 vt(,),Tx\x
2=1 k=1
Yl[yt \zi>nm>PDJV
oc
-------
Explicitly:
n[z,|Z,_,rz,pz]x
i=1
\m\[pdIxWIpz]
[z,n,PD,tx,T',Tz,pz\x,r\*.
N f x\
nfe
N ( rY ^
nfe
TX N N,
eXPl"V^^^
z i=1 k=1
exp] - vS k - U(0 + Z< + P"l)- )]2
i =1
expi Y^/kt
(rz)2 Ar(pz)ฎAJ
pz )ฎ AP. IZ !> x
Full Conditional Distributions
Variable: Z, e R
where
\z,\-]=n{a->b,a->)
A = TXNX +Tr +TZ
B -TX [xt - N?{lt^]+ Tr [)', - + /?"/} )]+
Variable: //(eR
where
[^( I"]= A'1)
A = zxYJNf +rYNT +tm
ielt
B=T*Jd\x,-N?z]+
z'g/j
?'Yh,-(z,+PDD$+T'ef
islt
-------
Variable: f3D g R
\pd\-]=n(a1b,a1)
where
N
a, = ''Zn"d^d,>+sซ'd
i=1
B, = rIf1Dll[>>,-(fWl + Z,)]+TDffD
i=1
Variable: zx
[zx \-\=G(A,B)
where
A = \t"?+r*
^ i=1
54ilk-(A(0 + Z,)f+^
Z z=l Ar=l
Variable: zY
\z'\-\ = G(A,B)
where
2
B = -(fl,i,l + Z,+PDD$+S<
^ i=1
Variable: rz
[tz\-\=G(A,B)
where
2
-------
B = ^h:z,(z,-n;)+S'
2 z i=i
Variable: pz
[pZ \~h Z(a^bz)(pZtl-(pZ)2}2N (W ^expj-^rf(pz)zi(zi-Aif(pz))J
This full conditional is not a recognized form, so it has to be sampled using a Metropolis-
Hastings step.
Jump:
p'~ Ap'~pz )
Acceptance:
mini
^bZfp'ii-(p7)
-N \N -1
exp|" 2 S T' ('y)Z> (Z> ~ ฃ (p')
1"(
P
Z VI 2
N \N -1
exp|" 2 S (pZ K fe " (p2
-------
-------
Section B:
Validation SAS Program Example
* *
Program Name: ValidationTemplate.sas
Purpose: Summarizes the model parameters.
* *.
9
*===================================================================
* Inputs
* This program requires the following
* Hierarchical Bayes Model (HBM) output files:
* Chain.csv
* SurfaceModel.csv
*
* Kriged surface output SAS dataset:
*
* Monitor data for validation SAS dataset:
*
* This script provides statistics on the differences between
* the HBM, Kriged, and CMAQ predicted surfaces and the actual monitor data, on a log scale;
* (technically, CMAQ is modeled, not predicted).
* Statistics computed for each monitor location are 'mse' and 'bias'
* where mse is squared error for each monitor location, averaged either daily or by site;
* and bias provides the raw error for each location.
* (The error is the difference between the prediction and the monitor measured value).
* Statistics are computed by day ('time'), by site, and for all the data overall.
*
* Performance is indicated by the average differences in mse,
* and by how often the HBM mse is smaller than the Kriged or CMAQ surfaces.
* Specifically:
* Average differences between the Kriged mse and HBM mse,
* and between CMAQ mse and HBM mse,
* indicate whether the HBM squared-error is smaller or larger on average
* relative to Kriging or CMAQ.
* HBM performance is also rated by how often the daily HBM mse
* is smaller than those of the Kriged or CMAQ predictions;
* and by how often site-by-site mse is smaller.
* Finally, the prediction intervals of both the HBM model and the Kriging method
* are checked to determine how often the intervals
* include the true value measured by the monitor;
%let LocSurfaceData = $$SurfaceFileFolder$$;
%let LocChainData = $$ChainFileFolder$$;
%let LocMonitorData = $$MonitorFileFolder$$;
%let LocKrigeData = $$KrigeFileFolder$$;
-------
/* HBM data (Hierarchical Bayesian Model) */
%let inChainFile = $$ChainFileName$$;
%let model SurfaceFile = $$SurfaceFileName$$;
/* Kriged data*/
%let krigedSurface = $$KrigeFileName$$;
/* Monitor data: Combined Improve and STN for PM2.5; CASTNet for 03*/
%let monitorData = $$MonitorFileName$$;
* 'afterDate' is the day before the start date;
%let afterDate = '$$AfterDate$$'d;
/* Report Title */
%let rptTtl = $$ReportTitle$$;
/* Define the range */
%let minX = $$xMin$$;
%let maxX = $$xMax$$;
%let minY = $$yMin$$;
%let maxY = $$yMax$$;
/* These are the reading columns for data */
%let monRdg = $$monRdg$$;
%let logRdg = $$logRdg$$;
%let Valcols = &monRdg. &logRdg.;
libname sd "&LocMonitorData";
libname sd2 "&LocKrigeData";
filename rptFile '$$ReportFile$$';
filename logFile '$$LogFile$$';
/* macro to determine if a column exists in a dataset */
%macro ColumnExist(DS=,Var=);
%global found;
%let dsid = %sysfunc(open(&DS,i));
%if &dsid le 0 %then %do;
%put WARNING: %sysfunc(sysrc()) ~ %sysfunc(SysMsg());
%let found = -1;
%end;
%else %do;
%let dsname = %sysfunc(DsName(&dsid));
%put NOTE: Opened dataset: &dsname;
%let NVar = %sysfunc(attrn(&dsid,NVARS));
-------
%putNOTE: There are &Nvar Variables in &dsname;
%let found = 0;
%do i = 1 %to &Nvar;
%let vn&i = %sysfunc(varname(&dsid,&i));
%if %upcase("&&vn&i") = %upcase("&VAR") %then %do;
%let found = 1;
%goto ColdStop;
%end;
%C old Stop:
%end;
%if &found = 1 %then %put NOTE: Variable &Var was found.;
%else %put NOTE: Variable &Var was not found.;
%end;
%let rc = %sysfunc(close(&dsid));
%mend ColumnExist;
/* end macro */
proc printto log = logFile print = rptFile New;
run;
* Calculate the summary statistics on the HBM model parameters
proc import out = chain
datafile = "&LocChainData.\&inChainFile"
dbms = csv replace;
/* tau is the precision parameter in HBM
x denotes FRM monitor precision
y denotes CMAQ 'precision'
z denotes precision for Conditional autoregressive (CAR) process
These are estimated from the characteristics of the HBM Markov chain
seTauX is se_tauX; etc.
*/
proc univariate data = chain plot;
var tauX tauY tauZ;
output out = sumParms
mean = mnTauX mnTauY mnTauZ
stdErr = seTauX seTauY seTauZ
n = nTauX nTauY nTauZ;
proc print data = sumParms;
title "Summary Statistics for &rptTtl";
-------
run; quit;
/* create a couple of macro variables */
/* posterior variance to be computed from x and z precision ******/
data _null_;
set sumParms;
call symput("sigmaX", 1 / mnTauX);
call symput("sigmaZ", 1 / mnTauZ);
run;
/* dump values to log */
%put &sigmaX;
%put &sigmaZ;
* Comparison summaries *;
* Read in HBM predicted surface, referred to as modelsurface;
proc import out = hbm
datafile = "&LocSurfaceData.\&modelSurfaceFile"
dbms = csv replace;
data hbm_surfacePreds(keep = time xCoord yCoord predAvg predStd computerData test
rename = (xCoord = col yCoord = row predAvg = predSurface computerData =
cmaqSurface));
set hbm;
* Remove the data where a prediction was not made *;
time = time - &afterDate.;
if predAvg A= -999;
test = predStd * predStd;
* This step exponentiates the results back to normal units *;
array vars [4] predAvg biasAvg monitorData computerData;
do i = 1 to 4;
if vars[i] ne -999 then vars[i] = exp(vars[i]);
end;
* The bias is transformed *;
if vars[2] ne -999 then vars[2] = (vars[2] - 1) * vars[l];
drop i;
* For the validation results, any monitor result greater than 600 change to -999 *;
if vars[3] > 600 then vars[3] = -999;
/* covarData is the 15th variable (and last) in the record. Check to see if it contains
a carriage return (ascii code 13. or 'OD'x). If not, take all of the 15th variable as
the value of covarData. If it does contain a carriage return then only use the data up
-------
to, but not including, the carriage return as the value of covarData */
if index(varl5,'0D'x) = 0 then covarData = input(varl5, 8.);
else covarData = input(substr(varl5, 1, index(varl5, 'OD'x) - 1), 8.);
proc sort data = hbmsurfacePreds;
by time col row;
data check;
set hbm surfacePreds;
diff = predSurface - cmaqSurface;
proc sort data = check;
by time;
proc means data = check noprint;
by time;
var diff;
output out = sumt(drop = type freq_)
mean = mnDiff
stdErr = seDiff
n = nDiff;
run; quit;
proc sort data = check;
by col row;
proc means data = check noprint;
by col row;
var diff;
output out = sumcr(drop = type freq)
mean = mnDiff
stdErr = seDiff
n = nDiff;
run; quit;
* Kriging data for comparison *;
data krige_predict;
set sd2.&krigedSurface.;
krigePred = exp(krige_predict);
run; quit;
proc sort data=krige_predict;
by time col row;
run;
* IMPROVE, STN, or CASTNet, data for comparison *;
-------
%ColumnExist(DS=sd.&monitorData.,Var=site);
%let foundsite = &found;
%ColumnExist(DS=sd.&monitorData.,Var=site_code);
%let foundsitecode = &found;
%ColumnExist(DS=sd.&monitorData.,Var=logpm);
%let foundlogpm = &found;
data tempValidation;
set sd.&monitorData.;
if &foundsite = 1 then;
rename site = sitelD;
run; quit;
data tempValidation;
set tempValidation;
if &foundsitecode = 1 then;
rename site code = sitelD;
run; quit;
data tempValidation;
set tempValidation;
if &foundlogpm = 1 then;
pm = exp(logpm);
run; quit;
data surfvalidation (keep = time col row site &Valcols.);
set tempValidation (rename = (sitelD = site));
where (col ge &minX.) and (col le &maxX.) and (row ge &minY.) and (row le &maxY.);
run; quit;
proc sort data = surf validation;
by time col row;
run;
*
* Combine the model prediction, Kriging results, and the IMPROVE/STN data
* NOTE: The bias is calculated as bias = validation value - predicted value
*
9
data combined;
merge krige_predict
surf validation (in = invalidation)
hbmsurfacePreds (in = inPred);
-------
by time col row;
* Data to keep in the file *;
if invalidation;
if predSurface A= .;
if cmaqSurface = -999 then cmaqSurface = .;
if predSurface = -999 then predSurface = .;
* Calculate the summary stats *;
* regular units *;
msemodel = (predSurface - &monRdg.) **2;
mse krige = (krigePred - &monRdg.) **2;
mseCMAQ = (cmaqSurface - &monRdg.) **2;
msediff = mse krige - mse model;
mse_diff_2 = mse_CMAQ - mse_model;
biasmodel = &monRdg. - predSurface;
bias krige = &monRdg. - krigePred;
bias CMAQ = &monRdg. - cmaqSurface;
* log units *;
mselmodel = (log(predSurface) - &logRdg.) **2;
mse_lkrige = (krige_predict - &logRdg.) **2;
mse_lCMAQ = (log(cmaqSurface) - &logRdg.) **2;
* compare MSE of Krige to HBM and of CMAQ to HBM;
mseldiffKrige = mselkrige - mse lmodel;
mse ldiffCMAQ = mselcmaq - mse lmodel;
biaslmodel = &logRdg. - log(predSurface);
bias lkrige = &logRdg. - krige_predict;
bias lCMAQ = &logRdg. - log(cmaqSurface);
run;
* Summarize performance by time *;
proc sort data = combined out = combined_pltnt;
by time;
run;
* Average daily MSE for HBM, Kriging, and CMAQ *;
* avg mseldiff... > 0 implies HBM error is smaller on average;
* semse... provides measure of variation in the magnitude of square errors;
proc means data = combined_pltnt noprint;
-------
by time;
var mselmodel mselkrige mselcmaq mseldiffkrige mseldiffcmaq
biaslmodel biaslkrige bias lCMAQ;
output out = mse_pltnt (drop = type freq)
mean = mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ
stdErr = semselmodel se mse lkrige semselcmaq
semseldiffkrige semseldiffcmaq
sebiaslmodel se bias lkrige se bias lCMAQ
n = nmselmodel nmselkrige nmselcmaq
nmseldiffkrige nmseldiffcmaq
nbiaslmodel n bias lkrige n bias lCMAQ;
* Annual MSE for HBM, Kriging, and CMAQ. Bias summary for HBM, kriging, and CMAQ
9
proc means data=combined_pltnt noprint;
var mse lmodel mse lkrige mse lcmaq mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
output out = overall_mse_pltnt (drop = type freq_)
mean = mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
proc print data = overall_mse_pltnt label;
var mse lmodel mse lkrige mse lcmaq
bias lmodel bias lkrige bias lCMAQ;
label mse lmodel = 'Overall MSE HBM'
mse lkrige = 'Overall MSE Krige'
mse lcmaq = 'Overall MSE CMAQ'
bias lmodel = 'Overall Bias HBM'
bias lkrige = 'Overall Bias Krige'
bias lcmaq = 'Overall Bias CMAQ';
format mse lmodel mse lkrige mse lcmaq
bias_lmodel bias_lkrige bias_lCMAQ 5.2;
run; quit;
* Summarize performance by site *;
proc sort data = combined_pltnt;
by site;
proc means data = combined_pltnt noprint;
by site;
var mse lmodel mse lkrige mse lcmaq
mse ldiffkrige mse ldiffcmaq
bias lmodel bias lkrige bias lCMAQ;
-------
output out = mse_pltnt2
mean = mselmodel mselkrige mselcmaq
mseldiffkrige mseldiffcmaq
biaslmodel bias lkrige bias lCMAQ;
* Comparing HBM to CMAQ and Krige based on MSE *;
* mse_pltnt is daily statistics (by 'time');
data mse_pltntl;
set mse_pltnt;
retain suml sum_2;
if mse ldiffkrige > 0 then do;
krige_larger = 1;
sum_l + 1;
end;
if mse ldiffcmaq > 0 then do;
cmaq_larger= 1;
sum_2 + 1;
end;
proc means data = mse_pltntl noprint;
var sum_l sum_2 time;
output out = Improve_Time(drop = type freq_ tl t2)
max = krigeworse CMAQworse
n = tl t2 ndays;
data ImproveTime;
set Improve Time;
Pct_Time_Krige_worse = (krigeworse / ndays) * 100;
Pct_Time_CMAQ_worse = (cmaqworse / ndays) * 100;
proc print data = Improve Time;
title 1 "&rptTtl: % Improvement over Kriging and CMAQ";
var ndays pct_time_krige_worse krigeworse pct_time_CMAQ_worse cmaqworse;
format pct_time_krige_worse pct_time_CMAQ_worse 7.2;
run; quit;
* mse_pltnt2 is statistics by site;
data mse_pltnt3;
set mse_pltnt2;
retain sum l sum_2;
if mse ldiffkrige > 0 then do;
krige_larger = 1;
sum_l + 1;
-------
end;
if mseldiffcmaq > 0 then do;
cmaq_larger= 1;
sum_2 + 1;
end;
run; quit;
proc means data = mse_pltnt3 noprint;
var sum_l sum_2 _freq_;
output out = Improve_site(drop = type freq_ tl t2)
max = krigeworse CMAQworse
n = tl t2 nsites;
data Improvesite;
set Improve site;
Pct_site_Krige_worse = (krigeworse / nsites) * 100;
Pct_site_CMAQ_worse = (cmaqworse / nsites) * 100;
proc print data = improve site;
title 1 "&rptTtl: % Improvement over Kriging and CMAQ";
var nsites pct_site_krige_worse krigeworse pct_site_CMAQ_worse cmaqworse;
format pct_site_krige_worse pct_site_CMAQ_worse 7.2;
run; quit;
*
* 95% Prediction Interval : this interval is calculated using modelstd=sqrt(sigmaZ+sigmaX)
* MCMC Prediction Interval: this interval is calculated using
modelstd=sqrt(predStd*predStd+sigmaX)
* here begin to calculate the prediction interval
*
9
* pmsurfacePreds is HBM predicted surface;
* sigmaX is posterior variance of FRM data;
* sigmaZ is posterior variance of CAR process;
data compare;
merge krige_predict(keep = time col row krige_predict stdErr)
hbm_surfacePreds(keep = time col row predSurface cmaqSurface predStd)
surf_validation(in = Validation);
by time col row;
if Validation;
if predSurface A= .;
sigmaX = & sigmaX;
sigmaZ = & sigmaZ;
modelstd = sqrt( sigmaZ + sigmaX);
-------
* keep track of confidence interval coverage of prediction vs. monitor measurement;
if ((&logRdg. > log(predSurface) - 2 * modelstd) and (&logRdg. < log(predSurface) + 2 *
modelstd)) then modelCI = 1;
else modelCI = 0;
if ((&logRdg. > krige_predict - 2 * stdErr) and (&logRdg. < krige_predict + 2 * stdErr)) then
krigeCI = 1;
else krigeCI = 0;
proc means data = compare noprint;
var modelCI krigeCI predStd;
output out = PredictionInterval(drop = TYPE FREQ)
mean = modelCI krigeCI predStd;
proc print data = predictioninterval;
var modelCI krigeCI predStd;
titlel "&rptTtl: Average Prediction Interval";
run; quit;
proc printto;
run;
-------
Section C:
Validation Report Example
The SAS System 14:36 Friday, April 24, 2009 1
The UNIVARIATE Procedure
Variable: TauX
Moments
N
Mean
Std Deviation
Skewness
Uncorrected SS
Coeff Variation
50
75.002386
0.00890517
-0.1340276
281267.899
0. 01187318
Sum Weights
Sum Observations
Variance
Kurtosis
Corrected SS
Std Error Mean
50
3750.1193
0.0000793
-0.8349036
0.0038858
0. 00125938
Basic Statistical Measures
Location
Variability
Mean 75.00239 Std Deviation 0.00891
Median 75.00090 Variance 0.0000793
Mode 74.99030 Range 0.03710
Interquartile Range 0.01490
NOTE: The mode displayed is the smallest of 3 modes with a count of 2.
Tests for Location: Mu0=0
Test
-p Value-
Student's t
Sign
Signed Rank
59554.95
25
637 . 5
Pr >
Pr >=
Pr >=
I t |
I M |
I S |
<.0001
<.0001
<.0001
Quantiles (Definition 5)
Quantile
Estimate
100
99%
95%
90%
75%
50%
25%
10%
5%
1%
0
Max
Q3
Median
Q1
Min
75.0184
75.0184
75.0152
75.0138
75.0105
75.0009
74 . 9956
74 . 9905
74.9881
74.9813
74 . 9813
-------
The SAS System
14:36 Friday, April 24, 2009 2
The UNIVARIATE Procedure
Variable: TauX
Extreme Observations
-Lowest-
Value
Obs
Highest
Value Obs
74 . 9813
74.9872
74 . 9881
74 . 9903
74 . 9903
1
23
44
34
33
75.0141
75.0146
75.0152
75.0154
75.0184
36
48
28
37
4
Stem Leaf # Boxplot
75018 4 1 |
75016 |
75014 1624 4 |
75012 604 3 |
75010 0455688 7 + +
75008 3558 4 | |
75006 2 1||
75004 | |
75002 0101 4 I + I
75000 153 3 * *
74998 0347908 7 | |
74996 689 3 | |
74994 22456 5 + +
74992 8 1
74990 3376 4
74988 1 1
74986 2 1
74 984
74 982
74980 3 1
+ + + +
Multiply Stem.Leaf by 10**-3
-------
The SAS System 14:36 Friday, April 24, 2009 3
The UNIVARIATE Procedure
Variable: TauX
Normal Probability Plot
75.019+ ++
I + +
| *
| *** +
| **** +
| *** + +
| * + +
I + + +
|
I +**
| * + +
| * +
I + +
I + +
74.981++++*
+ + + + + + + + + + +
-2 -1 0 +1 +2
-------
The SAS System
14:36 Friday, April 24, 2009 4
The UNIVARIATE Procedure
Variable: TauY
Moments
N
Mean
Std Deviation
Skewness
Uncorrected SS
Coeff Variation
50
0.16230594
0.00025038
-0.3695166
1.31716398
0.15426482
Sum Weights
Sum Observations
Variance
Kurtosis
Corrected SS
Std Error Mean
50
8 .115297
6.269 0 6E-8
-1.0537027
3. 07184E-6
0.00003541
Basic Statistical Measures
Location
Variability
Mean 0.162306 Std Deviation 0.0002504
Median 0.162334 Variance 6.26906E-8
Mode 0.162334 Range 0.0008990
Interguartile Range 0.0004300
NOTE: The mode displayed is the smallest of 3 modes with a count of 2.
Tests for Location: Mu0=0
Test
-p Value-
Student's t
Sign
Signed Rank
4583.72
25
637 . 5
Pr >
Pr >=
Pr >=
I t |
I M |
I S I
<.0001
<.0001
<.0001
Quantiles (Definition 5)
Quantile Estimate
100% Max
0.162701
99%
0.162701
95%
0.162670
90%
0.162592
75% Q3
0.162512
50% Median
0.162334
25% Q1
0.162082
10%
0.161920
5%
0.161907
1%
0.161802
0% Min
0.161802
-------
The SAS System
14:36 Friday, April 24, 2009 5
The UNIVARIATE Procedure
Variable: TauY
Extreme Observations
-Lowest-
Value
Obs
Highest
Value Obs
0.161802
0.161886
0.161907
0.161910
0.161915
0.162602
0.162615
0.162670
0.162682
0.162701
46
47
48
50
49
Stem Leaf # Boxplot
1627 0 1 |
1626 78 2 |
1626 02 2 |
1625 6778 4 |
1625 011134 6 + +
1624 66777 5 | |
1624 3 1||
1623 589 3 | |
1623 013333 6
1622 58 2 | |
1622 02 2 | |
1621 59 2 | |
1621 1 1||
1620 88 2 + +
1620 02 2
1619 778 3
1619 1122 4
1618 9 1
1618 0 1
+ + + +
Multiply Stem.Leaf by 10*^-4
-------
The SAS System
14:36 Friday, April 24, 2009 6
The UNIVARIATE Procedure
Variable: TauY
Normal Probability Plot
0.162725+ ++
| + + * *
I +* *
0.162275+ **+
I * *
| +*
| + + *
| ++ *
| ++ *
| k k k k
| * +
0.161825+ * ++
+ + + + + + + + + + -
-2 -1 0 +1 +2
-------
The SAS System
14:36 Friday, April 24, 2009 7
The UNIVARIATE Procedure
Variable: TauZ
Moments
N
Mean
Std Deviation
Skewness
Uncorrected SS
Coeff Variation
50
3.7075934
0.78433975
-0.0195176
717 .456695
21.154956
Sum Weights
Sum Observations
Variance
Kurtosis
Corrected SS
Std Error Mean
50
185.37967
0.61518885
-1.1976147
30.1442536
0.11092239
Location
Basic Statistical Measures
Variability
Mean 3.707593
Median 3.714860
Mode
Std Deviation 0.78434
Variance 0.61519
Range 2.637 31
Interquartile Range 1.34256
Tests for Location: Mu0=0
Test
-p Value-
Student's t
Sign
Signed Rank
33.42511
25
637 . 5
Pr >
Pr >=
Pr >=
I t |
I M |
I S |
<.0001
<.0001
<.0001
Quantiles (Definition 5)
Quantile
Estimate
100
99%
95%
90%
75%
50%
25%
10%
5%
1%
0
Max
Q3
Median
Q1
Min
5.00858
5.00858
4.90812
4.77865
4.38390
3.71486
3.04134
2.62880
2.48634
2.37127
2.37127
-------
The SAS System
14:36 Friday, April 24, 2009 8
The UNIVARIATE Procedure
Variable: TauZ
Extreme Observations
Lowest-
Highest-
Value
Obs
Value
Obs
2.37127
1
4
80453
46
2.43050
2
4
85586
47
2.48634
3
4
90812
48
2.54217
4
4
96257
49
2.60102
5
5
00858
50
Stem Leaf
50 1
48 0616
46 0505
44 4 95
42 2838
40 0617
38 405
36 3949
34 2728
32 0516
30 495
28 2839
26 0616
24 394
22 7
- + -
- + -
- + -
- +
Boxplot
+ +
+ +
Multiply Stem.Leaf by 10**-1
-------
The SAS System 14:36 Friday, April 24, 2009 9
The UNIVARIATE Procedure
Variable: TauZ
Normal Probability Plot
5.1+ +++
I * + ** *
I **** +
I *** + +
I *** +
| ** + +
I +
3.7+ ***
| + + **
| + + ***
| + + ***
| + ****
2.3+ * +++
+ + + + + + + + + + -
-2 -1 0 +1 +2
-------
Summary Statistics for Validation for Users Guide 10
14:36 Friday, April 24, 2009
n n n
Obs TauX TauY TauZ mnTauX mnTauY mnTauZ seTauX seTauY seTauZ
1 50 50 50 75.0024 0.16231 3.70759 .001259381 .000035409 0.11092
-------
Summary Statistics for Validation for Users Guide 11
14:36 Friday, April 24, 2009
Overall Overall Overall
Overall MSE Overall Overall Bias Bias
Obs MSE HBM Krige MSE CMAQ Bias HBM Krige CMAQ
1 0.10 0.04 0.06 0.03 0.06 -0.09
-------
Validation for Users Guide: % Improveme
Pct_Time_
Krige
Obs ndays worse krigeworse
1 333 3.60 12
over Kriging and CMAQ 12
14:36 Friday, April 24, 2009
Pct_Time_
CMAQ_worse CMAQworse
11.71 39
-------
Validation for Users Guide: % Improvement over Kriging and CMAQ 13
14:36 Friday, April 24, 2009
Pet site
Krige Pet site
Obs nsites worse krigeworse CMAQ worse CMAQworse
1 41 9.76 4 31.71 13
-------
Validation for Users Guide: Average Prediction Interval 14
14:36 Friday, April 24, 2009
Obs modelCI krigeCI predStd
1
0.97616
0.84095
0.4232558534
-------
Section D
Cells.txt
Edit
=t i
4"-
L
ROW
Cent
er
ฆ Sir. Corr
-
i,SE Corf
er;
HE corner
Cn* v.orr
er"
Longitude
Latitude
Longitude
Latitude
Longitude
Latitude
Longitude
Lstitf. de
Longiti de
Lctm de
1
i
-99
4595S
25.4^195
-93 5455"
25.41-12
- 53.i2-2-
2S.42005
_ ฆ
h 2051
" c
525-5
-99. 5510E
2 5 51::2
1
2
-99
49291
2S.5"So"
-99.5519S
25.52262
-93.-2051
25. 526~5
. -
.;; " '
If
6234y
- . c 5 * 9
23.62056
1
3
-HQ
^9625
25, 6S5-2
-99.55529
25.
23. o",349
-99
* :ป . U ^
2S
"4026
23.~?-22
1
i
-93
"9.-160
2S,~9221
-99.55552
_ 1. 2 : _
25.-4026
44030
2 5
S4-U7
-9=i. 55226
25. S4412
1
*
-q:i
502
25. ฃ9903
-99.55225
25. 5441?
-33,4-020
26.S4-Q""
:
25
95291
-99. 56 5-0
25. ^096
1
0
-93
5 06? 2
29.0u5S6
-99.56570
28.95096
25.95291
- ; -
4465^
29
060-S
-93. 56915
29.05~S2
1
~
-99
509-0
29.112'-
-99.56 ?ฑ 5
2 9. 0 5 - S 2
-99.4J6S"
2 050""ฃ
-99
45016
29
; 'i
-99.572ol
29.164-2
1
P.
51205
29.21969
-99.5_261
29.16^-2
-99.4 5016
29.16-65
-99
4 5 2-4 5
2 9
-99. 5"60ฃ
--
1
-99
51645
29. 32664
5'
29.2"166
-99. -152-15
29. 2"462
4 55-9
29
25159
-99. tr"9c-
2 ." 5 5"
1
10
5195S
29.-42362
-99.5-35-
29.3"S62
-99.455-9
29.3 5159
-99
ฆ>*6011
29
4S550
-99.5ฃ*ri?
2' -2'-;;
i
11
- 99
r -
2-1. 54064
->'<,
29.48563
-9J.46011
29.4SS60
_ ;
29
-99.5ฃoc6
2 . 2 2 2 3 2
1
12
- -i
29.64-69
-99.5 5656
2 J,5^265
-29.46345
29.59563
loo^O
2 r'
-02~0
Q9, 5900-
2 9. 639"2
1
i:
-99
5:01-1
2 -cr^--
-13 * -t00~
29.699"2
-99.465""9
29."02"0
-99
4'014
29
S09~3
- . . -.. :
29. 50651
1
l-
- ;"|C|
53255
29.56155
-99.59355
29.S06S1
- i-01-t
20. sO^"^
:
4 "* E S ri
2^
91692
-93. 5^11
29.91393
1
15
-99
5 2"02
29.96902
-99. 59"il
29.91392
-99.-1 ""350
29.91692
-
- :
20
02405
-93. 60065
30.02105
1
16
-99
5-050
20.0~619
-99.600o5
20.0210S
-99.4-635
20.02-08
-99
4&026
20
12126
-99.60420
20.12S2~
1
17
-99
C
20.1S22:?
-99.60-20
30.1ZS2"7
-99.4S026
20.12126
-99
- c : 54
20
": 5 - 5
-93.6'> 6
" ฆ. " ^ -i
1
13
-99
54"i4
'0,29^62
-99,6077 6
20.22 5-5
20.2254S
-i!9
20
-99.51122
1
1*
-99
5509?
30.39"S"
-99.ell 32
20.3-2"2
-99.45 04
30.^45-2
- ฆ
4904 5
30
4 5300
-99.614 90
30.44 999
1
in
:
55-442
20.50516
-99. ol-90
20.Ji9 99
-Q-j,4Q04 5
20.4 5200
-99
- "I 2 >2
20
55021
- 33. 51S4 5
2 25"2.
1
21
- jCi
3u.612-5
-99.615-3
20. 5 5 ~ 2 9
-c
20.56031
-:"2:
30
65754
-99.622^9
30.56462
1
41
-99
5614 5
20."1952
-Oct. 622'9 j
20.66462
-:. -; ฃ
30. 66764
-99
500-2
20
-93.62569
: " _
1
"J .3
-
56-* 97
20. 52"*20
-99.62563
20, ~"*19~
--">9. 500~2
DO. """500
_o:i
*.141-
20
55239
-93. 62321
30. S""936
Note that this file is specific to a year and grid size.
-------
-------
Section E:
Summary.CSV file description
The program generates a file named SUMMARY.CSV with each fit of the T-SpACE
simulation, which contains summary statistics on all parameter estimates in T-SpACE. Table E-
1 presents example rows from this file. Each row contains summary statistics on the estimates
which were generated for a particular model parameter (specified in the first column) within each
iteration of the MCMC simulation. These summary statistics include the minimum ("Min"),
maximum ("Max"), average ("Avg"), and variance ("Var") and are calculated on the set of
estimates associated with the specified model parameter. Thus, the number of estimates entering
into these statistics equals the number of iterations performed within the simulation. Also,
estimates generated within the first and last iteration of the simulation are presented in columns
labeled "First" and "Last," respectively. The average and variance are of greatest importance
within this file, as they represent unbiased estimates of the true mean and variance of the
marginal posterior distribution for the given model parameter. The last column (labeled
"Success") has a more technical interpretation; it represents the proportion of iterations for which
the model parameter successfully moved to a new sample value when the analysis proposed such
a move. This proportion is less than 1 whenever the algorithm incorporates a Metropolis-type
step in which at least one move to a new sample value was rejected.
Table E-1. Example Contents of the SUMMARY.CSV Data File
Name
First
Last
Min
Max
Avg
Var
Success
ZfOl
0.981941
-0.31232
-0.59761
0.981941
0.062989
0.081035
1
zm
0.685994
-0.45824
-0.47367
0.685994
0.057133
0.070435
1
Z[2669l
-0.08756
-0.2092
-0.78271
0.621141
-0.06417
0.081926
1
vfoi
0.449908
-0.29966
-1.73886
1.54585
0.00787
0.267134
1
vrn
1.29497
-0.23137
-2.1078
1.29497
0.003764
0.302053
1
V[2669l
-0.5677
-0.26227
-1.11112
1.14958
0.013141
0.23977
1
Mu
1.54973
2.22671
1.54973
2.72309
2.36964
0.039279
1
Eta
-21.4706
20.2539
-73.8361
76.3331
1.04448
1083.5
1
BetaC[0l
14.6051
7.58928
-72.5916
72.2254
-0.58191
763.101
1
BetaC[1l
43.5833
27.0894
-103.352
94.3341
-0.3937
902.584
1
BetaCri 11
16.7617
-35.1617
-71.7791
83.4948
2.95884
1037.14
1
BetaD[0]
-0.63135
-0.36124
-1.40918
0.204948
-0.75164
0.106207
1
Beta DMl
-0.65789
-0.34519
-1.07664
0.06656
-0.53966
0.054132
1
BetaD[271
0.000295
0.000613
-0.00033
0.000824
0.000334
5.73E-08
1
BetaV
2.26E-05
1.54E-05
-7.66E-05
7.05E-05
-2.13E-06
1.04E-09
1
TauX
22.5191
181.391
3.00131
1380.1
142.902
44200.4
1
TauY
1.00057
1.00137
0.998501
1.00475
1.001
1.23E-06
1
TauS
1.6368
0.505814
0.007201
8.80531
1.08206
1.47834
1
TauZ
2.24921
4.60805
2.24921
5.72256
4.622
0.336919
1
TauV
1.40102
1.06579
0.697571
2.01309
1.15445
0.074964
1
RhoZ
0.448629
0.017082
0.000195
0.448629
0.040497
0.00768
0.607
RhoV
0.47431
0.014791
0.000597
0.47431
0.048789
0.008882
0.727
-------
WfOl
2.62485
1.76478
1.46325
2.88756
2.17617
0.084941
1
W[1]
2.32892
1.61885
1.61885
3.00553
2.17031
0.072253
1
Wf26691
1.55533
1.86789
1.34492
2.78993
2.04901
0.08942
1
Bias[01
-0.53088
-0.53437
-1.12979
-0.32377
-0.76766
0.03328
1
Biasf 11
-0.50349
-0.5204
-1.06923
-0.30484
-0.73469
0.029905
1
Bias[26691
1.12976
1.1581
0.650178
1.57056
1.01738
0.030676
1
Covar[01
-21.0207
19.9543
-73.8786
76.3927
1.05235
1082.49
1
Covar[1l
-20.1756
20.0226
-73.663
76.4343
1.04824
1084.8
1
Covar[26691
-22.0383
19.9917
-73.9745
76.2039
1.05762
1086.94
1
WeeklyAvg[01
2.03077
2.00202
1.81888
2.5076
2.12928
0.023503
1
WeeklyAvg[1]
1.85091
1.8708
1.69304
2.53312
2.13047
0.026893
1
Wee klyAvg [5331
1.59642
1.96121
1.59642
2.45346
2.09582
0.021795
1
MonthlyAvgfOl
2.03077
2.00202
1.81888
2.5076
2.12928
0.023503
1
MonthlyAvgfll
1.85091
1.8708
1.69304
2.53312
2.13047
0.026893
1
MonthlyAvgf5331
1.59642
1.96121
1.59642
2.45346
2.09582
0.021795
1
YearlyAvgfOl
2.03077
2.00202
1.81888
2.5076
2.12928
0.023503
1
YearlyAvgfll
1.85091
1.8708
1.69304
2.53312
2.13047
0.026893
1
YearlyAvgf5331
1.59642
1.96121
1.59642
2.45346
2.09582
0.021795
1
-------
Section F:
4th Highest Concentration Surface
Within the time period of interest (such as a year), the T-SpACE software determines the fourth
highest concentration within each grid cell across all simulated air pollution concentration
surfaces that T-SpACE generates for each day in the time period. The results are averaged and
smoothed to generate a single air pollution concentration response surface (daily maximum 8-
hour average ozone O3 concentration). T-SpACE generates a *.CSV file, which contains the
cell-specific predictions that make up this response surface. However, only the following four
columns, which document the grids and the predictions, are included in this CSV file:
XCoord: The cell's x-coordinate (east-west) within the grid (where the coordinate value
increases by 1 with each grid cell as you move from west to east).
YCoord: The cell's y-coordinate (north-south) within the grid (where the coordinate
value increases by 1 with each grid cell as you move from south to north).
PredAvg: The (natural log-transformed) T-SpACE-predicted value of the fourth highest
concentration for the given grid cell.
PredStd: The (natural log-transformed) T-SpACE-predicted value of the standard error
for the fourth highest concentration for the given grid cell.
In particular, because the air pollution concentration response surface for the fourth highest
concentration represents the entire time period of interest (e.g., one year), a date column is not
necessary in this *.CSV file.
-------
-------
Section G:
Diagram of Preparation of Validation Files
Validation of Ozone Results
castnet_data_2Q01_20
02_daily_8hr_maxinnu
ms.csv
CoordinatesfGrid
sizc}_{year}. sasdbat
Step 1-Put Other
Ozone Network Data
on Grid.sas
monitor_Q3_{year},csv
{FHM data from
streamlined software)
castnet grid o3.g
bat
Step 2-Create
Ozone Krige
Surface, sas
krigo_prodict_ozone_
error.sasbdal
krlge_{year)_ozone.csv
ozonesurface{year}.
CSV
(From streamline
software-Predicted
Surface)
Step 3-Summarize
model parameters-
Ozone.sas
-------
Validation of PM2,5 Results
IMP20O4O53O.C5V
Coordinates(Grid
size}_{year}.sasdbat
Step1-Put Other
PM25 Network Data
on Grid.sas
AQS
monitor_{Pollutant}_{
Year},csv
(FRM data from
streamlined software)
irtiprovegrid _pm25sasdbat
Step 2-Create
PM25 Krige
Surface, sas
SurfacePM25{year}
_{grid>.c5v
(From streamline
software-Predicted
Surface)
krigepradict _pm25_e
rror.sasbdat
krige{year}_pm25.csv
Step 3-Summarize
model parameters-
PM25.sas
-------
Section H: Visualization of T-SpACE Three-Dimensional Coordinate System
-------
Step 6: Visualize Surface
This step allows the user to visualize the estimated T-SpACE air pollution concentration surface
in a three-dimensional (3-D) coordinate system (latitude, longitude, concentration).
Surf ace 2xsv - AqVปs4 - Combined Data for Space Time Models
Date 1/2/2001
Rewind
Previous
Predicted
Kriging
Monitors
CMAQ
Concentration
3.41 19.99 43.38 86.77 90.16 113.56
Monitor and CMAQ Concentration
)000000 >000000 OOOOOOl :000000 DOOOOOC DOOOOOC
Covar Concentration
-------
This T-SpACE graphic displays PM2.5 monitor concentration values (white spheres) and CMAQ
air quality model estimates (colored surfaces) for the north eastern portion of the US.
25% Median
10.06 15.61
10.60 15.90
9.23 15.96
Predicted
Monitors
CMAQ
Concentration
27.78 55.56 83.33 111.11
Monitor and CMAQ Concentration
-------
Section I: T-SpACE Utilities
-------
T-SpACE Utilities
The seven T-SpACE utilities that are provided to assist the user in generating air pollution
concentration surfaces are provided here in summary format.
Tool #1: Compare AQS Raw Data files: This tool determines which monitor locations are
contained in one AQS file but not in the other AQS file.
y* .....
"I? AQS File Compare
Select 2 AQS Data Files to Compare
File 1
I I Browse ...
File 2
I Browse ...
Compare
Save
Results
Close
-------
Tool #2: Compare Airsites TSV (Tab-Separated Values) files: This tool determines which
monitor locations are contained in one TSV file but not in the other TSV file.
airsites,tsv - Notepad
File Edit Format View Help
010010001
-86.
472778
+32.437222
6
010010002
-86.
44 3611
+32.428333
6
010010003
-86.
791667
+32.332222
6
010030002
-87.
706944
+30.552222
6
010030003
-87.
713611
+30.555278
6
010030010
-87.
881389
+30.497778
6
010050001
-85.
135278
+31.891111
6
010050002
-85.
6062 34
+31.664137
6
010070001
-87.
233056
+32.903889
6
010090001
-86.
464444
+ 33.956944
6
010090002
-86.
473056
+ 33. 949444
6
010150001
-85.
824444
+33.656667
6
010150002
-85.
837778
+33.685556
6
010210001
-86.
637778
+32.839444
6
010210002
-86.
628889
+32.839444
6
0102 30001
-88.
051944
+32.209167
6
010230002
-88.
053611
+32.205556
6
0102 50001
-87.
781944
+31.966389
6
010250002
-87.
773333
+31.694167
6
010270001
-85.
802222
+33.281111
6
010310002
-87.
259167
+31.056667
6
010330001
-87.
886667
+ 34.736944
6
010330007
-87.
665000
+34.760000
6
010330008
-87.
64 3056
+34.760833
6
0
-------
Tool #3: Create ESRI GIS Shape File: This tool converts T-SpACE model-generated surface
files to the ESRI GIS format for input into other software.
~ Lx|
Files to be created:
Shape File:
.shp
Indeac File:
.shx
dBase File:
dbf
Projection File:
ฆPti
Create Shape File
Create ESRI GIS Compatible Shape File
Please Choose T-SpACE Surface File:
Please Specify Output File Name {No Director/ or Extension):
-------
Tool #4: Create Google Earth File: This tool converts T-SpACE model-generated surface files
to the proper format for visualization in Google Earth.
ฆ Create Google-Earth Compatible KM I Files
-ln|x||
Please Choose T-SpACE Surface File:
I 1
1
Please Specif1/ Output File Name (No Director/ or Extension):
I '
File to be created (KM L files are archived to a single KMZfile):
KMZ File: .kmz
Create KML Files |
-------
Tool #5: Visualize Surface File: This tool launches the AQVis application, which is the same
utility that is used in T-SpACE Model Step 6.
w
Run AqVis
jd.
dl
1
3lease Choose T-SpACE Surface File:
1
a
Run AqVis
-------
Tool #6: Average 4th Highest Surface Files: This tool averages three consecutive annual 4th
highest daily maximum 8-hour average ozone (O3) concentration surface files into a single
(separate) 4th highest surface file.
dlj Average 4th Highest Surface Files
~EM
Select 3 Fourth-Highest Surface Files to Average
File 1
File 2
File 3
Average File (Output)
Average
Browse
Browse
Browse
Browse
Clos
CMAQ Example of 4th highest surface - Note: This graph is not provided in T-SpACE.
-------
Tool #7: Convert CAMx File to CMAQ Format: This tool converts a 'raw' CAMx surface
file to the CMAQ (NetCDF) format, allowing it to be used in T-SpACE.
Please Choose CAMx Surface File:
Convert CAMx to CMAQ-Format
Please Specify CMAQ-Formatted Output File Name:
Hd
Select Year:
3
Convert File
-------
SCIENCE
------- |