This provides an update of a backward air trajectory analysis project conducted for the OTAG Air Quality Analysis Workgroup by VT DEC. Previous status reports have summarized results of HY-SPLIT back trajectories calculated 4 times/day, June - August, 1989 - 1995 for 23 ozone monitoring sites distributed throughout the OTAG domain. The current report responds to suggestions and questions raised by other AQA workgroup members in the following areas: effects of "seasonality" on trajectory results; effects of sub-regional grouping on multi-site trajectory results; and relative influence from intra-regional and inter- regional transport. Also summarized here are results of inter-site and inter-regional correlations among afternoon ozone concentrations at the 23 trajectory sites, and areas of convergence among these and other OTAG AQA studies.
Several commenters have asked about possible influences of seasonal variation in ozone levels on the trajectory results. Conceivably, there could be concurrent seasonal variations in ozone concentrations and synoptic scale meteorological flows, such that spatial differences between "clean and dirty" trajectories might be more reflective of seasonal patterns than of source-receptor relationships. Rao et al. (JAWMA 45: 57-61, 1995); Porter et al. (http://capita.wustl.edu/otag/reports/StatChar/otagrep.htm) and co-workers have developed methods (KZ filter) for statistical decomposition of ozone (and other) data which allows the seasonal component to be separated from the synoptic scale and long-term trend components. Although our trajectory sorting did not utilize the KZ filter to remove seasonal influence, we did attempt to minimize the influence of seasonal variation in ozone by limiting our "window of view" to only the summer season (June through August). In addition, we have "standardized" the ozone data from different sites and different hours of day by converting the raw ozone data to "Z-scores"
(C = MG * SGZ, where C is concentration,
MG is geometric mean and SG is standard
geometric deviation). Figure 1 shows the strong seasonal variation
for a few years of daily maximum ozone at Whiteface Mtn. NY. Figure
2 is based on our Summer-only Z-scores (1989-1995) for the Whiteface
site. Clearly, the seasonal variation is minimized when only Summer
months are considered.
Given the large resource requirements for re-sorting the more than 50,000 individual trajectories in this analysis, and the minimal degree of seasonal influence on the summer-only Z-scores, it is neither feasible nor warranted to re-examine all results with seasonality removed (i.e. with KZ filter). However, the potential influence of the (small) degree of seasonal variation on the trajectory results can be further evaluated by examining trajectory results for the Whiteface site (for which Rao et al. have kindly provided a file of their hourly "short-term" data - with seasonal and long-term trend components removed). Figure 3 displays incremental residence-time probabilities for upper 25th percentile ozone concentrations at Whiteface Mtn. for the past 7 summers (See Poirot and Wishinski 11/7/96 OTAG AQA status report). In the left hand side of Figure 3, trajectories were sorted by our original Z-score metric. On the right, the trajectories were sorted using the KZ-filtered short-term component.
Clearly, the differences (influences of seasonality on Summer
trajectory results) are very minor. This "convergence"
between different approaches also lends indirect support to the
work of Rao, Porter and co-workers, as their short-term component
is specifically intended to represent synoptic-scale influences.
The trajectory-based assessment is also intended to identify synoptic-scale
influences. The results of these two very different approaches
are quite consistent with each other.
Back to Contents
Site Selection and Site Grouping
The trajectory analysis has been conducted in an iterative manner,
beginning with a selection of 6 high-elevation ozone monitoring
sites, and gradually expanding to a total of 23 sites based on
recommendations from AQA workgroup members. The first 6 sites
(predominantly located along the spine of the Appalachian Mountains)
were chosen initially because they were relatively remote, and
because they exhibited minimal diurnal variation. This allowed
combination of ozone concentrations and trajectories without regard
for the time of day. As the analysis was expanded to include lower-elevation
sites, alterative ozone metrics (Z-Scores) were applied to "standardize"
the ozone data at different hours-of-day and at different sites.
Site selection criteria included: regional representativeness,
spatial distribution throughout the OTAG domain, and reasonably
complete data coverage over the Summers of 1989-95. All sites
except the roof of the World Trade Center in NYC were either "rural"
or suburban". Two sites (Grafton, WI and Seaford, DE were
moved slightly during the 7-year period. About one third of the
sites are in counties currently designated as non-attainment,
and all of them have experienced at least occasional hourly concentrations
of 120 ppb or higher. Site locations are displayed in Figure 4
and summarized in Table 1.
In previously reports, trajectories from multiple sites have been
combined in an attempt to condense information, provide a degree
of "triangulation", and show "regional" influences.
For example, incremental residence-time probabilities for upper
5 percentile ozone concentrations for selected site groups are
displayed in Figure 5. While certain of these site groupings help
provide a useful degree of triangulation (ie. identify locations
upwind when ozone is high at a number of widely scattered sites),
they are not necessarily logical groupings from a "regional
influence" perspective (particularly for the "high-elevation"
and "Southern" site groups.
Back to Contents
Site Selection for OTAG "Subregions"
In an attempt to improve on the regional relevance of the site
groupings, we have regrouped selected sites into 6 smaller subregions
- generally described as: Southeast, Southwest, Northwest, Midwest,
New England, and Mid-Atlantic (Table 2 and Figure 6). This subregional
grouping was based on geographical proximity and also on inter-site
ozone correlations (among 3 PM Z-scores). Within each subregion,
intersite correlations ranged from moderate (r = 0.3) to high
(r = .8). We also calculated a mean 3 pm Z-score based on all
sites in a subregion and considered each individual site's correlation
to the sub-regional mean. These correlations ranged from 0.7 to
0.9, indicating that the sites are appropriately grouped (for
sample size of about 650, r > 0.1 is significant at .01 level).
Distances between subregions (calculated as average distance between
all pairs of sites in each pair of subregions) and inter-regional
ozone correlations (from 3 PM Z-Scores) are displayed in Table
3. The New England and Mid-Atlantic subregions are relatively
close together (about 300 miles) and exhibit a moderate inter-regional
correlation (about 0.4). The Midwest and Northwest are similarly
close together and also exhibit a moderate inter-regional correlation.
All other Subregions are separated by at least 400 miles and exhibit
inter-regional correlations of less than 0.3 (consistent with
the more comprehensive evaluation of inter-site correlation (of
short-term component) by Rao et al. and Porter et al.).
|De Soto, MS (0.79)||Boone Co., KY (0.86)||Port Huron, MI (0.88)|
|Greg Co., TX (0.80)||Fortville, IN (0.80)||Grafton, WI (0.88)|
|Iberville Par., LA (0.82)||Nilwood, IL (0.79)|
|Bennington, VT (0.83)||Look Rock, TN (0.87)||Ancora, NJ (0.92)|
|Mr. Greylock, MA (0.87)||Granville Co., NC (0.72)||Seaford, DE (0.82)|
|Port Clyde, ME (0.80)||Lithia Spr., GA (0.81)||World Trade, NY (0.81)|
|Rye, NH (0.83)|
Figure 6. Locations of Ozone/Trajectory Sites included in 6
Back to Contents
Figure 7. Areas of greatest Local Residence-Time Influence
for OTAG "Sub-regions" as defined by
Grouped Trajectory Sites
Various "local" characteristics of these 6 selected
subregions are displayed in Figures 7 - 10. In Figure 7, we attempt
to define "areas of greatest local importance" for each
subregion. An "everyday residence-time probability field"
is calculated for each subregion. Then, grid squares are identified
which have a higher residence-time probability for one subregion
than for any other subregion (see Figure 7). For example, areas
shaded light blue in Figure 7 represent grid squares for which
the residence-time probabilities are higher for the grouped Mid
Atlantic sites (Ancora, NJ, World Trade Ctr., NY, and Seaford,
DE) than for any other subregional group. It is emphasized that
this exercise is not intended to objectively define logical geographic
boundaries of subregions in OTAG, but rather to show - as a function
of preselected site groups and trajectory results - those areas
most probabalistically upwind of each predefined subregion. These
areas might be considered as representing regions of unique local
importance for each subregion.
These unique local areas appear generally logical, with the notable
exception of the "Southwest", which has higher everyday
probabilities extending surprisingly far to the North and East
of the Southwestern sites. This is likely due to inclusion in
the Southwest group of sites extending further South than any
in the Southeast group, and further West than any sites in the
Midwest Group. Also, a few grid squares with highest probabilities
in the Northwest region appear to be geographically misplaced.
Figure 8 provides an indication of the "seasonality"
associated with the ozone Z-scores in each subregion. In each
plot, the 3 PM data for all 7 summers are sorted from June 1st
on the left through August 31 on the right (ie, the 1st 7 data
points are all June 1 - 1989 - 1995). Figure 9 displays the incremental
residence-time probabilities for upper 50th percentile ozone for
Figure 8. 'Seasonality' in 3 PM Ozone Z-Scores from Six Sub-Regions: June - August, 1989-95
Figure 9. Incremental Residence-time Probabilities for Upper
50% Ozone for Six Subregions
The ozone Z-Scores (see page 1) have the unique statistical
characteristics of an (arithmetic) mean of 0 and standard deviation
of 1. Assuming an approximately log-normal distribution to the
original ozone data, a Z value of zero is approximately equal
to the median ozone concentration, 68% of the values will fall
within Z = + 1, and 95 % of the values within Z = +
2. Note the occasional very low (negative) Z-scores for some subregions
in Figure 8. There is no equivalent Z value for a concentration
of zero (Z approaches negative infinity as concentration approaches
zero). The large negative Z-scores represent very low ozone concentrations,
and most likely occur on rainy days. The large annual seasonal
variation in raw ozone data displayed in Figure 1 is minimized
in Figure 8 when only summer data is considered. The strongest
(but not very strong) seasonal variations are evident in the Mid-Atlantic
region (upper right and the Southwest (Gulf Coast) region (lower
right). The seasonal pattern in these subregions is most evident
in the lowest concentrations (and is not apparent in the higher
concentrations - above Z = 0). The Mid-Atlantic (Coast) region
exhibits lower (negative Z's) concentrations in the early and
latter periods of the (June - August) Summer season, while the
Southwest (Gulf Coast) region exhibits lowest concentrations (highest
negative Z's) in mid-Summer. This mid-summer dip in concentrations
along the Gulf Coast is consistent with, and described in additional
detail in Porter et al.
Back to Contents
Incremental Residence-Time Probabilities (Distribution Among Subregions)
Incremental residence-time probabilities for upper 50th percentile
ozone concentrations for each sub-region are displayed in Figure
9. Using the trajectory-based definition of "areas of greatest
local influence (Figure 7), the percentage of incremental probability
associated with each region's "local influence" area
can be determined. These percentages range from a low of about
14% in New England to a high of about 57% in the Midwest (see
Figure 10. Percent of Upper 50% Incremental Probability in
The full distribution of incremental residence time probabilities
for upper 50% ozone concentrations among the 6 subregions is summarized
in Table 4. This is based on the areal definition of subregions
displayed in Figure 7. For the New England region for example,
14% of its incremental probability is located within the pink
area labeled "New England" in Figure 7, and 43.2% New
England's incremental probability is located within the light
blue area labeled "Mid-Atlantic". It is emphasized that
this distribution of incremental probabilities is in no way intended
to represent a quantitative estimate of inter-regional ozone contributions.
Rather, the objective is to provide a more quantitative estimate
of the graphic patterns displayed in Figure 9. Percentages of
> than 15% are highlighted for emphasis.
Table 4. Distribution of Upper 50% Incremental Residence-Time Probability among Subregions
Back to Contents
Duration of Ozone Episodes (Subregional Ozone Auto-Correlations)
The site grouping of subregions described in Figures 6, 7 and
8 and Tables 2 and 3 was based on geographical proximity and inter-site
correlation. In Figure 11, the 3 PM ozone Z-scores in each subregion
are correlated with the 3 PM Z-scores in the same subregion with
lag-times of 1, 2 and 3 days (forward or backward - it doesn't
matter when a region is correlated with itself). This provides
an indication of the duration of ozone episodes in each subregion.
With a 1-day lag-time, all regions exhibit moderate to high (r
= 0.4 to 0.7) auto-correlations (with this sample size of about
650, an r value of about 0.1 is significant at the 0.01 level).
After 2 days, autocorrelations in the New England, Mid-Atlantic
and Northwest subregions become insignificant, while significant
correlations persist even after 3 days for the Southwest and Southeast
subregions. Clearly, the duration of ozone episodes increases
(from 1 to 3+ days) as one moves from Northeast to Southwest in
the OTAG region.
Figure 11. Lagged Auto-Correlations for OTAG Subregions (3
PM Ozone, 1989-95)
Back to Contents
Lagged Inter-Regional Correlations
In Figure 12, average 3 PM ozone Z-scores from each subregion
are correlated with the 3 PM Z-Scores from other subregions on
the same day, and 1 and 2 days earlier. For example, afternoon
ozone in the Northwest region (Figure 12a) has positive,
significant correlations (r > 0.1 for n = 644) with ozone measured
in the Midwest and Southwest 1 day earlier, and with the Southwest
2 days earlier. New England ozone (Figure 12b) correlates
with ozone measured 1 day earlier in the Mid-Atlantic, Northwest
and Midwest regions 1 day earlier, and also with Northwest and
Midwest ozone measured 2 days earlier. Note also that the Southwest
and Southeast subregions - which exhibited the highest lagged
auto-correlations (Figure 11) - generally exhibit the poorest
inter-correlations with other subregions. Significant, positive,
lagged correlations from Figure 12 are summarized in Table 5.
Note: Color indicates Regions where lagged correlation
exceeds same day Correlation
It should be cautioned that although the seasonality in the Summer-only
subregional Z-scores is relatively minor, some degree of seasonality
remains, and may influence these inter-regional correlations
(Ideally, this potential influence might be eliminated by
applying the KZ filter). Also, it should be noted that near-by
(and some more distant) subregions exhibit significant correlations
on the same day (without lag-time). Ideally, the potential influences
of this "auto-correlation" should be removed (by
someone more clever than us). Presumably, the lagged inter-regional
correlations might improve if the data were stratified by wind
direction and wind speed. In this case the correlations are presented
independent of meteorology to provide an independent test on the
trajectory results. Despite these limitations, the results of
the inter-regional correlations are very consistent with the back
trajectory results (Figure 9 and Table 4). The comparative results
of the Inter-regional incremental residence-time probabilities
and the inter-regional lagged correlations are summarized in Table
with Northwest (Compared to Other Regions 0, 1 and 2 Days Earlier)
with New England (Compared to Other Regions 0, 1 and 2 Days Earlier)
with Midwest (Compared to Other Regions 0, 1 and 2 Days Earlier)
with Mid-Atlantic (Compared to Other Regions 0, 1 and 2 Days Earlier)
with Southwest (Compared to Other Regions 0, 1 and 2 Days Earlier)
with Southeast (Compared to Other Regions 0, 1 and 2 Days Earlier)
Table 6. Comparative Results of "Inter-Regional Transport" as derived from:
A. Trajectory Residence Time Analysis and B. Lagged Inter-Regional Ozone Correlations
A. Greater than 15% of upper 50% ozone incremental probability in "source region"
B. Significant correlation with ozone measured 1 or 2 days
earlier in "source region"
Back to Contents
1. Degree of Seasonality is small in Summer-only ozone data measured
at trajectory sites.
2. Seasonality has no discernable influence on trajectory residence-time
3. Trajectory sites in close geographical proximity (within 350
miles) exhibit moderate to high (r = 0.3 to 0.8) same-day ozone
correlations, and can be logically grouped into 6 geographic OTAG
"subregions" described as: New England, Mid-Atlantic,
Northwest, Midwest, Southwest and Southeast.
4. Each subregion has a unique area of local importance. With
the exceptions of the Northwest and New England subregions, the
highest fraction of a receptor region's high ozone incremental
residence-time probability is contained within the region's "local
5. Time-lagged auto-correlations within each subregion indicate
that the average duration of ozone events is shortest (1 to 2
days) in the Northwest, New England and Mid-Atlantic Regions,
and longest (2 to 3 days) in the Southwest and Southeast.
6. Neither residence-time analysis nor time-lagged inter-correlations suggest substantial inter-regional transport contributions between the following sub-regions: to NW from NE, MA or SE: to MW from NE or MA; to MA from SW; to NE from SE or SW; to SW from NW, NE or MA; to SE from NE or MA.
7. Both residence-time analysis and time-lagged inter-correlations suggest substantial inter-regional transport contributions between the following sub-regions: to NW from MW; to MW from SW and SE; to MA from SE, NW, and MW; to NE from MA, NW, and MW; to SW from SE; to SE from MW.
|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.|