Back to Contents
This summarizes past reports, current status, and work in progress
pertaining to applications of backward trajectory analyses conducted
for the OTAG Air Quality Analysis Workgroup by VT DEC. The approach
employs the NOAA HY-SPLIT model to calculate several thousand
backward air trajectories (4/day for June-August, 1989-95) for
each of 23 non-urban ozone monitoring sites with reasonably complete
data capture over the past 7 Summers. Spatial characteristics
of this long-term trajectory/ ozone data base are examined through
sorting and aggregation techniques ('residence-time analysis').
The objective is to provide contextual information on 'ozone transport'
by identifying locations which, over the long-term, have upwind
statistical associations with high (or low) downwind ozone concentrations
at ozone monitoring sites throughout the OTAG domain.
Back to Contents
The general methods for trajectory 'residence-time analysis' employed
in the current project have been previously described in: Poirot
and Wishinski (1985) in AWMA Spec. Conf. on Receptor Methods
for Source Apportionment, T.G. Pace, Ed.; Wishinski and Poirot
(1986) in AWMA Spec. Conf. on Visibility Protection: Research
and Policy Aspects, P.S. Bhardwaja, ED; and Poirot and Wishinski
(1986) Atmos.. Environ., 20: 1457-1469. These earlier assessments
were based on the NOAA ARL-ATAD trajectory model (Heffter (1980)
NOAA Tech. Mem. ERL-ARL-81), while the current results are based
on the NOAA HY-SPLIT model (Draxler (1992) NOAA Tech. Mem. ARL-195).
Otherwise, current methods for processing trajectory results are
essentially the same as reported in these earlier studies. 'Residence-time
analysis' involves tracking space/time characteristics of trajectories
on a grid of 1440 80x80 km squares. Resultant trajectories and
pollutant concentrations are sorted, aggregated and plotted in
one of two general ways:
1. Concentration-Based Sorting: Trajectories are first sorted
into subsets based on receptor site pollution concentrations (low,
high, very high, etc.). For each subset, we track the time that
associated trajectories reside over each grid square, and plot
locations characterized by the largest number of hours in each
subset. A standard plotting routine involves bounding the smallest
areas that account for 25%, 50% and 75% of the residence time
hours for a given subset of trajectories (see Figures 5 and ,
for example). These Residence-time probability plots address the
question : "if the concentration at this site was high (or
low or very high), where did the air come from?"
2. Location-Based Sorting: A descriptive statistic (mean, median,
etc.) is calculated for each grid square based on receptor site
concentrations associated with all trajectories arriving at the
receptor which have passed through that square. Calculation of
an average value for a square is weighted by each trajectory's
residence time over the square. Average values have little meaning
for squares traversed by few trajectories (few residence-time
hours, small sample size, etc.), so results are only reported
for squares which exceed a "threshold" minimum number
of residence-time hours (see Figure 7, for example). Resulting
plots display average concentrations at the receptor as a function
of prior airmass location, and address the question: "if
the air has previously been here (or there), what's the average
concentration at this (or that) receptor?"
Back to Contents
The initial phase of the current OTAG project (presented at the
2/96 OTAG meetings, and posted by Wishinski and Poirot (1996)
at http://capita.wustl.edu/otag/Reports/Restime/Restime.html focused
on a group of 6 high elevation ozone sites, located primarily
along the spine of the Appalachian Mountains. The NOAA HY-SPLIT
model (Draxler (1992) NOAA Technical Memorandum ARL-195) was run
in a 'backward' mode using a subset of the NGM (Phillips (1975)
NOAA Tech. Rpt. NWS-22) meteorological data, which covers most
of North America. Initial trajectory starting "elevations"
were input as "pressure-heights", as determined by the
formula: Pressure (mb) = (1000-Height above MSL (m))/10. An alternative
method of selecting a starting elevation (in meters above the
terrain) was not employed because the model's gridded terrain
is substantially lower than the actual elevation for the mountain-top
sites. For lower elevation sites the difference between model
terrain and actual elevation is minimal. Vertical trajectory motion
is determined by the NGM omega vertical velocity fields (as opposed
to an isobaric (pressure-following) or isosigma (terrain-following)
mode, without exercising an option to interpolate meteorological
sub-layers within the model's first sigma layer (based on recommendations
from B. Stunder and R, Draxler, NOAA-ARL).
The maximum trajectory duration is 106 hours (4 days, 10 hours),
although many trajectories are truncated sooner, if they encounter
"holes" in the NGM data or exceed the model's spatial
domain. Additional truncation is imposed by the 'trajectory-tracking"
grid of 1440 80x80 km squares, and by the plotting routines for
concentration-based and location-based residence-time plots (see
discussion of "Trajectory Duration" under "Response
to OTAG Comments"). The net effect of this truncation limits
the average duration of the utilized portion of the trajectories
to a typical range of 2.5 to 3.5 days.
The first 6 high-elevation sites were selected because (with the
exception of the roof of the World Trade Center in NYC) they were
inherently remote from local source influence (ozone concentrations
result predominantly from "transport"). They also exhibit
minimal diurnal variation in ozone concentrations (such that a
"high" (or low) ozone concentration is equally likely
to occur at any hour of the day). This latter characteristic allowed
use of trajectories and associated ozone concentrations at all
hours of the day, without the need to consider whether a "high"
(or low) concentration threshold at 3 AM should be different from
a threshold at 3 PM, etc.
Figure 1. Average 7-Summer Diurnal Cycles at High
Elevation Sites
Because ozone formation requires sunlight, high ozone levels at
night or early morning provide more "relevant" information
on transport than mid-afternoon levels (a measured concentration
at sunrise represents ozone formed at least a day earlier). Results
were not included from the World Trade Ctr., NYC (#360610003)
and Greenbriar Cnty., WV(#540250001) sites because ozone levels
at these sites exhibited significant diurnal variation, and the
absolute ozone concentration at different hours of day were not
"comparable" (i.e. levels in the mid-PM were typically
higher than early AM).
Figure 4. Residence-Time Probability Plot for
All Whiteface Mtn. Trajectories: Summers 1989-95
(25% of trajectory hours in each separately
shaded area)
An example trajectory arriving at Whiteface Mtn., NY and the 1440-square
'trajectory tracking' grid are displayed in Figure 2. The portion
of all trajectories arriving at Whiteface during the past 7 summers
residing over this 1440 k grid are displayed in Figure 3 (note:
a plotting error truncates these trajectories at 100 degrees longitude,
rather than at the 105 degree western edge of the grid).
Figure 3. All Whiteface Mtn. Back Trajectories:
June - August, 1989 - 1995
Figure 2. Example Back Trajectory for Whiteface
Mtn. on 1440 80x80 km square 'Trajectory-Tracking' Grid
Figure 4 is a residence-time plot based on the same 2,120 trajectories
displayed in Figure 3. These trajectories collectively resided
for a total of 153,737 hours over the 1440-square grid. The separately
shaded areas each contain 1/4 (38,434) of the total hours, and
bound the smallest areas accounting for 25% (pink), 50% (grey)
and 75% (tan) of the hours.The least probable 25% of the trajectory
hours are distributed in the unshaded, white area of the map.
This plot provides a trajectory-based answer to the question:
"Where is the Summertime air at Whiteface Mtn. Most likely
to have previously resided?"
Figures 5 and 6are examples of concentration-based sorting for
Whiteface trajectories. 50% of the long-term ozone "dose"
at Whiteface results from ozone concentrations above 51 ppb (and
50% from concentrations below 51 ppb). Figure 5 displays residence-time
probabilities for 'low ozone' trajectories (< 51 ppb) and Figure
6 shows probabilities for 'hi ozone' trajectories (>51 ppb).
Figure 6. Whiteface Mtn., NY Summers 1989-95 Residence-Time
for Ozone > 51 ppb (797 Traj.)
Figure 5. Whiteface Mtn., NY Summers 1989-95
Residence-Time for Ozone < 51 ppb (1323
Traj.)
Figure 7. Average Ozone by Prior Trajectory Location
for (lowest of) 4 Mountaintop sites: Summers 1989-95
Figure 7 displays an example of 'location-based' trajectory sorting. In this case, an average ozone concentration was first calculated for each grid square for all trajectories passing through that square en route to each of 4 mountaintop receptor sites: Whiteface Mtn., NY, Mt Greylock, MA; Shenandoah NP VA; and Look Rock, TN. Thus, for each grid square, there were 4 sets of average ozone concentrations, one for each receptor. In Figure 7, the lowest of the 4 average concentrations is plotted, such that for each separately shaded area, average ozone concentrations are at least as high as the indicated value for all trajectories passing through that square and arriving at any of the 4 mountaintop sites. Green (outer) > 45 ppb; pink > 47.5 ppb; tan > 50 ppb; purple > 52.5 ppb; red (inner) > 55 ppb.
An average grid square value is not displayed for squares traversed
by less than 100 trajectory hours.
Figure 8. Ozone Sites for OTAG Back-Trajectory
Analysis
Back to Contents
While high-elevation monitoring sites provide ideal platforms
from which to observe ozone transport aloft, they are limited
primarily to the Appalachian Mountains and provide minimal representation
of lower elevation exposures or to other areas of the OTAG region.
Based on recommendations from members of the OTAG Air Quality
Analysis workgroup and Ad Hoc Air Trajectory group, similar long-term
trajectory data sets were developed for 17 additional low-elevation
sites distributed throughout the OTAG domain. Locations for the
current total of 23 high and low elevation sites are displayed
in Figure 8 and Table 1.
Code Site Name Latitude Longitude Elev. (m) AIRS Site #
wfmn Whiteface Mtn., NY 44.36 73.90 1480 360310002
mglm Mt. Greylock, MA 42.64 73.17 1140 250034002
wtcn World Trade Ctr., NYC 40.71 74.01 503 360610063
shen Shenandoah NP, VA 38.52 78.44 1073 511130003
grbw Greenbriar County, WV 37.82 80.51 829 540250001
grsm Gt. Sm. Mt. NP, TN 35.63 83.94 793 470090101
benn Bennington, VT 42.90 73.25 216 500030004
ptcl Port Clyde, ME 43.92 69.26 9 230130004
rynh Rye, NH 43.00 70.75 10 330150012
ancr Ancora, NJ 39.67 74.86 35 340071001
seaf Seaford, DE 38.65 75.61 10 100051001(2)
graf Grafton, WI 43.43 87.92 299 550890008(5)
mktw Mark Twain SP, MO 39.47 91.79 213 291370001
nilw Nilwood, IL 39.40 89.81 201 171170002
fort Fortville, IN 39.94 85.84 265 180590003
boon Boone Cnty., KY 38.92 84.85 171 210150003
pthr Port Huron, MI 42.95 82.46 186 261470005
gran Granville Co., NC 36.14 78.77 91 370770001
semi Seminole Co., FL 28.75 81.31 18 121171002
lith Lithia Springs, GA 33.74 84.63 300 130970002
deso De Soto Co., MS 34.83 89.99 117 280330002
iber Iberville Par., LA 30.20 91.10 9 220470002
greg Gregg Co., TX 32.38 94.71 103 481830001
While none of the selected low-elevation sites were urban (all
were rural or suburban), they all exhibit substantial diurnal
variation. See, for example, the Figure 9 comparison of long-term
average diurnal
Figure 9. Average Sumer 1989-95 Diurnal Ozone at Mt. Greylock, MA and Bennington, VT
ozone levels at high-elevation MT. Greylock, MA and the nearby
low-elevation Bennington, VT site. To compare ozone levels and
associated trajectories at different hours of the day from the
low-elevation sites, we first calculated a long-term (7 summer)
mean concentration for each site for each hour of day (3 AM, 9
AM, 3 PM, 9 PM), and then re-expressed each hourly ozone value
as the deviation (in ppb) from the diurnal mean for that hour.
Figure 10 displays the hourly ozone levels at Bennington and
Mt. Greylock during a 3-day period of increasing ozone concentrations
during June, 1991. Concentrations increase at both sites, and
both sites exhibit similar mid-afternoon peak levels. But while
the mountaintop concentrations increase "smoothly",
a strong diurnal pattern is evident at the Bennington site. The
mid-afternoon concentration at Bennington on 6/24 is higher than
the midnight concentration on 6/26. In Figure 11, the data from
both sites have been re-expressed as the deviations from their
diurnal mean concentrations. Now both sites exhibit relatively
smooth increases; their nighttime levels are more comparable;
and the 6/26 midnight value at Bennington is higher (and a positive
deviation) than the mid-afternoon value on 6/24 (a negative deviation).
Re-expressing (standardizing) the hourly data in this manner allows
use of data (and trajectories) from different hours and from different
sites on a more directly comparable basis.
Figure 10. Hourly Ozone at Mt. Greylock , MA
and Bennington, VT on 6/24-26/91
Figure 11. Ozone Deviations from Diurnal Means
at
Mt. Greylock, MA and Bennington, VT on 6/24-26/91
Having "standardized" the ozone concentrations for
each site as deviations from the average (for that site and hour
of day), we next employed "trajectory-based sorting"
for each receptor site to calculate average ozone deviations as
a function of prior trajectory location. That is, for each of
the 1440 grid squares, we calculate an average receptor site ozone
deviation for all trajectories passing over that grid square en
route to the receptor. The average is time-weighted, such that
a trajectory residing for 8 hours over a square is given twice
as much weight as a trajectory residing over the square for 4
hours. Average values are for squares characterized by less than
100 hours of total trajectory time (approximately equivalent to
a minimum of 25 trajectories) are not displayed.
Currently available results using these methods (as presented
at 7/96 OTAG meetings)have been processed into a series of 23
movie animations (one for each receptor site) which plot locations
associated with trajectories resulting in low and high deviations
from the site's mean ozone levels. Each movie has 62 frames; begins
with locations (if any) associated with large negative deviations
(15 ppb below the mean) and ends with locations (if any) associated
with large positive deviations (15 ppb above the mean). Frame
31 of each movie shows all locations with trajectories associated
with negative deviations (below average ozone concentrations)
at the receptor, while frame 32 shows all locations associated
with positive deviations (above average ozone) at the receptor.
Movies are created in .avi format (viewable with CAPITA movie
program or Microsoft Media Player). Movie names follow the form
xxxxdev.avi, where xxxx is the 4-character site code listed in
the first column of Table 1 (benndev.avi for Bennington, VT).
All 23 single-site movies are compressed into the zipfile: 1sitedev.zip
(1.2 Megs compressed; 3.3 Megs uncompressed) and posted on the
OTAG AQA website.
Example frames 31 and 32 from the benndev.avi movie are pasted
below in Figure 12. These show locations upwind of Bennington,
VT for which trajectories are associated with below average (left)
and above average (right) ozone deviations at Bennington.
Figure 12. Locations Associated with Negative
(left) and Positive (right) Ozone deviations at Bennington, VT
Figure 13 a, b, c and d display example frames from 'Average Ozone
Deviation' Movies for 4 sites:Pt. Huron, MI; Rye, NH; Mk. Twain
SP, MO and Gt Smk. Mtn., NP, TN - in diverent sections of the
OTAG domain (from 4sitedev.avi movie).Clockwise from upper left,
frames show locations for which trajectories arriving at specified
receptors are associated with ozone levels: a. Less than Average;
b. Greater than Average; c. At least 5 ppb Greater than Average;
d. At Least 10 ppb Greater than Average. Low ozone concentrations
are associated with areas external to (North, South, East and
West) of the OTAG domain, while high ozone levels are associated
with trajectories internal to OTAG, with areas in the industrial
Midwest being upwind of high ozone deviations at all 4 sites..
| Figure 13a. Locations with
Ozone Deviations of < Average |
Figure 13b. Locations with
Ozone Deviations of > Average |
|---|---|
 |
 |
| Figure 13c. Locations with
Deviations at least 5 ppb > Avg. |
Figure 13d. Locations with
Deviations at least 10 ppb > Avg. |
 |
 |
Back to Contents
Most of the results presented on the preceding pages have been
presented for review at meetings of the OTAG Air Quality Analysis
Workgroup and Modeling and Assessment Subgroup (and have been
posted on the OTAG AQA Website. Various comments and suggestions
have been received and will be addressed in a final report. Following
are some preliminary responses, and an outline of the future plans
for completing this analysis project.
Back to Contents
It has been suggested that we provide some discussion of the uncertainty
inherent in trajectory calculations. A backward trajectory is
a meteorological estimate of the history of airmass motion of
air arriving at a pre-specified location and time. Sources of
uncertainty in a trajectory calculation include the meteorological
data which drives the model and the physical assumptions by which
the trajectory model operates on that data. In this work, we have
applied the NOAA HY-SPLIT (Hybrid Single-particle Lagrangian Integrated
Trajectories) model, driven by meteorological data archived
from the National Meteorological Center's (NMC) Nested Grid Model
(NGM). Draxler has estimated a HY-SPLIT-NGM trajectory error
(after 24 hours) in the range of 20 to 30 % of trajectory distance,
based on model evaluation during the CAPTEX (Draxler (1987) J.
Appl. Meterol. 26:1577-1578) and ANATEX (Draxler (1991) J.
Appl. Meterol. 30:1466-1467) tracer experiments.
At an average wind speed of about 5 meters per second, a trajectory would travel roughly 400 km after 24 hours (and would pass through about 5 of our 80 km grid squares. At this 400 km distance, an error of 20 to 30 % would be in the range of 80 to 120 km, and thus the "actual" trajectory position might be off by 1 or 2 grid squares. At a distance of 1600 km (about 4 days), a 20-30% error might displace the "actual" trajectory position by 320-480 km - equivalent to 4-6 of our 80 km grid squares. As the trajectory error increases with distance, it should also be considered that the potential for a trajectory to reside over any particular location decreases. There are more than 100 80x80 km grid squares at a distance of 400 km from a receptor, and hence a probability of less than 1% that a random trajectory will reside over any particular square at this distance. With large trajectory numbers, the probability that many of them will have erroneously passed over an individual distant square decreases with distance. These potential errors are assumed to apply to any individual trajectory. Unless there is a systematic bias in the trajectory estimates (ie. They always move too fast and always turn too far to the right) we assume that the errors for large numbers of trajectories (in our case more than 2000 for each of 23 receptor locations) tend to be offsetting.
If there are systematic biases in the HY-SPLIT calculations, it
is unlikely that such biases would somehow tend to "force"
the trajectories associated with high (or low) ozone levels to
pass over any specific locations prior to arriving at a large
number of different receptor locations.
Another potential source of trajectory error is the presumption
that a single trajectory represents the path of all the pollution
molecules experienced at the receptor. HY-SPLIT tracks the backward
path of a single "particle" and does not explicitly
incorporate mixing into the trajectory calculations. Under certain
atmospheric conditions the "representativeness" of a
single trajectory pathway is questionable. The CAPITA Monte Carlo
model (Schichtel and Husar (1995) Regional simulation of atmospheric
pollutants with the CAPITA Monte Carlo model. J. Air & Waste
Manage. Assoc. Accepted for publication), simulates effects of
atmospheric mixing through release of multiple particles and application
of vertical and horizontal mixing algorithms which result in multiple
trajectory pathways. Schichtel and Wishinski (1996 ) conducted
a detailed comparison of HY-SPLIT and CAPITA Monte Carlo Results(posted on OTAG Website at http://capita.wustl.edu/otag/Reports/Trajcomp/trajcomp.html).
They found "individual trajectories at times compared very
well and at other times predicted substantially different airmass
pathways". When they applied residence time analysis to three
months of trajectories, they found that "there were no critical
differences between the residence time plots from the HYSPLIT
model and the Monte Carlo model. This is an indication that there
are no systematic differences between the back trajectories from
the two models, and the differences between the individual trajectory
tend to average out when aggregated over "long" time
periods".
It should also be emphasized that the NGM meteorological grid
(composed of 182.9 km squares) is particularly inappropriate for
evaluation of local-scale influences (use of higher resolution
surface winds would be more effective). Also, our trajectory -tracking
grid (80 km squares) is too coarse for evaluation of local-scale
effects. Our 'location-based sorting techniques are also
inappropriate for identifying local influences. The emphasis here
in on identifying influences over larger, synoptic scales.
Back to Contents
Several reviewers have commented on the "excessive length" of our trajectory calculations. The initial trajectory calculations were conducted for a maximum duration of 106 hours (4 days, 10 hours). It should be noted that this is the nominal trajectory duration, and that a majority of the initial trajectories are effectively terminated after shorter durations. This truncation (implemented as a function of distance rather than time) occurs for the portion of the trajectory which exceeds our 'trajectory tracking' grid of 1440 80x80 km squares. The radial distance for this grid is approximately 1000 miles. For our 'concentration-based sorting residence-time plots (see Figures 4, 5, 6), the locations of the outermost 25% of residence time hours (the least probable trajectory locations) are not plotted, imposing a further truncation on the 'effective" length of the trajectories. Figures 3 and 4 (all trajectories for Whiteface Mtn. are initially based on 153,737 hours from 2,210 trajectories (an average of about 72 hours per trajectory). Figure 4 excludes the outer 25% of these trajectories, and consequently represents an average trajectory length of 2 to 3 days.
For our location-based residence-time plots (see Figures 12 and
13), additional truncation of trajectory length is imposed by
exclusion of grid squares traversed by fewer than 100 trajectory
hours. In Figure 5, the plot displays locations of all squares
with greater than 100 trajectory hours (234 squares < average
and 195 squares > average. Thus the utilized portion of trajectories
within the grid is limited to about 30% of the total 1440 square
grid area. When these 100 hour limits are examined for all 23
sites, the plotted areas have a radial distance of about 600 miles,
and reflect average trajectory durations in the range of 2.5 to
3 days. This "effective truncation" is conducted on
the basis of distance, rather than time, and so there are some
very slow-moving trajectories which extend back the full 106 hours
included within all of these plots.
We do not believe there is an objective basis for terminating
a trajectory calculation at any specific length or time. The potential
trajectory error increases with distance, incrementally, and not
suddenly at some specific distance. It is not logical to "trust"
a trajectory completely at 72 hours, but mistrust it completely
at 73 hours. Truncation of a trajectory at a specific time step
does however, have the effect of pre-determining the maximum transport
time. If we truncate them at 12 hours, we have pre-determined
that transport does not occur (or that we will disregard its effects)
over longer time periods.
The effects of arbitrarily trajectory truncation at 106, 88, 70
and 52 hours have been determined for the Mark Twain receptor
site, and are summarized in a series of movies submitted to the
OTAG website. Some example frames from the movie mktw1836.avi
are pasted in Figure 14. From upper left, to lower right, the
frames show (as in figure 13) locations with average ozone deviations
of <average, > average, at least 5 ppb > average, and
at least 10 ppb > 0. In each frame, the 4 panels show, from
upper left to lower right, results when trajectories are truncated
at 106 hours, 88 hours, 70 hours and 52 Hours.
 |
 |
 |
 |
| Figure 14. Effects of Trajectory Truncation
(106, 88, 70, 52 hours) on Average Deviation Plots for Mark Twain SP,MO |
|---|
With each progressively severe truncation, the implicated areas
associated with low and high deviations become slightly smaller.
Truncation clearly has minimal effect on the directional characteristics
of the results. With the exception of the most extreme 52 hour
truncation, the spatial extent and locations of the plotted areas
remain similar. A certain extent of this "shrinkage"
is due to our 100 hour minimum threshold for plotting (This effect
can be seen in the upper 2 frames which together show all locations
with at least 100 trajectory hours. So part of the difference
is due to a reduction of the number, rather than the length of
the trajectories considered in each square. Even when we pre-determine
that ozone transport cannot occur beyond 2 days (or that we will
disregard it) the results are similar. At about 3 days, the results
show very minor differences from our standard procedure based
on an initial maximum 106 hour calculation.
Back to Contents
Figure 15. Comparison of Location Based Sorting
using absolute deviations (left) and geometric deviations (right)
Following presentation of the "average deviation" results
at the July, 96 OTAG meetings, there was some discussion of the
ozone metrics employed for calculation of average ozone deviations
as a function of prior trajectory location. For each site and
each hour of day we calculated an arithmetic mean ozone concentration,
and re-expressed each hourly concentration as an absolute deviation
(in ppb) from that mean. A number of alternative metrics were
discussed including use of "percent deviations", and
use of the geometric mean and standard geometric deviation. These
metrics are primarily intended to provide an objective answer
to the simple questions: "What is a high (or low) ozone concentration?"
and "How high or low is it?"
We have tested several alternative metrics and believe that the
use of geometric mean and standard deviation represents a statistically
superior way to "standardize" the ozone data for different
hours of day and for different sites (although it is more difficult
to explain). For log-normally distributed data (which ozone approximately
is), the mean is approximately equal to the median. Using the
geometric mean as a threshold between "high" and "low"
essentially divides the long-term data set in half. We can also
re-express each hourly concentration as a "Z-score"
by the formula: Concentration = Mg sgZ,
solving for Z. A given Z-score also relates statistically to
a specific percentile of the population. For example, the 98th
percentile concentration is approximately 2 standard deviations
above the mean.
We have subsequently re-expressed the ozone data for each site
and each hour of day as Z scores, and plan to use this metric
in future determinations of "high", "low",
"how high", etc. While the metric and threshold are
different from, and not directly comparable to, the previously
applied "absolute deviation" concept, the effects on
the results appear to be minor. This is illustrated by the movie
ancrcomp.avi (posted on OTAG Website) which compares deviation
plots based on absolute ozone deviations to comparable plots based
on geometric means and standard geometric deviations (see Figure
15). While this alternative metric essentially has no effect on
the patterns for location-based sorting for individual sites,
we believe it provides for a more equitable standardization among
data from different sites - which varies considerably in terms
of absolute concentration. The ancrcomp.avi movie and several
"multsiite" movies (which combine geometric deviations
from multiple receptors are posted on the OTAG Website in the
file geomdev.zip. Example frames from several of the multi-site
aggregations are pasted below. Note that while the spatial areas
are similar in all 4 frames, the Z values are quite different
(0.2 for the Southern sites, 0.3 for the Midwestern sites, 0.4
for the Northeastern sites and 0.5 for the high elevation sites.
 |
 |
 |
 |
| Figure 16. Example Plots from
multi-site Geometric deviation movies for clusters of 6 receptors |
|---|
Conversely, if equal Z values were compared for the different
groups of sites, progressively smaller areas have high deviations
as on moves from Northeast to Midwest to South . High (or low)
ozone levels at the Compared to the Northeast, Southern sites
all appear to be characterized by substantially more ambiguous
"source region influence", and few "source
regions" are appear to contribute persistently to
high (or low) ozone concentrations at multiple sites
in the South. ...to be continued...
Back to Contents
One reviewer suggested that our residence time results may be
influenced by the seasonal variation that occurs between the middle
and ends of the 3-month (June-August) "Summer" period
examined in our 7-Summer data set. We are exploring the separate
potential effects of "seasonality" (and long-term trends),
through application of residence-time methods to different components
of ST Rao's statistically ozone data for the Whiteface Mtn. site.
Rao and co-workers have developed methods to statistically separate
seasonal, long-term trend and short-term synoptic variation from
the raw data. We feel this approach will help evaluate the influence
(if any) of seasonality and trend on our results, and may also
provide a useful perspective on the extensive body of work already
conducted by Rao et al.
Back to Contents
| Submit your comments, feedback, questions, and ideas pertaining this page. Your input will be automatically added to the existing annotations. In order to add a new comment, you must be registered with the OTAG/AQA Peoples Page. |
