00627-
UAPSP-
113
OOOR86106
Comparison of Weekly and Daily
Wet Deposition Sampling Results
UAPSP113
Contract U101-1
Supplemental Report
December 1986
Prepared by
Combustion Engineering Environmental Monitoring and Services, Inc.
Camarillo, California
Keywords:
Acid Precipitation
Precipitation Chemistry
UAPSP
Sampling Frequency
UTILITY ACID PRECIPITATION STUDY PROGRAM
-------
Comparison of Weekly and Daily Wet Deposition
Sampling Results
UAPSP 113
Contract U101-1
Supplemental Report, December 1986
Prepared by
COMBUSTION ENGINEERING
ENVIRONMENTAL MONITORING AND SERVICES, INC.
4565 Calle Quetzal
Camarillo, California 93010
Principal Investigators
L. E. Topol
M. Lev-On
SYSTEMS APPLICATIONS, INC.
101 Lucas Valley Road
San Rafael, California 94903
Principal Investigators
A. K. Pollack
T. J. Permutt
Prepared for
U.S. Environmental Protection Agency
Environmental Monitoring Systems Laboratory
Research Triangle Park, North Carolina 27711
EPA Project Manager
W. J. Mitchell
and
Utility Acid Precipitation Study Program
1111 Nineteenth Street, NW
Washington, DC. 20036
and
Electric Power Research Institute
3412 Hillview Avenue
Palo Alto, California 94304
UAPSP and EPRI Project Managers
p. K. Mueller U.S. Environmental Protection Agency
M. A. Allan GLNPO Library Collection (PI-12JJ
77 West Jackson Boulevard,
Chicago, 1L 60604-3590
-------
ORDERING INFORMATION
Requests for copies of this report should be directed to UAPSP Report Center, P.O. Box 599,
Damascus, Maryland 20872, (301)670-0100. General information on UAPSP maybe
obtained through Robert Hathaway, Ohio Edison Company, 76 South Main Street, Akron,
Ohio 44308, (216) 384-5812. For additional technical information, contact John Jansen,
Southern Company Services, Inc., P.O. Box 2625, Birmingham, Alabama 35202, (205) 877-
7698.
All rights reserved
NOTICE
This report was prepared by the organization(s) named below as an account of work sponsored by the United States
Environmental Protection Agency (USEPA) and by the Utility Acid Precipitation Study Program (UAPSP), which is
administered by the Edison Electric Institute (EEI) and technically managed by the Electric Power Research Institute
(EPRI) Neither USEPA/UAPSP/EEI/EPRI, members of USEPA/UAPSP/EEI/EPRI, the organization(s) named below,
nor any person acting on behalf of any of them (a) makes any warranty, express or implied, with respect to the use
of any information, apparatus, method, or process disclosed in this report or that such use may not infringe privately
owned rights, or (b) assumes any liabilities with respect to the use of, or for damages resulting from the use of, any
information, apparatus, method, or process disclosed in this report
Prepared by
Combustion Engineering Environmental Monitoring and Services, Inc
Camanllo, California
and
Systems Applications, Inc
San Rafael, California
-------
UAPSP PERSPECTIVE
The Utility Acid Precipitation Study Program (UAPSP) was established in 1981 to
ensure that daily precipitation chemistry data of quantified accuracy and precision
would be available. The use of such data would include the evaluation of temporal
and geographic variabilities and trends in relation to emissions. Other
precipitation sampling networks in the United States generally collect either
daily or weekly samples. The choice of sampling frequency has always been a
difficult one often determined by available resources. It has been assumed that
weekly sampling is sufficient for determining long-term trends and ecological
exposures, and that daily sampling is only necessary if one wished to explore
source-to-receptor relationships. Because combining data sets would enhance their
usefulness, it is important to determine the comparability of these two sampling
frequencies.
Although a number of recent studies have compared weekly and daily samples,
interpreting the results is complicated by the use of different equipment,
operators, and laboratories analyzing the samples. Therefore, beginning in
October 1983, the Utility Acid Precipitation Study Program (UAPSP) and the U.S.
Environmental Protection Agency jointly sponsored a collocated field study to
provide a direct assessment of the effect of sampling frequency. Specifically,
the objectives of this study were to determine the efficiency of the collectors,
the precision and composition of daily and weekly samples, and to compare the
concentration and deposition obtained from daily and weekly samples.
This study was designed to overcome previous limitations and focused on the
differences in sampling frequency only. It is unique in several ways. First,
three UAPSP sites (Georgia, Kansas, and Vermont) were selected to represent
different geographical, climatic, and chemical (i.e., rain chemistry) regions.
Second, the study was conducted for one full year. Third, four identical
precipitation samplers (two daily and two weekly) were operated at each site to
determine the precision of both sampling frequencies as well as the comparison
between daily and weekly measured chemistry. Fourth, collocated rain gauges were
11
-------
operated at two sites (Kansas and Vermont) to determine the precision of
precipitation depth and deposition. Fifth, all chemical analyses were performed
in the same laboratory. Finally the concentration data were screened for extreme
values and missing data prior to their analysis. However, a method accounting for
missing daily deposition values remains to be successfully established.
According to the findings, the differences in concentrations reported for weekly
and daily sampling are not large. Nevertheless, detectable biases exist which
deserve attention when weekly and daily data are combined. To what extent such
biases may be exacerbated when other procedural differences among networks are
also considered remains to be studied. The ability to detect annual trends in
precipitation composition is enhanced by daily sampling. Moreover, daily data
appear to be more reliable for estimating differences among seasons. Taken
individually, daily and weekly sampling will provide consistent spatial
distributions but, when combined, may alter or add noise to these distributions.
Therefore, a network that changes its sampling frequency would create a bias that
could interfere with the detection of trends.
This first study of how sampling frequency alone influences precipitation
chemistry data, should be useful to data analysts. In addition, network managers
will now be in a better position to assess the effectiveness of their sampling
protocol in meeting network objectives in comparison with costs.
John J. Jansen
Southern Company Services, Inc.
Chairman, UAPSP Technical Review Committee
IV
-------
LIST OF UAPSP AND EPRI PUBLISHED REPORTS ON PRECIPITATION CHEMISTRY
This report is one of a series of reports on the Utility Acid Precipitation Study
Program (UAPSP). To give the reader an easy reference to other reports on UAPSP
and related EPRI reports, the following list is presented. UAPSP reports can be
obtained through the UAPSP Report Center (see copyright page for further ordering
information), unless indicated otherwise. Copies of the EPRI reports can be
obtained through the EPRI Research Reports Center (RRC), Box 50490, Palo Alto,
CA 94303, (415) 965-4081.
Report Project
Number Number
Title
Pub!ication
Date
UAPSP Reports
100 U101-00
101 U101-90
102 U101-01
103
104
105
106
107
108
U101-01
U101-01
U101-01
U101-01
U101-02
U101-01
Proceedings: Advisory Workshop on Methods
for Comparing Precipitation Chemistry Data
1982 Annual Summary Report
UAPSP Laboratory Standard Operating
Procedures
UAPSP Precipitation Data Displays for
January 1, 1979-June 30, 1982
Volumes 1 & 2
The Utility Acid Precipitation Study Program:
Field Operator Instruction Manual
The Utility Acid Precipitation Study Program:
Network Description and Measurements for
1981 and 1982, Volumes 1 & 2
Plan for Controlling the Quality of Measure-
ments and Data Base During the Utility Acid
Precipitation Study Program (UAPSP)
1982 Quality Assurance Summary Report for
the Utility Acid Precipitation Study Program
Description and Format of Data Base for
EPRI (UAPSP) Acid Precipitation Measurements
February 1983
February 1983
March 1983 (1)
August 1983
January 1983 (1)
November 1986
September 1982
(2)
March 1983 (1)
April 1983 (2)
-------
109
111
112
113
U101-90
U101-04
U101-03
U101-01
UAPSP Second Summary Report
The Influence of Meteorological Factors on
Precipitation Chemistry
Statistical Analysis of Precipitation
Chemistry Measurements over the Eastern
United States
Comparison of Weekly and Daily Wet
Deposition Sampling Results
EPRI Related Reports
EA-1914 1630-01
(vol 2)
EA-1751 1630-02
EA-4663 1630-52
Precipitation Scavenging Chemistry for
Sulfate and Nitrate from the SURE and
Related Data
Precipitation Chemistry Measurements in
the SURE Region
Proceedings: Methods for Acidic
Deposition Measurement
September 1984
November 1986
June 1986
November 1986
February 1983
(in press)
August 1986
(1) Available on request from EPRI Project Manager
(2) Available on request as a working document from the EPRI Project Manager.
Subject to revisions.
-------
ACKNOWLEDGEMENTS
The authors would like to express their appreciation to Peter K. Mueller (Elec-
tric Power Research Institute) and William J. Mitchell (Environmental Protection
Agency) for their guidance and advice throughout this project; to Richard J.
Schwall and R. Vijayakumar (Cumbustion Engineering) for technical assistance; to
Mithra M. Moezzi and Howard P. Beckman (Systems Applications, Inc.) for computing
support and editorial assistance, respectively; to Cynthia S. Hirtzel (Rensselaer
Polytechnic Institute) for technical review; and to the U.S. Military Academy at
West Point for providing emissions data. Several peer reviewers provided con-
structive criticisms, which are much appreciated.
VI1
-------
TABLE OF CONTENTS
SECTION PAGE
1 INTRODUCTION 1-1
2 SAMPLING SITES AND PERIOD 2-1
Site Selection Criteria 2-1
Site Descriptions 2-2
3 MEASUREMENT METHODS 3-1
Site Equipment and Operations 3-1
Laboratory Operations 3-6
4 DATA MANAGEMENT AND STATISTICAL METHODS 4-1
Quality Control Screening 4-1
Data Processing 4-2
Statistical Testing 4-6
5 RESULTS 5-1
Frequency of Rain Events 5-1
Collection Efficiencies 5-3
Comparison of Daily and Weekly Concentration
Measurement Precision 5-5
Derived Versus Measured Weekly Concentrations 5-21
Comparison of Daily and Weekly Deposition 5-39
6 CONCLUSIONS 6-1
APPENDIX A - WEEKLY AND DAILY SAMPLER BIAS SUMMARY
STATISTICS BY PRECIPITATION TYPE
APPENDIX B - WEEKLY MEASURED AND DERIVED ANALYTE
CONCENTRATIONS FOR EACH SITE AND
PRECIPITATION TYPE
APPENDIX C - WEEK-BY-WEEK COMPARISON OF DEPOSITION AMOUNTS
CALCULATED FROM DAILY AND WEEKLY SAMPLERS
FOR MAJOR IONS (H+, SO" NOl, NH* Ca2+)
-
IX
-------
ILLUSTRATIONS
NUMBER PAGE
5-1 Frequency of rain events by week for the
1983 to 1984 study period 5-2
5-2 Distributions of collocated differences for
precipitation mass 5-6
5-3 Distributions of collocated differences for
hydrogen 5-7
5-4 Distributions of collocated differences for
sulfate 5-8
5-5 Distributions of collocated differences for
nitrate 5-9
5-6 Distributions of collocated differences for
ammonium 5-10
5-7 Distributions of collocated differences for
calcium 5-11
5-8 Ratios of weekly to daily pooled standard
deviation from collocated samplers 5-19
5-9 Ratios of weekly to daily median absolute
collocated difference 5-20
5-10 Derived weekly minus measured weekly data for
hydrogen and sulfate ions 5-24
5-11 Derived weekly minus measured weekly data for
nitrate and ammonium ions 5-25
5-12 Derived weekly minus measured weekly data for calcium
ion in mg/1 and precipitation mass 5-26
5-13 Weekly percent relative method bias for hydrogen and sulfate 5-28
5-14 Weekly percent relative method bias for nitrate and ammonium 5-29
5-15 Percent relative method bias 5-31
-------
TABLES
NUMBER PAGE
2-1 Coordinates and elevation of collocation study sites 2-2
2-2 Source emissions inventory for sites 2-4
3-1 EMSI laboratory methods, detection limits, precision
and accuracy for precipitation sample analyses 3-8
5-1 Frequency of precipitation types at each site 5-1
5-2 Average collection efficiency for rain, snow and
mixed (rain & snow) samples at three monitoring sites 5-4
5-3 Daily and weekly concentration sampler bias 5-12
5-4 Measurement system precision for daily and
weekly samples 5-15
5-5 Summary of median and mean site concentrations 5-22
5-6 Method bias for all samples 5-30
5-7 Comparison of site mean and relative concentration
bias for rain and all samples 5-33
5-8 Derived weekly minus measured weekly concentration bias
by season 5-36
5-9 Comparison of mean and relative concentration bias
for rain and all samples by season 5-39
5-10 Rain gauge bias by precipitation type calculated
from daily measurements 5-40
5-11 Rain gauge precision by precipitation type for
daily samples 5-41
5-12 Seasonal differences in rain gauge bias 5-42
xm
-------
NUMBER PAGE
5-13 Daily deposition sampler bias for all
precipitation types combined 5-44
5-14 Daily deposition measurement precision 5-45
5-15 Site annual depositions based on daily
and weekly measurements 5-47
5-16 Seasonal depositions derived from daily and
weekly measurements 5-48
xiv
-------
SUMMARY
The objectives of this study were to evaluate the changes in chemical composition
that may occur when precipitation is stored up to seven days in collectors, and
to assess the significance of the variability of those changes from site to site
and among seasons. In a one-year field program, two pairs of identical wet depo-
sition samplers were deployed at three sites (in Georgia, Kansas and Vermont),
and for each predesignated pair either daily samples or weekly samples were col-
lected. The Kansas and Vermont sites also had a pair of collocated weighing-
bucket rain gauges. The results of this study provide a direct assessment of
differences in sampling schedule since common procedures were used at all three
field sites and all samples were analyzed at the same laboratory.
The data collected were screened and validated prior to data analysis. Samples
were considered invalid if there was obvious contamination. Concentrations were
also missing for low-volume events and when there was a sampler malfunction. In
order to ensure that the results are not overly influenced by extreme values, an
outlier detection and rejection scheme, based on three standard deviations around
the mean of the collocated sampler differences at each site, was utilized. When-
ever possible, nonparametric statistical techniques, which are less sensitive to
outliers than parametric techniques, were used to assess statistical signifi-
cance.
The data obtained were used to determine differences between daily and weekly
sampling for (a) collection efficiency (by reference to the same rain gauge) for
each site and by precipitation type, (b) precision of daily and weekly monitor-
ing, (c) concentration bias, expressed as the relative difference between derived
weekly and measured weekly concentrations, overall and by season, and (d) depo-
sition bias, defined as the difference between the deposition calculated from the
daily measurements and that from the weekly measurements. The major results of
this investigation are given below.
S-l
-------
The collection efficiencies were highest for rain and about
equal for snow and mixed precipitation with a value of about
1.1 for rain and 0.75 for mixed and snow. Generally small
differences were observed between weekly and daily sampling and
among sites for the same precipitation type.
The precision of ionic concentration determinations is either
comparable or better for daily than for weekly samples. Over-
all measurement precision for both sampling protocols (with the
exception of potassium) is less than 20 percent of ionic
concentrations.
The measured weekly concentrations were, in most cases,
approximately 10 percent higher than the weekly concentrations
derived from daily samples.
The differences between weekly and composited daily
concentrations varied by constituents, but they were in general
less than 10 percent of the mean concentration. Also, the
individual ion concentration biases are of similar magnitudes
and direction among the sites with the exception of acidity and
sodium.
There is seasonal variation in the concentration bias between
daily composited and weekly samples. Concentrations of all
ions except ammonium were significantly higher in the fall for
the weekly measurements. Concentrations of calcium and nitrate
were significantly higher in weekly measurements in the
spring. No statistically significant biases were observed for
the summer quarter.
Differences in paired rain gauge depths at the two sites with
collocated rain gauges were greatest for snow samples. Over-
all, the paired rain gauges at the Vermont site measured
significantly different precipitation amounts; the differences
were greatest in the winter quarter. The significant differ-
ences in rain gauge depth resulted in significant differences
in daily deposition for all ions except potassium at the
Vermont site.
Precision of daily depositions (calculated from the two daily
collectors and the two rain gauges) was better at the Kansas
site; for most ions the precision was 5 to 15 percent. At the
Vermont site, where the rain gauge differences were larger, the
precision of ion depositions were 10 to 30 percent (as measured
by relative absolute collocated difference).
Deposition bias, the difference between deposition calculated
from daily and weekly measurements, remains to be examined. A
method must first be established to adjust daily data for
missing values.
Calculated deposition amounts vary by season at each site.
Peak sulfate deposition occurred in the spring at the Kansas
and Georgia sites, and in the summer at the Vermont site. Peak
acidity deposition values occurred in the summer at the Vermont
and Georgia sites, and in the spring at the Kansas site.
S-2
-------
In conclusion, the comparison of measurement errors demonstrated that ionic
concentrations of daily collected samples can be determined with better precision
than those of samples collected weekly for most constituents. Therefore, daily
sample collection provides narrower confidence intervals for concentration means
and allows earlier detection of significant trends in precipitation composition.
In addition, measured weekly concentrations were generally higher than weekly
concentrations derived from daily samples. However, the seasonal variability in
the results obtained from daily versus weekly sampling was not consistent either
in magnitude or direction. This indicates that daily and weekly sample collection
provides consistent spatial distribution results (at least for the three sites
studied), while temporal distributions obtained from weekly samples might be
incompatible with those obtained from daily samples. Therefore, for a network
that changes from daily to weekly sampling, the bias in the results will interfere
with trend analysis. Finally, it must be noted that in this study all field and
laboratory procedures were identical for daily and weekly sample collection.
Different results might be obtained in comparing data from two different networks
where differences in both sampling schedules and field and laboratory procedures
exist.
S-3
-------
Section 1
INTRODUCTION
Precipitation sampling networks in the United States generally collect either
daily (UAPSP, MAP3S) or weekly (NADP/NTN) samples. Under a weekly schedule,
samples can remain in the collector under ambient conditions for up to seven
days, possibly resulting in more chemical changes than might occur if the samples
remain in collectors for at most 24 hours. It is important to know the occur-
rence and magnitude of such changes when comparing chemical composition and
trends calculated from weekly and daily deposition networks.
To determine the importance of any such chemical changes, a collocated sampling
study was performed by Environmental Monitoring and Services, Inc. (EMSI) at
three sites of the Utility Acid Precipitation Study Program (UAPSP) network from
October 1983 to October 1984. The sites--at Uvalda, Georgia; Lancaster, Kansas;
and Underhill, Vermontwere selected to represent the southeastern, central
(west of the Mississippi River), and northeastern regions of the United States.
The objectives of the study were to determine the efficiency of the collectors,
to quantify the precision and composition of daily and weekly samples, and to
compare the concentrations and depositions obtained from weekly and daily sam-
ples. For this study a daily sample is the total precipitation in a 24-hour
period, starting at about 0900 1ST. A weekly sample is the total precipitation
collected in the period from 0900 of the sample pick-up day, generally Monday, to
the same time and day of the following week.
Four studies comparing weekly versus daily (or event) samples have recently been
performed using the same type of samplers as in the present study: at one MAP3S
site at Pennsylvania State University (I), at one NADP site at North Carolina
State University (2), at Argonne National Laboratory (3), and at one Florida site
(4). The Penn State study utilized two samplers, one collecting daily samples
and the other, weekly samples. All analyses were performed by Battelle Pacific
Northwest Laboratory. The weekly samples had lower concentrations and deposition
values for all the ions analyzed, acidity (H+), sulfate (SOp, nitrate (NOI),
1-1
-------
ammonium (Nhlt), calcium (Ca ), and magnesium (Mg ). The study at North
Carolina State involved four daily collectors and six weekly collectors of wet
precipitation. The data from this study were used primarily to compare col-
lection efficiencies as a function of the collector's distance from the rain
gauge. However, the data indicate that gross chemical changes, as evidenced by
pH and specific conductance, did not occur over collection periods of one week,
although the possibility of chemical changes in other analytes cannot be ruled
out by the available data. In the Argonne study two samplers activated by a
single sensor were operated for approximately two years. This approach assured
simultaneous opening and closing of both collectors. The ions measured were the
same as in the Penn State study; the weekly samples were analyzed by the Illinois
State Water Survey, whereas the event samples were analyzed by Argonne. This
study found that weekly samples had significantly less NH^ and H+ in all seasons
and more SO. in every season but summer. Weekly samples had significantly more
Ca and Mg during seasons with little precipitation. No significant differ-
ences between weekly and event NO, were evident. In the Florida study three
samples collected wet precipitation on a daily, weekly, and biweekly schedule for
one year. All analyses were performed at the University of Central Florida with-
in ten days of collection. No statistically significant differences in precipi-
tation composition with sampling interval were found.
Interpreting the results of these and other (older) studies is complicated by the
fact that most studies used different equipment and operators, and different
laboratories doing the measurements. The present study was designed to control
these complicating factors and focus on the differences in sampling period
only. This study is unique in several ways. Three regular UAPSP monitoring
sites were equipped with four identical precipitation samplers, two collecting
daily samples, and two, weekly samples. This allows precision data to be calcu-
lated for both sampling schedules as well as a comparison of the chemical compo-
sition of the daily and weekly samples. All analyses were performed in the same
laboratory. Except for the sampling schedule, all procedures were identical.
Two sites had collocated rain gauges so that the precision of precipitation depth
and deposition could also be determined.
A discussion of the sampling sites and sampling period is given in Section 2.
The site equipment and operations and the laboratory operations are described in
Section 3. Data management, including data processing, outlier screening, and
statistical tests, is covered in Section 4, and the results are presented and
discussed in Section 5. The conclusions of this study are presented in Section 6.
1-2
-------
REFERENCES
1. R. G. de Pena, K. C. Walker, L. Lebowitz, and J. G. Micka. "Wet Deposition
MonitoringEffect of Sampling Period." Atmos. Environ. Vol. 19, pp. 151-
156, 1985.
2. L. J. Schroder, R. A. linthurst, J. E. Ellson, and S. F. Vozzo. "Comparison
of Daily and Weekly Precipitation Sampling Efficiences Using Automatic Col-
lectors." Water, Air, Soil Poll. Vol. 20, pp. 1-11, 1984.
3. D. L. Sisterson, B. E. Wurfel, and B. M. Lesht. "Chemical Differences Be-
tween Event and Weekly Precipitation Samples in Northeastern Illinois."
Atmos. Environ. Vol. 19, pp. 1453-1469, 1985.
4. B. C. Madsen. "An Evaluation of Sampling Interval Length on the Chemical
Composition of Wet-Only Deposition." Atmos. Environ. Vol. 16, pp. 2515-2519,
1982.
1-3
-------
Section 2
SAMPLING SITES AND PERIOD
In this section the monitoring sites are described and their conformance to the
UAPSP site selection criteria is discussed. To demonstrate site conformance to
the criteria, a source emission inventory is included.
SITE SELECTION CRITERIA
Criteria used for site selection for precipitation monitoring are outlined in the
U.S. EPA Quality Assurance Manual for Precipitation Measurement Systems (1.).
Primary criteria adopted for the UAPSP network were the following:
No large point emission or urban sources located within 20 km;
Grassy areas at the sites to minimize splash;
No obstructions, including overhead wires, that might cast rain or wind
shadows on the site and serve as sources of contamination (nonrepresent-
ative data caused by obstructions can be avoided by ensuring that the
distance between any tall object and the collector or rain gauge is at
least twice the difference between the height of the object and the
collector);
Fairly flat land to assure level positioning of equipment and to mini-
mize snow drift problems;
Proximity to the operators' homes or jobs and ready accessibility to
facilitate the required daily visits;
No storage of agricultural products, fuels, or other foreign materials
within 100 m.
Secondary considerations were
Location at least 20 m from all mobile or small local sources, such as
roads, crop fields, or grazing animals, or at least upwind of these
sources; and
2-1
-------
The presence of trees to act as a windbreak.
These siting criteria are similar to those of the MAP3S network (2), the National
Atmospheric Deposition Program (NADP) (3), and the EPRI-SURE study (4). One
exception is that NADP requires a 5 m separation between the collector (1.5 m in
height) and rain gauge; this criterion is more severe than for other obstruc-
tions, for which NADP accepts a separation distance equal to the height of the
object. The 1.5 m spacing between collocated samplers and between samplers and
rain gauges was used in this study. The conformance of the UAPSP sites to the
siting criteria is discussed below.
For this study three sites from the UAPSP network that typified different regions
and climates of the eastern United States were selected. These sites were
Uvalda, GA (Site 14), in the Southeast; Lancaster, KS (Site 18), west of the Mis-
sissippi River; and Underbill, VT (Site 20), in the Northeast. The coordinates
and elevations of the three sites are listed in Table 2-1.
Table 2-1
COORDINATES AND ELEVATION OF COLLOCATION STUDY SITES
Site No. Location Latitude Longitude Elevation (m)
14 Uvalda, GA 32° 03' 18" 82° 28' 25" 64
18 Lancaster, KS 39° 34' 10" 95° 18' 17" 346
20 Underbill, VT 44° 31' 42" 72° 52' 08" 442
SITE DESCRIPTIONS
In order to determine whether the study sites met the source distance criterion,
important point sources within 50 km of each site were identified using the U.S.
EPA's 1977-78 National Emissions Data Systems (NEDS), as updated by MAP3S (5).
The industrial categories for point sources are utilities, mining, paper and
allied products, chemicals and allied products, petroleum refining, concrete and
related products, metal industries, "other" manufacturing, and miscellaneous.
The sources listed in the NEDS inventory include only those that emitted at least
250 tonnes (250,000 kg) per year of SOX (as S02), NOX (as N02), hydrocarbons (as
2-2
-------
methane), or particles in 1977-1978. The sources near the sites, their distance
(km) and heading (angle from the site), and their 1978 emission data are presen-
ted in Table 2-2. In addition, cities with populations greater than approxi-
mately 25,000 within 50 km of the site are listed. No data are available for
soil dust, another important contamination source for precipitation chemistry.
The sites are briefly described below. Features that do not conform to the sel-
ection criteria, as well as important point sources, are noted.
Uvalda, GA (Site 14)
The site is located 75 m south of a pond and 53 m northwest of a grove of pine
trees. Some livestock are nearby (approximately 25 m away). No other sources
are present near the site. Prevailing winds are from the west-southwest.
Lancaster, KS (Site 18)
The site is on pasture land with trees in all directions. No animals or obstruc-
tions are present. The largest sources are power plants, 43 km northeast and 37
km north-northwest. The closest emission source, 5.2 km south, is a grain stor-
age facility and is a large source of particles. However, since the prevailing
winds are from the west, this source should generally have little effect.
Underbill. VT (Site 20)
The site is in a hilly area and is about 19 km east of the town of Burlington,
VT, which has a population of 38,600. The collectors are mounted on a level
wooden platform. The only large source is a paper products plant, 36 km north-
west, which emits hydrocarbons. There are power lines 13 m from the collector,
and these do not conform to the obstruction criterion. This site has been chosen
as a National Trends Network site.
2-3
-------
X 00
M n r-i
x o o o o o
X l» O O O _t
O 41 in *+ n
z c -
C ci **
Q
ss
"
2
o ~
I KX OOOOOOOOOOOOO O3>
7 -i 01 oomoooooooooo <*>
i,, 3« f*1*1 "fl^miNoBi^nrtwr*
> U C - I
I -§
*^ u *j a.
« o
00 £ 5.
o -.
I i O< «
co c
LU
O «
Di
O
III).
O III «*««
c «--i«-<-<-«-<«ixo<**«o«. oo
o -«x««-Hr »<«ii<«ir ^ *e ^ < -< « "u
J; a.uo. U4J ^_.^*u e*u>«O(pa« < TJ » M > M " c
n w aucjc 3 aej<«eue*e«e«e«c-< ""
^4 41 V'O'D4f-*«U'O^3'Q'O-4U'4>^t(a*^<^'<^'^*^H4'^>«4CJ
e e xoos>ooooo>oo
-------
Sampling Period
Collocated sampling was performed at the three sites for approximately one year,
as follows:
Georgia, Oct. 9, 1983 - Aug. 15, 1984
Vermont, Oct. 9, 1983 - Oct. 10, 1984
Kansas, Oct. 12, 1983 - Oct. 1?, 1984
REFERENCES
1. Quality Assurance Handbook for Air Pollution Measurement Systems. Vol. V.
Manual for Precipitation Measurement Systems, Part I. Quality Assurance
Manual. EPA-600/4-82-0422a. Research Triangle Park, NC: U.S. Environmen-
tal Protection Agency, 1982.
2. "The MAP3S/RAINE Precipitation Chemistry Network: Quality Control." Rich-
land, WA: Pacific Northwest Laboratory, 1980.
3. D.S. Bigelow. "Instruction Manual--NADP/NTN Site Selection and Installa-
tion." Fort Collins: Natural Resources Ecology Laboratory, Colorado State
University, 1984.
4. L. E. Topol and R. Schwall. "Precipitation Chemistry Measurements in the
SURE Region." Draft Final Report. Palo Alto, CA: Electric Power Research
Institute, 1986.
5. C.N. Benkovitz. "Compilation of an Inventory of Anthropogenic Emissions in
the United States and Canada." Atmos. Environ. Vol. 16, p. 1551, 1982.
2-5
-------
Section 3
MEASUREMENT METHODS
This section describes the site equipment and the site and laboratory opera-
tions. The site operations include sampling procedure, shipping, and quality
control. The laboratory operations discussed here include sample check-in,
analysis, storage, and quality control, including analytical precision and accu-
racy.
SITE EQUIPMENT AND OPERATIONS
The site equipment and operations are summarized below. A more detailed descrip-
tion is given in the UAPSP "Field Operator Manual" (1) and the UAPSP Interim
Report (2).
Equipment
Each site had the following items:
Four automatic precipitation collectors, manufactured by Aerochem
Metrics (3) (with a peaked snow roof for the Vermont site).
One or two Universal Recording Weighing Bucket Rain Gauges (4) with
an eight-day spring-powered clock and strip chart recorder. The
chart is graduated in inches of rainfall and is wrapped around a
vertical cylinder that rotates. An event pen marker is electri-
cally connected to the samplers and notes the sampler lid opening
and closing times on the strip chart; gauge capacity is 12 inches
(30.5 cm); sensitivity is 0.02 inches (0.05 cm).
A triple beam balance (5); capacity 2610 g, accuracy 0.1 g.
3-1
-------
A digital pH/temperature meter (6) and combination pH electrode (for
rain or low ionic strength solutions) (7); range 0-14 pH, accuracy
0.02 pH.
A thermistor (8); temperature accuracy 0.5°C.
A conductivity meter (9) and dip cell (10), temperature compensated;
range 0-10, 0-100, 0-1000, 0-10,000 and 0-100,000 uS/cm; accuracy 1%
of full scale.
At each site, one pair of samplers was collected daily and the other pair collec-
ted weekly. Two rain gauges were installed at the Kansas and Vermont sites,
while the Georgia site had one gauge.
The samplers were a modified HASL type (11) and are currently manufactured by
Aerochem Metrics; they contain two 3.5-gallon plastic buckets (one for wet
deposition and one for dry deposition), which are inert to inorganic con-
stituents; a common lid, driven by a motor; and a rain sensor. When the sensor
is wet, it activates a motor that moves the lid from its normal position,
uncovering the wet deposition bucket and covering the dry deposition bucket.
When the sensor is dry again, the lid returns over the wet bucket.
The rain gauge has a double traverse pen with a range of 0 to 6 inches on the
first scale and 6 to 12 inches on the second. The gauge is readily calibrated to
±0.02 inch (±0.05 cm) except for an interval around the 6-inch level, where the
calibration is good to only ±0.1 inch (±0.25 cm).
The conductivity meter is calibrated with 75 uS/cm KC1 standard solution. Thus
the calibration range is 0-100 uS/cm, which is used for most of the precipitation
samples.
The useful life of the combination pH electrodes for precipitation samples in the
UAPSP has been three months to one year, with an average of six months. The pH
meter and electrode are calibrated at pH = 4.0 and 7.0 with standard buffer solu-
tions.
3-2
-------
Field Procedures
All operators were trained in site operations based on the U.S. EPA "Operation
and Maintenance Manual for Precipitation Measurement Systems" (12) and the UAPSP
"Field Operator Instruction Manual" (I). Daily site visits were generally made
at about 0900 LSI to check the equipment and remove any precipitation collected
by the daily samplers.
Daily samples were weighed and a portion removed from each for pH and conduc-
tivity measurements. These measurements were made about an hour after sample
removal from the collector to allow for temperature equilibration with the stan-
dard solutions required for meter calibration and a pH quality control check
solution. When the sample was frozen or contained snow, it was allowed to thaw
first, prior to removal of the aliquot for field analysis. If the rest of the
sample after pH and conductivity measurement was greater than 20 g, a 500 ml
sample (or whole sample if less than 500 ml) was transferred to a clean, labeled,
plastic bottle and sealed; the rest of the sample was discarded. If less than
20 g, the sample was discarded and thus was lost from the weekly composite of
daily results. The effect of this procedure on the comparisons with weekly
samples will be discussed in Section 4.
Weekly samples were removed on Monday from the Georgia and Vermont sites and on
Tuesday at the Kansas site. The same procedures were followed with the weekly
samples as with the daily samples, except that the weekly samples were shipped on
the day of pickup and were not refrigerated prior to shipment. However, both
daily and weekly samples were shipped together to the analytical laboratory at
low temperature. If no event occurred within a week, the buckets were rinsed
with deionized water to remove any dust that may have deposited and then
reused. A protocol to accomplish this was implemented throughout the UAPSP
network (.UH). The rinse water was shipped to the central laboratory for
analysis as a dynamic blank.
Each sample was identified by site, sampler, date, and weight. To prevent con-
fusion of the sample identity, since there were four samplers at a site, the
sampler buckets were inscribed with an identification number. The information
was recorded both on the sample bottle and on a data sheet. Also listed on the
data sheet were the site values of the pH and conductivity, the rain gauge pre-
cipitation reading(s) in inches per event, the number of lid openings, and any
3-3
-------
pertinent observations. The data sheet thus served as a checklist to record all
station data and observations.
Shipping
Daily samples were preserved by maintaining them at about 4°C in a refrigerator
and using cold packs when shipping to the laboratory. The cold samples were
packed directly from the refrigerator into an insulated Styrofoam box with four
frozen gel packs. Weekly samples were packed together with the daily samples for
cold shipment to the laboratory. The carton was shipped by air freight each week
on Monday or Tuesday after packing of the weekly samples so that delivery would
be made before the weekend. It generally arrived at the laboratory on Wednesday
and rarely later than Friday.
Quality Control
The objectives of the quality control procedures in the field were to maximize
capture of uncontaminated samples, to identify and document them, to preserve
their integrity until their arrival at the laboratory, and to obtain each site's
measurement of precision and accuracy of pH and conductivity. (The pH and con-
ductivity were used to indicate if precipitation samples had degraded between the
field and laboratory measurements.) All field equipment was tested in the labora-
tory and any problems were corrected before the equipment was sent to the field.
To minimize contamination of the samples, measurements were made on sample ali-
quots and never on the whole sample. In addition, periodic rinses of the sampler
bucket were sent to the laboratory for analysis to yield dynamic blanks. Each
operator was contacted weekly by EMSI personnel to confirm sample shipment and
problems. To minimize downtime of equipment, each operator was given a schedule
of tests to perform on a routine basis (1,12). Collector malfunction was evi-
denced by large differences in the amounts of precipitation collected by the col-
located samplers or differences between the sampler and rain gauge; by the event
pen markers (indicating late or premature lid opening or closing, or lid cyc-
ling); the presence of precipitation in the dry bucket after an event; and direct
observation. Rain gauge problems included inoperative pens, clock malfunction,
and incorrect response to calibration weights.
3-4
-------
Each site was audited semi-annually, and the auditor inspected the site for pro-
blems. The operator was questioned and observed as he went through his routines,
including a test sample measurement. Any discernible equipment malfunction was
corrected, the rain gauge was recalibrated and adjusted, and operator questions
and problems were addressed at this time. An audit report was written.
To help assure routine correct measurement of pH, each site was supplied with a
dilute acid sample of known pH for measurement with the precipitation samples.
If the measured pH of the check sample differed from the labeled value by 0.10
unit or more, it alerted the operator and the laboratory that a problem was pre-
sent. A new electrode generally corrected the problem.
To evaluate the site's precision and accuracy for pH and conductivity, test solu-
tions were sent monthly to the stations for analysis; the chemistry of the test
solutions was unknown to the site operator. These are not true environmental
precision and accuracy results since the test solutions were not precipitation
samples but consisted of dilute acid plus salt solutions with rain-type pH and
conductivity values.
The test solutions were prepared at the EMSI laboratory, and the pH and conduc-
tivity of each sample were measured just before the sample was sent to the
field. For calculating accuracy the EMSI measurements were assumed to be the
standard. Station operators took readings on five aliquots of the solution,
reported the values to the laboratory, and returned the remainder of the sam-
ple. After their return to the laboratory, the samples were remeasured for pH
and conductivity to assure that they had not degraded. If the laboratory mea-
surements after the sample's return matched those before shipment to the field
(within 0.1 pH unit and 10% in conductivity), the samples were assumed to be
stable; if not, the measurements were not used to assess precision and accuracy.
Accuracy is based on the absolute difference between the average field value
and the average laboratory value. Field precision for pH and conductivity is
obtained from the standard deviation of the measurements made on the five ali-
quots of each sample. Generally, the standard deviation for a well-performing
electrode is less than 0.05 pH unit, and for conductivity less than 2 uS/cm.
3-5
-------
The average accuracy for pH at the Georgia, Kansas and Vermont sites was respec-
tively 0.08, 0.17 and 0.05 pH units and for specific conductance 8.7%, 13.3%, and
3.0%. The poor accuracy in pH for the Kansas site was due to two bad electrodes,
which were subsequently replaced. The precision for pH for the Georgia, Kansas,
and Vermont sites was respectively 0.02, 0.04, and 0.05, and less than 1 uS/cm
for conductivity.
LABORATORY OPERATIONS
The laboratory operations are summarized below. A more complete discussion is
given in the UAPSP Interim Report (2) and in the "Laboratory Standard Operating
Procedures Manual" (13).
Sample Check In
When a sample arrived at the laboratory, the temperature of the box interior was
measured. Each sample was logged in, and the temperature and any pertinent codes
were recorded on the data sheets. Approximately every twentieth sample was
marked for analysis in duplicate as a quality control measure. All the samples
were assigned consecutive EMSI laboratory numbers and the chemist performing the
analysis in the laboratory had no awareness of either the collection schedule or
the identity of the collocated samples.
Analyses
If sufficient sample (over 30 ml) was present, the following order of analysis
was taken to minimize chance of degradation:
pH, conductivity;
S0=, N0~, CT;
Ca2+, Mg2t Na+,
3-6
-------
The pH and conductivity were measured within one week of receiving the sample.
Other observables were analyzed as the instruments became available, generally
within six weeks.
If insufficient sample was available for complete analysis, the order of analysis
followed the importance of the constituents. The priority and the required vol-
ume for analysis were:
pH, conductivity (10 ml)
SOj, NO', Cl" (5 ml);
NHj (3 ml);
Ca2t Mg2t Nat K+ (6 ml).
All samples were analyzed without filtration. Filtration has the advantage of
removing both bacteria, which can degrade nitrogen and phosphorus compounds as
well as organic acids, and soil particles, which are generally basic and react
with hydrogen ion. However, filtration has the disadvantage of being a source of
contamination and is costly. To minimize degradation, the samples were kept cold
after removal from the collector, during shipment, and before and after analysis
at the laboratory. In addition, portions of the sample that were removed for
analysis were decanted to eliminate sedimented particles.
The methods of analysis and their detection limits are given in the first columns
of Table 3-1. For most analytes the detection limit is equal to twice the stan-
dard deviation of the baseline noise. For those analytes for which responses are
read from a strip chart (NH* SO", N0~ and Cl"), there is no real measure of
baseline noise, and the concentration corresponding to twice the baseline pen-
width was used as the detection limit.
Sample Storage
The samples, when not being analyzed, were stored in a refrigerator at 4°C.
After all the analyses were completed, the remaining sample, if larger than 100
ml, was stored at 4°C for one year for possible reanalysis at some later date.
3-7
-------
Table 3-1
EMSI LABORATORY METHODS, DETECTION LIMITS, PRECISION AND
ACCURACY FOR PRECIPITATION SAMPLE ANALYSES
Method Analyte
pH meter pH
Ion
chromatography SO^
NOj
CI-
Autoinated
colorimetry NH,
Atomic
Absorption Na+
K+
Ca2+
Mg2+
(a) From 1983-1984 data
Precision
Detection
Limit
(mg/l) N RMSDa
150 0.013d
0.020 150 0.029
0.020 145 0.027
0.002 134 0.010
0,010 112 0.0054
0.001 105 0.0022
0.010 111 0.0024
0.020 126 0.0042
0.002 123 0.0007
for split samples
(mg/l) Recovery
Range of
Cone, at Standard
10% CVb N (mg/l)
0.040 32 <1.0
346 1.0 - 10.0
0.025 18 <1.0
329 1.0 - 10.0
0.010 20 <0.2
345 0.2 - 2.0
0.025 43 <0.5
246 0.5 - 5.0
0.025 405 <1.0
0.025 413 <1.0
0.025 402 <1.0
0.005 388 <0.2
17 0.2-2.0
(%)
Mean + SEC
...
100.0 ± 0.4e
101.0 ± 0.1
101.2 ± 0.6
101.2 ± 0.1
102.7 ± 0.8
101.1 ± 0.2
100.4 ± 0.3
99.4 ± 0.1
99.4 ± 0.1
99.5 ± 0.1
100.7 ± 0.1
99.7 ± 0.1
100.7 + 0.2
T N 2 11/2
Root Mean Square Deviation = H (X,., - X,.,) /2N
[ft Jl 21 J
where X^ and X2.j
(b) CV = coefficient of
are the duplicate analyses and
variation
N is the number of sample pairs.
f N ? "11/2
(c) Standard error of the mean = £(R - R) /N(N - 1)
(d) pH units
3-8
-------
The reanalysis of "old samples" and comparison of results for analytical con-
sistency are a feature of the quality control system employed in the EMSI labora-
tory (14).
Quality Control
All analytical instruments were calibrated at least once a day. In addition,
reagent blanks, and split samples were analyzed daily. Instrument response was
sent directly to the computer for many of the analyses, thus eliminating errors
in transcription of data. When data were input to the computer manually, double
entry was used. A full description of the quality control program implemented
for UAPSP by EMSI is provided elsewhere (14).
Analytical Precision
The tolerance bounds of the laboratory data are measured by their precision.
Precision is derived from the data for duplicate or split (not collocated) sample
analyses. The precision is calculated from the equation for root mean square
deviation:
[N 2 "I
Root Mean Square Deviation = £) (X^ - X2i) /2N
where X^- and X2-j are the duplicate analysis values of the ith sample. The lab-
oratory results for over 100 split samples analyzed in 1983 and 1984 are shown in
Table 3-1. The precision values in Table 3-1 are generally similar to or better
than those found in the 1979-1980 SURE study (.15). Since the measurement error
increases with concentration, the coefficient of variation (the ratio of the
standard deviation to the mean concentration) is generally a more meaningful
value. The coefficient of variation, CV, is much more constant than the mean or
standard deviation over large concentration ranges, but generally increases rap-
idly with decreasing concentration below a certain level. The concentration
levels below which the CV increases above 10% are shown in Table 3-1. Concentra-
tions equal to or greater than these levels can be determined with good preci-
sion, whereas the relative uncertainty for concentrations below these values is
3-9
-------
large. Thus the 10% CV concentration is the limiting concentration for attaining
good analytical precision.
Analytical Accuracy
Laboratory accuracy was assessed by analyzing prepared samples of known concen-
tration. The concentrations of these laboratory standards was unknown by the
laboratory technician. Accuracy is measured as percent recovery, defined as the
percent ratio of the measured concentration to the true standard concentration.
A variety of laboratory standards were prepared across a range of concentrations
for each constituent.
The accuracy results are presented in Table 3-1. For one or two concentration
ranges for each analyte, the table lists the mean percent recovery and the stan-
dard error about the mean. The means range from a minimum of 99.4 to a maximum
of 102.7 percent recovery. For those analytes where two ranges of standards were
prepared, the percent recovery values are generally more variable in the more
dilute solutions, as would be expected. The mean recovery values are similar to
those reported for the SURE data (15).
In 1983-1984 EMSI was involved in five interlaboratory comparison tests with
Research Triangle Institute (RTI). The RTI results, which are presented in its
quality assurance reports (16), generally showed agreement to within 5% for ions
between the two laboratories. Other laboratory quality controls include use of
the computer to automatically flag the following conditions: spike data out of
control, below-detection-1imit data, calibration constants out of tolerance, and
inconsistencies in precipitation collection amounts, in cation-anion equivalents,
and in measured versus calculated conductivities.
REFERENCES
1. L. E. Topol. Field Operator Instruction Manual for Utility Acid Precipita-
tion Study Program (UAPSP). UAPSP report 104. Newbury Park, CA: Environ-
mental Monitoring & Services, Inc., 1983.
2. L. E. Topol and R. Schwall. "The Utility Acid Precipitation Study Program,
Network Description and Measurements for 1981-1982." UAPSP report 105.
Washington, DC: Utility Acid Precipitation Study Program, 1986.
3-10
-------
3. Manufactured by Aerochem Metrics, Miami, Florida, Models 201 and 301.
4. Belfort Instrument Co., Baltimore, Md., Model 5-780.
5. Ohaus Model 700.
6. Extech Model 671, Markson Science, San Diego, CA.
7. Orion Model 9162.
8. YSI, Model 701.
9. Markson Science, San Diego, CA, Model 10.
10. Markson Science, San Diego, CA, Model 1100.
11. H.L. Volchok and R.T. Graveson, "Wet/Dry Fallout
Collection," Proceedings of Second Federal Conference on Great Lakes,
1976, pp. 259-264.
12. Quality Assurance Handbook for Air Pollution Measurement Systems. Vol. V.
Manual for Precipitation Measurement Systems, Part II. Operation and Main-
tenance Manual. EPA-600/4-82-042b. Research Triangle Park, NC: U.S. Envi-
ronmental Protection Agency, 1981.
13. L. E. Topol, L. M. Carl in, T. Long, and S. Ozdemir. "UAPSP Laboratory Stan-
dard Operating Procedures." UAPSP report 102. Newbury Park, CA: Environ-
mental Monitoring & Services, Inc., 1983.
14. L. E. Topol. "[Working Draft] Plan for Controlling the Quality of Measure-
ments and Data Base in the Utility Acid Precipitation Study Program
(UAPSP)." UAPSP report 106. Newbury Park, CA: Environmental Monitoring &
Services, Inc., 1982.
15. L. E. Topol and R. Schwa!1. "Precipitation Chemistry Measurements in the
SURE Region." Draft Final Report. Palo Alto, CA: Electric Power Research
Institute, 1986.
16. W.C. Eaton and E. D. Estes. "Laboratory Systems and Audit Results: Utility
Acid Precipitation Study Program (UAPSP) (October 1982-March 1984)." UAPSP
internal report. Research Triangle Park, NC: Research Triangle Institute,
1984.
3-11
-------
Section 4
DATA MANAGEMENT AND STATISTICAL METHODS
This section describes data management procedures pertinent to data quality con-
trol, screening for rejection of outliers, data processing, and statistical test-
ing.
QUALITY CONTROL SCREENING
Study data were invalidated for the following reasons:
Obvious physical contamination of a sample
Instrument malfunction
Evidence of erroneous data entry
Obvious physical contamination of samples was indicated on the field sheets by
the site operators, verified by the receiving clerk at EMSI, and the appropriate
code entered into the data base. The contaminated samples were not analyzed. A
code was also entered in the data base for documented instrument malfunctions
indicated by the site operator. The various codes and data flags utilized in the
data base have been documented (I) and all data sheets are archived in bound
notebooks by site and date.
A double entry procedure with 100 percent verification was used for all data
manually entered into the computer. In this procedure, computer-generated out-
puts document the differences, if any, between the two corresponding entries.
All discrepancies are resolved before the data are permanently recorded in the
data base. Laboratory chemists manually check approxiately 5 percent of all
data, concentrating on extreme observations of concentrations and ion balance
4-1
-------
data, thus verifying both instrument calibration and the integrity of the analyt-
ical process. On the basis of these checks some samples were reanalyzed and the
results reentered into the data base. No extreme observations were deleted from
the main data base if they passed all the checks described here. For the purpose
of this study, values below detection limits were set equal to zero. Further
details on data handling and validation can be found in other EPRI reports (2,
3).
DATA PROCESSING
The collocated sampler data were used to determine random measurement error (pre-
cision), mean and median concentrations of daily and weekly sampling, and the
bias between the measured weekly and the derived weekly values, composited from
the corresponding daily results. All derived weekly concentrations were calcu-
lated separately for each collector from the daily values as the precipitation-
weighted mean, C = (E.J C^P^\/(z^ P^), where C^ and P^ are the daily concentration
and sample weight. The weekly precipitation values were derived from the sum of
the daily precipitation weights for the week for each collector. The weekly
deposition values were calculated from the product of the measured weekly concen-
tration for each collector and the total daily precipitation depths from the rain
gauge, i.e., D = 10CW E.J R^, where Cw is the weekly measured concentration in
mg/1, and Rn- is the daily rain gauge depth reading in cm. The derived weekly
values were calculated from the total of the daily deposition values, i.e., D =
10 E. C-jRj. Annual deposition values were calculated by summing the weekly depo-
sition values.
The data were classified by sampling site, precipitation type (rain, snow, or
mixed: rain + snow + ice), and by meteorological season (spring March-May;
summer June-August; fall September-November; and winter December-Feb-
ruary). A weekly sample that consisted of one or more rain events and one or
more snow events was classified as mixed.
Only events having a rain gauge depth greater than 0.51 mm (0.02 inch) were
included in this study to ensure that a sufficient sample was available for anal-
ysis. Thirteen daily events with such low-volume precipitation were not
analyzed. At the Georgia site there were two weeks with one low-volume event
each; each events represented less than four percent of the total precipitation
4-2
-------
for the week. There was only one low-volume event at the Kansas site; in the
week of this event there were two other events and the low-volume event rep-
resented 12 percent of the total weekly precipitation. Most of the low-volume
events occurred at the Vermont site, where there were 10 such events in seven
weekly periods. For five of those weeks, the one or two low-volume events rep-
resented less than three percent of the total weekly precipitation. In one week
there were four snow or mixed events, one of which was a low-volume snow event
representing 15 percent of the total precipitation. One week in late June had
four rain events, three of which were low volume; the fourth event that week was
small, and the three low-volume events accounted for 11 percent of the weekly
precipitation. For the Georgia and Kansas sites the very small number of low-
volume events would not be likely to bias the method comparison results. For the
Vermont site, where the majority of low-volume events occur (13 percent of the
weeks at the Vermont site have one or more low-volume events), there may be a
bias in the weekly concentrations derived from the daily events, especially if
such small precipitation events have unusually high concentrations. Further
investigation of possible effects of low-volume events in Vermont on the compari-
son of weekly versus daily methods is beyond the scope of the present study.
There were also low-volume samples with precipitation amounts greater than
0.51 mm but not enough for complete chemical analysis, which required 24 ml of
sample. As indicated in Section 3, a prescribed order of analysis was
followed. At a minimum, pH and conductivity were measured; this required 10 ml
of sample. If an additional 5 ml was available (i.e., total sample volume at
least 15 ml), the anions (S0~ NOI, and Cl~) were analyzed next. Ammonium was
next, requiring 3 ml (total sample volume at least 18 ml).
Measurements are missing, then, if there was a sample less than 24 ml. There are
also missing samples if a sampler was malfunctioning. In addition, samples with
visible organic material contamination (e.g., bird droppings) were considered
invalid and are not used in the analysis. For weeks containing one or more
missing daily samples having a rain gauge depth greater than 0.51 mm, a derived
weekly sample was not calculated.
The distributions of collocated differences in the daily and weekly data for the
three sites and for the three precipitation types are roughly symmetric with long
tails. Outliers contain important information about the measurement process, but
can also be the result of errors or unusual events extrinsic to the measurement
4-3
-------
process, such as contaminations. A small fraction of data are in the tails and
their elimination removes their disproportionately large influence on testing for
statistical significance of observed differences using parametric methods. In
this study, outliers were identified as those collocated differences beyond three
standard deviations from the mean for each observable. Collocated sample pairs
were considered outliers if one or more of the following criteria were met:
(a) One of the four major ion differences (H+, SO^, NO^, and NH^)
of that sample pair was outside the 3o acceptance limits,
(b) At least three of the minor ion differences of the pair were outside
the acceptance limits,
(c) One of the minor ion differences of the pair was outside the 3a accep-
tance limits, and three out of the four primary ions for the same sam-
ple were at either extreme of the collocated pair distribution.
Thus, all or none of the ion measurements in a collocated pair are considered to
be outliers. Outliers in collocated differences of precipitation mass (deter-
mined by weighing the Aerochem Metrics buckets) were identified separately.
Collocated precipitation differences were considered outliers if they were beyond
three standard deviations of the mean collocated difference.
A screened data set was created after removal of outliers. If a weekly sample
was identified as an outlier, it was discarded for the screened data set. How-
ever, if a daily sample identified as an outlier represented less than 20 percent
of the total precipitation mass for the week, then the sample was used to calcu-
late a derived weekly value, but was not used to calculate bias and precision of
the daily data. If a daily sample was identified as an outlier and it repre-
sented more than 20 percent of the total precipitation mass for the week, then no
derived weekly value was calculated for the screened data set, and the sample was
not used to calculate daily sampler bias and precision. This procedure resulted
in a screened data set with the following numbers of samples:
_Dai1y Measured Meekly
Unscreened Screened Unscreened Screened
Georgia
Kansas
Vermont
67
86
142
63 (94%)
72 (84%)
133 (94%)
35
38
52
33 (94%)
37 (97%)
47 (90%)
4-4
-------
In both the screened and the unscreened data sets, when one of two paired samples
was missing or rejected as an outlier, no collocated difference was obtained.
However, the valid collocated measurement was taken as the mean for the pair.
For the parametric statistical tests, which are based on assumptions of normality
of the distributions, the reduced data set (outliers removed) was used. Nonpara-
metric techniques are used to test the significance of the observed differences,
and since they are relatively insensitive to the data in the tails of the distri-
bution, the complete data set (i.e., including outliers) was used for these anal-
yses. Outliers were deleted for the parametric statistical analyses so they do
not exert undue influence on the results; where possible, however, nonparametric
methods were applied to the unscreened data set. There are alternative methods
for identifying outliers and for statistical analysis of data sets containing
outliers. For example, rather than rejecting all ions together or none at all,
one could delete only those measurements corresponding to collocated differences
beyond three standard deviations from the mean for the appropriate distribu-
tion. One could also use a more robust estimate of scale other than the sample
standard deviation (which is disproportionately affected by outliers); acceptance
limits can then be based on the robust scale estimates. One example of a robust
scale estimate is the sample interquartile range, which is the difference between
the 75th and the 25th percentiles of the sample distribution.
An alternative to nonparametric techniques or to deleting outliers before apply-
ing parametric techniques is to use statistical techniques that accommodate out-
liers. These techniques, which have been developed mostly within the last 10
years, are known as robust/resistant methods. They do not depend upon assump-
tions of normality and are not heavily influenced by outliers, yet they have much
greater power than standard nonparametric techniques. The book edited by Hoag-
lin, Hosteller, and Tukey (4) is a good reference for these procedures. Although
robust/resistant procedures are powerful and applicable in a wide variety of
situations, they can be computationally intensive and have not yet come into
general use outside the statistical community.
4-5
-------
STATISTICAL TESTING
Preliminary evaluation of the analyte data for collocated concentration differ-
ences showed that the distributions appeared to be symmetric but not normally
distributed as demonstrated in Section 5 (Figures 5-2 through 5-7). The distri-
butions of weekly derived minus weekly measured concentration data showed similar
behavior (Figures 5-10 through 5-12). These plots show distributions with out-
liers in both the upper and the lower portions. Because the distributions are
not skewed (i.e., outliers in one half of the distribution only), no monotone
transformation can be applied to make the distributions more normal.
Although the design of the study would suggest that the statistical analysis
proceed with a nested (random effects) analysis of variance model, this model was
not applied because of the reliance on assumptions of normality, which are not
met. Instead, a nonparametric Wilcoxon signed-rank test was used to evaluate the
statistical significance of differences in paired data. For example, suppose
there appears to be a systematic difference between measured and derived weekly
concentrations at the same site. The Wilcoxon test is used to test the signifi-
cance of this apparent method bias. The measured and derived concentrations for
a given week and site form a pair. The difference between each pair of numbers
is calculated, and the absolute differences are ranked. The Wilcoxon statistic
is the sum of the ranks of the positive differences minus the sum of the ranks of
the negative differences. If there were no systematic difference, positive and
negative differences would be about equal in number and of about the same size,
and the rank-sum would be near zero. A large, positive rank-sum indicates that
either (1) more differences are positive, or (2) the positive ones are larger, or
(3) both, i.e., a systematic positive difference. Similarly, a highly negative
rank-sum indicates a systematic negative difference.
The influence of outliers is limited by the use of ranks instead of raw data in
the Wilcoxon test. Large differences still count more than small ones, however,
because they have bigger ranks. The probability of the Wilcoxon indicating a
significant difference when there is no real difference is independent of any
distributional assumptions. The probability of the Wilcoxon test correctly indi-
cating a small, real difference is not much less than with the t-test for normal
distributions and much greater for some outlier-prone distributions. Throughout
this study, a significance level of a = 0.05 was used.
4-6
-------
Several robust parametric tests available from the BMDP statistical routines
developed by the University of California (5), were also used in the data analy-
ses. These included testing for the equality of the means of the collocated pair
averages and collocated pair differences among the sites for all analytes. For
parametric tests, in general, establishing levels of significance for the equal-
ity of the mean observations at each site is strongly dependent on the normality
of the parent population from which the data are drawn and the equality of the
variances of the distributions. To overcome the latter difficulty, the Levene
Test (6), a robust test of the equality of variances, was used. Welch's test is
used for certain problems in one-way analysis of variance. For example, are
there significant differences in method bias (i.e., daily versus weekly collec-
tion protocols) from site to site? As in the usual F-test, a (weighted) variance
of the estimated biases for the three sites is divided by an estimate of what it
would be if there were no difference. However, the usual assumption of equal
variances within groups (sites) is obviated by not pooling the variances and
using a generalization of Satterthwaite1 s approximation to the degrees of freedom
(5).
The precision of measurements for paired samplers is estimated two ways: pooled
standard deviation and median absolute collocated difference. The pooled stan-
dard deviation is
where the dn- are the pair differences and N is the number of pairs. This
estimate of precision includes variability both in the chemical measurements and
in the samplers (i.e., sampler bias), both of which are random errors that cannot
be controlled. This is a standard calculation (sometimes known as Youden's
method) and thus the results can be compared with other studies that use the same
measure of precision. Furthermore, precision so defined has a direct
relationship to the usefulness of the data in detecting trends if linear methods
(t-tests) are used to test for trends. That is, if daily data have a pooled
standard deviation half that of weekly data, then a study with daily data would
be able to detect a trend of half the magnitude of the smallest trend that could
be detected with the same number of weekly samples. Of course, this approach
makes no allowance for the fact that there will probably be more daily than
weekly samples.
4-7
-------
On the other hand, the pooled standard deviation is very sensitive to outliers
and therefore to the method chosen to eliminate outliers. This fact makes com-
parison across studies meaningless unless the same outlier-deletion protocols are
used, which is very unlikely. Furthermore, the presence of extreme outliers
suggests that linear methods may not be the best way to test for trends anyway,
so that the relationship between pooled standard deviation and the sensitivity of
the t-test may be irrelevant.
The other precision statistic used is the median of the absolute values of the
collocated paired differences (MACD), which is not sensitive to outliers. Thus,
comparison with previous daily collocation studies, such as EPRI-SURE (2) and
UAPSP (3), and other studies reporting median absolute differences is straight-
forward. The MACD, like the pooled standard deviation, is an estimate of the
variability of the distribution of differences, and thus can also be used to
compare daily and weekly sampling protocols in the ability to detect long-term
trends.
REFERENCES
1. L. E. Topol, R. Schwall, and J. Silverstein. Description and Format of Data
Base for EPRI Acid Precipitation Measurements. Ad Hoc Report for EPRI pro-
ject RP1630-2. Newbury Park, CA: Rockwell International EMSC, 1982.
2. L. E. Topol and R. Schwall. Precipitation Chemistry Measurements in the
SURE Region. Draft Final Report. Palo Alto, CA: Electric Power Research
Institute, 1986.
3. L. E. Topol and R. Schwall. "The Utility Acid Precipitation Study Program,
Network Description and Measurements for 1981-1982." UAPSP report 105.
Washington, DC: Utility Acid Precipitation Study Program, 1986.
4. D. C. Hoaglin, F. Mosteller, and J. W. Tukey, eds. Understanding Robust and
Exploratory Data Analysis. New York: Wiley, 1983.
5. BMDP Statistical Software. Los Angeles, CA: University of California
Press, 1983.
6. M. B. Brown and A. B. Forsythe. "Robust Tests for the Equality of Vari-
ances." J. Amer. Statist. Assoc., Vol. 69, pp. 364-367, 1974.
4-8
-------
Section 5
RESULTS
In this section we present the results of the statistical comparison of the daily
and weekly precipitation samples. First, the frequency and types of rainfall
events are described. Second, the collection efficiencies of the two sampling
methods are compared. Then, the precision of the measurements (determined from
collocated samplers) is discussed, with particular attention to differences
between daily and weekly sampling in their ability to detect long-term trends.
Next, measured and derived weekly concentrations are compared; method bias is
examined by precipitation type, site, and season. Finally, seasonal and annual
depositions calculated from the daily and weekly measurements are compared.
FREQUENCY OF RAIN EVENTS
The number of events at each site for each week during the study period is shown
in Figure 5-1. The total number of rain events is 67 in Georgia, 86 in Kansas,
and 142 in Vermont. Although the Vermont site has the largest number of rain
events, the average rainfall amount in each event is the lowest among the three
sites. Table 5-1 shows the distribution of precipitation types at each site. At
the Georgia site there are no snow or mixed events; at the Kansas site events are
predominantly rain; at the Vermont site about 40 percent of the events are snow
or snow mixed with rain.
Table 5-1
FREQUENCY OF PRECIPITATION TYPES AT EACH SITE
Precipitation Type
Site Rain Snow Mixed Total
Georgia
Kansas
Vermont
67
72
88
0
3
29
0
11
25
67
86
142
5-1
-------
7
6 .
5 .
4 .
3 .
2 .
1 .
1
N
T
_.
T
1
T
i
U SEP OCT NOV DEC JAN
M
B
E 7
6 .
o 5 .
f 4 .
E 3 .
V 2 .
E
Nl .
T
r
s '
_-
_
T
1
i
I
Georgia
_
n n
T
T
T
r n
_
| |
FEB MAR APR MAY JUN JUL AUG SEP OCT
1
Kansas
n
m
_
TI r
i i i
SEP OCT NOV DEC JAN FEB MAR APR MAY JUN JUL AUG SEP OCT
7
6 .
5 .
4 .
3 .
2 .
1 .
Vermont
i i i i i i i i r r i i i i
SEP OCT NOV DEC JAN FEB MAR APR MAY JUN JUL AUG SEP OCT
FIGURE 5-1. Frequency of rain events by week for the 1983 to 1984
study period.
5-2
-------
COLLECTION EFFICIENCIES
Sampler performance is measured by the precipitation collection efficiency, which
is the ratio of the average sample depths of the collocated buckets to the
standard rain gauge depth; i.e.,
C.E. = sampler depth/rain gauge depth
for each event. The sampler depth is derived from the sample weight and the area
of the sampler bucket, and the rain gauge depth is read directly from the rain
gauge stripchart. The daily and weekly mean collection efficiencies for rain,
snow and mixed samples for the three sites are presented in Table 5-2. Also
included are the standard error of the mean and the number of samples in each
category.
The mean collection efficiencies for rain events are higher than the mean collec-
tion efficiencies for snow and mixed events, which are relatively similar. This
is as expected since the sampler sensor does not respond readily to dry snow, and
snow capture losses also occur either when snow blows past the bucket or is blown
out by wind. Collection efficiency is especially low for Kansas snow and mixed
events; this may be due to drier snow or windier conditions at this site than at
the Vermont site. The mean collection efficiencies for rain, which are slightly
larger than unity, indicate that the samplers collect more precipitation than the
gauges, possibly due to precipitation splashing off the moveable bucket lid.
The collection efficiencies for the daily and weekly mean values show little dif-
ference. The only large difference is for Vermont rain samples; the mean collec-
tion efficiency is significantly higher in the daily samples than in the weekly
samples. Since the daily and weekly sample depths are both referenced to the
same rain gauge, it appears that some precipitation is lost from the weekly
samplers. This difference is hard to explain by evaporation alone; it might be
due to different threshold sensitivities of the lid sensors, or slightly
different bucket dimensions. Note that for Kansas there are four daily snow
events, which correspond exactly to the four weekly samples; the lower mean
collection efficiency for the weekly samples indicates a different sensor
response, possibly due to evaporation and/or blow-away.
5-3
-------
Table 5-2
AVERAGE COLLECTION EFFICIENCY FOR RAIN, SNOW AND
MIXED (RAIN & SNOW) SAMPLES AT THREE MONITORING SITES
Daily Sampling Weekly Sampling
Precipitation
Type
Mean
Coll. Eff.
SEM3
N Coll. Eff.
SEM3
N
Uvalda, GA
Rain
Mixed
Snow
Rain
Mixed
Snow
Rain
Mixed
Snow
1.059
1.090
0.770
0.736
1.164
0.948
0.943
0.021
0.033
0.078
0.079
0.032
0.066
0.060
61
0
0
Lancaster,
60
8
4
Underbill ,
74
24
25
1.063
KS
1.103
0.831
0.607
VT
0.999
0.939
0.970
0.040
0.045
0.072
0.170
0.033
0.029
0.040
33
0
0
24
9
4
26
17
4
(a) SEM = standard error of the mean = SD//N.
5-4
-------
COMPARISON OF DAILY AND WEEKLY CONCENTRATION MEASUREMENT PRECISION
Sampler Bias
The distributions of the daily and weekly collocated differences for precipita-
tion mass and the concentrations of major ions (H+, SO^, NO^, NH^J, and Ca ) are
shown in Figures 5-2 to 5-7 for each site. In these figures the extreme bars
contain all the data at the given abscissa value and beyond, and the curve
represents the Gaussian distribution that has the same mean and standard devia-
tion as the data. The means and standard deviations of the pair differences are
shown in Table 5-3 as a percentage of the mean measurement. The pair differences
were tested for systematic error (sampler bias) using the Wilcoxon signed-ranks
test (I). Statistically significant bias between paired daily or weekly samplers
is denoted in the tables with a W superscript. The same set of statistics for
daily and weekly samples of each precipitation type are listed in Appendix A
(Tables A-l, A-2, and A-3 for rain, snow, and mixed samples, respectively).
Most of the relative differences are below 7 percent. For daily sampling the
significant differences in paired samples were as follows: precipitation amounts
for all three sites; sulfate, nitrate, sodium, and chloride for Uvalda, Georgia;
and ammonium for Lancaster, Kansas.
The bias for all the observables at the Georgia site, though statistically
significant, is small (under 3 percent). The larger analyte concentrations,
together with the larger precipitation amounts found in the same sampler, could
be explained if this sampler opened sooner and captured more of the initial
precipitation, which generally contains more wash-out of constituents. Concen-
trations of sodium and chloride could be increased by human contamination, but it
is difficult to account for differential contamination between paired samplers.
The larger precipitation amount in this sampler argues against evaporation as the
cause of the higher concentrations. The ammonium bias for Kansas is 7.6 percent,
which suggests some contamination or biodegradation has occurred.
For weekly sampling, statistically significant sampler biases were found for
precipitation at the Kansas and Vermont sites, and for nitrate at the Vermont
site. The precipitation biases, though statistically significant, are small,
5-5
-------
WEEKLY
DAILY
p
E
R
C
E
N
T
0
f
V
A
1
L>
u
E
S
40 . 40 1
35 .
30 .
25 .
20 .
15 .
10 .
5 ,
35 .
Georgia 30 .
25 .
t*
M »* ^^^^
20 .
" 15 .
« *
1 10 '
...^- - - 5 .
FTTK4]
Georgia
si
n _wn r
*
s
\
V
Tiri;
-0.240 -0.120 0.000 0.120 0.240 -0.240 -0.120 0.000 0.120 0.240
40 40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
35 .
Kansas 30 .
25 .
mm
m
r f
t
A
1 20 .
3 1S '
V 10
\ 5
/
/
/S
K
«*/[
*
Kansas
j
1
|
1
I
-0.240 -0.120 0.000 0.120 0.240
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
Vermont
40
35
30
25
20
15
10
5
-0.240 -0.120 0.000 0.120 0.240
9
Vermont
-0.240 -0.120 0.000 0.120 0.240
-0.240 -0.120 0.000 0.120 0.240
FIGURE 5-2. Distributions of collocated differences for precipitation mass in kg.
The curve represents the Gaussian distribution with the same mean and standard
deviation as the displayed data. The extreme bars represent the values at the
given abscissa and beyond.
5-6
-------
WEEKLY
DAILY
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
R -
I «,
N 35 .
30 .
o 25 .
f 20.
V 15 .
L" »
U 5 .
E
f*
40 ,
35 .
Georgia 30 .
KJ&
25 ,
20 .
tn
i\ ls
**
s
iv 5
2
* U| Tl
1 Georgia
n
016 -0.008 0.000 0.008 0.016 -0.016 -0.008 0.000 O.OOB 0.016
40 .
N
n 3s
m
«
i
A
Kansas 30 .
25 .
20 .
15 .
'\ 10 .
Ikn
a -
W 1
9
Kansas
N
-0.016 -0.008 0.000 0.008 0.016
-0.016 -0.008 0.000 0.008 0.016
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
Vermont
40
35
30
25
20
15
10
5
Vermont
-0.016 -0.008 0.000 O.OOB 0.016
-0.016 -0.008 0.000 0.008 0.016
FIGURE 5-3. Distributions of collocated differences for hydrogen in mg/1. The
curve represents the Gaussian distribution with the same mean and standard deviation
as the displayed data. The extreme bars represent the values at the given
abscissa and beyond.
5-7
-------
WEEKLY
DAILY
40 .
35 1
30
25 .
20 .
15 ^
10 ,
5
P
R -°
C
E 40
N 35 .
T 30.
o 25 ,
f
V 15 .
U 5 .
E
S
-0
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
40
35 .
a
n
N XR
i M
Georgia 30 .
25 .
20 .
15 .
q
'k. *
m
n ccrfmr
s Georgia
>
.500 -0.250 0.000 0.250 0.500 -0.500 -0.250 0.000 0.250 0.500
40
35 .
Kansas 30 .
25 .
20
RM M^f .
i-Af' 1 (I
1 15 .
fl ]0 '
^ 5
Kansas
s
* f* *i
trfr^^ "
n
- XI _ _
IrTrFK-n
500 -0.250 0.000 0.250 0.500 -0.500 -0.250 0.000 0.250 0.500
40
35 .
Vermont 30 .
25 .
20 .
mr
_5irff
11 1S '
s 10
s
Vermont
r
J'
4
tf
* «!
n
-0.500 -0.250 0.000 0.250 0.500
-0.500 -0.250 0.000 0.250 0.500
FIGURE 5-4. Disbributions of collocated differences for sulfate in mg/1. The
curve represents the Gaussian distribution with the same mean and standard
deviation as the displayed data. The extreme bars represent the values at the
given abscissa and beyond.
5-8
-------
WEEKLY
DAILY
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
E
"
r»
""" ^X***
JJ___^x^[]
40
35 .
Georgia 30 .
25 .
20 .
15 .
10 .
M
^k. - 5 .
TTl>X-~-rJl
1 ^
8
1
|
/
I
jf
'
in
H Georgia
j
L
K
m
III
i .
R -0.500 -0.250 0.000 0.250 0.500 -0.500 -0.250 0.000 0.250 0.500
C
E 40 i
N 35 .
T 30.
o 25 .
f
20 .
V 15 .
A
L 10 '
U 5 .
E
S
40
35 .
Kansas 30 .
01
** *
N /^ O
JSnmn
25 .
20
n is
10 .
o
RKJi
Kansas
.
*
n
n J>^'
LfTfn - [[
7*
~
"n
-0.500 -0.250 0.000 0.250 0.500
-0.500 -0.250 0.000 0.250 0.500
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
40
35 .
f
f
,
jHJjf
i
T Vermont 30 .
25 .
20 .
15
I. 10 .
TrSsi 5 '
«
C*
n
f
I
J
t
t
Vermont
\
-J
fti
-0.500 -0.250 0.000 0.250 0.500
-0.500 -0.250 0.000 0.250 0.500
FIGURE 5-5. Distributions of collocated differences for nitrate in mg/1. The
curve represents the Gaussian distribution with the same mean and standard
deviation as the displayed data. The extreme bars represent the values at the
given abscissa and beyond.
5-9
-------
WEEKLY
DAILY
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
**>
"
n ^m
40
35 .
Georgia 30 .
25 .
20 .
15 .
10 .
ifr^n
E ....
c
f
/
/
wti
sl
I V
m
Georgia
.
y
\
k
R -0.240 -0.120 0.000 0.120 0.240 -0.240 -0.120 0.000 0.120 0.240
C
E 40 i
N 35 .
T
30 .
o 25 .
f 20.
V 15 .
A
I 30
L
U 5 .
E
40
35 .
Kansas 30 .
25 .
20 .
V
f
ran i c
10 .
^
En
Kansas
n in
" /f '
\ jf
s
"
b.
-0.240 -0.120 0.000 0.120 0.240
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
Vermont
-0.240 -0.120 0.000 0.120 0.240
40 8
35 .
30 . Vermont
25 .
20 .
15 .
10 .
5 »
-0.240 -0.120 0.000 0.120 0.240
-0.240 -0.120 0.000 0.120 0.240
FIGURE 5-6. Distributions of collocated differences for ammonium in mg/1. The
curve represents the Gaussian distribution with the same mean and standard
deviation as the displayed data. The extreme bars represent the values at the
given abscissa and beyond.
5-10
-------
WEEKLY
40 .
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
R -°
C
E 40 i
N 35 .
T 30.
o 25 .
f a.
V 15 .
t «
U 5 .
E
S
-0
40 .
35 .
30 .
25 .
20 .
15 ,
10 .
5 ,
/
/
if
1
*
Georgia
M
I
\
\
fin
160 -0.080 0.000 0.080 0.160
Kansas
n
-JL-.
X *N N N
160 -0.080 0.000 0.080 0.160
S
n
J'
Vermont
40
35
30
25
20
15
10
5
DAILY
8
Georgia
-0.160 -0.080 0.000 0.080 0.160
40
35 J
30
25
20
15
10
5 .
Kansas
-0.160 -0.080 0.000 0.080 0.160
40
35
30
25
20
15
10
5
Vermont
-0.160 -0.080 0.000 0.080 0.160
-0.160 -0.080 0.000 0.080 0.160
FIGURE 5-7. Distributions of collocated differences for calcium in mg/1. The
curve represents the Gaussian distribution with the same mean and standard
deviation as the displayed data. The extreme bars represent the values at the
given abscissa and beyond.
5-11
-------
Table 5-3
DAILY AND WEEKLY CONCENTRATION SAMPLER BIAS
Daily Sampling
Weekly Sampling
Analyte
Precipitation
H+
S0=
NO^
NHj
Cl"
Na+
K+
Ca2+
Mg2+
Precipitation
H+
so;
NOg
NHj
Cl"
Na+
K+
Ca2+
Mg2+
RD (%)a
-1.7W
-4.1
-2.4W
-2.8W
-3.3
-0.8W
? RW
-£ .0
-7.0
-2.7
0.0
-1.3W
-6.1
-2.6
-0.2
7 fiW
-/ .0
-7.0
-11.4
-7.0
0.3
4.3
RSD (%)b
3.9
12.2
5.9
5.4
21.3
13.0
14.0
58.8
21.7
18.5
5.1
18.2
11.7
21.7
18.6
44.2
41.8
71.9
20.7
19.6
N
Uvalda, GA
63
57
57
57
57
57
57
57
57
57
Lancaster,
71
69
69
69
69
69
69
69
69
69
RD (%)a
-0.0
0.0
0.0
-4.5
-2.9
1.8
1.3
-2.8
1.5
-2.3
KS
0.4W
12.5
-2.8
-1.2
-3.3
-5.0
-3.0
-20.4
-7.1
-14.5
RSD (%)b
9.2
17.4
13.6
22.5
46.3
8.3
17.2
30.6
15.0
11.8
4.1
37.5
9.9
9.0
22.2
24.4
15.1
102.0
22.9
48.2
N_
34
32
32
32
32
32
32
32
32
32
38
36
36
36
36
36
36
36
36
36
5-12
-------
Table 5-3 (Concluded)
Daily Sampling
Weekly Sampling
Analyte
Precipitation
H+
S0
NH
Cl
RD
RSD
RD
RSD
Na1
+
K
Ca
Mg
2+
2+
-1.5
0.0
-0.6
-0.3
-2.7
-1.8
-5.8
-4.0
-2.5
0.0
W
Underbill
4.5
7.9
4.7
5.5
22.4
22.5
25.2
52.0
20.2
21.0
132
126
124
124
121
124
117
118
117
117
, VT
-3.2W
-4.1
-0.5
-4.2W
1.8
-4.7
-2.5
0.0
1.8
0.0
15.3
20.4
7.4
30.4
16.2
52.1
53.2
109.1
20.0
28.6
50
48
48
48
48
48
48
48
48
48
(a) RD = (d/C) x 100, where d and C are the mean collocated
diffe£ence and average concentration.
(b) RSD = (SD/C) x 100, where SD is the standard deviation of
the collocated differences.
W = Sampler bias for the collocated pair significantly different from zero
based on the Wilcoxon signed-rank test.
5-13
-------
less than 4 percent. Bias for a single ion suggests contamination problems, but
the nitrate bias at the Vermont site is less than five percent.
Measurement System Precision for Paired Samples
The pooled standard deviations (SD), the relative standard deviations (RSp, the
Sp divided by mean concentration), the median absolute collocated difference
(MACD), and the relative MACD (RACD, the MACD divided by median concentration)
are listed in Table 5-4. Most of the RACD values are less than 10 percent except
for Na+ and Ca at the Georgia site and K+ at all three sites. The RS values
are generally less than 20 percent except for K+, which is as high as 50 percent
at the Georgia site. The large variation in pair differences for K+ may be due
to contamination or adsorption-desorption phenomena from the plastic containers.
Dynamic blanks for the collector buckets frequently yielded K+ concentrations
similar to those of the precipitation samples. Similar results were found by
Sisterson et al. (2).
The MACD precision estimates for daily concentrations in Table 5-4 are representa-
tive of the MACDs for collocated daily samplers in the EPRI-SURE and UAPSP networks
(3,4). For example, in the EPRI-SURE and UAPSP networks (1978 to 1981), MACDs for
daily H+ concentrations ranged from 0.0008 to 0.0106 mg/1 and SO^ MACDs ranged from
0.035 to 0.055 mg/1. The MACDs for the three sites in this study are within those
ranges; the same is true for all of the major ions and most of the minor ions.
Since the EPRI-SURE and UAPSP networks cover the northeastern United States, pre-
cision of the three sites in this study can be considered generally representative
of precision of collocated monitors in the Northeast.
As stated in Section 4, the two estimates of precision can be used to compare the
number of samples required under daily and weekly sampling to detect a specific
trend in time if linear techniques (i.e., t-tests) are used to test for trends. If
the daily Sp is the same as the weekly Sp for a particular analyte, then the same
number of daily samples as weekly samples would be required to detect the same
trend. If the daily Sp is half that of the weekly Sp, then one-quarter as many
daily samples as weekly samples would be required to detect the same trend over
time. To put it another way, if the daily Sp is half that of the weekly Sp, then a
study with daily data would be able to detect a trend of half the magnitude of the
smallest trend that could be detected with the same number of weekly samples. Of
5-14
-------
CO
LU
Q.
s:
-
i
UJ
LU
Q
"ZZ.
C
r
Q.
E
co
r5
ft\
\U
cu
3
«
a
Qi
0
£
2:
-~
^^
Q-
CO
o;
Q-
CO
OOrHLDI CTl O CO 1
, 1
«*
O «d- i i 2 O O O
oooooooo
LOOOOOOOO
CM
co f~~i Li^ f~*i co cy^ c^ ^~
r*^. OJ CTi ^D CM LO CM r~t
t i r- < ro r i CM
00
CM CO 00 ID O ID 00
t lOi it (OOOO
CD OOOOOOOOO
r--.
ro ^H
r~~ in
IT)
o o
o o
O 0
co «d-
O CO
, 1
ID
^- ro
O 0
0 0
I
LO
et
Q
cc
O
cn
E
O
ii
CO
II
o
LU
CC
LU
I
CO
CO
o;
co
LU
Q.
E
(O
co
ro
o
a
C_)
et
o
Q
o
eC
CO
a;
CV100CMCMLf>
O
C\J
CM
CTl
O
0
0
0
01
ro
o
0
CO
CM
o
o
CM
1 1
0
o
CM
1 1
0
o
CTi
o
0
o
Lf>
0
o
0
ID
0
O
0
CM
O
0
O
O
o
VO
oo
01
ff.
o
cu
o
co
I CO+ »d- |
o a:
CM CM
+ (O cn
^02:
5-15
-------
(1)
I
LO
-Q
(O
O1
c
i
"a.
e
ro
CO
^
QJ
Ol
3
cn
^
Q.
E
ro
CO
^
ro
CD
«
g
o;
g
CO 0
CM
1
*
CO
CO
0
o
CM
1^
CM
CO
i I
O
LO
ro
CM
VD
O
O
,3-
CO
O
r
11
O
CO
*
i t
vo
0
0
^.
vo
CM
O
i I
O
CO
^_
r-.
LO
0
o
CO
LO
, i
o
LO
CM
O
CM
1
i-H
CO
O
0
^
in
i-H
LO
1
O
o
UD
1
i 1
CO
o
o
1 1
,-j.
T-H
CM
CO
o
CD
CM
cn
o
i-H
0
0
^
1^.
1 1
i-H
CM
O
o
<*
cn
o
i-H
0
0
.H-
i I
CO
LO
«^-
o
CD
LO
CO
LO
0
o
o
CO
o
, H
r
O
o
o
LO
cn
LO
o
o
o
«f
o
CO
^~
CM
0
0
cn
^
LO
O
O
0
ID
CM
1
ID
ro
0
0
ID
,3.
i I
LO
0
o
o
^
CD
LO
cn
CM
0
o
LO
LO
I-H
r-.
LO
O
o
^
VO
i-H
CM
CO
O
o
CO
IO
1 1
CM
o
o
LO
,_!_
i-H
CM
CO
O
o
CO
, 1
I-H
CO
CO
0
0
0
I 1
LO
CO
in
^*
, t
0
o
ID
ID
CM
CM
0
0
CD
i i
,-f.
i i
LO
*n
CD
0
O
O-
o
o o
co 2:
i +
i ro +
CM CM
rocn
5-16
-------
Ol
c:
Q.
E
to
3
Q
O
O
0
<:
o.
to
CO CO
CM **
CM
O
O O
I-- 0
CM
CTl LO
0 *
i-H i 1
i 1
0
i-i O
> * 0
* 1 1
CO
CO
CT>
LO
O
O
CM
LO
LO
0
T 1
o
o
CO
o
o
LO
I 1
CM
0
CO
CO
o
CM
LO
v 1
< 1
O
0
I-H
1 1
CM
co
0
0
LO
0
0
0
ID
LO
CO
CT)
co
0
o
CM
CM
i i
CO
o
o
o
CM
CO
LO
I-H
o
o
1 1
CM
CO
O
o
0
CO
LO
i 1
o
o
1 1
CO
1 1
CO
o
o
o
0
i-H
1 1
o
o
CM
CO
. 1
CM
1 1
o
0
0
o
o
CM
LO
CO
o
o
o
-o
C
ea
01
o
01
Q.
in
ai
O)
o
O)
t_
QJ
""*
rt 1
ty
Z3
*"""
~^
~
O
O
«g-
'
LO
OI
**}
Ti
1
1
C7I
C
r
a
E
tv
>^
r
r
(O
Q
T3
, .
*4
O
t_j
e£
OL
0
Q
O
>1
ro
c
e^
^«
t_
01
T3 I-H O 1 CTi
^3 CO ^J- CM CM
r-
ii r->. CM
O CO ^
CO O O O
0 O 0 0
I 1
fo LO co cyi
CO LO CO CO
CO
CM 0 CO
CD l^-~ ^O
r~~ o o o
o o o o
CM
f »
in
e
ro
£_
a>
o.
f
(J
OI II ^ 1 CO
t- + O O
CL. ?Z IS) Z.
CT» LO *-H CO LO O
co r-- ^- . i CM o
i 1 CM i 1 i-H
CTl
CO LO CO CO LO O
O O O O O O
0 0 O O 0 O
o o o o o o
CTI cr* CM r~- co oo
LO LO CO LO *i- «1"
I-H I-H i-H CO I 1 I-H
CO
CO CO CT) CTl r~, CM
^1- I-H O O i i O
O O O 0 0 O
O O O O O O
-f- +
+ =1-1 + CM CM
3: i ro + re CD
z. <-> -z. ^. u s:
.,_
T3
^_
ro
Q.
T3
01
4_>
ro
o
o
r
i
O
o>
r-
4_>
(/)
.,
o
01
t_
01
_j^*
^
CSJ
1 1
*" "*
2:
C\J
***N.
ro
u
o
o
o
<4_
0
t_
01
_o
E
3
c
f
4->
ro
^
c:
OI
o
c
o
(J
OI
OI
re
t-
>
ro
OI
.c
4_j
in
>
Ijj
o>
t_
01
^-
^
o"
o
I-H
X
, s
|o
Q.
to
II
x .
S^
1
a.
to
OL
X .
-D
in
01
O
C
OI
t_
01
ro
o
o
r
O
u
o>
"o
^
ro
4_
O
C
ro
i
T3
QJ
£
II
Q
0
SI
,^ ^
O
O
o
I 1
x.
f s
c
0
r
4_>
ro
t-
4-*
c
O)
-------
course, in a given time period there are more daily samples than weekly samples, so
this must be taken into account. In the three sites in this study, there are two to
three times as many daily samples as weekly samples; i.e., precipitation events
occurred two to three times per week at the sites. So if the pooled standard
deviation of the daily data is the same as that of the weekly data, but there are
twice as many daily events as weekly events in a study, then a trend in time of a
specified magnitude will be detected in daily samples in approximately one-quarter
the time it would take to detect a trend from weekly samples.
The relative efficiency of the two sampling protocols to detect a trend of a certain
magnitude with a specified probability is thus estimated by the ratio of the weekly
and daily pooled standard deviations, or alternatively, by the ratio of the weekly
and daily median absolute collocated differences. These ratios are shown in Figures
5-8 and 5-9, respectively, for each parameter at each site. The figures show that
the majority of the weekly to daily scale ratios are greater than one, i.e., that
the daily samples are more precise than weekly samples. The two figures, however,
do not always tell the same story. In the case of NO" at the Georgia site, for
example, the ratio of the weekly to the daily pooled standard deviation is 3.63, but
the ratio of the weekly to the daily MACD is 0.679. This suggests that differences
in precision between daily and weekly measurements may have more to do with exactly
how precision is defined than with any practically significant difference between
the two methods.
Both the ratio of weekly to daily samples and the number of precipitation events
per week (or number of daily samples per week) must be considered to determine
how much more quickly a trend can be detected with daily samples than with weekly
samples. These calculations are listed in the table below for the ratio of
weekly to daily precision values that are typical in this study.
Ratio of
Weekly-to- Number of Daily Samples Per Week
Daily Precision 1.0 1.5 2.0 2.5 3.0
1
1
1
1
.0
.2
.4
.6
1
0
0
0
.00
.69
.51
.39
0
0
0
0
.44
.31
.23
.17
0
0
0
0
.25
.17
.13
.10
0
0
0
0
.16
.11
.08
.06
0.
0.
0.
0.
11
08
06
04
5-18
-------
0
O
w
d t- ^
° I =
°" o £
2 n H>
D C "O
> D ^-
V//////////
o
V////////////A
0'9
5'9 0'9
ere
O'L Q'O O'O
s-
O)
Q.
E
re
O)
O
O
O
o
o
S-
QJ
Q
T3
i-
"O
re
-l->
l/l
-O
CD
O
O
Q_
re
o
OJ
OJ
O
O
re
oo
i
in
01
S ^l!
5-19
-------
O
O
D
3 _J
0)
4->
w .c
o
0)
V///////
o
Y///////////
o
V////////////////
o
J I I L
J I L
J L
OT
Q'Z
O'Z Q'L
QOVIA]
O'L
Q'O
Ol
S-
Ol
M-
<*-
Q
-o
CL1
rO
O
O
r
O
,
O)
O)
to
o
OJ
5-20
-------
Thus we see, for example, that if there are two daily samples per week and weekly
sampling is 40 percent less precise than daily sampling (ratio = 1.4), then a
trend of a specified magnitude can be detected from daily samples in only 13
percent of the time it would take to detect the same trend from weekly samples.
A ratio of weekly-to-daily precision of 1.4 is about the average across the three
sites for H+ and SOT, the two most important ions. Although the daily samples
will reveal trends at a much faster rate, cost must be considered as a factor.
Since daily sampling is considerably more expensive than weekly sampling, the
costs may outweigh the benefits of daily sampling.
The number of monitors in a geographical area is another factor to consider when
designing a network. One may be able to more easily detect trends in an area
with more weekly monitors than with a smaller set of daily samplers. This study
was not designed to determine how the number of monitors in an area affects trend
estimation.
DERIVED VERSUS MEASURED WEEKLY CONCENTRATIONS
The median and mean daily, weekly, and derived weekly concentrations for each
site are summarized in Table 5-5. In all cases the mean values are larger than
the median values, indicating that the ion concentration distributions are
skewed, with long positive tails. The Vermont site has the highest mean H+,
SOT, and NOZ concentrations. Kansas has the lowest H+, the highest NH^, and,
except for Na+, the greatest metal ion concentrations. The Georgia site has the
lowest concentrations of acid anions and ammonium, but the highest Na concentra-
tions. The median, mean, and standard error of the mean of the site concentra-
tions for both measured weekly and derived weekly samples for rain, snow, and
mixed samples are given in Appendix B (Tables B-l, B-2 and B-3, respectively).
Method Bias By Site and Type of Precipitation
Method difference is defined as the derived weekly minus the measured weekly
value. Figures 5-10 through 5-12 contain histograms of the differences between
derived weekly and measured weekly concentrations of the major ions and precipi-
tation at each of the three sites. The curves in the figures represent the Gaus-
sian distribution with the same mean and standard deviation as the displayed
5-21
-------
Table 5-5
SUMMARY OF MEDIAN AND MEAN SITE CONCENTRATIONS (mg/1)
Analyte
Precip (grams)
Georgia
Kansas
Vermont
H+
Daily Measured
Median Mean
759.
573.
329.
1285.
785.
620.
Weekly Measured
Median Mean
1832. 2312.
1025. 1544.
1179. 1538.
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
0.019
0.011
0.040
1.212
1.745
1.386
0.025
0.016
0.058
1.494
2.056
2.363
0.015
0.014
0.045
1.132
1.690
1.521
0.022
0.016
0.047
1.329
1.781
1.990
Derived Weekly
Median Mean
1848. 2362.
891. 1498.
1170. 1561.
0.013 0.022
0.016 0.017
0.048 0.050
0.951 1.261
1.477 1.729
1.543 2.014
NH+
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
0.726
1.386
1.466
0.132
0.398
0.215
0.948
1.704
2.056
0.201
0.596
0.334
0.628
1.405
1.271
0.117
0.425
0.205
0.802
1.591
1.755
0.175
0.469
0.157
0.627 0.772
1.362 1.543
1.285 1.818
0.125 0.175
0.387 0.472
0.157 0.278
5-22
-------
Table 5-5 (Concluded)
Analyte
CT
Georgia
Kansas
Vermont
Daily Measured Weekly Measured
Median Mean Median Mean
0.284 0.452
0.125 0.151
0.071 0.120
0.304 0.415
0.100 0.114
0.065 0.106
Deri ved Weekly
Median Mean
0.302 0.400
0.110 0.120
0.068 0.110
Georgia
Kansas
Vermont
0.160
0.055
0.021
0.260
0.088
0.052
0.210
0.051
0.024
0.240
0.057
0.040
0.160
0.054
0.022
0.220
0.060
0.040
Georgia
Kansas
Vermont
Ca
2+
Mg
2+
0.029
0.036
0.013
0.049
0.060
0.041
0.029
0.035
0.014
0.031
0.048
0.021
0.030
0.035
0.010
0.034
0.044
0.016
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
0.077
0.317
0.047
0.025
0.035
0.009
0.120
0.597
0.129
0.039
0.049
0.022
0.088
0.342
0.057
0.029
0.032
0.008
0.114
0.465
0.111
0.035
0.039
0.018
0.065
0.403
0.049
0.026
0.030
0.007
0.093
0.460
0.098
0.032
0.037
0.015
5-23
-------
HYDROGEN
SULFATE
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
N /P
ml
40
35 .
Georgia 30 .
25 .
20 .
15 .
v «
YL ID
fl
E ....
Georgia
n
,
A'-
0
-
"ftk
R -0.016 -0.008 0.000 0.008 0.016 -0.500 -0.250 0.000 0.250 0.500
C
E 40 i
N 35 .
T
30 .
o 25 .
f
20 .
V 15 .
A
L
U 5 .
E
r-
40
3 35 .
PI
_ -. j"i *<< | ru
iJiftrnrlr
Kansas 30 .
25 .
20 .
15 .
10 .
4 CW*
TfC -* **
||]p^_ji
Kansas
«*
nfi
-fTfTn I "
4 PI
1 |M (M
ItrmrnUJ
-0.016 -0.008 0.000 0.008 0.016
-0.500 -0.250 '0.000 0.250 0.500
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
40
35 .
Vermont 30 .
25 .
20 .
li
J-
i^ff
15 .
inn 10
«
Iff
n
mrV-J]
-0.016 -0.008 0.000 0.008 0.016
-0.500 -0.250 0.000 0.250 0.500
FIGURE 5-10. Derived weekly-measured weekly data for hydrogen and sulfate
ions in mg/1. The curve represents the Gaussian distribution with the same
mean and standard deviation as the displayed data. The extreme bars represent
values at the given abscissa and beyond.
5-24
-------
NITRATE
AMMONIUM
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
E
40 1
r*
_J
n 35.
Georgia 30 .
25 .
] 20 .
1 1S '
Q 10 .
'
n
a A
Georgia
1
il
S
^
P -0.500 -0.250 0.000 0.250 0.500 -0.240 -0.120 0.000 0.120 0.240
C
E 40 1
N 35 .
30 .
o 25 .
f
20 .
V 15 .
A
L 10 '
Uc
J ,
E
S
40
35 .
Kansas 30 .
25 .
20 .
(V W
ftffnrf
" 15 .
*
10 .
MM
^hrrsJ
^R
Kansas
i-xflh
1 i
n
» i*
T P
w
hf^J
-0.500 -0.250 0.000 0.250 0.500
-0.240 -0.120 0.000 0.120 0.240
40
35 ,
30 .
25 ,
20 .
15 .
10 .
5 .
40
35 .
Vermont 30 .
25 .
4
n^T
20 .
15 .
10 .
WWW N
mflkJj
a
"
1
1
-A
Vermont
'
y.
\fi
-0.500 -0.250 0.000 0.250 0.500
-0.240 -0.120 0.000 0.120 0.240
FIGURE 5-11. Derived weekly-measured weekly data for nitrate and ammonium ions
in mg/1. The curve represents the Gaussian distribution with the same mean and
standard deviation as the displayed data. The extreme bars represent values
at the given abscissa and beyond.
5-25
-------
CALCIUM
PRECIPITAION
40 .
35 .
30 .
25 .
20 .
15 .
10 .
5 .
P
E
V
0
/
A
^
Georgia
n
*
^ -0.160 -0.080 0.000 0.080 0.160
C
E 40 i
N 35 .
T
30 .
o 25 .
f
20 .
V 15 .
A
L 10 '
U 5 .
E
S
Kansas
t-
NN
* ^* m
n pi I IT "
P
N
J _
""itm
-0.160 -0.080 0.000 0.080 0.160
40
35 .
30 .
25 .
20 .
15 .
10 .
5 .
ai
p
If
I
1
1
Ji\
Vermont
\
A_
i
I
IL
40
35
30
25
20
15
10
5
Georgia
-0.240 -a 120 0.000 0.120 0.240
40
35
30
25
20
15
10
5
Kansas
-0.240 -0.120 0.000 0.120 0.240
40
35
30
25
20
15
Vermont
-0.160 -0.080 0.000 0.080 0.160
* S. N N
.3 .
-0.240 -0.120 0.000 0.120 0.240
FIGURE 5-12. Derived weekly-measured weekly data for calcium ion in mg/1 and
predipitation mass in kg. The curve represents the Gaussian distribution with
the same mean and standard deviation as the displayed data. The extreme bars
represent values at the given abscissa and beyond.
5-26
-------
data. The collocated sampler differences (Figures 5-2 through 5-7), appear to be
symmetric with more values clustered about the center and longer tails than
Gaussian distributions. The median or the mean method difference may be called
the method bias. Relative differences between derived weekly and measured weekly
concentrations for the four major ions are shown in Figures 5-13 and 5-14. Most
of the relative differences are within plus or minus 20 percent. From a
comparison of these plots with those in Appendix C, it appears that the larger
relative differences occur at low depositions.
The median or mean method difference may be called the method bias. These means and
medians and relative bias (mean bias divided by mean derived weekly concentration)
for all the precipitation types combined are listed for the three sites in
Table 5-6. Statistically significant biases according to the Wilcoxon test are
noted in the table, and Figure 5-15 provides a graphical summary of the relative
bias for each analyte at each site. The biases are statistically significant for
most ions at the Georgia site. All of the significantly different concentrations
are higher for the weekly samplers. Possibly the ion concentration is increased
by some evaporation of precipitation from the weekly collectors or the warmer
Georgia climate causes weekly samples to be more unstable. On the other hand, the
weekly samplers in Georgia actually collected slightly more precipitation, so the
differences may be attributable also to differences in collection efficiency. But
this factor would be independent of the sample collection time, and would reflect
a possibility of catching more of the initial rainfall which is often more
concentrated in constitutents.
At the Kansas site the biases are both positive and negative. Although some of
the relative biases are large, none is statistically significant because of the
large measurement variability associated with these ions. For hydrogen ion the
mean concentration, 0.016 mg/1, for the Kansas site (Table 5-5) is the smallest of
the three sites. In general, the lower the hydrogen ion concentration the lower
is the specific conductance of the solution. Low conductivity increases the
difficulty to make accurate pH measurements and yields poor precision (Table 5-4).
For potassium, the other ion of large bias, many of the concentrations are near
the analytical detection level which also results in poor precision for the
measurement. At the Vermont site, the daily samplers contained slightly more
precipitation than the weekly samplers (Table 5-2); the difference is statisti-
cally though not practically signficant. There are statistically significant nega-
tive biases only in the concentrations of potassium, calcium, and magnesium ion.
5-27
-------
p
E
R
C
E
N
T
p
1\
L
T
I
V
E
B
I
A
S
HYDROGEN Key
50 o x »o o o o o o + Georgia
40 .
30 .
20 .
10 .
0
-10 .
-20 .
-30 .
-40 .
-50 .
x o Kansas
x Vermont
o
0 0
*
0+ *° $X* v
x x x++ x + 0 x x
X P n T«^ * Q. X x X n
O ** T v ^ X O
V * V JA ft" A
x
X
0 * X
+ X
0
+ n n
1 1 1 1 1 1 I 1 I 1 1 I 1 1
SEP OCT NOV DEC JAN FEfl MAR APR MAY JUN JUL AUG SEP OCT
SULFATE
50
40 .
30 .
20 .
10 .
0
-10 .
-20 .
-30 .
-40 .
-50
t
0
0 °
o
X oxxx * ° o xXxx
v v* ° * + 4?xX * vX 0
* * u° +* * O * v»' Ok TT i O \f
o x *+ * 000xxx^»°f xx
x* n XX X* * * ° °
f ° x 0 X
+ x o
XX o
I- + 0
X 0
o
nv
i i r i r i i i i I i i i i
SEP OCT NOV DEC JAN FEB MAR APR MAY JUN JUL AUG SEP OCT
FIGURE 5-13. Weekly percent relative method bias for hydrogen and
sulfate. Differences beyond 50 percent are truncated.
5-28
-------
p
E
R
C
E
N
T
R
1
k»
T
I
V
E
B
I
A
S
NITRATE Key
50 + Georgia
40 .
30 .
20 .
10 .
0
-10 .
-20 .
-30 .
-40 .
-50 .
x o Kansas
x Vermont
X 0 o X
X + X ° X .
..J °* V 0 « n + + °S X » X X ,
"n + j.fl+ * _ + X *v*X '"8 x -T-
X * + y4.° nX *- « X * *+ X n OX*
0 + ° X 0 * *
0 +X * 0+ * °°
0
Xx v
+ 0
*
y
£>
I 1 1 1 1 1 I 1 I I 1 1 I I
SEP OCT NOV DEC JAN FEfl MAR APR MAY JUN JUL AUG SEP OCT
AMMONIUM
"l
40 j +
30 .
20 .
10 .
0
-10 .
-20 .
-30 .
-40 .
-50
v
O
XX *
X* ° *
x o
X OX
x ° o + « * o *x xo x x-
X °x x°° ,x° x xx
« « 00+ X
+ ° v + Ov+ - t X0 +" °
x x t 6 T
J. V + ^ O
x ° * x
* * x 0 + 0
X *+
o
x + +
X
f + V V V
III 1 1 I I 1 I I I I I I
SEP OCT NOV DEC JAN FEB MAR APR MAY JUN JUL AUG SEP OCT
FIGURE 5-14. Weekly percent relative method bias for nitrate and
ammonium. Differences beyond 50 percent are truncated.
5-29
-------
<-O
1
in
0)
JD
fO
1
to
LU
J
O_
SI
*I
t/j
j
J
tf
or
o
Li-
* *
^
o^
E
*fc-~*
2T
O
i i
1
-
_J
UJ
LU
S
^J
LU
Q£
ZD
55
^
LU
27
1
^~
LU
LU
3
O
LU
>
LU
^^
"*~~'
t/}
^
CO
o
:r
i
LU
3E_
O) -
? ^ *t r-
TO « * ^< in o
r ^-^
4-> CM *-O
TO OO 1 *-H
i TO
Q) !-
Cr- CO
^
in CM
TO c i" o
CO TO TO ^- O
C O) -
TO ET CO r O
^ CO
'
c O
TO oo o
r- TO CM O
O) CO O O
SI
TO
OJ '~-
> »* CO r-~
4_> O ' '
TO W 1 '
, TO
OJ -i
C£ CD
TO C5
i- 0
cn c w CM o>
t_ TO TO
O Ol *i~ CTi O
01 s: co ^H i
CO
c o
TO OO O
i- TO in o
OJ CD *3- O
s: co i
'
^_^
in
E
TO
OJ 01
1 < ^ J
>l D.
TO O
1= OJ
Q- 3:
co
f 1
1
10
CO
0
o
1
CM
0
o
o
r^
1 1
1
0
CO
o
0
1
1 1
^>
o
o
1
CM
,3-'
1
CO
in
o
o
3
CO
o
o
1
II «r
o
CO
o
IO
o
0
0
in
o
o
o
en
CM
1
-3-
0
o
1
o
CM
o
o
r-H
p^
1
in
m
o
o
i
3
CO
o
o
'
1 CO
o
z
o
o
o
,1
o
1
^J.
o
o
o
o
1
in
o
o
o
in
3-
i
oo
0
o
0
1
2
t i
( (
O
o
1
+ *&
3:
zz
r-~
Q
I
^H
O
0
o
1
CO
o
o
o
in
l-H
CM
O
O
o
CO
o
o
o
in
CO
1
«3-
i*
0
o
1
3
01
o
0
o
1
1
,
o
0
CO
1
I 1
0
0
o
1
1 1
o
0
o
in
o
1
CO
0
o
o
o
i
CO
o
o
0
o
1 1
oo
i
oo
i i
o
0
1
2
l-H
O
o
0
1
+
TO
Zi
co
CO
CO
1
in
o
0
o
1
3
CM
O
O
o
i
**
o
CM
CTi
O
O
0
1
^ t
o
o
o
1
CM
r~*
CM
o
o
0
0
o
o
o
1
+
^
CO
r-~
1
oo
o
0
o
1
3
o
r-*
|
10
1 t
o
o
o
1
3
01
o
o
o
o
1
t f
CM
i 1
1
in
o
0
o
i
«3-
i i
O
O
o
1
T-<
in
i
ID
, <
O
O
0
1
00
0
o
o
o
1
CM
en
s:
.
4-^
oo
OJ
4J
CO
c
TO
c_
o
01
en
oo
c
o
X
0
u
r
5
c
o
"^
CD
in
fO
.0
00
c
o c
r- O
TO 4J
t- TO
(-> U
C 4->
OJ C
O OJ
C 0
o c
u o
X)
OJ >>
(_ 0)
T3 3
-c
oo OJ
> >
>> C
r OJ
OJ
OJ C
S *- II
o oo
TO
*-> -r-
C J3
TO
O CJ
r- >
<4- T
C TO
O) r-
,- 01
CO C£
II '
TO
IZ * '
5-30
-------
c
0)
u
c
o
u
Ol
OJ
T3
O)
i
O)
TD
ro
OJ
fC
-Q
CO
£Z
ro
01
O
O
l/l
ro
T3
o
QJ
Ol
>
O)
D:
OJ
o
S-
O)
Ol
S-
5-31
-------
The percent biases in Figure 5-15 are predominantly negative; i.e., the concen-
trations are higher in the weekly samplers. The only case in which concentra-
tions are much larger in the daily samples is for H+ at the Kansas site (16 per-
cent). The largest relative biases are observed for K+, especially at the Kansas
site (-20.4%) and Vermont site (-33.8%). At all three sites concentrations of
Ca + and Mg are also relatively large in the weekly samplers, typically about
10 percent larger, which might be attributable to contamination of weekly samples
by re-suspended soil particles. Relative precipitation bias is practically
insignificant at all three sites, no larger than 2.5 percent. However, small
rainfall events (less than 0.51 mm) are not included in the derived weekly pre-
cipitation amounts but they are included in the measured weekly precipitation
amounts, so there is actually a smaller bias in precipitation amount than is
indicated in Table 5-6. Evaporation alone cannot explain the observed method
bias. Other feasible explanations are contamination during the seven-day
storage, potential contribution of highly concentrated trace samples (which are
included in the weekly samples but not in the daily composites), and instability
of weekly samples.
Table 5-7 compares the mean and relative mean bias for rain samples separately
and all precipitation types combined. The Georgia site has relative bias values
under 10 percent for all the observables except Ca2+ for which the bias is 11.5
percent. For Kansas, the relative bias for rain sample values is 6.0 percent or
less, except for H+, for which it is 18 percent. If all the precipitation types
are considered, results similar to those for rain are obtained, except for Cl~,
K+, Ca , and Mg ; for Cl~, the relative bias is now smaller, whereas for the
cations it is larger. For Vermont, the relative bias values for rain samples and
all precipitation types combined are similar, and except for K+, are less than 11
percent. One-way ANOVAs with the Welch test indicate that for rain samples there
are significant differences in the mean bias among the sites for Na+ only, where-
as when all precipitation types are combined, significant differences among the
sites occur both for Na+ and for H+. These results indicate that there is little
difference among the sites in the bias for any of the observables except for Na+
for rain samples and for Na+ and H+ for all precipitation types. In addition,
except for K+, the differences between weekly and daily composite samples are
generally under 12 percent, and for the major ions (H+, NH., SO" and NOI) they
are typically much less (H+ for Kansas is the exception). These results indicate
that errors in annual average concentrations due to method bias are likely to be
5-32
-------
oo
UJ
i I/"
C
01
CD O
o o
O OJ
z: >-
o o
o o
o o
et
OO
or
o
+J
c
c
E
!_
OJ
r-.
i
in
O)
XJ
ro
oo
f.
CO
2:
o
I
I
z.
UJ
CJ
2:
o
UJ
ti
t
<
_l
UJ
C£
Q
et
LU
I
00
U-
o
o
oo
eC
Q.
in
c:
ro
ro
01
O
Ol
dj ^~-
> 5*
4->
re in
i ro
CO -,-
a: co
c in
re re
r *
4->
ro t/1
i fC
c
o
i I i 1 ID ID «d~
o o in oo CM
o o o o o
o o o o o
1 1 1
O O CM P-. ID
CO ID CM I-H i-H
i-H ^H 1 1
CM CM CO O CM
O O 00 OO CM
O O O O O
O O O O O
1 1
P-- CM r- 1
I-H "=f r--
1 1 1
o oo in
o m in
o o o
000
i i i
c c c
ro i re i ro
rv e^ fV «=C fv
II ** 1 00
+ 0 O
:n oo -ZL
oo
o
10
o
o
0
CM
O
0
1
5
i i ^~ CTi r^~
CM CM O O
1 1 1
CO r I-H I-H
O O O O
O O O O
o o o o
1 1 1
CM
o i i o in
o o m I-H
i i
i i «d~
O O P^ CM
O O O O
o o o o
o o o o
1 1
in in
«a- oo
CO *
0 r-l
0 0
0 0
i i
c c
ro r ro i
Q: =c o; <:
+ «d- i
z: o
^- o
ID OO
OO i-H
O 0
0 O
o o
in in
-H 0
i
oo
-H O
0 0
0 0
0 0
1
I-H
CO
CO
! 1
0
o
1
c:
ro r-
o; =c
ro
5-33
-------
o
LO
' O)
o o
o o
o o
en
oo
-4_J
C
o
t_
01
>
o
0)
T3
3
O
f
r-j
LO
Cli ^3
LO
r- c
1 n
LO ^
OJ
TO
TO
TO
r~
01
O
O)
CO
Ol
>
"-
^J
TO
i
01
cz
C.
01
s:
OJ
>
TO
i
Ol
C£.
C
01
0)"
>.
.,
-4_>
TO
r
QJ
C
TO
01
2:
. .
^S
LO
)
r~
TO
C
H.
U3 CO r}- 00 LO to
» I CO LO 1 CT* CZ3
CO CO II 1 <-l
1 1 1
<£> LO r-. OO CM CM
00 0 O O 0
00 0 O 0 O
o o o o o o
II II II
LO *3" LO CO O t~f
CM O LO O LO CM
1 CM 1 i I 1 . I
1 1 1
i < O1 CO CO CM LO
O O CM =* O O
00 00 O 0
O O O O O O
II II II
CM LO i 1
r ti LO
1 , 1 1
1
CM r-l CM
0 <-l 0
O O O
000
1 1 1
CZ C C
I r *f r ! r
re i re ' re i
Qi < o; <: oi <
CM CM
-)~ -|-
+ TO CT>
^ O ^
C
o
r
TO
+J
C"
<_>
O
o
>^
r
vx
d)
OJ
3;
>, 0)
C -r-
O t-
01
c -c
TO C
C- TO
01
0) E
t_ -\
O) LO
5 TO
TO _O
r
CJ> CI
£_ TO
O Ol
O) E
X
c
r- O
0
T3 i-H
Ol
!-> II
U
OJ ^ *
! J^
O
O LO
re
LO -i
Ol _O
1
Q. Ol
E >
TO !-
LO +->
TO
i Ol
^t f*S
*~-~ * »
03 _a
- -
^"F
o
C
LO
O)
o
T3
O
i
j~
3
*i
4_)
LO
O)
^
o
"oj
^
0)
"^
>^
o
o
Ol
^_>
o
0)
Ol
Ol
t-
TO
CO
ro
1
_o
=c
c: >
re o
O) Z
e
O) 10
O Ol
C 0
0) C
«- TO
Ol -i-
<4- TO
i- >
-D
t
j-> TO
c rs
ro O"
O O!
r
l»- 0)
.- e
C ZJ
en LO
i- LO
t/) TO
^ »
o
*
5-34
-------
small, and therefore do not seem to have much practical significance. Further-
more, since method bias is more or less constant from site to site, conclusions
about the geographic pattern of concentrations are not likely to be affected by
the choice of method if these sites are representative.
Method Bias By Season
Although there is obviously some confounding of the effects of precipitation
type, site (there is no snow in Georgia), and season the effect of season on the
concentration bias was also examined for the three sites individually and for all
sites combined. For the fall quarter, the 1983 and 1984 data were combined. The
biases for the weekly derived minus weekly measured concentrations together with
the respective Wilcoxon test results are presented in Table 5-8 for all precipi-
tation types and all sites combined. Significant bias is seen for nitrate and
calcium in the spring but for no analytes in the summer or winter quarters. How-
ever, there is significant bias for H+, SO" Cl~, Na+, K+, Ca2+ and Mg2+ in the
fall. The measured weekly concentrations are larger than the derived ones in all
cases of significant bias except for H+ in the fall quarter. Higher Ca and
Mg +, which generally come from basic soil dust, are expected to yield lower H+
in the weekly samples. Even if the Na+ and H+ results are confounded by site
effects, the results still indicate that a majority of the analytes show signifi-
cant bias in the fall. The significant fall season biases are difficult to explain.
It is expected that the largest number of freeze-thaw cycles would occur in the fall
and that may affect some gas and ion solubilities.
In Table 5-9 the bias and relative bias for rain and for all precipitation
samples are summarized by season for all sites combined. In accord with the
results in Table 5-8, the largest relative bias for all the observables except
potassium in rain occurs in the fall. For the four major ions, the relative
biases are under 7 percent in rain samples and 13 percent in all samples. ANOVA
tests show that there are no significant differences in bias due to season for
all observables except Na+ and Mg in rain samples and Cl~ in all precipitation
samples. Thus the concentration bias between weekly and daily samples is similar
across seasons for most of the constituents.
5-35
-------
00
I
in
_a
ro
o
oo
c
ro
CU
c
ro
T3
CD
E:
CO
ro
.,
CO
CO
ro
03
CO
ro
i
EC
CO
ro
>r_
cQ
CO
ro
>r
CO
CO
ro
03
CO
ro
>r
CQ
V)
ro
i
02
OJ
^
ro
^j^
CM .-(
O
LO O
CM
O
3
** CM
0
0 O
O
O1
i l i l
O
=3- O
O
1
CO O
o
0 0
1 I
o
o
0 0
0
i l O
1 1
1 0
ro
0
LO O
o
1 (
1 O
. r
1 i l
O
^ 0
1 1
1 0
1
o
OO CD
0
"3- O
1
O
CO
E
ro
CJ)
CL
U
OJ
£- +
a- m
CM
CM
0
*
O
-^J
oo
0
0
01
LO
0
*
o
1
LO
oo
o
o
^o
I 1
f 1
*
o
1
CO
CM
o
o
o
o
o
CD
1
CM
CM
CD
*
CD
1
II *
O
^>
CM
t 1
0
1
f^
CM
0
O
^J-
1
O
o
1
I <
CM
O
o
1
CM
f-
0
O
1
3
i l
^J-
O
CD
O
LO
CD
CD
oo
o
o
o
1
i oo
o
z
o
0
o
1
^.
o
o
0
LO
ro
o
0
1
1 1
. 1
0
0
1
^J-
0
o
o
1
o
r->
o
o
o
. 1
CD
0
1C
0
o
0
~t~ *3"
3:
z.
r~~
oo
0
o
1
o
1 1
o
0
ro
o
0
o
1
CM
0
o
0
^3-
o
o
«
CD
1
oo
0
0
o
1
LO
0
CD
*
0
«^-
0
o
o
1
,
CJ
oo
CM
0
o
1
CM
o
o
0
1
1 1
o
o
0
1
1 1
o
0
o
ID
o
o
0
1
0
o
o
o
0
CD
o
1
o
0
o
0
+
ro
z.
CO
CM
o
o
LO
o
o
o
CM
r-^
o
o
l 1
o
o
o
o
o
o
%
o
r-H
o
o
o
ro
o
o
0
1
CM
0
o
o
+
N^
to
IO
0
o
1
3T
CO
1 1
c
0
I
co
VO
o
o
00
0
0
c
1
II
CM
o
CD
1
3
01
o
o
o
o
0
o
1
oo
0
o
o
+
CM
(C
0
CO
o
o
o
1
CO
0
0
o
CO
. 1
o
o
1 I
0
0
0
CM
o
o
o
1
1 1
o
0
o
1
o
CD
CD
0
1
oo
CD
0
o
o
1
+
CM
01
s:
-p
a>
CD
oo
ro
-------
CO
c-o
CO -
ro LO
O> O
co
O
Cr.
c
O ^~-
E
=£
O O
O O
O O
O O
co co
O CD (DO
Z >- >- "Z.
O
00
oo
>-
Q-
s:
00
Oi
>
CTi
IT)
(1)
=t
or
E
oo
tI
CD
O
II
<;
T-
or cc
c co
ro ro
O) -r-
s: CQ
QJ
ro co
r ro
01 -
ct cc
c to
ro ro
'
O
1
O
o
CD
0
O
O
,
"*
LO
ID
1
LO
O
o
1
00
CO
^J-
f-^.
CD
O
1
LO
Cvj
'
«3-
CVJ
O
O
CVJ
r-H
1
CO
0
0
o
1
c
ro
o;
1 00
o
2:
o
CTi
t
^~
cvj
i t
o
1
r-.
to
CVJ
r^.
0
0
LO
Cvj
1
CD
LD
0
O
l
^
CO
ro
t
CVJ
.1
O
O
0
r--.
LO
00
o
o
1
ro
oo
LO
UD
oo
o
, i
CD
f-H
O
O
O
c
ro
o;
+ «d
~r~
^
^i-
CVJ
r-<
'
O
^J-
o
0
1
0
. 1
1
0
! 1
o
o
1
^f
^t-
0
1 1
o
0
,
=c
LO
^f-
, 1
^J-
ro
o
o
1 t
v 1
ro
0
o
0
1
*±
o
r 1
0
o
o
1
^f-
00
1 1
o
o
c
ro
o:
i
P
o
ro
i
i i
i
r*^
oo
0
o
1
ro
CVJ
'
^}-
0
0
0
1
1 1
CsJ
LO
0
0
o
_
«=c
CT) LD
IO O1
i l i l
1 1
CVJ CO
CVJ CVJ
0 0
0 0
1 1
CVJ
( t
1
1 1
o
0
o
1
vD CTi
^J- LO
1 1
IO vO
o o
o o
0 0
1 1
LO ro
CD ro
1
1 1 *d~
0 0
o o
0 O
1
c
ro i
+
ro
^.
5-37
-------
-a
cu
T3
o
o
cn
!
UD
CU
CO
c
oc
<4- CO -
4 to ^
i- 0) 0
O CO
O
CT.
cn e
r- C ^~-
to E
et
CU
> ^
.r- as
rc co
_O i ro
i 0.' -r-
r cr cc
ro
C CO
fO 'O
r cc
a1
r- S*
i * ^_^i
rc co
(C r - to
£- CU !-
cu or cc
E
3!
CO
C CO
re ro
a !-
21 CQ
a>
r- S«
rc co
p to
cn cu -r-
c: ctr cc
r
£_
a.
CO
C CO
to to
2EI CC
(DO
r- 9<
rc CO
L. i rc
CU CU T-
+-> or CQ
c:
r
3.
C CO
CU -r-
2: CQ
CU
1
jQ
to
>
t_
CU
CO
o
o
0 0
2^ Z
r- o
CM ai
CM i-H
1 l-H
1
LO CO
O CM
o o
0 O
1 1
CM
CM
to
CM
r-.
o
o
ro in
CM i-H
1
1-H 0
o o
o o
o o
l-H **
-H CM
i I i 1
I
CM CO
O O
o o
O 0
1
c
1 r
tO r
o; =t
+
CO
o o cu o
z: 'z. > z.
CO ^f tO CM
CTi CO r~~ .-i
l-H CO l-H CO
II II
LD to to CO
^10 O O
O O O O
0 O 0 O
II II
f ^ ^^
tO r-1
-H CM
CO CO
tO l-H
0 O
O 0
* LO CTl CO
CO CO IT) CO
II II
Ol i-H CM CM
O CM O O
o o o o
0 O O 0
II II
i-H <~- i-H CO
ir> co I-H CM
1 t 1
co r-~
CO to O O
o o o o
o o o o
o o o o
1 1 1
cr c
r- r *i r
fO r tO I
CC. <=t. cr: =£
CM CM
-^- -j-
to cn
o s:
5-38
CO
c
>, 0)
.- E
c cu
O t-
C CO
i- ro
t_ E
CU i
t_ r
CU 10
5 1-
t- «i-
§co
CTl
1 1
3
CO T3
c
QJ to
-C
^ CO
CO
c cn
r- r 1
cu e
-(-> !-
O o
CU E
i O
r U
o
O to
4_>
CO tO
cu -o
r
CX r
E r
fO fO
CO M-
r CU
i x:
cC r-
^ «. x^
rd ^3
- -
c
o
r
p
T3
I_
+J
c
cu
o
0
1
_£
cu
cu
-rj
cu
r
S-
CJ
"O
c
ro
cu
E
CO
to
>
-O
c
ro
OJ
E
X
0
o
II
f X
^^
X f
CO
to
tf
&
cu
>
1
+J
to
r
cu
^ *
o
.p
o
CO
cu
o
T3
.c
o
.,_
-C
3
^J1
CO
cu
4->
jr
o
3
QJ
^3
>)
_o
-rj
cu
0
cu
^
cu
-^u
cu
to
CO
to
>f_
.0
=c
c >
to o
cu z.
E ^
cu cu
.c .c
c t.
i O
(4_
(/I
cu co
O d)
c o
cu c:
t- to
CU !-
t- t-
tf- (O
t- >
c
4-> ra
c: 13
1C O"
o cu
1
!*_ CD
r- E
C 3
cn oo
r- CO
CO IO
ff ^
T3
-------
COMPARISON OF DAILY AND WEEKLY DEPOSITION
Deposition is the mass of a constituent deposited per unit area, and it is of
prime importance in the study of effects on the environment. It is not a mea-
sured quantity, but the product of two measurements, the analyte concentration
and the precipitation depth. Thus errors in deposition values arise from errors
in both measurements. The deposition results discussed next consist of precision
data and an evaluation of annual and seasonal values from both daily and weekly
concentration measurements. The precision of the weighing bucket rain gauge was
also calculated to help analyze the deposition results.
Rain Gauge Precision
Collocated rain gauges were installed at the Kansas and Vermont stations. These
sites were chosen to test the rain gauge response to the three different precipi-
tation types. Only daily data were evaluated since weekly precipitation depths
were obtained by summing the daily values. Summary statistics for rain gauge
bias and precision are presented in Tables 5-10 and 5-11, respectively. At both
sites, snow samples show the greatest bias, relative bias (10 to 15 percent), and
RSD (11 to 15 percent), as expected. Snow capture losses occur from wind
effects, as with the samplers, and these losses are not the same for each
gauge. On the basis of one-way ANOVA, the Kansas site shows no significant dif-
ferences among the three precipitation types for either bias or standard devia-
tion, while the Vermont site has significantly different paired differences among
the different precipitation types. The bias in the precipitation depth measure-
ments for the collocated gauges at the Vermont site is significant for all pre-
cipitation types combined but not for any specific precipitation type. These
results indicate that there was a problem with at least one of the gauges at the
Vermont site.
The effect of season on rain gauge measurements for both sites is shown in Table
5-12. Comparisons based on separate variance t-tests were made between the two
sites for each season, and between seasonal pairs for each site. Results of
these comparisons showed a significant difference between the distributions of
the two sites for winter only. No significant seasonal differences were detected
for the Kansas site. Significant seasonal differences in amount of precipitation
5-39
-------
Table 5-10
RAIN GAUGE BIAS (cm) BY PRECIPITATION TYPE
CALCULATED FROM DAILY MEASUREMENTS
Precipitation
Type
Bias
RD
SD"
RSD (%)
Lancaster, KS
Rain
Snow
Mixed
All
70
4
14
88
0.0109
0.0381
0.0145
0.0127
0.9
10.5
1.3
1.1
0.1363
0.0790
0.0667
0.1251
11.0
21.8
6.4
10.7
Underfill 1, VT
Rain
Snow
Mixed
All
87
30
24
141
-0.0321
-0.0694
-0.0096
-0.0362W
-3.1
-14.5
-1.0
-3.9
0.1015
0.0763
0.0937
0.0966
9.8
15.9
8.7
10.5
(a) Bias = d, mean collocated difference.
(b) RD = relative difference, d/R, where R is the average rain
gauge depth.
(c) SD = standard deviation of the differences.
(d) RSD = relative standard deviation, SD/R".
W = Bias significantly different from zero based on the
Wilcoxon signed-rank test.
5-40
-------
Table 5-11
RAIN GAUGE PRECISION (cm) BY PRECIPITATION TYPE
FOR DAILY SAMPLES
Precipitation
Type
Rain
Snow
Mixed
All
Rain
Snow
Mixed
All
RS
MACD
Lancaster, KS
0.0960 7.8
0.0554 15.3
0.0466 4.5
0.0884 7.6
Underhill, VT
0.0749 7.2
0.0722 15.0
0.0652 6.1
0.0728 7.9
0.0000
0.0254
0.0000
0.0000
0.0507
0.0508
0.0254
0.0266
RACD (%)
0.0
7.1
0.0
0.0
6.6
16.0
4.5
4.8
(a) SD = £ d;
kl/2
, pooled standard deviation.
(b) RSp = S /R, where R is the average rain gauge
measurement.
(c) MACD = median absolute collocated pair difference.
(d) RACD = MACD/median rain gauge depth.
5-41
-------
Table 5-12
SEASONAL DIFFERENCES IN RAIN GAUGE BIAS
Kansas
Season
Winter
Spring
Summer
Fall
Sig. Seasonal
Diff.d
RD(%)a
3.0
0.2
4.4
-0.9
No
RSD(/
6.6
3.0
13.2
1.5
^b N
13
32
18
24
Vermont
RD(%)
-8.8
0.6
-5.5
-2.9
Yes
RSD(%;
7.5
8.0
2.6
8.4
) N
37
36
27
41
Significant
Site Diff.c
Yes
No
No
No
(a) RD = bias/R, where R is the average rain gaujje depth.
(b) RSD = standard deviation of the differences/R.
(c) Pairwise comparisons based on separate variance t-test.
(d) Significant differences detected by a one-way ANOVA using the
Welch test.
5-42
-------
collected by the two gauges were detected at the Vermont site. As expected, the
differences are greatest in the winter quarter, which contains primarily snow and
mixed events. The observed differences at the Vermont site, particularly in the
winter, can be attributed in part to the location of the site on a hillside plat-
form, which is subject to more drainage winds down the face of the hill. Snow
capture efficiency, especially with high winds, is much worse than that for
rain. All but one of the weekly samples in the winter in Vermont were either
snow or mixed.
Daily Deposition Sampler Bias and Precision
The bias and precision of the daily deposition data was calculated from the dif-
ferences between the product of the values of collector 1 and rain gauge 1, and
the product of the values of collector 2 and rain gauge 2. Summary statistics
for sampler deposition bias and precision are given in Tables 5-13 and 5-14,
respectively. For the Kansas site, all the ion biases have a negative sign.
This occurs even though rain gauge No. 1 collects 1.1% more precipitation than
rain gauge No. 2 (see Table 5-10). For the Vermont site, the ion biases, except
for potassium and magnesium, are also negative and are in accord with the rain
gauge bias.
Table 5-13 identifies bias values that different significantly from zero, as
detected by the Wilcoxon test. Collocated deposition values for all ions except
K+ are significantly different at the Vermont site; this may be attributable to
the significant differences in the rain gauge amounts (Table 5-10). No median
deposition differences at the Kansas site are significantly different from zero.
Daily deposition precision, using the relative absolute collocated difference
(RACD) as the measure, is generally poorer than daily concentration precision
(see Table 5-4). This is to be expected since deposition differences are affec-
ted by both concentration differences and rain gauge differences. This is not
always the case at the Kansas site, however, where the median difference between
the two rain gauges is 0 (Table 5-4). At Kansas the daily concentration pre-
cision (RACD) is higher than the daily deposition precision for H+, Ca , and
Mg2+.
5-43
-------
Table 5-13
DAILY DEPOSITION SAMPLER BIAS (mg/1) FOR ALL
PRECIPITATION TYPES COMBINED
Analyte
H+
so;
3
NH4
Cl"
Na+
K+
CA2+
2+
Mg
Site
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Bias3
-0.032W
-0.553
-1.185
-°-294w
-0.983W
-0.145
-0.123
-0.068
-0.044
0.135
-0.027
-0.022
-0.018
-0.092
-0.028
-0.0095
0.014W
Std.
Dev.
0.043
0.114
2.367
2.502
2.339
2.207
0.629
0.897
0.596
0.232
0.793
0.125
0.141
0.325
1.151
0.468
0.080
0.259
RD (%)b
-5.0
-7.0
-2.5
-5.8
-1.9
-7.3
-2.7
-4.4
-4.5
-5.6
-13.9
-8.7
-4.5
-10.3
-2.0
-3.3
-2.4
9.2
RSD (%)c
19.7
25.1
10.9
12.2
15.3
16.5
12.0
31.8
39.3
28.4
82.2
40.2
28.6
183.0
24.7
55.2
20.2
171.0
N
72
127
72
125
72
125
72
119
72
125
72
115
72
114
72
114
72
114
(a) Bias = d, mean collocated deposition difference.
(b) RD = relative difference, d/D, where D is the average daily
deposition amount.
(c) RSD = relative standard deviation, standard deviation/If.
W = Significant deposition bias based on the Wilcoxon signed-
rank test.
5-44
-------
Table 5-14
DAILY DEPOSITION MEASUREMENT PRECISION (mg/1)
Analyte
H+
S04
3
NHj
Cl"
Na+
K+
Ca2+
Mg2+
Site
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
Kansas
Vermont
S a
0.031
0.084
1.71
1.95
1.65
1.70
0.45
0.64
0.42
0.17
0.56
0.09
0.10
0.08
0.81
0.33
0.056
0.18
RSp (%)b
14.2
18.3
7.9
9.5
10.8
12.6
8.6
22.6
27.7
20.3
58.6
28.8
20.3
45.7
17.4
38.9
14.2
120.4
MACDC
0.007
0.029
0.744
0.833
0.46
0.85
0.31
0.11
0.10
0.07
0.28
0.03
0.04
0.02
0.21
0.06
0.019
0.008
RACD (%)d
6.6
14.6
5.2
11.6
4.0
12.0
11.1
11.5
14.6
17.9
13.2
32.6
19.5
30.8
7.7
23.3
9.1
20.5
/ N 2 \l/2
(a) sp = \JL dt/2NJ , pooled standard deviation.
(b) RSp = relative S , S /D", where TJ is the average daily
deposition amount.
(c) MACD = median absolute collocated deposition
difference.
(d) RACD = MACD/median daily deposition amount.
5-45
-------
Annual and Seasonal Deposition
Deposition amounts were calculated for the one-year study period for Kansas and
Vermont, and for approximately 11 months for Georgia, by summing both the mea-
sured weekly and derived weekly deposition amounts. The averages of the two
daily and the two weekly sampler concentrations were used, but only the routine
UAPSP rain gauge at each site was used to calculate deposition amounts. This was
done to control for rain gauge variability at a given site and to enable assess-
ment of bias in the derived deposition that can be attributed solely to different
sampling schedules. The calculated deposition amounts are unadjusted for missing
data and are therefore not directly comparable. The missing values generally
occur in the daily deposition data. In addition, trace events are included in
the measured weekly depositions but not in the daily deposition data. The
measured weekly values are therefore a more reliable indication of actual
deposition amounts. The total depositions are listed in Table 5-15 for the three
sites. Measured weekly and derived weekly deposition values for the four major
ions and calcium are plotted on a weekly basis in Appendix C. There is more
acidic deposition at the Vermont site than at the Georgia and Kansas sites, as
evidenced by the relatively high H+, SCT, and NO" values. Deposition of Na+ is
by far the highest at the Georgia site, while the Kansas site has the highest
Ca deposition amounts. In general, the daily values are less than the weekly
values, but this may be attributable to missing daily data. We are at a loss,
however, to explain the very large difference between the daily and weekly totals
for Cl~ at the Kansas and Vermont sites.
Daily and weekly derived deposition values calculated separately for each site
for each of the four seasons are summarized in Table 5-16. Deposition varies
significantly by season, and the seasonal patterns differ by site. For SO^, for
example, peak values occur in the spring at the Kansas and Georgia sites, but in
the summer at the Vermont site. Peak deposition amounts for H+ occur in the sum-
mer at the Vermont and Georgia sites, and in spring at the Kansas site.
REFERENCES
1. W. J. Conover. Practical Nonparametric Statistics, 2nd ed. New York:
Wiley, 1980. Chapter 5.
2. D. L. Sisterson, B. E. Wurfel, and B. M. Lesht, "Chemical Differences Between
Event and Weekly Precipitation Samples in Northeastern Illinois." Atmos.
Environ. Vol. 19, 1985, pp. 1452-1469.
5-46
-------
Table 5-15
SITE ANNUAL DEPOSITIONS (mg/m2) BASED ON DAILY
AND UEEKLY MEASUREMENTS3
Analyte Site
+
H
S0=
NHj
Cl
Na+
K+
Ca2+
Mg2+
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
Georgia
Kansas
Vermont
17. 9
13.0
51.9
1188.3
1482.8
2333.6
691.8
1022.3
1647.0
211.1
366.5
304.5
414.5
112.5
97.7
225.9
72.7
36.5
47.4
35.6
30.9
111.3
346.4
85.4
34.4
28.9
14.8
18.6
13.9
53.5
1243.5
1719.3
2408.7
744.0
1218.7
1654.9
205.4
413.3
311.5
323.8
264.2
207.6
244.7
71.9
38.1
38.5
43.8
29.1
124.1
392.0
108.0
35.7
34.3
18.0
(a) For Georgia the depositions are for approximately 11 months.
Deposition amounts are not adjusted for missing weeks.
5-47
-------
LO
s:
LU
LU
UJ
O)
13
i
LO
(V
JD
re
-*
CD
OJ
re
o
^
01
CU
a
>
ii
a:
UJ
a
en
co
O
a.
LlJ
D
O
OO
CL
co
^t
ai
OO
CO
1
CM
r--
LO
LO
O
CM
CO
CO
CM
T (
CO
CM
LO
10
LO
CO
LO
r_
cn
t_
O
d)
CD
II «*
O
CO
l-~
CO
i I
*
ro
O
1 1
ro
LO
CM
^3-
, 1
LO
00
CO
rH
0
ro
LO
0
rH
^
LO
CM
CTl
i 1
LO
0
i 1
co
CO
o
1 t
CO
re
CO
c
re
LO
ro
LO
LO
CM
CTl
CO
"*
LO
LO
(~^
o
1 1
1 1
CTl
O
, 1
i i
CO
CTl
LO
LO
O
CM
OO
LO
ro
o
o
CM
OO
CO
o
CM
^
r;
t_
CO
rH
«£j-
00
<*
ro
LO
o
o
CM
O
p^
CO
rH
LD
O
LO
OO
O
t-H
ro
ro
rH
rH
LO
rH
O
CTl
CO
! 1
re
r
cn
C-
o
OJ
1 CO
o
2^
ro
CTl
CTl
CM
r-.
CM
, t
CM
LO
! 1
r^»
i i
o
j-j-
VO
1 1
r-~
CTl
CO
LO
LO
rH
LO
LO
CO
OO
o
1 1
1_1
^1-
CTl
CO
re
c
re
*
to
co
o
ej-
LD
LO
LO
CO
ro
CO
o
LO
ro
O
CM
LO
LO
CO
ro
o
LO
o
ro
00
01
LO
CO
ro
LO
"*
P
O
£_
OJ
>
CTl
CO
CTl
,^
CM
C
CO
LO
CO
CM
LO
LO
CO
, 1
CTl
CO
r I
r--
o
CO
CO
o
co
re
cn
t_
o
O)
cs
+ <*
DC
"Z.
LO
r~-
CTl
CTl
^-
t^.
^.
^-
ro
CO
,3-
ro
CTl
,3-
LO
CM
CO
0
OO
CM
LO
LO
CM
CTl
LO
CM
CO
re
CO
c
re
^
co
o
1
LO
LO
LO
CO
CO
^3-
1 1
LO
CM
LO
! 1
VO
0
r
,-f.
LO
ro
CM
CM
CO
. 1
oo
^
r;
O
£_
0)
>
r^
CM
CM
^_
ro
ro
CTl
CO
oo
CO
1^,
CO
^
CO
CO
1 1
1 1
« 1
CM
CO
CO
f^v
CM
cj.
CM
i I
re
r
co
£_
O
OJ
C3
1
r
O
ro
LO
LO
ro
^
CM
i 1
LO
CM
r-~
LD
rH
LO
r-.
LO
rH
rH
i I
VO
CM
U3
t l
LO
r
CO
re
CO
c
re
^
CTl
ro
^~
i
co
\ i
,3.
r--.
00
0
LO
CM
LO
^t-
«^-
CTl
OO
rH
r~-
1 1
ro
0
LO
OO
p
c:
o
E
L_
. CO
LO VO
f I
r^> CTI
r-~ co
rH *3-
(Yi fn
^\j wi
CO CTl
CO I--
co
CTl CM
rH 1
«H^
o *
r~~ LO
1 1
CTl !
LO ro
i i
^
CO C
TO O
CO £
c t-
(C CU
^ >
5-48
-------
13
C/)
o
o>
3
U
o
o
1JD
CU
%
o|
0)
Ol
Q.
OO
Q
01
+J
to
Q
01
IO
t-H
,_
"
CM
LD
^
LD
O
CM
CM
CO
CO
IO
LD
1^
r-.
(O
r-
cn
O
CM
I 1
CM
ID
^_
"*
LD
"*
1
CM
CM
cn
CM
CM
^.
CM
0
CM
Kansas
00
o
r 1
CO
**"
^
in
cn
r-.
rH
cn
CO
CM
1
1
CM
LD
i i
Vermont
ID
IO
CO
^
i (
CO
1 1
10
rH
00
t 1
il
^
0
CM
1
OO
t-H
LD
CM
LD
CO
CO
ID
CM
CM
. c
CM
^
O
o
CM
cn
cn
CM
CO
cn
CM
l/l
(tl
c
(O
l~~
i i
i
CO
CM
^.
CM
^
CO
ID
O
CM
cn
cn
i i
^
CO
^
OO
03
cn
t_
o
CM
cn
E:
1
CO
1 1
LD
CO
CM
CO
**
^
CO
1 1
1
, 1
cn
CM
^
CM
nl
c
to
CO
'
^
CM
ro
o
CM
CO
vo
^
LD
,-J-
co
^.
CM
CM
CM
Vermont
c
o
T3
CU U1
O en
c
en <-
c. i/i
i t/i
TO £-
E O
o z:
o
a
n3 c
-i-j re
(O
-o t-
o>
O
O
5-49
-------
3. L. E. Topol and R. Schwall. "Precipitation Chemistry Measurements in the
SURE Region." Draft Final Report. Palo Alto, CA: Electric Power Research
Institute, 198.6.
4. L. E. Topol and R. Schwall. "The Utility Acid Precipitation Study Program,
Network Description and Measurements for 1981-1982." UAPSP report 105.
Washington, DC: Utility Acid Precipitation Study Program, 1986.
5-50
-------
Section 6
CONCLUSIONS
The study described in this report compared the results of weekly and daily
precipitation samples. The following conclusions were drawn from the analyses:
The collection efficiencies were highest for rain and about
equal for snow and mixed precipitation with a value of about
1.1 for rain and 0.75 for mixed and snow. Generally small
differences were observed between weekly and daily sampling and
among sites for the same precipitation type.
The precision of ionic conentration determinations is either
comparable or better for daily than for weekly samples. Over-
all measurement precision for both sampling protocols (with the
exception of K+) is less than 20 percent of ionic
concentrations.
The measured weekly concentrations were, in most cases,
approximately 10 percent higher than the weekly concentrations
derived from daily samples.
The differences between weekly and composited daily
concentrations varied by constituents, but they were in general
less than 10 percent of the mean concentration. Also, the
individual ion concetration biases are of similar magnitudes
and direction among the sites with the exception of H+ and Na+.
There is seasonal variation in the concentration bias between
daily composited and weekly samples. Concentrations of all
ions except NH^ were significantly higher in the fall for the
weekly measurements. Concentrations of Ca^+ and NOo were
significantly higher in weekly measurements in the spring. No
statistically significant biases were observed for the summer
quarter.
Differences in paired rain gauge depths at the two sites with
collocated rain gauges were greatest for snow samples. Overall,
paired rain gauges at the Vermont site measured significantly
different precipitation amounts; the differences were greatest
in the winter quarter, when there was more blow-away of
6-1
-------
snow. The significant differences in rain gauge depth resulted
in significant differences in daily deposition for all ions
except K+ at the Vermont site.
Precision of daily depositions (calculated from the two daily
collectors and the two rain gauges) was better at the Kansas
site; for most ions the precision was 5 to 15 percent. At the
Vermont site, where the rain gauge differences were larger, the
precision of ion depositions were 10 to 30 percent (as measured
by relative absolute collected difference).
Calculated deposition amounts vary by season at each site.
Peak SOT deposition occurred in the spring at the Kansas and
Georgia sites, and in the summer at the Vermont site. Peak H
deposition values occurred in the summer at the Vermont and
Georgia sites, and in the spring at the Kansas site.
The conclusions of this study favor daily sampling over weekly sampling to detect
trends in concentration or deposition more readily. Our conclusions indicate
that, for most ions, weekly samples yield higher concentrations than daily
samples. Therefore, a network that changes its sampling schedule would see a
bias that interferes with trend analysis. Also, comparison of concentrations
from networks with different sampling collection schedules will be influenced by
these biases. Finally, the seasonal differences in deposition strongly favor
monitoring and data analysis for a whole year and preferably for complete sea-
sons.
Additional analysis remains to be done on three items. First, it is not known
how the exclusion of trace events from the daily concentration data fould alter the
method bias. It has been suggested that ionic concentrations are particularly
high in trace events. We can compare the measured and derived concentrations for
only those weeks where no trace events occurred. Second, deposition method bias
remains to be estimated. Calculated annual deposition amounts presented in this
report were unadjusted for missing data (trace samples or partial samples) in the
daily measurements. Deposition method bias can be calculated by comparing
measured and derived weekly deposition values for only those weeks where there
are no missing concentrations. Third, the sensitivity of results to different
outlier rejection schemes is unknown. Those results based on nonparametric
techniques applied to the unscreened data (i.e., no outliers removed) would not be
affected. But precision, as measured by the pooled standard deviation, can be
significantly altered by choice of outliers rejected. The results of these
tasks should provide additional insight into the comparability of daily and
weekly precipitation chemistry measurements.
6-2
-------
Appendix A
WEEKLY AND DAILY SAMPLER BIAS SUMMARY STATISTICS
BY PRECIPITATION TYPE
-------
Appendix A
BASIC SUMMARY STATISTICS
Sampler concentration bias, estimated by the average collocated difference, is
examined in Section 5; relative bias estimates and standard deviations are
presented in Table 5-3 for all precipitation types combined. In this Appen-
dix, we list relative sampler concentration bias and standard deviation by
precipitation type at each site. Statistical significance of the reported
biases is tested with the Wilcoxon signed-rank test; those biases which are
significant are noted in the tables.
A-l
-------
Table A-l
RELATIVE PERCENT DIFFERENCES (RPD) AND RELATIVE STANDARD DEVIATIONS
(RSD) FOR COLLOCATED RAIN SAMPLES IN UVALDA, GA
Daily Sampling Weekly Sampling
Analyte RD (%)a RSD (%)b N_ RD (%) RSD (%)
Precip -1.7W
H+ -4.1
$04 -2.4W
NO 3 -2.8W
NH4 -3.3
Cl" -0.8W
+ w
Na -2.8W
K+ -7.0
Ca2+ -2.7
Mg2+ 0.0
3.9
12.2
5.9
5.4
21.3
13.0
14.0
58.8
21.7
18.5
63
57
57
57
57
57
57
57
57
57
-0.0
0.0
0.0
-4.5
-2.9
1.8
1.3
-2.8
1.5
-2.3
9.2
17.4
13.6
22.5
46.3
8.3
17.2
30.6
15.0
11.8
34
32
32
32
32
32
32
32
32
32
(a) RD = (d/C) x 100, where d and C are the mean collocated
difference and average concentration.
(b) RSD = (Sd/C~) x 100, where S^ is the standard deviation of
the collocated differences.
W = Sampler bias for the collocated pair significantly different from zero
based on the Wilcoxon signed-ranks test.
A-2
-------
Table A-2
RELATIVE PERCENT DIFFERENCES (RD) AND RELATIVE STANDARD DEVIATIONS
(RSD) FOR COLLOCATED SAMPLES IN LANCASTER, KS
Daily Sampling
Weekly Sampling
Analyte
Precip
H+
S0=
N03
NHj
Cl"
Na+
K+
Ca
Mg
2+
2+
Precip
H+
S0=
mt
Cl"
NaH
K+
Ca
Mg
2+
2+
D n ( °i \ ^
K I) \ h )
-1.3W
-6.1
-0.6
-0.9
-5.5W
-6.5
-7.6
-5.9W
1.1
2.2
-18.4
15.4
-19.8
-4.2
-45.2
23.3
6.3
120
-7.8
0.0
RSD (%)b
3.6
18.2
9.4
10.0
13.8
36.6
32.9
23.5
9.0
15.2
9.6
19.2
17.7
5.1
59.7
38.6
37.5
193
5.0
22.8
N
Rain
60
58
58
58
58
58
58
58
58
58
Snow
4
4
4
4
4
4
4
4
4
4
RD (%)
0.2
7.1
-3.1
-2.2
-1.7
-0.8
-4.0
2.4
-3.9
-4.5
14.8
2.3
-8.3
0.9
-36.8
-5.5
43.5
-88.9
-11.2
-26.7
RSD (%)
1.3
21.4
11.1
11.1
17.5
14.2
12.0
43.9
19.9
27.3
11.1
4.5
25.4
5.5
87.6
15.1
60.9
111
39.2
46.7
N
24
23
23
23
23
23
23
23
23
23
4
4
4
4
4
4
4
4
4
4
A-3
-------
Table A-2 (Concluded)
Daily Sampling Weekly Sampling
Analyte RD (%)a RSD (%)b N RD (%) RSD (%) N
Precip
H+
S0=
N03
NHj
Cl"
J_
Na+
K+
Ca2+
Mg2+
0.8
-13.8
-14.7W
6.6
-16.2
-15.2
-36.8
-22.9
-3.3
-9.0
Mixed
0.8
-13.8
-14.7W
6.6
-16.2
-15.2
-36.8
-22.9
-3.3
-9.0
12.9
82.8
16.5
58.8
33.1
70.4
61.7
110
42.7
36.0
8
8
8
8
8
8
8
8
8
8
0.5
25.0
-1.6
-0.7
0.7
-14.3
0.0
-39.7
-14.9
32.3
7.6
56.2
5.7
9.4
1.1
36.6
17.4
115
26.7
73.1
(a) RD = (d/C) x 100, where d and C are the mean collocated
difference and average concentration.
(b) RSD = (Sd/C) x 100, where Sj is the standard deviation of
the collocated differences.
W = Sampler bias for the collocated pair significantly different from zero
based on the Wilcoxon signed-ranks test.
A-4
-------
Table A-3
RELATIVE PERCENT DIFFERENCES (RD) AND RELATIVE STANDARD DEVIATIONS
(RSD) FOR COLLOCATED SAMPLES IN UNDERBILL, VT
Daily Sampling
Weekly Sampling
Analyte
Precip
H+
so=4
NOg
NHj
Cl"
Na+
K+
Ca2+
2+
Mg
Precip
H+
S04
NOi
NHj
Cl"
Na+
K+
Ca2+
Mg
RD (%)a
-0.6
0.0
-0.2
-0.5
-3.7
-0.9
-4.8
-3.3
-0.7
0.0
-9.6W
-1.9W
-2.4
0.4
2.5
-3.4
-7.9
-21.0
2.0
0.0
RSD (%)b
2.8
9.1
3.8
5.2
21.8
19.9
24.0
52.5
8.9
16.0
10.3
5.7
9.9
4.2
12.7
12.8
18.4
52.6
19.6
15.4
N
Rain
81
77
76
76
76
76
75
76
75
75
Snow
27
25
24
24
21
24
20
20
20
20
RD (X)
-0.6
-3.9
-1.5
-2.6
1.9
-4.9
-12.2
-4.2
3.0
4.3
2.3
-1.7
-1.7
-1.4
9.1
-2.2
1.6
0.0
3.3
0.0
RSD (%)
3.6
11.8
4.7
8.0
16.4
19.6
31.7
133
16.7
26.1
7.3
7.0
6.9
7.2
14.6
13.8
3.2
13.8
6.7
1.0
N
26
24
24
24
24
24
24
24
24
24
4
4
4
4
4
4
3
3
3
3
A-5
-------
Table A-3 (Concluded)
Daily Sampling
Weekly Sampling
Analyte
RD (%)a RSD (%
o/\b
RD (%) RSD (%) N
Precip
H+
S0=
NHj
Cl"
Na+
K1
Ca
Mg
2+
2+
1 fiW
-l.o
-2.2
-1.7
-0.8
0.5
-2.2
-9.8
4.9
12. 5W
-8.7
6.3
8.6
6.5
6.1
11.8
39.3
36.0
29.3
26.3
26.1
Mixed
24
24
24
24
24
24
23
23
22
22
3.2W
2.6
3.7
3.1
0.6
5.6
8.7
0.0
2.4
8.7
3.1
12.8
12.8
10.5
9.5
14.2
26.1
29.4
16.9
17.4
17
17
16
16
15
16
16
16
15
15
(a) RD = (d/C) x 100, where d and C are the mean collocated
difference and average concentration.
(b) RSD = (Sd/C x 100, where Sj is the standard deviation of
the collocated differences.
W = Sampler bias for the collocated pair significantly different from zero
based on the Wilcoxon signed-ranks test.
A-6
-------
Appendix B
WEEKLY MEASURED AND DERIVED WEEKLY ANALYTE
CONCENTRATIONS FOR EACH SITE AND PRECIPITATION TYPE
-------
Appendix B
CONCENTRATION STATISTICS
Mean and median ionic concentrations are listed for each site in Table 5-5 for
all precipitation types combined. In this Appendix, mean and median concentra-
tions are listed by precipitation type (rain, snow, and mixed) at each site.
These tables allow a comparison of mean and median concentrations as well as a
comparison of derived and measured weekly concentrations.
B-l
-------
Table B-l
ANALYTE CONCENTRATIONS (mg/1) IN RAIN FOR WEEKLY MEASURED
AND WEEKLY DERIVED SAMPLES
Measured Weekly
Deri ved Weekly
Analyte
Median
Mean
SEMa
Median
Mean
SEM
Uvalda, GA
Precip (grams)
pH (pH units)
H+
504
NO-
NhJ
cr
Na+
K+
Ca2+
Mg2+
Precip (grams)
pH (pH units)
H+
S04
NO-
NHj
cr
Na+
K+
Ca2+
Mg2+
1832
4.84
0.015
1.132
0.628
0.117
0.304
0.212
0.030
0.088
0.029
1147
5.06
0.0088
1.722
1.340
0.383
0.100
0.057
0.035
0.416
0.035
2312
4.65
0.023
1.329
0.802
0.175
0.415
0.241
0.031
0.114
0.035
1690
4.84
0.015
1.808
1.396
0.451
0.112
0.061
0.039
0.501
0.040
330
0.07
0.004
0.155
0.102
0.026
0.055
0.032
0.005
0.020
0.005
Lancaster
377
0.08
0.003
0.169
0.113
0.050
0.013
0.008
0.006
0.069
0.005
1848
4.87
0.014
0.951
0.627
0.125
0.302
0.156
0.030
0.065
0.026
, KS
975
5.02
0.0096
1.421
1.325
0.359
0.124
0.055
0.036
0.439
0.035
2362
4.65
0.022
1.261
0.772
0.175
0.398
0.220
0.034
0.093
0.032
1650
4.88
0.013
1.696
1.349
0.441
0.117
0.062
0.040
0.511
0.040
326
0.06
0.004
0.161
0.100
0.027
0.057
0.032
0.007
0.017
0.005
423
0.07
0.002
0.188
0.120
0.051
0.014
0.009
0.007
0.075
0.005
3-2
-------
Table B-l (Concluded)
Measjred Ueekly
Deri ved Weekly
Analyte
Median
Mean
SEMd
Median
Mean
SEM
Precip (grams)
pH (pH units)
H+
504
N03
NHj
cr
Na+
K+
Ca2+
Mg2+
1096
4.29
0.052
2.410
1.185
0.284
0.065
0.019
0.014
0.066
0.009
1096
4.29
0.052
2.410
1.185
0.284
0.065
0.019
0.014
0.066
0.009
1559
4.29
0.051
2.653
1.474
0.359
0.102
0.041
0.024
0.132
0.023
Underhil
292
0.05
0.006
0.359
0.175
0.069
0.030
0.016
0.005
0.042
0.006
11, VT
1073
4.28
0.052
2.107
1.285
0.271
0.063
0.017
0.015
0.052
0.008
1561
4.25
0.056
2.744
1.509
0.372
0.110
0.047
0.019
0.129
0.021
293
0.05
0.006
0.379
0.186
0.082
0.035
0.019
0.004
0.046
0.007
(a) Standard error of the mean = SD//N.
B-3
-------
Table B-2
ANALYTE CONCENTRATIONS (mg/1) IN SNOW FOR WEEKLY MEASURED
AND WEEKLY DERIVED SAMPLES
Measured Weekly
Deri ved Weekly
Analyte
Precip (grams)
pH (pH units)
H+
$04
3
NHj
cr
Na+
K+
Ca2+
Mg2+
Precip (grams)
pH (pH units)
H+
$04
3
NH4
cr
f
Na
K+
Ca2+
Mg2+
Median
157
4.62
0.024
0.444
2.247
0.259
0.063
0.010
0.033
0.134
0.017
356
4.21
0.061
0.642
3.508
0.110
0.152
0.066
0.014
0.057
0.005
Mean
186
4.58
0.026
0.503
2.555
0.285
0.073
0.012
0.027
0.143
0.015
408
4.24
0.058
0.859
3.221
0.164
0.181
0.062
0.015
0.060
0.010
SEMa
Lancaster
62
0.09
0.006
0.110
0.653
0.089
0.025
0.005
0.007
0.036
0.004
Underbill
159
0.12
0.016
0.210
1.070
0.053
0.044
0.017
0.003
0.012
0.003
Median
, KS
156
4.68
0.021
0.381
2.090
0.219
0.085
0.016
0.012
0.190
0.017
, VT
322
4.19
0.064
0.677
3.438
0.084
0.153
0.060
0.010
0.036
0.006
Mean
178
4.58
0.026
0.474
2.663
0.275
0.099
0.016
0.027
0.181
0.014
409
4.24
0.058
0.799
3.203
0.126
0.177
0.060
0.010
0.042
0.008
SEM
84
0.11
0.008
0.142
0.957
0.078
0.044
0.003
0.018
0.059
0.004
166
0.13
0.017
0.162
1.130
0.039
0.045
0.017
0.002
0.010
0.002
(a) Standard error of the mean = SD//N.
B-4
-------
Table B-3
ANALYTE CONCENTRATIONS (mg/1) IN MIXED (SNOW & RAIN) WEEKLY
MEASURED AND WEEKLY DERIVED SAMPLES
Measured Weekly
Derived Weekly
Analyte
Precip (grams)
pH (pH units)
H+
504
3
NHj
cr
Na+
K+
Ca2+
Mg2+
Precip (grams)
pH (pH units)
H+
504
NO-
NH4
cr
Na+
Ca2+
Mg2+
Median
1602
4.85
0.014
2.020
1.405
0.559
0.142
0.053
0.040
0.397
0.035
1674
4.60
0.025
0.862
0.896
0.103
0.057
0.024
0.011
0.050
0.008
Mean
1758
4.80
0.016
2.283
1.640
0.595
0.140
0.069
0.081
0.521
0.047
1761
4.46
0.035
1.284
1.379
0.170
0.071
0.023
0.015
0.087
0.012
SEMa
Lancaster
510
0.13
0.005
0.436
0.249
0.099
0.029
0.018
0.030
0.115
0.010
Underhill
252
0.09
0.007
0.260
0.316
0.044
0.016
0.004
0.003
0.028
0.003
Medi an
, KS
1543
4.70
0.020
1.947
1.334
0.475
0.105
0.056
0.036
0.415
0.035
, VT
1824
4.59
0.026
0.917
0.970
0.112
0.066
0.025
0.006
0.034
0.006
Mean
1742
4.66
0.022
2.280
1.582
0.622
0.139
0.074
0.058
0.445
0.039
1816
4.42
0.038
1.273
1.690
0.186
0.090
0.023
0.012
0.064
0.008
SEM
485
0.12
0.006
0.429
0.269
0.103
0.033
0.019
0.018
0.089
0.008
257
0.09
0.008
0.273
0.415
0.046
0.024
0.005
0.003
0.017
0.002
(a) Standard error of the mean = SD//N.
3-5
-------
Appendix C
WEEK-BY-WEEK COMPARISON OF DEPOSITION AMOUNTS
CALCULATED FROM DAILY AND WEEKLY SAMPLERS
FOR MAJOR IONS (H+, SO^, N0~, NH*, Ca2+)
-------
Appendix C
WEEK-BY-WEEK PLOTS OF DEPOSITION
In this appendix we present week-by-week deposition of H , SO., NO,, NH.,
2+ 'tot
and Ca as calculated from the daily and weekly measurements at each site.
The plots show seasonal variation in deposition for these major ions. Weekly
deposition values are derived from the daily data by summing the daily deposition
amounts. Some deposition amounts are missing from the daily samples but not from
the weekly samples because individual trace events (which may contain relatively
high ionic concentrations) were not chemically analyzed. Since the derived
weekly deposition amounts are not adjusted for the missing events, the derived
and measured weekly deposition values are not directly comparable. However,
large differences seem to occur only rarely.
C-l
-------
a
a
u.
a
Ul
5
a
u
a
in
0> 01 Ifl Hi
(9
a
o
»
O £
I > L'J Ct O iD kl Z " C E 'I- '..'E
C-2
-------
in
tn
a
ui
tf
01
a
ui
o
Q
H
CO
a
T3
01
H U]
o cn
a a
Q CO
O ^
u
a M
c o)
CO AJ
ca
t^ cd
rH CJ
ai cd
cu ,j
T3 o
0) H-l
(U U
6 -H
to
"4-1 O
o o.
D H O Hi U Z - C EiS'-.'E
C-3
-------
a
a
u.
a
E
01
JC
111
111
a
ui
K
en
a
ui
o
tc
0)
c-l
Hi
>,
*H
H
T3
a
Tl iH
C H
(fl -H
^
U £
UJ
rD
C£>
;- Q a o
z - c
C-4
-------
y-
w
a
a
u.
a
en
t-
111
IU
a
ui
ct
< CO
H T3
O M
(U O
Vi u-i
3
CO C
rt o
(U -H
6 ^
H
u-i tn
o o
ex
a
-------
a
a
n
a
E
01
J-
X
Ul
m
a
ui
a
in
E
U
D
IU
10
ID
a
j
3
Z
V
a
DC
D.
II
EC
a
r
Q)
111
IJL
Z
a
D
UJ
D
D
Z
u
D
n.
LJ
10
ro
(0
T3
rl
in co
O «
a M
E C
8:2
T3
C M
n) 0)
u
>, co
^H n)
M o
CD c
a) n)
S .-I
T3 IJ
cu o
M <"
a
w c
'!
E
-------
a
a
u.
a
E
cn
ui
ui
a
ui
5
en
a
ui
Ill
r-t
I
OJ
05
H
CO
T3
a
cu
4J
H JJ
(0 C
O O
O CU
o >
T3
C rH
(0 rH
r-H 1-1
X 0)
(11 T)
(U C
73 M
0) O
M 14-4
SC
cfl O
CD -H
B *~*
H
u-i cn
o o
a
fi 0)
O T3
CO
i^ 0)
nj ccj
CX I4H
E '-i
O 3
u cn
in
i
C-7
-------
f
M
a
a
u.
a
z
01
3C
UJ
a
ui
tt
01
a
UJ
H
(8
13
W
0
O. CO
a -H
O 60
CJ »J
O
13 HI
C O
cd
r-j 13
^i rH
» p
-a i-i
-------
a
a
u.
a
cn
ui
UJ
a
en
n
r
o
o
01
1-
u
D
&.
HI
W
ID
a
J
z
111
K
a
E
01
111
u.
z
a
u-
LU
D
I-
u
D
CL
LJ
10
0)
n)
T3
OJ
4-)
H
to cn
o «
ex cn
e c
T3
C
tu c
cu n)
S i-I
T3 M
tu o
cn a
rt o
0) -rt
E J->
r(
M-l CO
O O
a
c C
CO
Z M H E CI h Id
C-9
-------
a
a
u.
a
Ul
Ul
a
UJ
a
m
I-
u
D
Ul
10
10
a
j
z
D
u.
a:
E
01
U.
. a
D
ui
0
u
D
D.
Ul
10
c-1
1
Q
r-l
tfl
T3
T3
OJ
CO C
O O
a 6
B M
O OJ
o >
T3
C r-H
cfl ^H
l-l
(U
1U T3
(U 0
O (J
(U O
LM ra
o o
a
c a)
o -o
CD
H. ai
|J 4J
nJ cd
a. M
e w
O -H
en
i
Z M !- i (I 1-
»- C
C-10
-------
a
a
U.
a
r
01
UJ
a
ui
K
ui
a
o
cd
T3
T3
(U
CO
O CD
B, -rl
B 60
O !-i
O O
01
TJ O
IT) -
4^ CB
(U >
0) D
T) O
11 M-l
M
3 d
oi o
ca -H
0) J->
B -H
CO
M-l O
O P.
-------
a
a
u.
a
01
LU
7
a
ui
tt
en
a
UJ
a
0>
4-1
H en
0) Cfl
O a]
a c
E n)
o w
-O l-l
c cu
at <->
ca
>, tfl
rH O
J^ C
OJ cfl
0) nJ
l-l
73 O
01 <4-l
u
3 C
en o
(0 -H
01 J->
E -rl
en
14-1 O
o a.
01
c -a
o
en 6
H a
M -H
ct) C
H
O n)
,D
Z Z O Z i-i D
'
> C E i!t '..
C-12
-------
a
a
u.
a
E
on
JC
ui
ui
a
ui
d
a
o
o 1-1
C rH
cd -H
>> M
-I
-------
>-
J
M
a
a
u.
a
01
t-
,
j
1C
in
in
a
Ul
if
U.
01
a
UJ
0
J 0> IB
w u> :#
:-
.
*
h.
'^ t
li
^1
*K
jT
Jt
, i fjp r
O '! ^ " ' f~
-" 124^
*^ftSu^
^ i " ri"
QH"^ ^
' ' i i.l
^rf-
^--^
^
\
S
'9
f
°<
C3
>_
*
r
r
_
L
i
.
O I'd d4 (0 iD
d4 rt i?J
i-
u
D
H.
UJ
10
13
3
J
3
f
3
-)
U.
tt.
Q.
[I
Ct
a
E
§
Z
-)
D
Ul
a
>
b
z
t-
U
D
D.
UJ
10
J
0)
CD
r-f
1
1
ff
m
0)
^4
>>
H
cd
a
t3
D
a
CU M
n o
3 u-i
01
cO C
CU O
6 w
..-I
0 «
0 0
a
C (U
O T)
01
rt e
M' 3
cd -H
a. o
e -H
o cd
u a
ro
1
o
UJ
a;
o
KJ3
U_
C-14
-------
M
a
a
u.
a
01
U
UJ
a
Ul
a
ui
a
o
H
n)
-O
T)
01
JJ
H
ca en
o
-------
"I
a
u.
a
r
JC
tu
Uj
a
ui
a
01
a
UJ
(5
rl4
a)
o
a
11
en
o
O OJ
o >
T3 -
C M
CO i-f
------- |