WAG AN Emphron Report

Oxides of Nitrogen and Particulate
Monitoring at Wagerup Refinery
An Analysis of Influences Detectable
in Yarloop
Mervyn Thomas
[email protected]
Emphron Informatics Pty Ltd
Report Number 2005.14
www.emphron.com
April 4, 2005
Contents
List of Figures
3
List of Tables
5
Glossary
6
Executive Summary
8
1
Background
10
2
Oxides of Nitrogen and Airborne Particulate Data
12
2.1
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
2.2
Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . .
13
2.3
Monthly Variation in Wind Direction, Oxides of Nitrogen and Airborne Particulates . . . . . . . . . . . . . . . . . . . . . . . . . .
17
Joint Analysis of Meteorological Conditions . . . . . . . . . . . .
17
2.4.1
Analysis of Oxides of Nitrogen . . . . . . . . . . . . . .
21
2.4.2
Analysis of Particulate Concentrations . . . . . . . . . . .
26
2.5
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
2.6
Technical Appendix . . . . . . . . . . . . . . . . . . . . . . . . .
27
2.4
3
Alkalinity of Airborne Particles
31
1
4
5
3.1
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
3.2
Estimation of Peak Concentrations . . . . . . . . . . . . . . . . .
31
3.3
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
Environmental Measurements and Community Complaints
34
4.1
Objectives and Background . . . . . . . . . . . . . . . . . . . . .
34
4.2
Descriptive Statistics for Complaints Data . . . . . . . . . . . . .
35
4.3
Overview of Analysis . . . . . . . . . . . . . . . . . . . . . . . .
37
4.4
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
39
4.5
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
39
4.6
Technical Appendix . . . . . . . . . . . . . . . . . . . . . . . . .
44
Conclusions
47
2
List of Figures
2.1
NOx Distribution at Boundary Road . . . . . . . . . . . . . . . .
16
2.2
Proportion of Day with Wind from North by Month, at Boundary
Road . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
2.3
log Nitric Oxide Concentrations (ppb) by Month, at Boundary Road. 18
2.4
log Nitrogen Dioxide Concentrations (ppb) by Month, at Boundary Road. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
log Total Oxides of Nitrogen Concentrations (ppb) by Month, at
Boundary Road. . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
2.6
log PM10 (µgm−3 ) by Month, at Boundary Road . . . . . . . . .
20
2.7
PM2.5 (µgm−3 ) by Month, at Boundary Road. . . . . . . . . . . .
20
2.8
Effect of Wind Direction on Probability of Peak Nitric Oxide Boundary Rd . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
Effect of Wind Direction on Probability of Peak Nitrogen Dioxide
- Boundary Rd . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
2.10 Effect of Wind Direction on Probability of Peak Nitrogen Oxides
- Boundary Rd . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
2.11 Effect of Wind Direction on Probability of Peak Nitric Oxide Upper Dam . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
24
2.12 Effect of Wind Direction on Probability of Peak Nitrogen Dioxide
- Upper Dam . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2.13 Effect of Wind Direction on Probability of Peak Nitrogen Oxides
- Upper Dam . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2.5
2.9
3
2.14 Effect of Wind Direction on Probability of Peak PM10 at Boundary Rd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27
2.15 Effect of Wind Direction on Probability of Peak PM2.5 at Boundary Rd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
2.16 Effect of Wind Direction on Probability of Peak TSP at the South
RDA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
2.17 Effect of Wind Direction on Probability of Peak TSP at the South
West RDA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
3.1
Peak Daily Estimated NaOH Equivalents by Month Boundary Rd
33
4.1
Distribution of Number of Complaints made by Each Complainant 36
4.2
Incidence of Complaints by Issue Type and Month. . . . . . . . .
40
4.3
Incidence of All Complaints by Oxides of Nitrogen, Airborne Particulates and Meteorological Variables . . . . . . . . . . . . . . .
41
Incidence of Odour Complaints by Oxides of Nitrogen, Airborne
Particulates and Meteorological Variables . . . . . . . . . . . . .
42
Incidence of Health Related Complaints by Oxides of Nitrogen,
Airborne Particulates and Meteorological Variables . . . . . . . .
43
4.4
4.5
4
List of Tables
2.1
Distribution of NO (ppb) Boundary Rd by Month . . . . . . . . .
2.2
Distribution of NO2 (ppb) Boundary Rd by Month . . . . . . . . 14
2.3
Distribution of NOx (ppb) Boundary Rd by Month . . . . . . . .
15
2.4
Distribution of PM10 (µgm−3 ) Boundary Rd by Month . . . . . .
15
2.5
Distribution of PM2.5 (µgm−3 ) Boundary Rd by Month . . . . . .
16
4.1
Number of Complaints by Issue and by Complainant Frequency .
37
4.2
Bootstrap Confidence Intervals for Effect of Oxides of Nitrogen,
Airborne Particulates and Meteorological Variables on Complaints
46
5
14
Glossary
NO
Nitrogen Oxide Concentration (ppb)
NO2
Nitrogen Dioxide Concentration (ppb)
NOx Total Nitrogen Oxides Concentration (ppb): equals NO + NO2
PM10 Airborne particulate matter less than 10 microns in diameter µgm−3
PM2.5 Airborne particulate matter less than 2.5 microns in diameter µgm−3
Blocked Bootstrap A version of the bootstrap re-sampling procedure which ameliorates the effects of serial correlation in the samples.
Bootstrap A non-parametric procedure for estimating confidence intervals or
performing significance tests, used when the theoretical distribution of an
estimator is uncertain or intractable. The bootstrap procedure is computationally intensive, and is based on re-sampling from the data.
Box and Whisker Plot A box and whisker plot provides summary information
about the distribution of measurements. The central box represents the
25th and 75th quantiles. The middle line is the mean of the observations.
The whiskers represent the largest and smallest observations within a cut
off distance of the top and the bottom of the box. The cut off distance is
set as 1.5 times the width of the box. Observations outside the range of the
whiskers are considered outlying observations, and are displayed as points.
Generalised Linear Model A very versatile approach to statistical analysis, which
allows data with non-normal error distributions to be analysed. The approach encompasses many of the common distributional forms of data,
including Poisson models for count data, and Binomial models for success
/ failure data.
6
Kernel Density Estimate A non-parametric smooth estimate of the statistical
distribution of some quantity. A kernel density estimate provides information about the shape of a distribution.
logit
A transformation applied to a probability to map the (0,1) range onto the
p
).
interval (−∞, ∞) The logit function is defined as logit(p) = log( 1−p
lowess An algorithm for fitting a smooth curve to data. It uses a local regression,
in which the line is fitted to points within a moving window about each
data point. The local fit is based on a robust weighted regression.
RDA Residue Drying Area
TSP Total Suspended Particulates µgm−3
7
Executive Summary
This report describes an analysis of Oxides of Nitrogen and airborne particulate
data obtained from the Boundary Rd Yarloop monitoring station, and from air
quality monitoring stations within the Wagerup refinery complex. It seeks to answer the following questions:
1. Is there any objective evidence of refinery operations influencing Oxides of
Nitrogen and airborne particulates in Yarloop?
2. Are peak concentrations of particulate matter or Oxides of Nitrogen sufficient to cause irritant effects in Yarloop? Is it possible that taking hourly or
daily average values masks transient peaks in these parameters, which are
sufficient to cause irritation?
3. Is there any objective evidence of a relationship between refinery operations
and the frequency of complaints?
An analysis of the effect of wind directions on peak concentrations of Oxides of
Nitrogen and particulate matter was performed. The analysis revealed that:
1. Both Oxides of Nitrogen and airborne particulate concentrations tend to be
higher when the wind is blowing from the direction of the refinery complex;
2. Oxides of Nitrogen appear to be elevated when the wind is from the direction of the refinery stacks;
3. Airborne particle concentrations appear to be elevated when the wind is
blowing from the direction of the Residue Drying Area.
4. There is evidence that other sources, not located within the refinery, contribute to the concentrations of Oxides of Nitrogen and airborne particulates. In particular, winds from the direction of Yarloop result in increased
PM2.5 concentrations and increased Nitrogen Oxides.
8
The distribution of PM10 during periods in which the wind blew from the Northern sector was used to estimate the total alkalinity of airborne particles. Even after
making pessimistic (ie upwards) adjustments to the estimated NaOH concentrations there was no evidence of alkalinity sufficient to cause an irritant response.
The incidence of complaints is related to the proportion of each day with a northerly
wind. Complaint incidence also increases with NOx concentrations. Elevated
NOx concentrations probably serve as a marker for instances in which the refinery stack plumes reach ground level. This provides some objective evidence that
complaints are related to refinery operations, and the most likely cause of complaints is odour associated with the plume.
9
Chapter 1
Background
Alcoa operate an Alumina refinery at Wagerup. The refinery is situated about 5
km north of the town of Yarloop. The local community has often complained of
nuisance associated with the site. Reports include unpleasant smells, transitory
irritant effects, and more controversially, the development of “multiple chemical
sensitivity”. Complaints are more common during the winter months.
The Wagerup refinery has an extensive environmental monitoring programme
which includes the following data sources:
• Particulate concentrations of air sampled at various locations within the refinery, and at Boundary Rd, Yarloop;
• Oxides of Nitrogen sampled within the refinery, and at Boundary Rd Yarloop;
• Measurements of total suspended particulate alkalinity at Boundary Rd Yarloop;
• Records of complaints by issue, date and by complainant.
Despite community concerns, there has been little objective evidence to date of
any influence of refinery operations on Oxides of Nitrogen or airborne particulates
in Yarloop. In particular the concentrations of particulates and Oxides of Nitrogen
are well below relevant health standards for hourly averages, daily averages or
annual averages [3, 2]
This report seeks to address several questions:
1. Is there any objective evidence of refinery operations influencing Oxides of
Nitrogen or airborne particulates in Yarloop?
10
2. Are peak concentrations of particulate matter or Oxides of Nitrogen sufficient to cause irritant effects in Yarloop? Is it possible that taking hourly or
daily average values masks transient peaks in these parameters, which are
sufficient to cause irritation?
3. Is there any objective evidence of a relationship between refinery operations
and the frequency of complaints?
The relationship between refinery operations and Oxides of Nitrogen and airborne
particulates was investigated by an analysis of meteorological records from Bancell Rd and Oxides of Nitrogen and airborne particulates records from Boundary
Rd Yarloop, and from sites within the refinery. This analysis is described in Chapter Two.
The potential irritant effect of peak concentrations was evaluated by considering
the distribution of peak particulate concentrations, and the alkalinity of total suspended particulates. This analysis is described in Chapter Three.
The relationship between refinery operations and complaints data was investigated
by a joint analysis of:
1. the history of complaints;
2. meteorological records from Bancell Rd and
3. Oxides of Nitrogen and airborne particulates records from Boundary Rd
Yarloop.
This analysis is described in Chapter Four.
Chapters Two, Three and Four each provide a non–technical overview of the results and the conclusions of the analyses. Chapters Two and Four also have technical appendices, which detail the analyses performed.
Chapter Five provides a summary of the results, and conclusions of the analyses.
Throughout this report, reference to the ‘refinery’ should be understood to mean
the refinery plant, the Residue Drying Areas and the stockpiles. It includes all the
area within the refinery complex.
11
Chapter 2
Oxides of Nitrogen and Airborne
Particulate Data
2.1
Background
There is an automated meteorological station at Bancell Rd, and automated atmospheric measurement stations are located at: Boundary Rd, Yarloop; The Upper
Dam, The South Residue Drying Area (RDA) and the South West RDA. The following measurements are available at the measuring stations:
Bancell Rd Wind Direction (degrees), Wind Speed at 10m and at 3m (km/h),
Temperature (◦ C), Solar Radiation, Barometric Pressure (mBar), Rainfall
and Relative Humidity;
Boundary Rd Yarloop Nitric Oxide concentration (NO), Nitrogen Dioxide concentration (NO2 ), Total Oxides of Nitrogen (NOx ), PM10 and PM2.5 ;
Upper Dam Nitric Oxide concentration (NO), Nitrogen Dioxide concentration
and Total Oxides of Nitrogen (NOx );
South West RDA Total Suspended particulates (TSP);
South RDA Total Suspended particulates (TSP);
All of these variables are available as 6 minute averages. Unless otherwise qualified, any further reference to these variables should be understood to refer to 6
minute average values.
12
Any influence of plant operations on Oxides of Nitrogen or airborne particulates
in Yarloop must involve wind transport. If the plant is a major source of Oxides
of Nitrogen and particulate matter, we might expect to see higher values of NO,
NO2 , NOx , PM10 and PM2.5 during those months in which the wind is often from
the North (since the monitoring station at Boundary Rd is South of the refinery).
2.2
Descriptive Statistics
Analysis of wind direction was restricted to those records containing a full set of
monitoring data - ie the period March 2002 to January 2004 inclusive.
All of the Oxides of Nitrogen and airborne particulates parameters follow a longtailed distribution. That is, most measurements are relatively low, but there are a
few measurements which are many times the average value. These larger measurements are generated by transient peaks in the measured concentration. For
illustrative purposes only, Figure 2.1 shows a histogram of the distribution of natural log NOx concentrations at Boundary Rd. The very long tail of the distribution
is apparent (even after taking logarithms).
Summary statistics (Median, 75th , 95th and 99th percentiles and maximum) of the
concentrations of NO, NO2 , NOx , PM10 and PM2.5 at Boundary Rd are shown
in Tables 2.1, 2.2, 2.3, 2.4 and 2.5 respectively. Each table displays the summary
statistics by month.
13
Month Median 75 Percentile 95 Percentile 99 Percentile Max
January
1.2
3.1
3.7
5.5 21.9
February
1.1
2.3
5.8
16.2 48.2
March
0.0
0.2
0.9
3.6 36.1
April
0.2
0.5
2.0
14.5 46.3
May
0.1
0.3
1.7
4.8 29.7
June
0.0
0.2
1.4
17.9 71.7
July
0.0
0.1
0.8
2.9 20.8
August
0.0
0.1
1.0
5.2 47.0
September
−0.2
0.0
0.4
3.4 53.7
October
0.0
0.2
0.6
2.7 98.2
November
0.1
0.4
0.9
3.2 33.6
December
0.2
0.7
3.4
4.2 30.3
Table 2.1: Distribution of NO (ppb) Boundary Rd by Month. Note the negative
median value for September - this is an artifact of the measurement process for
NO. Very low concentrations of NO may give rise to negative measured values.
Month Median 75 Percentile 95 Percentile 99 Percentile Max
January
1.1
2.0
5.3
9.8 28.5
February
1.6
3.5
9.7
14.8 18.1
March
1.5
2.9
7.1
11.3 18.0
April
2.0
3.4
6.8
14.0 22.2
May
1.8
3.2
6.9
10.8 18.4
June
1.3
2.2
5.7
8.7 17.8
July
1.4
2.3
6.1
9.0 15.4
August
1.1
1.7
5.0
9.0 18.3
September
1.0
1.6
3.6
6.8 23.9
October
1.0
1.8
4.1
8.2 17.0
November
1.3
2.2
4.7
7.8 41.5
December
1.4
2.5
4.9
8.0 17.4
Table 2.2: Distribution of NO2 (ppb) Boundary Rd by Month
14
Month Median 75 Percentile 95 Percentile 99 Percentile Max
January
3.0
4.3
7.5
14.5
34.4
February
3.0
4.6
16.3
26.4
53.1
March
1.5
3.1
8.1
14.6
46.2
April
2.2
3.9
8.8
23.6
51.6
May
1.9
3.6
8.4
14.5
39.8
June
1.4
2.4
7.1
27.3
83.7
July
1.4
2.4
6.7
10.8
33.7
August
1.1
1.9
6.1
14.5
53.4
September
0.7
1.3
3.7
10.2
66.3
October
1.0
1.9
4.5
11.4 108.8
November
1.5
2.5
5.4
10.2
44.9
December
2.0
3.8
6.2
10.9
36.1
Table 2.3: Distribution of NOx (ppb) Boundary Rd by Month
Month Median 75 Percentile 95 Percentile 99 Percentile
Max
January
17.7
25.8
51.2
120.7
676.7
February
17.0
25.0
43.3
94.9
593.6
March
18.2
28.3
55.8
117.6 1071.7
April
10.2
16.7
34.2
62.7
870.3
May
13.2
20.5
47.8
132.9 1342.4
June
9.5
14.3
24.9
50.6
297.6
July
9.7
14.2
22.4
30.7
93.3
August
9.5
14.4
24.8
39.5
404.8
September
10.4
15.6
27.1
47.3
187.5
October
11.2
17.4
40.0
115.6
447.9
November
15.7
24.8
58.4
120.7 1073.7
December
19.6
28.5
59.3
126.2
677.9
Table 2.4: Distribution of PM10 (µgm−3 ) Boundary Rd by Month
15
Month Median 75 Percentile 95 Percentile 99 Percentile
January
6.9
9.7
17.4
29.6
February
7.3
10.0
16.1
29.4
March
7.1
10.3
22.8
37.0
April
5.8
8.9
15.2
25.1
May
7.1
12.0
22.1
40.6
June
5.1
7.2
12.0
17.5
July
5.1
7.1
11.2
17.2
August
5.5
7.6
12.0
18.1
September
5.3
7.5
11.8
16.6
October
5.4
8.0
15.0
26.6
November
6.8
10.7
24.0
52.7
December
7.2
10.3
19.2
37.2
Max
94.2
129.8
149.8
93.5
132.3
50.7
56.7
68.4
65.9
126.0
551.9
137.8
1.0
0.0
0.5
density
1.5
2.0
Table 2.5: Distribution of PM2.5 (µgm−3 ) Boundary Rd by Month
2
3
4
5
6
7
NOx_Boundary_Rd_ppb
Figure 2.1: NOx Distribution at Boundary Road, log ppb. The solid line is a kernel
density estimate of the distribution.
16
2.3
Monthly Variation in Wind Direction, Oxides of
Nitrogen and Airborne Particulates
Figure 2.2 shows box and whisker plots for the proportion of each day in which
the wind is from the North (350◦ to 25◦ ) by month. It is clear that there are
substantial differences between months in wind direction. Throughout the winter
months (May to August) the wind is often from the North; throughout the summer
months the wind is rarely from the North.
Figures 2.3, 2.4 and 2.5 show the mean natural log concentration of NO, NO2 and
NOx by month, respectively. All three measurements show a pronounced reduction during the winter months. That is, considered on a monthly basis, Oxides of
Nitrogen (NO, NO2 and NOx ) are lowest during those seasons when the wind is
blowing from the direction of the plant.
Figures 2.6 and 2.7 show similar plots for the log mean values of PM10 and PM2.5
respectively. Again PM10 and PM2.5 values tend to be lower during the winter
months, when the wind is blowing from the direction of the plant. It is, however,
likely that drier conditions during the summer increase the lift off of dust.
If the effects of wind direction are considered in isolation from the effects of other
meteorological conditions, it appears unlikely that the plant is a major source of
the Oxides of Nitrogen or particulate matter detected at the Boundary Rd Yarloop
monitoring station. These measurements are lowest when the wind is blowing
from the refinery towards the monitoring station.
Wind transport, however, is a complex multi-factorial process; and most of the
determining factors (eg temperature, wind direction, wind speed, humidity) show
marked seasonal trends. It is possible that the effect of wind direction on Oxides
of Nitrogen and airborne particulates is masked by relationships with other seasonally varying parameters. This issue can be resolved only by a joint analysis,
which investigates the simultaneous effects of several meteorological variables.
2.4
Joint Analysis of Meteorological Conditions
Details of this analysis are provided in the technical appendix to this chapter. This
section provides only an informal overview of the approach.
17
0.8
Proportion of Day wind from North by Month
●
0.4
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.2
Proportion of Day
0.6
●
●
●
●
●
●
●
●
●
●
0.0
●
●
●
●
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Month
Figure 2.2: Proportion of Day with Wind from North by Month, at Boundary
Road. The wind is often from the North during the winter months.
●
4.5
●
●
●
●
3.5
●
3.0
4.0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2.5
Log Concentration
●
●
Jan
Feb
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Mar
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Apr
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
May
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Jun
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Jul
Aug
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Sep
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Oct
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Nov
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Dec
Month
Figure 2.3: log Nitric Oxide Concentrations (ppb) by Month, at Boundary Road.
Note the apparent decrease in Nitric Oxide concentration during the winter
months.
18
4.0
●
3.0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Jun
Jul
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Sep
Oct
Nov
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2.5
Log Concentration
3.5
●
●
●
●
●
●
●
●
●
Jan
Feb
Mar
Apr
May
Aug
Dec
Month
Figure 2.4: log Nitrogen Dioxide Concentrations (ppb) by Month, at Boundary
Road. Note the slight decrease in concentration during the winter.
●
●
●
3.5
4.0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
May
Jun
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2.5
3.0
Log Concentration
4.5
●
●
Jan
Feb
Mar
Apr
Jul
Aug
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Sep
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Oct
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Nov
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Dec
Month
Figure 2.5: log Total Oxides of Nitrogen Concentrations (ppb) by Month, at
Boundary Road. Note the decrease in concentration during the winter.
19
●
8
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
4
6
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2
Log Concentration
●
●
Month
Figure 2.6: log PM10 (µgm−3 ) by Month, at Boundary Road. Note the decrease
in concentration during the winter.
●
●
●
8
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Jan
Feb
Mar
Apr
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
4
6
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
2
Log Concentration
●
●
●
Month
Figure 2.7: PM2.5 (µgm−3 ) by Month, at Boundary Road. Note the decrease in
concentration during the winter.
20
Since irritant effects (and therefore the most likely environmental nuisance associated with the refinery) are sensitive to short–term peak concentrations, we analyse
the probability of a peak value of each Oxide of Nitrogen or airbone particulates
parameter, rather than the mean value. A peak value is defined as being in the
top 5% of all values. The rationale for this definition is given in the technical
appendix to the Chapter.
The analysis was based on a generalised linear model for the probability that a 6
minute Oxides of Nitrogen or airborne particulates measurement will be a peak
value. The model included terms for month, time of day, wind speed, temperature
and terms for a smooth periodic function of wind direction (the periodic function
was used to ensure that the estimated effect of a 360◦ wind direction was the
same as the estimated effect of a 0◦ wind direction). The purpose of this model
was to allow an analysis of the joint effect of these meteorological conditions.
Specifically, we were looking to see whether or not there was any effect of wind
direction, after having accounted for the effects of other meteorological variables.
The wind direction terms are then used to display the effect of wind direction on
logit probability of a peak value (the logit transformation stretches out the (0, 1)
interval for a probability to ± infinity). A 95% confidence interval for the wind
direction effect is calculated using a blocked bootstrap procedure. Details are
given in the technical appendix.
2.4.1
Analysis of Oxides of Nitrogen
Figures 2.8, 2.9 and 2.10 show the change in logit probability of a peak value
with wind direction for NO, NO2 and NOx respectively at Boundary Rd. In each
plot the pink-shaded region corresponds to wind directions from 0◦ to 15◦ . These
wind directions encompass the calciners, liquor burner, and powerhouse stacks.
The blue-shaded region (from 300◦ to 345◦ ) contains the RDAs, and the greenshaded region (from 160◦ to 220◦ ) contains most of the residential properties in
Yarloop. The black lines represent the upper and lower 95% bootstrap confidence
intervals for the wind direction effect. The red lines represent the best (see the
technical appendix) estimate of the effect.
The wind direction effect is estimated much more poorly for NO than for NO2
or NOx ; the confidence intervals are much wider, and give little or no information about the wind directions associated with the highest probability of peak NO
values.
21
For NO2 and NOx , however, there is a clearly defined wind direction effect, with
relatively tight confidence intervals. Both of these parameters show a well defined
peak direction associated with the direction of the refinery stacks (calciner, liquor
burner and powerhouse). Both variables also show a peak which is associated
with the direction of residential property in the town of Yarloop, and a peak in the
direction of the RDAs (300◦ to 345◦ ). In both of these figures, there seem to be
several subsidiary peaks in the region 100◦ to 220◦ . These subsidiary peaks are almost certainly an artifact of the modelling process, and represent harmonics in the
fitted periodic function. These subsidiary peaks are poorly defined in comparison
with their confidence intervals.
Nitrogen Oxides measurements are also available from the Upper Dam monitoring
station, within the refinery. This is some five kilometres north of the Boundary
Rd station, and a strong directional signal from this source provides a way of
localising possible point sources of Nitrogen Oxides. Figures 2.11, 2.12 and 2.13
show plots of the wind direction effect on logit probability of a peak value at the
Upper Dam, for Nitrogen Oxide, Nitrogen Dioxide and total Oxides of Nitrogen
respectively.
In these plots, the pink region encompasses the direction of the calciner, liquor
burning and powerhouse stacks (200◦ to 210◦ ), and the green region represents
the direction of most of the Yarloop properties (185◦ to 200◦ ). Again, the black
lines represent the upper and lower 95% confidence intervals, and the red line is
the best estimate of the wind direction effect.
The most striking feature of these graphs is that all show a peak in the region
200◦ to 210◦ . That is, the highest probability of a peak concentration of oxides
of Nitrogen is found when the wind is blowing from the direction of the refinery
stacks. At this location, however, the direction of the town of Yarloop and the
refinery stacks are not well resolved. These plots also suggest elevated Oxides of
Nitrogen concentrations associated with winds from 300◦ to 345◦ .
Although the measurements at the Upper Dam cannot resolve the effect of refinery stacks and the Yarloop properties, when the Upper Dam results are considered
in conjunction with the Boundary Rd results the implication is clear. The best
defined peaks at the Upper Dam and at Boundary Rd both point in the direction of
the refinery stacks. At each site, there also seems to be some elevation in Nitrogen oxide concentrations associated with winds from the residential properties at
Yarloop. Both sites also provide some evidence of elevated concentrations associated with wind directions in the region of 300◦ to 360◦ . This would be consistent
with a distant, or more diffuse source of Oxides of Nitrogen.
22
0.6
0.4
0.2
0.0
−0.2
−0.4
Change in logit Probability of Value > 95th Percentile
−0.6
0
50
100
150
200
250
300
350
Wind Direction
1.5
1.0
0.5
0.0
−0.5
−1.0
Change in logit Probability of Value > 95th Percentile
Figure 2.8: Effect of Wind Direction on Probability of Peak Nitric Oxide - Boundary Rd. Confidence intervals are wide relative to the size of the effect on logit
probability, and any wind direction effect is poorly characterised.
0
50
100
150
200
250
300
350
Wind Direction
Figure 2.9: Effect of Wind Direction on Probability of Peak Nitrogen Dioxide Boundary Rd. The pink region represents winds from the direction of the refinery
stacks. The green region represents winds from the direction of Yarloop properties, and the blue region represents winds from the direction of the RDAs. The
curves suggest a well defined source in the direction of the refinery stacks, and
more diffuse sources in the direction of the RDAs and Yarloop.
23
1.5
1.0
0.5
0.0
−0.5
Change in logit Probability of Value > 95th Percentile
−1.0
0
50
100
150
200
250
300
350
Wind Direction
2
1
0
−1
Change in logit Probability of Value > 95th Percentile
Figure 2.10: Effect of Wind Direction on Probability of Peak Nitrogen Oxides Boundary Rd. The pink region represents winds from the direction of the refinery
stacks. The green region represents winds from the direction of Yarloop properties, and the blue region represents winds from the direction of the RDAs. The
curves suggest a well defined source in the direction of the refinery stacks, and
more diffuse sources in the direction of the RDAs and Yarloop.
0
50
100
150
200
250
300
350
Wind Direction
Figure 2.11: Effect of Wind Direction on Probability of Peak Nitric Oxide - Upper
Dam. The pink region represents winds from the direction of the refinery stacks.
The green region represents winds from the direction of Yarloop properties. The
apparent peak at 300◦ cannot be linked to any potential source on the refinery
complex.
24
1.5
1.0
0.5
0.0
−0.5
Change in logit Probability of Value > 95th Percentile
−1.0
0
50
100
150
200
250
300
350
Wind Direction
2.0
1.5
1.0
0.5
0.0
−0.5
−1.0
−1.5
Change in logit Probability of Value > 95th Percentile
Figure 2.12: Effect of Wind Direction on Probability of Peak Nitrogen Dioxide Upper Dam. The pink region represents winds from the direction of the refinery
stacks. The green region represents winds from the direction of Yarloop properties. The apparent peak at 360◦ cannot be linked to any potential source on the
refinery complex.
0
50
100
150
200
250
300
350
Wind Direction
Figure 2.13: Effect of Wind Direction on Probability of Peak Nitrogen Oxides Upper Dam. The pink region represents winds from the direction of the refinery
stacks. The green region represents winds from the direction of Yarloop properties. The apparent peak at 360◦ cannot be linked to any potential source on the
refinery complex.
25
2.4.2
Analysis of Particulate Concentrations
Figures 2.14, 2.15, 2.16 and 2.17 show wind direction profiles for PM10 and
PM2.5 at Boundary Rd, and for Total Suspended Particulates at the South and
South West RDAs, respectively.
At Boundary Rd PM10 and PM2.5 values show a very similar wind direction profile. Peak values of both parameters are associated with winds in the direction
275◦ to 360◦ . These directions include both the RDAs and the refinery stacks
(calciner, liquor burning and powerhouse). In addition, the PM2.5 values show
elevated probabilities of a peak value when the wind is in the direction of 200◦
to 275◦ . The Upper Dam site did not have records of PM10 or PM2.5 , and it is
therefore not possible to triangulate these directions to identify a geographical
source.
The wind profile for TSP at the south RDA is very variable, but there does seem
to be a peak at 350◦ ; this is consistent with winds passing over the RDA to the
monitoring station. At the South West RDA the wind profile is variable, but the
lowest probability of a peak value occurs with the wind at 250◦ - when it is blowing from the monitoring station towards the RDA. The RDA monitoring sites are
too close to the RDA to provide effective triangulation with the monitoring site at
Boundary Rd.
It seems likely that the peak values of PM10 and PM2.5 at Boundary Rd are associated with wind from the direction of the refinery. It is not possible to localise
this effect within the plant. The peak is consistent with a source from the RDAs,
from the calciner stacks, from a combination of the two or indeed from any source
in that direction.
2.5
Conclusions
Transport of Oxides of Nitrogen is a multifactorial process. When all available
meteorological variables are taken into account there is evidence of an increased
tendency towards peak concentrations in Yarloop when the wind is blowing from
the refinery stacks (liquor burning, calciner and powerhouse). This provides objective evidence that refinery operations produce detectable increases in Oxides of
Nitrogen concentration.
26
0.5
0.0
−0.5
−1.0
Change in logit Probability of Value > 95th Percentile
0
50
100
150
200
250
300
350
Wind Direction
Figure 2.14: Effect of Wind Direction on Probability of Peak PM10 at Boundary
Rd. The pink region represents winds from the direction of the refinery stacks.
The green region represents winds from the direction of Yarloop properties, and
the blue region represents winds from the direction of the RDAs. Winds from the
direction of the refinery complex are associated with increased PM10 concentrations.
There is also an increased tendency towards peak concentrations of Oxides of
Nitrogen when the wind is blowing from the direction of residential properties in
Yarloop, and from the directions 345◦ to 360◦ .
The observed peak concentrations of Oxides of Nitrogen are substantially below
those that are likely to be irritant [3, 2].
The probability of peak values of PM10 occurring is greatest when wind is coming
from the direction of the refinery - including the RDAs and stacks. The probability of peak values of PM2.5 occurring is greatest when wind is coming from the
direction of the refinery – including the RDAs and stacks. The probability of peak
values of PM2.5 occurring when wind is coming from the direction of Yarloop is
not substantially less, however.
2.6
Technical Appendix
Analysis of wind direction was restricted to those records containing a full set of
monitoring data - ie the period March 2002 to January 2004 inclusive.
27
0.5
0.0
−0.5
−1.0
Change in logit Probability of Value > 95th Percentile
−1.5
0
50
100
150
200
250
300
350
Wind Direction
−0.5
0.0
effect
0.5
1.0
Figure 2.15: Effect of Wind Direction on Probability of Peak PM2.5 at Boundary
Rd. The pink region represents winds from the direction of the refinery stacks.
The green region represents winds from the direction of Yarloop properties, and
the blue region represents winds from the direction of the RDAs. Winds from the
direction of the refinery complex are associated with increased PM2.5 concentrations. There is some evidence of elevated PM2.5 associated with winds from the
direction of Yarloop.
0
50
100
150
200
250
300
350
Wind Direction
Figure 2.16: Effect of Wind Direction on Probability of Peak TSP at the South
RDA. The confidence intervals are wide, and any wind direction effect is poorly
defined. There does seem to be a peak in the direction of 360◦ .
28
0.5
0.0
effect
−0.5
−1.0
0
50
100
150
200
250
300
350
Wind Direction
Figure 2.17: Effect of Wind Direction on Probability of Peak TSP at the South
West RDA.The confidence intervals are wide, and any wind direction effect is
poorly defined. There does seem to be a peak in the direction of 360◦ .
Daily and hourly averages of Oxides of Nitrogen and particle concentrations in
Yarloop are well below the established health limits for these variables [2]. Therefore, if there are any irritant effects of Oxides of Nitrogen or airborne particulates
they are likely to be caused by transient peak values in these variables. We therefore analyse the concentrations in terms of the probability of obtaining a peak
value. In these analyses a “peak” value of any concentration is defined as being
any value within the top 5% of all values. This definition represents a balance
between the need to identify effects on rare peak observations, and the decrease
in power (increase in variability) of the analysis as more extreme definitions of
“peak” values are adopted.
The analysis was based on a generalised linear model for the probability that each
six minute period represented a “peak” value of each Oxides of Nitrogen or airborne particulates parameter. This analysis looked at the effect of wind direction
simultaneously with the effect of other environmental conditions (vide infra). A
binary variable was constructed for each Oxides of Nitrogen or airborne particulates parameter - set “TRUE” if the parameter was in the top 5% of values. The
binary variable was analysed using a Generalised Linear Model [8], The analysis
was undertaken with a logit link, and a Binomial error model.
29
Wind direction data are circular. A wind direction of 360◦ is very similar to a wind
direction of 1◦ . The model therefore incorporates wind direction effects using a
Fourier series model of the form:
W =
k
X
i=1
αi sin(
2πix
2πix
) + βi cos(
)
360
360
Where W is the effect of wind direction x◦ , k is the number of Fourier terms
in the model, and α1...k and β1...k are parameters to be estimated from the data.
The Fourier model is a periodic function of wind direction, ie, it is constrained
to predict the same wind effect for 360◦ as for 0◦ . The number of Fourier terms
k was set as 12. A twelfth order series provides a flexible model for the wind
direction effect, and is readily estimated given the number of observations. The
value was limited to twelve because of the computational cost of the bootstrap
analysis (vide infra).
Other terms in the model for mean pollutant concentration included month of the
year (a factor with 12 levels), hour of the day (a factor with 24 levels), temperature
at 2 metres and at 10 metres, relative humidity, solar radiation, wind velocity,
barometric pressure and rainfall.
The coefficients α1...k and β1...k of the fitted model were extracted, and used to
identify the contribution of each angle of wind direction to the mean concentration
of the substance concerned.
Given the serial correlation between successive observations, confidence intervals
were calculated using a bootstrap procedure [6]. Serial correlation between observations was accommodated using a blocked bootstrap [5], with blocks set to
240 six minute observations (ie one day). Choice of block size is problematic; the
analyses were repeated with varying block sizes, and the results were very similar
for blocksizes of 480. For each analysis, 1,000 bootstrap simulations were performed, and used to generate point-wise 95% confidence intervals for the impact
of each degree of wind direction on the probability of pollutant levels.
30
Chapter 3
Alkalinity of Airborne Particles
3.1
Background
In a study conducted between 26th August and 2nd September 2002, Morawska
et al [9] measured the particle concentration at Boundary Rd, when the wind was
blowing from the Northern quadrant (315◦ to 45◦ ). The accumulated PM10 particle samples were titrated for alkalinity, and used to estimate an equivalent concentration of Sodium Hydroxide - which was found to be 0.255 ± 0.007µgm−3 ,
for the PM10 fraction. This concentration represents a weekly average - and peak
concentrations could be substantially greater.
3.2
Estimation of Peak Concentrations
During the period concerned, the mean PM10 concentration for wind directions
from the Northern Quadrant was 11.67µgm−3 . If we assume that transient equivalent concentrations of NaOH are directly proportional to the PM10 concentration
whilst the wind is blowing from the Northern quadrant, then we may estimate the
NaOH equivalent concentration at any value of PM10 as follows:
N=
Nmean P
Pmean
31
Where N is the estimated alkalinity (equivalent mols of NaOH), P is the value of
PM10 for which alkalinity is to be estimated, Pmean is the average PM10 whilst
the wind is from the Northern quadrant during the period 26th August 2002, to
2nd September 2002 (ie 0.255) and Nmean is the average alkalinity during the
period 26th August 2002, to 2nd September 2002. This expression, of course, is
relevant only for those values of PM10 measured whilst the wind is blowing from
the Northern quadrant.
Using this expression, we may estimate the peak NaOH equivalents each day,
from the peak Northern quadrant PM10 values each day. The peak estimate, however, is likely to be too low. This is because:
• The Morawska experiment captured particlate matter when the wind was
from the Northern quadrant - but the likely source of alkaline particles
should be over a angle of less than 45◦ . This will dilute the alkalinity estimate.
• The finest time resolution available is 6 minutes. It is possible that peak
values within this sampling interval are substantially greater than average
values.
If we double the estimated peak values for each of these factors, we have a four
fold increase in estimated NaOH equivalents.
Using this procedure, we calculate the estimated peak NaOH concentration each
day from the peak PM10 when the wind is in the Northern quadrant. The overall
maximum estimated NaOH concentration is 93µgm−3 . Figure 3.1 shows the daily
maximum estimated NaOH equivalent concentration plotted by month.
3.3
Conclusions
This concentration peak estimated NaOH equivalent concentration of 93µgm−3 is
substantially below the Threshold Limit Value (TLV) for NaOH of 2000µgm−3
[1]. It is therefore unlikely to produce irritant responses. Furthermore, the daily
estimated peak concentration is lowest during the winter months when there are
most complaints (see below). Daily peak concentrations during the winter months
are generally below 20µgm−3 , - around 1% of the TLV.
32
60
40
●
●
●
●
●
●
●
20
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
7
8
9
0
Daily Maximum Estimated NaOH
80
●
1
2
3
4
5
6
10
11
12
Month
Figure 3.1: Peak Daily Estimated NaOH Equivalents by Month Boundary Rd
On the basis of conservative estimates of peak alkalinity, (ie estimates which are
likely to over estimate alkalinity), there is no evidence of particles with alkalinity
sufficient to cause an irritant response. It is considered unlikely that complaints
are generated by a irritant response to airborne alkaline particles.
33
Chapter 4
Environmental Measurements
and Community Complaints
4.1
Objectives and Background
This chapter describes a study undertaken to investigate the relationship between
meteorological factors, Oxides of Nitrogen, airborne particulates and the incidence of complaints. In particular, the study seeks to establish whether or not there
is any relationship between wind direction, particle concentration, Oxides of Nitrogen and incidence of complaints. Such a relationship would provide objective
evidence of an association between refinery outputs and community complaints.
The nature of the relationship might also suggest possible mitigation strategies.
Wind direction is of interest because of the observed high frequency of winds from
the North during the winter months. Winds from this direction cross the refinery
before reaching the town of Yarloop - and complaints are more frequent in the
winter months (see below).
Particle concentrations are potentially interesting because particles originating
from the calciner stacks or the RDAs may be alkaline [9], and particles may make
a direct contribution to dust.
Nitrogen Oxide (NOx ) concentration is present at concentrations far too low to
generate irritant responses [2]; but it may act as a marker for the refinery plumes.
That is, peak measurements of NOx may indicate that a plume has reached ground
level near to the measuring station.
34
4.2
Descriptive Statistics for Complaints Data
Complaints records are available from 24th April 2000 to 18th September 2004.
Comprehensive Oxides of Nitrogen and airborne particulates data are only available for the period March 2002 to January 2004. Analysis of complaints which do
not refer to the relationship with air quality are for the entire period of complaints
data; but analyses relating complaints to air quality are restricted to the period for
which Oxides of Nitrogen and airborne particulates data exist.
Complaints are classified by the issue type. Only two issue types are sufficiently
common to warrant separate analysis: Health issues and Odour issues. Individual complainants have been assigned unique identifiers; this allows an analysis
of the distribution of the number of complaints made by each complainant. No
individual complainants are named in this report.
From the period 24th April 2000 to 18th September 2004 there were 3124 complaints, lodged by 250 complainants. The distribution of the number of complaints made by each complainant is shown in Figure 4.1. The mean number of
complaints per complainer was 12.5, and the variance 895.1. If each complainant
were equally likely to complain, and each complaint was independent, one would
expect the distribution of the number of complaints per complainant to follow a
truncated Poisson distribution. From elementary probability theory, the variance
to mean ratio for a truncated Poisson distribution is less than 1 for all values of
the mean. The observed data are therefore strongly inconsistent with a truncated
Poisson distribution. We may conclude that either:
1. All complainants are not equally likely to complain; or
2. Complaints are not independent (ie once a complaint has been made, a complainant is more likely to complain again); or
3. both 1. and 2. above.
35
200
150
100
0
50
Frequency
0
50
100
150
200
250
300
Number of Complaints Per Complainant
Figure 4.1: Distribution of Number of Complaints made by Each Complainant
36
More than half of the complaints were made by the 16 individuals who complained
on 50 or more occasions. For the purposes of discussion, we categorise individuals who have made more than 50 complaints as high frequency complainants, and
other individuals as low frequency complainants. The threshold of 50 complaints
was chosen so as to split the total complaints evenly between “high” and “low”
frequency complainants. Table 4.1 summarises the number of complaints by issue type (all issues, odour and health) and complainant type (all complainants,
high and low frequency complainants). All issue types show a similar pattern
- a relatively few individuals generate a large proportion of the total number of
complaints.
It is possible that high frequency and low frequency complainants are responding to different environmental, social or personal circumstances. To investigate
this potential difference, analyses were repeated for complaints made by low frequency complainants (those 234 complainants who complained on 50 or less occasions) and for complaints made by high frequency complainants (those 16 complainants who lodged more than 50 complaints).
All
Issue
(250 Complainants)
Total
3124
Odour
2687
Health
376
Complainant Type
High Frequency
(16 Complainants)
1665
1482
169
Low Frequency
(234 Complainants)
1459
1205
207
Table 4.1: Number of Complaints by Issue and by Complainant Frequency.
4.3
Overview of Analysis
Box and whisker plots for the number of complaints each day were plotted against
month.
For each complaint type, the natural log of the number of complaints each day
(plus 1 to avoid the problem with the log of zero) was plotted against each of the
meteorological, Oxides of Nitrogen and airborne particulates summary variables.
Given the low number of complaints per day, the relationships are noisy. The
underlying trend was illustrated by fitting a robust trend line using the lowess [4]
line (with span of 0.5).
37
A generalised linear model was used to investigate the relationship between daily
incidence of each complaint type, and meteorological, Oxides of Nitrogen and
airborne particulates conditions. Details of the generalised linear model are given
in the technical appendix to this chapter. Confidence intervals for each parameter
were constructed using a blocked bootstrap procedure [6, 5].
For each day, the following meteorological, Oxides of Nitrogen and airborne particulates summaries were constructed:
Wind.N The proportion of the day with the wind from the North (ie wind direction ≥ 350◦ and wind direction ≤ 25◦ ).
Wind.NW The proportion of the day with the wind from the North West (ie wind
direction ≥ 300◦ and wind direction ≤ 350◦ ).
PM10 .p The proportion of the day with an elevated PM10 measurement (where
elevated is defined as being above the 95th percentile).
PM2.5 .p The proportion of the day with an elevated PM2.5 measurement (where
elevated is defined as being above the 95th percentile).
NOx.p The proportion of the day with an elevated NOx measurement (where
elevated is defined as being above the 95th percentile).
Temp The daily mean temperature at a height of 2m.
Most of the quantities used in this analysis show pronounced differences between
months. The environmental variables are therefore quite strongly correlated with
each other. It is not possible to produce an un-ambiguous analysis of the effects of
each environmental variable - since the strength of relationship calculated for any
one environmental variable depends on which other variables are included in the
model. The results produced in this analysis allow an assessment of the effects of
each environmental variable when all of the other variables have been included in
the model.
Analyses were repeated using a factor to represent month (ie using 11 degrees
of freedom to fit the yearly pattern of response). Results were broadly similar
to those obtained without using month - but the confidence intervals were wider.
These results are not presented, since they provide little further information. It
seems unlikely that unknown seasonally variable environmental factors are masking the effect of the environmental variables which have been included in the
model.
38
4.4
Results
Figure 4.2 shows box and whisker plots for the daily incidence of each complaint
type. There are clear increases in total complaints and odour-related complaints
during the winter months, and there is some suggestion that health-related complaints increase during the winter. There is a fairly close agreement between the
monthly incidence of complaints, and the monthly proportion of each day with a
northerly wind (see Figure 2.2).
Figures 4.3, 4.4 and 4.5 show plots of log complaints incidence vs ambient measurements for total complaints, odour-related complaints and health-related complaints respectively. The red line on each plot is the estimated lowess fit to the
data [4].
There does seem to be evidence of a relationship between the incidence of complaints and the proportion of each day with a northerly or north westerly wind. The
relationship is apparent for total complaints and for odour-related complaints, but
the incidence of health-related complaints is too low to draw conclusions.
The bootstrap confidence intervals from the generalised linear model are presented
in Table 4.2. These results are obtained with a bootstrap block size of 7 days, but
the results were essentially unchanged as the block size varied from 4 to 14. For
total complaints, low frequency complainants show a statistically significant increase with PM2.5 , NOx and wind direction, and a just statistically significant
decrease with temperature; High frequency complainants however have a statistically significant relationship only for wind.
For odour complaints, low frequency complainants are statistically significantly
related to wind direction and NOx , but high frequency complainants are statistically significantly related only to wind direction.
Health-related complaints are statistically significantly related to wind direction,
but not to the other parameters, except for NOx for low frequency complainants.
4.5
Conclusions
Overall there is some evidence that the incidence of complaints is related to objective, measurable features of ambient conditions. It is not possible, however, to
rule out complaints being driven by other variables which are not available in this
study, but which show strong seasonal trends.
39
Odour Complaints
10
10 12 14
Total Complaints
●
●
8
●
●
●
●
●
6
●
●
●
4
count
8
6
count
●
●
●
4
●
●
●
●
●
●
●
●
●
2
●
2
●
●
0
0
●
01
03
05
07
09
11
01
Month
03
05
07
09
11
Month
3.0
Health Complaints
●
●
1.5
●
●
●
01
03
●
●
●
●
●
0.0
0.5
1.0
count
2.0
2.5
●
05
07
09
11
Month
Figure 4.2: Incidence of Complaints by Issue Type and Month. Total complaints
are dominated by odour issues, and both total complaints and odour issue complaints peak during the winter months. There is some suggestion that healthrelated complaints also increase during the winter months.
40
2.0
Count
0.0
1.0
2.0
1.0
0.0
Count
0.0
0.2
0.4
0.6
0.8
0.0
0.6
2.0
Count
0.0
1.0
2.0
1.0
0.0
0.1
0.2
0.3
0.4
0.0
0.2
0.3
0.4
0.5
0.6
2.0
Count
Proportion of Day with Peak PM2.5
0.0
0.0
1.0
0.1
1.0
2.0
Proportion of Day with Peak PM10
Count
0.4
Proportion of Day with North Westerly Wind
0.0
Count
Proportion of Day with Northerly Wind
0.2
0.0
0.1
0.2
0.3
0.4
0.5
5
Proportion of Day with Peak NOx
10
15
20
25
30
Mean Temperature
Figure 4.3: Incidence of All Complaints by Oxides of Nitrogen, Airborne Particulates and Meteorological Variables. The red line is a lowess fit. There is evidence
of an increase in complaint frequency with the proportion of the day having a
northerly wind and a north westerly wind.
41
2.0
0.0
1.0
Count
2.0
1.0
0.0
Count
0.0
0.2
0.4
0.6
0.8
0.0
0.6
2.0
1.0
Count
0.0
1.0
0.0
Count
0.0
0.1
0.2
0.3
0.4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
1.0
0.0
0.0
1.0
Count
2.0
Proportion of Day with Peak PM2.5
2.0
Proportion of Day with Peak PM10
Count
0.4
Proportion of Day with North Westerly Wind
2.0
Proportion of Day with Northerly Wind
0.2
0.0
0.1
0.2
0.3
0.4
0.5
5
Proportion of Day with Peak NOx
10
15
20
25
30
Mean Temperature
Figure 4.4: Incidence of Odour Complaints by Oxides of Nitrogen, Airborne Particulates and Meteorological Variables. The red line is a lowess fit. There is evidence of an increase in complaint frequency with the proportion of the day having
a northerly wind and a north westerly wind.
42
1.2
0.0
0.6
Count
1.2
0.6
0.0
Count
0.0
0.2
0.4
0.6
0.8
0.0
0.6
1.2
0.6
Count
0.0
0.6
0.0
Count
0.0
0.1
0.2
0.3
0.4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.6
0.0
0.0
0.6
Count
1.2
Proportion of Day with Peak PM2.5
1.2
Proportion of Day with Peak PM10
Count
0.4
Proportion of Day with North Westerly Wind
1.2
Proportion of Day with Northerly Wind
0.2
0.0
0.1
0.2
0.3
0.4
0.5
5
Proportion of Day with Peak NOx
10
15
20
25
30
Mean Temperature
Figure 4.5: Incidence of Health Related Complaints by Oxides of Nitrogen, Airborne Particulates and Meteorological Variables. The red line is a lowess fit. The
incidence is too low to provide much evidence for any relationship with environmental measurements.
43
Total complaints and odour-related complaints show a near identical pattern of
response - since odour complaints make up the majority of complaints. There
is evidence that complaints are increased when the wind is often from the direction of the refinery (the North and North West). There is also evidence that the
complaints are increased when much of the day shows elevated NOx measurements and elevated PM2.5 measurements. NOx is unlikely to be the causative
agent for odour-related complaints; but periods with increased incidence of elevated NOx concentrations are likely to be periods with increased impact from the
stack plumes. Peak PM2.5 levels may be associated with airborne material from
the stacks, from the RDA and possibly from Yarloop.
These results are not definitive, but provide some support for the hypothesis that
for low frequency complainants, odour-related complaints (and therefore the total
number of complaints) are influenced by airborne material from the refinery complex – either from the stacks or from the RDAs. For high frequency complainants,
the lack of a statistically significant relationship with complaint incidence and
Oxides of Nitrogen and airborne particulates may suggest that other factors are
involved.
4.6
Technical Appendix
The incidence of each complaint type was analysed using a generalised linear
model [7] with a Poisson error distribution and a logistic link. The model included all the environmental variables described above. Maximum likelihood estimates of all the linear model coefficients were obtained. There is one important
caveat with this analysis. The generalised linear model statistics are based on the
assumption that counts on successive days are independent - conditional on the
values of the predictor (meteorological and ambient) variables. This assumption
may be incorrect for two reasons:• The complaints data follow a contagious distribution - once a complaint has
been made, there is a transitory lower threshold for further complaints. In
this way, a single event may trigger not one, but a series of complaints over
several days.
• There are underlying environmental or social processes not included in the
analysis, but which show strong autocorrelation over time.
44
For this reason, standard errors for linear model parameters were obtained using
a blocked bootstrap procedure [5], to ameliorate the effects of serial correlation.
The analysis was repeated with the block size varying from 4 to 14 days. Results
were stable across the range of block sizes, and results for a block size of 7 are
presented.
Interpretation is complicated by the strong correlations between the various seasonally variable quantities. The approach used provides a test of the hypothesis
that each variable has no relationship with complaint incidence after having allowed for the effect of all other variables.
45
46
Environmental
Variable
Temp
Wind.N
Wind.NW
PM10
PM25
NOx
Temp
Wind.N
Wind.NW
PM10
PM25
NOx
Low
−0.10
1.74
0.37
−5.65
−0.31
−1.13
−0.11
1.66
−1.17
−2.53
0.43
0.71
Confidence Interval by Issue Type:
Total
Odour
Health
Est
High
Low
Est High
Low
Est
High
−0.04
0.01 −0.09 −0.04 0.02
−0.23 −0.08
0.05
3.00
3.94
1.51
2.74 3.73
2.94
5.02
6.91
1.65
3.05
0.57
1.77 2.97 −4.66
0.86
3.90
−1.27
1.65 −4.83 −1.03 1.71 −23.18 −4.70
6.51
1.58
5.73 −0.21
1.43 5.61 −6.56
3.60 14.49
0.50
2.56 −1.11
0.55 2.33 −5.79 −0.02
3.79
−0.06 −0.01 −0.12 −0.06 0.00
−0.13 −0.06 −0.01
2.51
3.36
1.50
2.51 3.48
0.90
2.56
3.70
0.22
1.55 −1.07
0.48 1.95 −3.30 −0.48
1.79
−0.02
2.45 −4.31 −0.88 2.32
−2.17
1.06
4.26
1.97
3.49 −0.53
1.73 4.01 −0.57
2.40
4.36
1.82
3.73
0.01
1.49 3.46
0.51
2.42
5.27
Table 4.2: Bootstrap confidence intervals for generalised linear model coefficients for environmental variables. The analysis
is from a blocked bootstrap with a block size of 7 days. The table shows confidence intervals for each environmental variable,
complaint issue and complainant type. Parameters in which the upper (High) and lower (Low) confidence intervals have the
same sign are statistically significant. The point estimates (best supported values) for the parameters are given in the column
headed “Est”
low
Complainant
Type
high
Chapter 5
Conclusions
It is clear that ambient Oxides of Nitrogen and airborne particulates parameters are
strongly influenced by wind direction, and that a wind direction from the refinery
increases both particulate and Oxides of Nitrogen concentrations at Boundary Rd
Yarloop. Although this is an observational study, not a designed experiment, it
does provide some support for the hypothesis that refinery operations generate
elevations in oxides of nitrogen and particulate matter. That conclusion cannot be
definitive without further experimental work.
Any elevations in concentrations of Oxides of Nitrogen seem to be of little health
or amenity significance. The main interest in these data is that they support the
approach used to localise (approximately) the source of emissions. Peak Oxides
of Nitrogen concentration are well below the levels associated with irritant effects
[2].
The particulate concentrations might be of greater potential significance. We have
established that wind directions from the refinery are associated with higher particulate concentrations (allowing for other variables such as temperature). This
lends some support to the hypothesis that the refinery contributes to ambient particulate matter at Boundary Rd. Furthermore, these wind directions occur much
more frequently during the winter months, when complaints are more frequent.
When the PM10 data and wind direction are used to estimate likely NaOH concentrations, however, we see no overall increase in likely NaOH concentrations during the winter months. Furthermore, even using pessimistic assumptions about
peak NaOH equivalent concentrations, particle alkalinity is unlikely to be sufficient to cause an irritant response.
47
In summary, it is possible that complaints are increased by airborne material from
the refinery. The source within the refinery cannot be localised. There is no
evidence that complaints are due to an irritant response to alkaline particles.
The analyses do not establish the relative importance of the RDAs or the stacks
(liquor burning, calciner and powerhouse) as sources of particulate material.
Complaints do seem to be more common when the wind is blowing from the
North, and they may be increased when there are elevated Oxides of Nitrogen concentrations. These elevated Oxides of Nitrogen concentrations are far too small
to be of physiological significance, but they may serve as a marker for the stack
plumes. In other words, days experiencing a higher proportion of time with peak
NOx levels are likely to be days in which the stack plumes ground in Yarloop.
Plume odour is the most probable cause of complaints (and indeed odour is the
most common issue for complaints).
48
Bibliography
[1] Documentation of the Threshold Limit Values (TLVs) for Sodium Hydroxide.,
Tech. report, American Conference of Governmental Hygienists, 2002.
[2] The National Environmental Protection (Ambient Air Quality) Variation Measure, Gazette S190 (2003).
[3] Alcoa Wagerup Environmental Department, Wagerup Ambient Air Quality
Monitoring Report, Tech. report, Alcoa Australia, December 2003.
[4] W. S. Cleveland, Robust locally weighted regression and smoothing scatterplots, J. Amer. Statist. Assoc (1979).
[5] A.C. Davison and D.V. Hinkley, Bootstrap methods and their application,
Wiley Series in Probability and Statistics, Cambridge University Press, 1997.
[6] B. Efron, The jackknife, the bootstrap and other resampling plans, Society for
Industrial and Applied Mathematics, 1982.
[7] P. McCullagh and J.A. Nelder, Generalised linear models, Chapman and Hall,
London, 1989.
[8] Charles E McCulloch and Shayle R Searle, Generalised, linear, and mixed
models, Wiley Series in Probability and Statistics, Wiley, 2000.
[9] Morawska, Alkalinity of airborne particulate matter, Tech. report, Queensland University of Technology, 2002.
49