EPA Report Number
                                                                      May  1,  1990
         ESTIMATION OF SEASONAL AND ANNUAL ACIDIC  DEPOSITION
          THROUGH AGGREGATION  OF THREE-DAY EPISODIC PERIODS
                                          by
                     Perry J. Samson, Jeffrey R. Brook and Sanford Sillman
                     Department of Atmospheric, Oceanic and Space Sciences
                                  University  of Michigan
                             Ann Arbor, Michigan  48109-2143
                          EPA Cooperative Agreement CR-814854-01
I
              Project Officer

              John F. Clarke
     Atmospheric Sciences Modeling Division
Atmospheric Research and Exposure Assessment Lab
  Research Triangle Park,  North Carolina 27709
    EJBD
    ARCHIVE         ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LAB
    EPA                    OFFICE OF RESEARCH AND DEVELOPMENT
    6 ° ° -                   U.S. ENVIRONMENTAL PROTECTION AGENCY
    2~ _                RESEARCH TRIANGLE PARK, NORTH CAROLINA 27709
    059

-------
                                    DISCLAIMER

The information in this document has been funded wholly or in part by the U.S. Environmental
Protection  Agency  under Cooperative Agreement  CR-814854-01  to the  Department of
Atmospheric, Oceanic and Space Sciences, University of Michigan, Ann Arbor, Michigan.  It has
been subjected to Agency peer and administrative review, and has been approved for publication
as an EPA document.
                                          ii

-------
 qo-                        US
 OS?

                                                             RepostoryMjrt
                                                            Permanent Collection
                                         ABSTRACT
    The Regional Acid Deposition Model  (RADM)  simulates the complex physical and  chemical
    processes involved in the formation of acidic deposition.  However, because of the complexity of
    RADM it has only been applied to selected episodes of a few days duration.  While the detail
    provided by RADM is  highly desirable to understand  the interactions between emissions and
    deposition on a storm-by-storm basis there is also a need for understanding  the probable long-
    term relationship  between changing emissions patterns and  deposition.  The cost of simulating
    and interpreting seasonal or annual deposition patterns using RADM directly would be high.

    A method for aggregating episodic deposition estimates has been developed and used to identify
    which meteorological  situations  merit  simulation  by  RADM  based on  their likelihood  of
    producing  sulfate (SO4=) wet deposition at multiple locations across  eastern North America,
    their frequency of occurrence, and their seasonally.  The aggregation approach  is based on four
    years (1982-1985) of  meteorological and precipitation  chemistry  data.
!«o  The aggregation approach is based on the. stratification  of three-day periods into categories, of
1;   similar 850 mb wind flow across eastern North  America  and subsequent selection of 30 RADM
^'   simulation periods from this stratification which represent the range pf storm patterns present
5   over eastern North America.  The program has also provided the RADM project with  scaling
    factors for use in weighing episodic simulation results to seasonal and annual deposition.  The
4-  approach  is shown  to improve  upon the use  of a  random selection process  in reproducing
*-   seasonal and annual wet deposition patterns.
                                              111

-------
                               CONTENTS

SECTION                                                             PAGE

DISCLAIMER  	ii
ABSTRACT    	iii
UST OF FIGURES	v i i i
USTOFTABLES	xv
EXECUTIVE SUMMARY	xvii
1.     INTRODUCTION	1
       1.1    Goals and Purposes	1
       1.2   History	2
       1.3   Conceptual Approach _	3
2.     METHODS	:	.'...:	6
       2.1    Aggregation Approach	6
             2.1.1         Meteorological  Stratification	6
             2.1.2         Selection of Stratification Parameters	8
             2.1.3         Categorization for Other Years	1 0
             2.1.4         Aggregation Procedures	1 0
             2.1.5         Selection Process	1 3
             2.1.6         Source/Receptor Relationships	1 5
                    2.1.6.1       Trajectory Calculations	15
                    2.1.6.2       Quantitative Transport Analysis	16
                    2.1.6.3       The Potential Source  Contribution
                                 Function	 18
       2.2    Description of Data Base	1 9
             2.2.1         Meteorological Data	1 9
             2.2.2         Precipitation Chemistry Data	20

-------
SECTION


       2.3    Testing	24
              2.3.1         Aggregation Versus Random Selection	2 4

              2.3.2         Statistical Regression	2 5

              2.3.3         Source-Receptor Relationships	2 7

3.     RESULTS	29

       3.1    Meteorological Categories	2 9
              3.1.1         Choice of Stratification Parameters	29

              3.1.2          Description of Categories	30
                     3.1.2.1        Wind Flow Patterns	31
                     3.1.2.2        Deposition Patterns	36

              3.1.3          Statistical correlations between category and
                            observed deposition	51
                     3.1.3.1        Three alternative statistical
                                   approaches	5 1
                     3.1.3.2        Time-of-Year Term 	52
                     3.1.3.3        Results	5 3
              3.1.4          Interannual variability	:	6 0

              3.1.5          Evaluation of the category-based aggregation
                            method	67

        3.2    Selection of Cases	6 9
        3.3    Comparison of Aggregate  Estimate with Long-Term
                     Deposition	70
              3.3.1          Annual Wet Deposition	70
                     3.3.1.1        Sulfate Wet Deposition	7 1
                     3.3.1.2        Nitrate Wet Deposition	74
                     3.3.1.3        Acidity Wet Deposition	77

              3.3.2          Seasonal Wet Deposition	7 7
                     3.3.2.1        Sulfate Wet Deposition	77
                     3.3.2.2        Nitrate Wet Deposition	77
                     3.3.2.3        Acidity Wet Deposition	88

              3,3.3          Uncertainties  in the Aggregate Estimates	8 8

        3.4    Source-Receptor Relationships	9 9

              3.4.1  Climatological Wind  Flow Patterns during
                            Precipitation	99

              3,4.2  Wind Flow Associated with Sulfate Wet Deposition	100

              3.4.3  Potential Source Contribution Functions	107
                                      VI

-------
SECTION                                                          PAGE


      3.5    The Variation in Source-Receptor Relationships between
                   Categories	1 1 2

      3.6    Selecting episodes to estimate source-receptor
                   relationships	1 25

      3.7    Source-receptor relationships with the 30 episodes	128

4.     SUMMARY	138

      4.1    Accomplishments	1 38

      4.2    Recommendations for Future Work	1 39

             4.2.1  Evaluation of the 30 Episodes and the Meteorological
                         Categories	1 39
                   4.2.1.1       Source-Receptor Relationships	140
                   4.2.1.2       Chemical and Meteorological Behavior
                                of the Categories	1 40
             4.2.2  Application of aggregation to other atmospheric
                         phenomena	1 4 1
                   4.2.2.1       Dry Deposition	1 4 1
                   4.2.2.2       Visibility	1 42

             4.2.3  Quantification of Aggregation Sensitivity	143

5.     REFERENCES	1 44

      APPENDIX A:  MEAN 850 MB WIND FLOW PATTERNS FOR THE 19
             CATEGORIES	A-1

      APPENDIX B: 85TH PERCENTILE WET SULFATE DEPOSITION PLOTS
             FOR THE 19 CATEGORIES	B -1

      APPENDIX C: SITE SPECIFIC TABLES OF TOTAL AND PERCENT OF
             TOTAL WET SULFATE DEPOSITION	C-1

      APPENDIX D: SITE SPECIFIC TABLES OF TOTAL AND PERCENT OF
             TOTAL PRECIPITATION	D -1
                                 VII

-------
                             LIST OF FIGURES


NUMBER
1      Illustration of the aggregation concept [[[ 5
2      NMC grid and subset of 48 grid points used to describe similar
       meteorological patterns [[[ 7
3      Location of the precipitation chemistry sites in the United States
       and Canada used for analysis of transport patterns ................................. 2 1
4      Beale and Calinsky criteria (F and C  statistics) versus the number
       of categories (step number), based on 850 mb wind field clusters
       and observed SO4° deposition for 1983 [[[ 3 1
5      Reduction in variance in SO4= wet deposition resulting from
       meteorological stratification [[[ 32
6      Comparison of frequency of occurrence of the 1 9 meteorological
       categories defined in this study between  1 979-1 98T and  1982-
       1985 [[[ 35
7a-f  Monthly mean  number of occurrences of meteorological categories
       during 1982 to  1985 [[[ 3 8
8a-d  Variation in amount of SO4= deposition between categories at four
       locations [[[ 4 1
9      The ratio of the percent of SO4= deposition to the percent of
       precipitation for each meteorological category ...................................... 43
1 0     The percentage of total SO4" wet deposition associated with each

-------
NUMBER                                                               PACE

1 3     Ts statistic for meteorological  category 7, indicating deposition-to-
       precipitation ratios significantly above or below the norm  at each
       site	58
14     Ts statistic for meteorological  category 11, indicating deposition-
       to-precipitation ratios  significantly above or below the norm at
       each site	59
1 5     Year-by-year variation in  sulfate wet deposition, nitrate
       deposition, sulfate deposition-to-precipitation ratio,  and
       precipitation at Turners Falls, Massachusetts	62
1 6     Year-by-year variation in  sulfate wet deposition, nitrate
       deposition, sulfate deposition-to-precipitation ratio,  and
       precipitation at Zanesville, Ohio	63
1 7     Year-by-year variation in  sulfate wet deposition, nitrate
       deposition, sulfate deposition-to-precipitation ratio,  and
       precipitation at Giles County, Tennessee	64
1 8     Year-by-year frequency of occurrence of events  of meteorological
       categories 6, 7 and 11, expressed as a fraction of all  events	65.
1 9     Comparison between annual sulfate deposition observed during
       1986 and 1987  at 13 northeastern and midwestern  sites with
       projected deposition based  on 1982-1985 observations and  1986-
       1987 meteorology	66
2 0    Comparison of selection and  aggregation techniques for S04= wet
       deposition  in northeastern North America.using  nineteen three-day
       episodes	68
 21 a   The mean annual wet sulfate deposition pattern across eastern
       North  America  for 1982-1985.   (Units=mgm"2)	72
 21 b   The aggregate estimate of the mean annual wet sulfate deposition
       pattern across  eastern North  America for 1982-1985	73
 22a   The mean  annual wet nitrate deposition pattern across eastern
       North  America  for 1982-1985	75
 22b   The aggregate estimate of the  mean annual wet nitrate deposition
       pattern across  eastern North  America for 1982-1985	76
                                      i x

-------
NUMBER

23a   The mean annual acidic wet deposition pattern across eastern North
      America for 1982-1985	7 8
23b   The aggregate estimate of the mean annual acidic wet deposition
      pattern  across eastern  North America for 1982-1985	79
24 a   The mean summer wet sulfate deposition pattern across eastern
      North America  for 1982-1985	80
24b   The aggregate estimate of the mean summer wet sulfate deposition
      pattern  across eastern North America for 1982-1985	81
24c   The mean winter wet sulfate deposition pattern  across eastern
      North America  for 1982-1985	82
24d   The aggregate estimate of the mean winter wet sulfate deposition
      pattern  across eastern North America for  1982-1985	83
25a   The mean summer wet nitrate deposition pattern across eastern
       North America  for 1982-1985	84
25b  The aggregate estimate of the mean summer wet nitrate deposition
      pattern  across eastern North America for  1982-1985	•.	85
25c  The mean winter wet nitrate deposition pattern  across eastern
       North America for  1982-1985	86
25d  The aggregate estimate of the mean winter wet  nitrate deposition
      pattern  across eastern North America for  1982-1985	87
26a  The mean summer wet acid (H*) deposition  pattern across eastern
       North America for 1982-1985	89
26b  The aggregate estimate of the mean summer wet acid (H*)
       deposition pattern across eastern North America for 1982-1985	90
26c   The mean winter wet acid (H+) deposition pattern across eastern
       North  America for  1982-1985	91
26d   The aggregate estimate of the mean winter wet  acid (H*) deposition
       pattern across eastern North America for 1982-1985	92
2 7    The spatial and temporal variation in the percent of annual wet
       sulfate deposition due to the categories selected	93

-------
NUMBER

2 8    The year to year variability in the annual wet sulfate deposition and
       in the total deposition due to the categories selected at two
       Midwestern sites	94
2 9    The year to year variability in the annual wet sulfate deposition and
       in the total deposition due to the categories selected at two
       northeastern sites	95
3 0    Uncertainty ranges due to the aggregate estimates of wet sulfate
       deposition	97
31    Uncertainty ranges due to the aggregate estimates of wet nitrate
       deposition	,	98
3 2    Uncertainty ranges due to the aggregate estimates of wet acidity
       deposition	98
3 3    The mean probability  for transport on precipitation days to  Big
       Moose,  NY,  during  the  period 1982-1985	101
3 4    The mean probability  for transport on. precipitation days to
       Gaylord, Ml,  during the-period  1982-1985....	102
3 5    The mean probability  for transport on precipitation days to
       Raleigh, NC,  during the period 1982-1985	103
3 6    The sulfur deposition weighted probability  for transport to  Big
       Moose,  NY during the  period 1982-1985	104
37    The  sulfur deposition weighted probability  for transport to
       Gaylord, Ml,  during the period  1982-1985	105
38    The  sulfur deposition weighted probability  for transport to
       Raleigh, NC, during the period 1982-1985	106
3 9    The bias for transport associated  with large S04= wet deposition at
       Big Moose, NY, during the period 1982-1985.   	108
4 0    The bias for transport associated with large  water deposition at  Big
       Moose,  NY, during  the period  1982-1985	1 09
4 1    The  bias for transport associated with large SO4= wet deposition at
       Gaylord, Ml, during the period 1982-1985	110
4 2    The  bias for transport associated with large water deposition at
       Gaylord, Ml, during the period 1982-1985	111

-------
NUMBER                                                              PAGE

4 3    The bias for transport associated with large SO4= wet deposition at
       Raleigh, NC,  during the period 1982-1985	11 2
4 4    The bias for transport associated with large water deposition at
       Raleigh, NC,  during the period 1982-1985	1 1 3
4 5    The ensemble bias for transport associated with large SO4= wet
       deposition at Big Moose, NY; Gaylord, Ml; and Raleigh, NC, during
       the period 1982-1985	1 14
46    Difference between climatological Q field (1982-1985) and the  Q
       field for category 20 (unbiased set of missing days)	11 5
47    Difference between climatological Q field (1982-1985) and the  Q
       field for category  1	1 1 6
48    Difference between climatological Q field (1982-1985) and the  Q
       field for category  5	1 1 7
49    Difference between climatological Q field (1982-1985) and the  Q
       field for category  6	1 1 8
50    .Difference between climatological Q field (1982-1985) and the  Q
       field for category  7	1 1 9
51    Difference between climatological Q field (1982-1985) and the  Q
       field for category  8	1 20
52    Difference between climatological Q field (1982-1985) and the  Q
       field for category  9	1 21
53    Difference between climatological Q field (1982-1985) and the  Q
       field for category  10	1 22
54    Difference between climatological Q field (1982-1985) and the  Q
       field for category  11	1 23
5 5    Percent RMS errors, ranked from smallest to  largest, for the
       random and category-based  ensembles using trajectories from all
       days in 30 episode sample	1 27
56    Percent RMS errors, ranked from smallest to  largest, for the
       random and category-based  ensembles using trajectories from
       precipitating days  only in 30  episode sample	1 27
                                   xi i

-------
NUMBER                                                               PAGE

5 7    The Q field for Big Moose, NY derived from both precipitating and
       non-precipitating days within the 30 episodes	129
5 8    The Q field for Gaylord, Ml derived from both precipitating and
       non-precipitating days within the 30 episodes.....;	130
5 9    The Q field for Raleigh,  NC derived from both precipitating and
       non-precipitating days within the 30 episodes	131
60    Bias in source receptor analysis resulting from  use of 30-event
       sample instead of 4-year sample at Big Moose, NY.  Bias is
       expressed as a percent of predicted Q-field from the 4-year
       analysis.  Both precipitating and non-precipitating  days during the
       30 episodes are  included in the analysis	1 32
6 1    Bias in source receptor analysis resulting from  use of 30-event
       sample instead of 4-year sample at Gaylord, Ml.  Bias is expressed
       as a percent of predicted Q-field from the 4-year analysis.   Both
       precipitating and  non-precipitating  days during  the  30 episodes are
       included  in the analysis	.•	.•	133
6 2    Bias in source  receptor analysis  resulting  from  use of 30-event
       sample instead of 4-year sample at Raleigh, NC. Bias is expressed
       as a percent of predicted Q-field from the 4-year analysis. Both
       precipitating and non-precipitating days during  the  30 episodes are
       included  in the analysis	1 34
63    Bias in source receptor analysis  resulting  from  use of 30-event
       sample instead of 4-year sample at Big Moose, NY.  Bias  is
       expressed as a  percent of predicted Q-field from the 4-year
       analysis.  Only days with precipitation reported at Big Moose are
       included  in the analysis	1 35
64    Bias in source receptor analysis  resulting  from  use of 30-event
       sample instead of 4-year sample at Gaylord, Ml. Bias is expressed
       as a percent of  predicted Q-field from the 4-year analysis. Only
       days  with precipitation reported at Big Moose are included in the
       analysis	1 36
                                    xi i

-------
NUMBER

65    Bias in source receptor analysis resulting from use of 30-event
       sample instead of 4-year sample at Raleigh, NC.  Bias is expressed
       as a percent of predicted Q-field from the 4-year analysis.  Only
       days with precipitation reported at Big Moose are included in the
       analysis	1 37
                                    xi v

-------
                             LIST OF TABLES

NUMBER                                                              PAGE

1      Sample aggregation	1 2
2      Name and location of the daily  (event)  precipitation chemistry
       sites	22
3      Meteorological factors used in initial cluster analysis	29
4      Synoptic description of the nineteen 850 mb wind flow categories	33
5      Matrix of percent of total deposition by site and category	45
6      Percent of total sulfate wet deposition by meteorological category	5 0
7      Results of statistical  regression	55
8      Initial selection of 30 days	:	:	71
9      Root-mean-square deviations between estimated and observed
       summer, winter and annual wet  deposition at 13 eastern sites	74
1 0    Description of source-receptor relationships  for selected
       meteorological categories	1 24
                                     xv

-------
                                EXECUTIVE SUMMARY

This report presents methods for developing estimates for long-term (climatological) patterns
of atmospheric wet deposition  based on in-depth knowledge of a finite number of samples.  These
methods are  designed  to complement  predictive models for physical processes in the
atmosphere, which characteristically provide detailed analysis  for a small number of days but
are unable to  represent  physical processes on  a  multi-year  time  scale  due  to computational
constraints.   Predicted  future effects  are  to  be  derived  based  on statistical  methods  of
aggregation  developed here,  in combination with results obtained from physical  models.  The
method  has  been applied to the  specific problem of estimating future rates of  wet deposition  in
the eastern United States in combination  with the  Regional Acid  Deposition  Model (RADM).

The method is based  upon the stratification of three-day periods into categories of similar wind
flow at 850 mb (roughly 1500  meters above  mean  sea  level).  The stratification has been used
to identify which meteorological situations merit  simulation by  RADM based on their likelihood
of producing sulfate  deposition  at  multiple locations  across eastern  North  America, their
frequency of occurrence, and  their seasonality.  A  specific ensemble of 3-day  events has been
recommended  for future  analysis by RADM, as part of ongoing  efforts  by the Environmental
Protection Agency (EPA) to develop  predictive  methods for  air quality.   In  addition  to even;
selection, the meteorological stratification has been used as the basis of a  statistical technique  to
derive aggregate estimates for  long-term  average  wet deposition. The aggregation technique has
been  simulated  using  four  years  of  precipitation  chemistry data   to demonstrate  that
uncertainties in  estimates of seasonal and annual sulfur deposition using  the  aggregation
technique is significantly  less  than those obtained  from random (non-stratified) sampling of the
data.

The following specific tasks have  been completed as part of this project:

      (1 )  An extensive data base with daily precipitation and wet deposition  of sulfate, nitrate
           and total  H+was  developed.   The data base  covered the years 1S82-1S65 and
           included 23 sites in  the northeastern, southern and midwestern U.S.  and southern
           Canada.  Meteorological information to be used  in subsequent analysis was  also
           collected.
                                          xvi i

-------
( 2)   Techniques of cluster analysis were  used to  stratify  3-day  events  over the 4-year
      period  into categories based on the meteorological  data described  above.  Four
      independent  stratification schemes  were investigated:  stratification by  850 mb
      wind field, stratification  by 850 mb geopotential and  mean surface pressure,
      stratification  by  850  mb  temperature  advection  and 500  mb vorticity,  and
      stratification by 500  mb vorticity advection and  850 mb  relative humidity. These
      four stratification schemes were evaluated to find the combination of meteorological
      parameters which best explained variations in sulfur  wet  deposition.

( 3)   A set of 19 meteorological categories was compiled based on similarity in wind flow
      at  850  mb.  These 19 categories were each subdivided into "wet" and "dry" cases,
      making 38 categories.  These 38 categories were used as the basis for selection of
      individual days and derivation of aggregated estimates for  annual deposition.

(4)  A  statistical correlation has been developed  between observed sulfate and nitrate
      wet deposition during  1982 through 1987 and wind flow category.  Results  show
      that  wet deposition rates over  eastern North  America show significant differences
      for events with  different 850 mb circulation.

( 5 )  Accuracy  and validity of  the 38-category stratification was  evaluated based on the
      •ability  to predict annual sulfate deposition at eastern and midwestern sites from a
      firjite sample of 3-day events.  We compared estimates for  annual deposition"
      derived from a random sample  of 3-day events with  estimates from a cluster-based
      sample of events. The standard deviation between estimated and  actual deposition
      was lower by 25% with cluster-based selection.

{ 6 )  A  preliminary  selection  of 30  individual 3-day  events for  in-depth  study with
      RADM was made. The selected events included representatives of a wide  variety cf
      circulation patterns,  including meteorological  categories that  account  for the bulk
      of wet deposition at  sites throughout the  eastern U.S.  A selection of relatively rare
      winter  circulation patterns  and  events without  rainfall  for evaluation of  dry
      deposition are  also included.   Aggregate  estimates for  sulfate  and nitrate  wet
      deposition have been derived based on the 30-event sample.  Deviations between the
      aggregate estimate and observed wet deposition during 1982-1985 are of  the same
      magnitude as the natural variation in wet  deposition from  one year to the next.

 ( 7 )  A  simplified model  for  identifying  emitted sources of  wet deposition  has been
      applied to the selected 30-event sample and also to  the  entire  1982-1985  period
      for selected sites in  eastern North America.   Differences  between predicted source-
      receptor relationships  based on the 30-event sample and  based on the entire 4-
      year period are assessed.
                                    xvi 11

-------
                                 1 .     INTRODUCTION
1.1.       GOALS AND PURPOSES

      The development of policies to address problems arising from acid deposition  requires a
thorough understanding of processes that affect acid deposition.  To this end the United States
Environmental Protection Agency (EPA) has supported investigations into the physics, chem-
istry and meteorology of acid deposition.  These efforts have culminated in the development of
the Regional  Acid  Deposition Model (RADM) (Chang  et  al., 1987), which  simulates the
processes involved in the formation of acid deposition and  other pollutants.

      In order to complete the evaluation of acid deposition processes it is necessary to establish
a  relationship between  individual  deposition  events  and the  long-term  climatology of acid
deposition.  The  relationship  between individual events  and  climatology is  difficult because
deposition varies greatly from one event to the next in terms of £>oth the amount of  deposition
and the relation between deposition and emission sources.  Establishment, of long-term deposi-
tion patterns  and  source-receptor relationships  requires analysis over an extended time period,
preferably including several years to account for meteorological variability.  Such a lengthy
analysis  is beyond  the capacity of the  resources available to the National  Acid Precipitation
Assessment Program.

      The goal of this study has been to provide methods for  representing aggregate long-term
patterns  of acid deposition from a  finite number of samples.  These  methods involve both the
selection of  individual  deposition  events  which  are   representative of recurring  climatic
patterns  and  the  calculation of climatic  averages based  on a  finite sample.  The aggregation
methods have been applied to the specific problem of representing the climatology of  acid depo-
sition based on in-depth analysis by RADM for a finite sample of events.  An array of 30 three-
day events has been identified which includes the major climatic patterns that contribute to acid
deposition over eastern  North America.  RADM is currently being  used to  analyze physical
processes during  these  30 events.   Assessment of the impact of future changes in precursor
emissions will be performed by (1) using RADM to predict  impacts for  the thirty  sample
events, and (2) estimating long-term impacts by aggregating from the 30 sample events, based
on the techniques described in this  report.  While designed specifically  in support of RADM  the
methods described  here  may have broad applicability to  problems that involve the  characteri-
zation of long-term  climatic impacts of human activities.

      Identification of representative deposition  events and methods of aggregation are based on

-------
meteorological and statistical  analyses of deposition events that occurred over a 4-year period
(1982-1985).   Events during this period were divided into categories  with similar three-day
patterns of 850 mb wind flow across eastern  North America.  The three-day time period  and
850 mb wind  fields  are  broadly representative of the evolution of distinctive  storm  patterns
that are likely to affect atmospheric transport and deposition rates.  Events have been  selected
for  analysis by  RADM from among the  meteorological  strata  based  on their likelihood of
producing sulfate (804*) wet deposition at multiple locations across eastern  North America,
their frequency of occurrence, and their seasonality.                                  - -

      The method developed here represents  a significant  improvement  over  the alternative of
selecting events for  RADM  analysis purely at random.  Analysis  of observed  wet deposition
during the  1982-1985 time period shows that deposition rates are statistically  correlated  with
meteorological category with greater  than 95% confidence  at most  sites in  eastern North
America. The aggregation technique has been used to obtain estimates for annual deposition at
21 sites in  eastern North America based solely on observed deposition during the selected 30-
day ensemble.  The  resulting estimates differ from the observed 4-year average wet deposition
by amounts comparable to  normal  year-to-year variations.  Aggregated  estimates based on
meteorological strata  are also significantly  more accurate  than estimates  from  randomly
selected events.

      The aggregation project has served two purposes:

      1.    to provide  the RADM project with  a list of simulation  periods which  represent the
           range of storm patterns present over eastern North America.  The selection of
           storm types was based on their  likelihood of producing SO4= wet deposition at
           multiple  locations across eastern North America, their frequency of  occurrence,
           and  their seasonality; and

      2.    to provide  the RADM project with scaling factors for use in weighing episodic
           simulation results to seasonal and annual deposition.
1.2.       HISTORY

      The  use of synoptic climatological methods (meteorological  stratification)  to  identify
homogeneous weather types  is a popular tool for use in studying climate-environment rela-
tionships.  Kalkstein and Corrigan (1986)  presented a methodology to characterize  air  masses
at a given locale  using principal component analysis to reduce the dimensionality of  surface
weather variables followed by cluster analysis on the most important components.  The means cf
the original variables and other parameters were  determined for each category and a physical
explanation was attached to the results.  Kalkstein, et at. (1987)  refined the method and
demonstrated it  for two other  sites.  Samson and Moody (1986, see also  Moody  and Samson,

-------
1989) used cluster analysis to stratify deposition events at  a  single location based upon back-
trajectories.  The concentrations of S04°, NO3", NH3* and I-T in precipitation were found to be
statistically different between  strata but  there was  also a considerable amount of within strata
variation  and overlap between strata.   Moody and Galloway (1988) studied the  influence of
atmospheric  flow  patterns  (isobaric  back-trajectories)  on  the precipitation chemistry  at
Bermuda.  They identified the  important wind flow categories  for both warm and cold seasons and
concluded that transport was an important but  not sole factor influencing the composition of
precipitation.

      Initially this project attempted to  cluster patterns of precipitation  across eastern North
America.  The stochastic nature of daily precipitation patterns lead  to the production  of cate-
gories which, in essence, represented the days when  a particular station  dominated the precip-
itation of the region.  This course was abandoned in favor of a categorization scheme based on
wind  flow patterns.

      Fernau (1988) categorized three-day periods of the vectors of trajectories at precipi-
tation chemistry locations in  the United  States  and Canada.  These categories were shown to
resemble  meteorological patterns common to  eastern North  America.  Eighteen categories of
three-day wind flow patterns  were found using  cluster analysis.  These categories represented
periods whose mean three-day  atmospheric transport patterns were statistically different  from
one another.  Fernau  (1988) also showed that the  behavior  of SOA=  wet deposition varied
between these categories and that the differences held statistical significance.  Nonetheless,
there was concern that the results may have been influenced by the lack of rawinsonde  observa-
tions over the Atlantic Ocean, It was therefore decided that the analysis should switch, to use of
the  National Meteorological  Center (NMC) global meteorological  data  base which  included
analyzed  meteorological  information  at selected  pressure surfaces over all  of the  Northern
Hemisphere.
 1.3.       CONCEPTUAL APPROACH

      The aggregation  technique is based on  the hypothesis that  there are recurring weather
 patterns  which  can be characterized  and described in terms  of  observable meteorological
 parameters (wind fields, precipitation,  etc.).   It is  further  hypothesized that  distinct weather
 patterns  will be associated  with  characteristic patterns of wet deposition, both  in  terms of
 spatial extent and deposition  amounts.  Given those  assumptions, long-term wet deposition
 processes may  be analyzed based on  a finite number  of individual deposition events, each of
 which is "representative"  of a major meteorological category.  The goal of  the  aggregation
 project has been to identify the degree to which weather patterns may be separated into identi-
 fiable categories and  to assess how this information would improve our  ability to extrapolate
 case study results to longer-term deposition  estimates.

-------
      The  initial  stratification of events  into meteorological categories  is  used both in the
selection of individual events for in-depth analysis with RADM  and in the calculation of aggre-
gated estimates  for total wet deposition from 30-event samples.   Event selection has  been
designed to include representatives from all major meteorological categories  and to  insure that
the selected ensemble includes events with significant wet deposition at sites throughout eastern
North America.   Estimates for long-term deposition  are derived from known deposition  rates
(either from observations or from RADM  analyses) for the 30-event sample.  The aggregation
concept, illustrated in Figure 1,  represents long-term deposition as the  sum  of deposition from
sample  events weighted by the  frequency of occurrence of each meteorological category.  The
aggregation technique has  subsequently  been modified to make use of statistical correlations
between wet deposition, category and other meteorological  parameters (precipitation, season,
etc.) in  deriving  estimates for long-term deposition.

      Testing the aggregation technique poses  a special problem because the most important
issue, response  of the method to changes in precursor emission rates, cannot be tested empiri-
cally. A variety of techniques have been used to explore aggregation accuracy.  These include the
establishment of  statistical correlations between  observed wet deposition,  wind flow pattern and
other meteorological  variables  and application  of  the aggregation technique  to arrays  cf
randomly selected events.  Discrepancies between aggregate estimates and observed deposition
are compared to  the natural uncertainty  arising from year-to-year variation in  wet  deposition.
Lastly,  an  abbreviated model for source-receptor relationships  is used  to  identify  similarities
and differences  between randomly selected 30-day ensembles and long-term averages.  These
combined  tests  should establish both the level of accuracy and limitations  of the aggregation
technique.

      The  presentation is divided into a discussion of methods (Section 2) and presentation cf
results  (Section  3).  The  discussion of methods includes a description  of  the meteorological-
stratification technique, procedures for deriving  aggregated estimates based on finite samples,
selection of sample events based on the meteorological stratification, and  methods used to test
the aggregation procedure.  A description of  the simplified  model for source-receptor rela-
tionships is also included.  Results  of the analysis are presented in Section 3.  These include
meteorological  descriptions of  individual wind flow  categories, results of statistical correla-
tions between wind flow and observed wet deposition, selection and identification of thirty 3-day
events  for RADM analysis, and aggregated estimates for wet deposition and  source receptor
relationships based on the chosen sample.

-------
  Group 1         Group 2
               Group 3
I
Group n
                                                                        r
                      D,(x,y)
                D,(x,y)
  Dn(x,y)
                                 N
                  Wt(x,y)=If1Q(x,y)   ±e(x,y)
                                n=1
Figure  1.   Illustration of the aggregation concept.  Each meteorological category occurs with
            some mean frequency, /„, and has associated with it a mean deposition pattern
            Dn(x,y). These are aggregated through a weighting scheme (though not
            necessarily the simple frequency weighting shown) to achieve an estimate of the
            spatial pattern of  deposition, Wt(x,y) with  an uncertainty e(x,y) which includes
            variability  in frequency and deposition patterns  from year-to-year.  The goal is
            to estimate WJx.y) accurately and keep e(x,y) less than the observed magnitude
            of interannual variation  in  deposition.

-------
                                    2.    METHODS
2.1.       AGGREGATION APPROACH

2.1.1.    Meteorological  Stratification

      The  method  for selection of representative  3-day deposition  events is  based  on a
stratification scheme that classifies all events into categories  based on  meteorology.   The
purpose of the meteorological stratification is to systematically identify the different types of
storm patterns that  contribute to acid deposition in the eastern United States.  Category-based
selection  will  then  insure  that   all major storm types  are  included in the  ensemble  of
representative  events.   Stratification schemes were attempted based  on  a  variety  of
meteorological  parameters,  including  sea  level  pressure,  temperature,   850  millibar
geopotential height, vorticity advection and wind fields.

      To portray  the "life-cycle" of a typical eastern North American  "storm"  the duration of
RADM simulations  has generally been  on the order of three days (not counting initialization
days). To  allow compatibility with RADM the categories have been based on three-day  periods.
This necessitated re-organizing the meteorological data into  running three-day  periods (record
one = days 1, 2, 3; record two = days 2.  3, 4; etc.) prior  to stratification.

      For purposes of categorization each  three-day event is evaluated based on the values
assumed by 2 independent  meteorological parameters over the three-day period  on  a  48-point
horizontal  grid covering eastern  North America.   Each three-day event is characterized by  arv
array of 268 parameters, which consist of  the values of  the 2 independent parameters on 3
successive days at 48 grid points.  Location of  grid points  is  shown  in Figure  2.   The 288
parameters represent the  maximum number of parameters that could  be included in  the
meteorological stratification, given computational constraints.  The choice of 48 grid points  and
two independent parameters were dictated by this constraint.

       Classifying all of the running three-day periods into  similar meteorological  categories
required computer  assistance.  The statistical method of cluster  analysis was selected for this
purpose.   Ideally  the data  from the entire time  period should have been clustered together,
however, due to the computer memory  constraints of the  statistical software used this was not
possible.   To circumvent this limitation, consecutive  periods (record 1 - days 1,2,3; record 2
« days 4,5.6; etc.)  from independent years were selected.  It was  anticipated that this would

-------
        10
65
Figure  2.   NMC grid and subset of 48 grid points used to describe similar meteorological
             patterns.

-------
minimize  the  effect  of  missing data  and  capture  a  broader cross-section  of situations.
Additionally, the limitation on computer memory allowed only the use of three  years of data.
Due to this limitation the  cluster analysis was used to  generate meteorological categories based
on the years 1979. 1981 and 1983 only.

      Ward's method of  cluster analysis (Ward, 1963) was selected to stratify the reduced set
of cases.  The goal of Ward's method, as with other cluster analysis techniques, is to divide a data
set into categories consisting of similar cases that are as distinct from the other categories as
possible.  Evaluation of  similarity between events  is based on the  euclidean distance between
events for a specified array of parameters, defined as
                                                   - Pa .1                            ( 1 )

where XAB is the euclidean distance between points A and B and PAj and P8j. represent the value
of the jtn  parameter for events A and B respectively.  Euclidean  distances are used to compare
both individual events and category means in Ward's method.

      Identification of individual categories follows  a hierarchical  sequence  beginning  with a
case in which each individual event is regarded as a separate category. Successive categorization
schemes  are then generated, with the  total number of categories reduced by one at each stage.
Category  selection is performed in a manner to minimize the  cluster variance ,  or  within-group
sum  of squares (£).  The cluster variance is the sum of squared Euclidean distances from each
data point to the  centroid (mean) for its assigned category, with  distance defined in Equation
(1).  Cluster variance provides a measure of  the degree of  similarity  within  each category in
terms of  the selected meteorological  parameters.  The minimal-variance criterion  insures that
events within the same category are most similar to each other.

2.1.2.    Selection  of  Stratification  Parameters

       Stratification based  on Ward's method generates optimal categorization schemes  for the
specified  base of  meteorological parameters and  for the  specified number of categories.
Completion  of the stratification process  requires  selection of  the  appropriate  number cf
parameters and  number of  categories.  A large number of individual  categories permits  finer
distinctions between meteorological events, but a large  number of categories becomes unwieldy
and lacks statistical robustness.

       As  discussed by Anderberg (1973, p. 194 and 212) evaluation of  clustering may be based
on its  effectiveness in  distinguishing an external  variable,  rather  than  the parameters  that
formed the  basis  of  the  initial stratification.   In this  instance, generation  of meteorological
strata is based on the meteorological  parameters discussed in the  preceding  section.  However
the evaluation of the proper  number of categories  is based on the ability of  the  strata to
distinguish between events with high and low S04* wet deposition.

                                             8

-------
      Observed SO4" deposition is available from 21 sites (See Section 2.2 below).  Deposition-
based differences between events are established by determining Euclidean distances (Equation 1
above) with S04° deposition  at the individual sites as the  independent parameter.  Cluster
variance, defined above,  may also be calculated based on  differences in SO4" deposition and
represents  similarity  (in  terms of SO4* deposition) of events within the same category.  Note
that the cluster variance based on SO4= deposition is different from the variance  based on
meteorological parameters which are used to define the meteorological categories.

      The separation of categories in terms of SO4* deposition was determined by comparing the
total within-groups  error sum  of  squares  (E).  If a category  of n, units  is merged with a
category of n2 units,  the within-group sum of squares for the merged categories equals the sum
of the within-group sum  of squares of the two  categories merged plus  n^n2l(n^ + n2) times the
square of the Euclidean  distance between the means of  the two categories.  A merger thus
increases the within-groups  sum  of squares  by  a quantity proportional  to the (distance)2
between the merged  categories.  So if Eg^  is the total within-groups error sum of squares when
there are g+1 categories  and Eg is the same quantity when two categories are  merged,

                                      ft-fii.  ,      .  ,       %
                                                                                     (2)

where na, nb, are sizes of the categories and Z~a and Z~b are the column vectors of  their means.

      Initially, when  each of  N data records is a  separate category there  is no within-group
error sum of squares; hence EN = 0.  When multivariate analysis of variance is carried out for g
categories  the total sum of squares  of the within-groups  matrix is Eg and the sum of squares of
the between-groups matrix Bg (used below) is defined as
The relative advantage gained by increasing the number of categories may be evaluated based on
the  reduction in within-group sum of squares  (Eg ) with increasing categorization.   Two other
measures  of the  marginal effectiveness  of  increasing numbers  of categories  are Eeale's
criterion  (Beale, 1969), computed at each step as:
                                                          • N
and the criterion of Calinsky and Harabasz  (1974), computed as
The Beale (F ) and Calinsky (C) criteria both provide measures for the  decrease in within-

-------
cluster variance and the increase in variance between clusters as the number of steps increases.
If Fis large, g categories are better than g-1. The "best" number of categories was obtained by
plotting F against g and choosing the value of g corresponding to large F. Similarly, the value of
g which maximized C was the best  value according to the C criterion. A monotonically decreasing
C indicated a hierarchical structure.  However,  C, rising to a maximum  at g, suggests the pres-
ence of g categories.


2.1.3.      Categorization  for  Other  Years

      As described above, meteorological categorization is developed based on data from three
individual years,  1979,  1981  and  1983.  Once  the categorization is established  it is necessary
to identify  appropriate categories  for running three-day periods for other years.  To classify a
particular case  into a  given  category, the 288 meteorological parameters that  define each
category are compared to  parameter values for the  individual 3-day event. This was done by
computing  the Euclidean distance  (Equation 1)  between the  event  and  the category  mean.  The
events are  assigned to the category producing the smallest Euclidean distance.


2.1.4.     Aggregation  Procedures

      The  goal of the aggregation  procedure is to estimate annual deposition of sulfate and other
species based on a finite number of episodes. Ideally, an aggregation procedure would be derived
based on observations that correlate changed deposition during  individual events with changes in
long-term  deposition  patterns.   Data of this sort  is  unfortunately  not  available from  trie
historical record.  However a reasonable approximation to the desired situation may  be obtained
by  developing procedures for estimated annual deposition based on finite samples  of randcrr.iy
selected events.  We postulate that future changes in deposition during an  individual event  rr.ay
be  correlated with changes in long-term deposition in the same  way that observed deposition for
a random ensemble may be correlated with observed long-term  deposition.

       The aggregation approach makes use of the varying frequency of occurrence and expected
S04* and  NO3" deposition within each category.  As  will be  demonstrated below (Sections 2.3.2
and 3.1.5)  selection of days by category provided a more accurate estimate of  annual deposition
than is provided  by purely random selection.

      An initial  estimate for annual deposition may be derived from  an array of  samples by
projecting the average rate of deposition over a 365-day period.  When an ensemble of days is
selected based on subjective  criteria (such as from extreme  events), projection over a  365-
day period is more likely to result  in a biased sample, since the selected categories ensure that
deposition  on the sample days will differ from average deposition.  The sample deposition must
be  adjusted by scaling factors in order to arrive at an unbiased estimate.

      In addition tojhe scaling factor, a weighting factor is applied in the summation of depo-

                                            10

-------
sition from  the individual days.  The weighting factor (wt) reflects the frequency of  occurrence
of categories, so that days which represent frequently occurring categories are weighted more
heavily than days which represent unusual meteorology.  The weighting factors are equal  to the
frequency of occurrence, /j, of each category divided by the number of days from the category,
HJ, included in the representative sample of days:
                                                                                       (6)
      The aggregated estimate for annual deposition has the following form:

                           DEPOSITION,  Jfrnnu*'   T 7  v^Di                         ( 7 )
where D\ is the observed deposition during the ith event of the sample to be used in the aggrega-
tion;  W| is a weighting factor that reflects the relative frequency of occurrence  of the meteor-
ological category associated with the Ith event;  Dannuai is the average annual deposition over the
period  1982-1985 at the site;  and  5, represents  a statistical  estimate for deposition  during
the ilh event, based  on available information  prior to selection of the individual event.

      The  statistical estimate for deposition  (b-t) varies depending on the nature of the sample to
be used in the aggregation and on the specific purpose of the aggregation.  When sample  events
are selected at random, b-t is equal to the average deposition per event, and the ratio of Danrvjaj to
^ w-tbt simply projects the sample deposition  over a  365-day period.  When  sample  events
are selected based on meteorological categories, 5; represents the average deposition for  events
within the  category associated with the itn event. The accuracy of the aggregate estimate  can
further be improved by using statistical correlations  with  other factors  in  estimating Dj.  For
example, it will be shown below that S04= and NO3' wet  deposition for individual  events  can be
statistically correlated  with precipitation, and that  an estimate  for deposition  may be developed
by linear regression:
 where  Bcat(j) represents  the  least-squares  estimate (i.e. average deposition-to-precipita:icn
 ratio)  over all  events within  the meteorological category  containing  event /', and P, is the
 precipitation amount  for the event.
                                             1 1

-------
TABLE 1. A SAMPLE AGGREGATION OF ANNUAL SO4" WET DEPOSITION CALCULATED FOR BIG
 MOOSE, N.Y., BASED ON A 30-EVENT SAMPLE. THE AGGREGATION PROCEDURE FOLLOWS
   EQUATION (7) WITH 6, EQUAL TO THE AVERAGE DEPOSITION ASSOCIATED WITH THE
      METEOROLOGICAL CATEGORY. CATEGORIES ARE DESCRIBED IN SECTION 3.1.
EVENT

4/16/82
1 1/3/85
5/30/83
3/18/84
7/13/85
9/10/83
10/12/85
1 1/2/83
8/22/84
6/14/84
4/3/85
9/1 0/84
9/15/83
9/25/82
8/5/83
1 1/1 7/85
9/27/85
7/16/84
12/16/82
4/30/82
9/14/85
7/23/85
12/24/84
1/2/84
1 1/6/83
2/2/82
12/10/83
1 1/5/85
3/30/85
1 1/14/83




CATEGORY

1W
3W
5W
5W
6W
6W
6W
7W
7W
7W
8W
9W
9W
9W
9W
10W
11W
11W
11W
7D
8D
8D
10D
18D
19D
2W
4W
12W
14W
17W
ZN A
1-Wi&i
y^.WiDj.
^annual =
estimated
WEIGHTING
FACTOR (Wj)
0.0397
0.0380
0.0463
0.0463
0.0419
0.0419
0.0419
0.0397
0.0397
0.0397
0.0545
0.0331
0.0331
0.0331
0.0331
0.0397
0.0391
0.0391
0.0391
0.0231
0.0231
0.0231
0.0314
0.0397
0.0397
0.0149
0.0066
0.0116
0.0182
0.0099
•B
=

deposition «
DEPOSITION FOR
CATEGORY (6 ,)
55.5
15.2
49.0
49.0-
35.8
35.8
35.8
56.1
56.1
56.1
33.0
35.5
35.5
35.5
35.5
48.6
42.5
42.5
42.5
1.0
2.6
2.6
8.1
3.8
1.4
8.4
18.1
19.9
10.1
17.6
33.6
31.4
3107 mgm'2
2910 mgm"2
DEPOSITION FOR
EVENT (Dj)
174.9
0.0
82.3
0.9
7.9
44.8
68.8
53.8
108.4
40.9
0.0
65.0
0.0
8.0
0.0
19.2
18.1
35.8
27.4
0.0
0.6
9.0
6.5
4.2
2.2
10.2
5.3
6.9
0.0
0.0




                                   12

-------
     The procedure for deriving estimates for annual deposition may also be modified to derive
an estimate for seasonal deposition (i.e. summer half-year, winter half-year, etc.).  Seasonal
estimates may be derived from a given ensemble  of events by application of equations 6 and 7
along with the following changes:

     (1)   Average annual deposition for 1982-1985 (Dannual) is replaced by the  average
           deposition for the season in question  (Dsummer, Dwinter).

     ( 2 )  The weighting factor (w-() is set equal to zero for all events outside the specified
           season, so that summer deposition is estimated based on summer events only, etc.
      ( 3 )  The estimated deposition ($ is derived based the 1982-1985  average for events
           from the specified season, rather than from all events.

      An example of the aggregation procedure is shown in Table 1. Aggregated annual deposi-
tion is calculated for Big Moose, N.Y., based on a 30-event sample. The aggregation procedure
follows  equation  (7)  with  6| equal to the average deposition for  events  associated with  each
category1.  Weighted sums are ^ .w,fi>,  and
2.1.5.    Selection  Process

      Selection of individual events for detailed analysis  by the RADM simulation meets, two
broad criteria.  First, the selection process must provide an ensemble of events that are broadly
representative of the range of meteorological conditions that contribute to wet and dry deposi-
tion in eastern North America.  Note  in particular  that the ensemble must include representa-
tive deposition events at sites throughout the region, rather  than at a few sites only.  Second, the
selection  process is based on rigidly defined objective criteria, rather  than  on  subjective
judgement.

      The selection process involved  generation of groups of individual events from the 1 982-
1985  period, each of which reflected desired characteristics  for the final event ensemble.  The
grouping of events in the selection process was based partly on the  meteorological categories
(Section 2.1.1) and partly on criteria discussed below.   Once the groupings were determined, a
series of event ensembles were chosen as candidates for final selection. The candidate ensembles
were each composed of a pre-determined number of events  from  each category, with the indi-
 The reader may note that the estimated deposition reported in Table 1 differs from the estimate
  for Big Moose shown in Figure 21 b below.  The final estimates shown in Figures 21-25 make
  use of observed correlations between deposition and precipitation ( Equation  8) to calculate
  Dj .  The simpler calculation illustrated  in Table 1 ignores precipitation and sets 5j.equal to
  the average deposition for the meteorological category.

                                            13

-------
vidual events selected randomly from within each group.  These candidate ensembles were then
put through a two-part  evaluation procedure to establish the final selection.  This evaluation
was based on calculation of aggregated estimates for annual SO4= deposition and  total annual
precipitation.

      The grouping of events  for selection was based primarily on meteorological categories
generated by the wind-based cluster analysis described  above.  The meteorological categories
identified by  the cluster  analysis- were  assumed to  represent characteristic atmospheric
circulation patterns over eastern  North America; and selection of events from these categories
assures that an appropriate range of circulation patterns are represented in the final ensemble.
In addition, the  initial cluster-based categories were further divided into subcategories that
reflect desired characteristics for the ensemble. Since our primary concern is to represent wet
deposition processes, it was decided to select the majority of events from groupings that include
substantial wet deposition.  To this end  the  initial categories were divided into "wet" and "dry"
subcategories based on SO4" wet deposition received at 13 eastern sites (sites are described
below).  The "wet" subcategories consist of  events with greater than SO4" wet deposition; the
"dry" subcategories consist of events with less than median SO4" wet deposition.

      Candidate  events were selected primarily from "wet" subcategories, although  a number of
extreme "dry"  events (with  little wet deposition  or  precipitation  at the  eastern  sites) were
prescribed  to permit evaluation  of dry deposition processes.  The ensembles of candidate events
(candidate  ensembles) generated based on these criteria are presumed to represent the desired
range of atmospheric circulation  patterns  as  represented by the cluster analysis.  However the
candidate ensembles do not necessarily represent typical patterns of wet deposition  or precip-
itation, as  neither deposition nor precipitation have been accounted  for by the meteorological
categories.   In particular, candidate ensembles may include abnormally high precipitation and
deposition at certain sites and abnormally  low deposition at others.

      Evaluation  of deposition  and  precipitation in the candidate ensembles is accomplished  by
aggregating estimates for annual and seasonal wet deposition and precipitation  from candidate
ensembles.  The aggregation procedure essentially represents a projection of annual deposition
based on (a) long-term average deposition  (or average deposition-to-precipitation ratios) for
the meteorological categories  from which events  were  selected; and (b) deposition received
during the  individual events of  the  candidate ensemble.  When deposition and precipitation for
candidate events are broadly representative  of average conditions for the associated  meteoro-
logical category,  the aggregation procedure  yields a reasonably accurate estimate for annual
deposition.  When the  candidate ensemble  includes extreme or  unrepresentative events, the
aggregation procedure yields large errors.

      The candidate ensembles are evaluated based on three separate aggregation procedures.
These are:
                                           14

-------
     (a)   Aggregated wet deposition, calculated based on the difference between ensemble wet
           deposition and average wet deposition within each meteorological subcategory;

     (b)   Aggregated wet deposition, calculated based on the difference between ensemble and
           average  wet deposition-to-precipitation ratios;  and

     (c)   Aggregated annual precipitation, based on the difference between ensemble precip-
           itation and average precipitation within each  subcategory

The three  aggregation evaluations tested the extent to which the candidate ensembles are char-
acterized  by:   (a)  wet deposition rates, (b) wet deposition-jo-precipitation  ratios, and (c)
precipitation rates that represent average values for the chosen meteorological subcategories.

     Accuracy of the aggregated estimates for each candidate ensemble is evaluated based on the
root-mean-square  difference  (RMSD) between estimated and observed deposition over 13
eastern sites, calculated by
                                              ,1  PJ - o,r                         o)

where b-. .is the aggregated estimate for annual deposition at the /n site from Equation (7) and
Dj.is the observed annual deposition (1982-1985).  In the final selection process 20 candidate'
ensembles  were generated and the candidate with the lowest RMSD for in the three aggregation
tests was selected for use by RADM.

2.1.6.     gource/ReceDtor  Relationships

2.1.6.1.     Trajectory Calculations—
      The transport of air to measurement locations was estimated  using the NOAA ARL-ATAD
trajectory model  (Heffter, 1980).  This model calculates the mean upwind  track  of the air
within  the  "mixed-layer" of the atmosphere  (a layer of variable  height  extending  from  the
surface to roughly 1500  m above the surface in the  summer and 800 m above the surface in the
winter).  Rawinsonde measurements of wind,  temperature and  humidity  in the vertical were
used to calculate the trajectories.  The upper  air observing network recording these data over
North America has a spatial resolution of about 500 km and a temporal resolution of 12 hours
with measurements taken at  OOZ and 12Z (Z=GMT).  The trajectory  model interpolates the
upper data in time and space (1/r2), utilizing  all information within 400 km of the  receptor or
previous trajectory endpoint, to determine the backward motion of the air. The model outputs a
new position (latitude and longitude) for  the air parcel after each  three-hour  period  out to  72
hours  (24  three-hour segments).   Four  trajectories,  arriving  at  the  receptor at  OOZ,  06Z,
12Z and 18Z, are calculated for each  day.
                                           15

-------
      Determining the mean wind velocity  along each trajectory segment required a value for
the height of the mixed-layer at that location and time.  This value was estimated for OOZ and
122 by scanning the vertical temperature profile for the height of the first non-surface based
inversion. When no such inversion was found the top of the layer was set at 3000 meters above
the elevation  of the  nearest  rawinsonde station.  The OOZ  and 12Z estimates  are linearly
interpolated in time to yield a mixing height for  the trajectory segment in question.  The wind
velocity measurements taken below this height were then used to calculate the mean motion of
the mixed-layer  over the next three-hour period.  This procedure is repeated up  to  24 times
(72 hours back  in time) or until there is  no data within 400 Km.   As with all methods for
calculating  trajectories those employed  here contain uncertainties which arise due  to the fact
that the data are limited (cf. Kahl and Samson, 1988) and because it is not possible to model in
detail all of the complex motions within the  atmosphere.  ARL-ATAD trajectories were compared
to data from the Cross-Appalachian  Tracer Experiment (CAPTEX).  The results showed lateral
displacement uncertainties  on the order of  100 to 200 Km at 24 h upwind and 100  to 400 Km
at 72 h  upwind  (Kahl and Samson, 1986; Draxler, 1987).  Relative to southwest  trajectories
arriving at Big Moose, NY these errors amount to an an area about the size of the state of Ohio.
These uncertainties  are  not significantly different than those  resulting from other methods  of
calculating  trajectories.   Given the availability of the mixed-layer  trajectories,  they  were
selected for the  assessment of source-receptorrelationships  within the context of this study.  !n
using them it was assumed that they  represent the most probable path of any air parcel advected
with the  mean motion of the mixed-layer.

2.1.6.2.     Quantitative  Transport  Analysis—
      The transport potential for a given sampling time includes the mean transport computed
using the trajectory model  plus the  horizontal spread imposed by atmospheric dispersion and
uncertainty in  transport estimates (Samson,  1980).   The probability of a reactive, depositing
tracer arriving  at a point x  at a time  t, Arfx.t), can be  expressed as

                                    t  + 00 +00
                         Af(x,{) -  J  J   | T(x.t\x',n  dx' dr                   (10)
                                   t-T -oo  -oo

where 7"(jr,fljr',f), the potential  mass transfer function in  two dimensions, is  defined by

                    Tfx.tix-.n  -  Q(x,t\x:n  Ra\n  Dfx-.n  Afx-.n               n 1)

and O(x,flx',0  is the transition  probability density function of an  air parcel located at x' at
time t1 arriving  at a  receptor  x at  time t, R(t\f)  is  the probability  of  the  tracer not being
reacted to  another species from time t' to  time t, D(x',tr) is  the probability that the tracer was
not dry deposited at (x',0  and A(x',0 
-------
      The  transition  probability density function,  Q(x,t\x',tl),  is  most important in  under-
standing source-receptor relationships since it describes the motion of the air transporting the
pollutants.   If a  particular  source is  located at x' at time t' then the probability that it will
impact upon the receptor at x at time t, is given by  ,Q(x,i\x',t*).   The  last three probability
functions on the  right side of (7) are not well defined  for soluble elements.  In this work they
are assumed to equal one so that the effects of dry and wet deposition and chemical reactions are
neglected.  The main drawback to these  simplifications is that they eliminate the possibility  of
determining the relative importance of local versus distant sources.

      The  transition  probability  density  function  can be approximated  from the  upwind
trajectories.  The axis of the computed trajectory is assumed to represent the  highest proba-
bility at any time  upwind for contributing to the trace element composition at the monitor. The
spatial distribution  of the transition probability  function away from the axis of the  trajectory
depends upon the  degree of vertical mixing,  coupled  with the magnitude of the wind velocity
shear.  As a first approximation, it has been assumed that the "puff" of transition probability  is
normally distributed about  the  trajectory with a standard  deviation which  increases  linearly
with time  upwind  (Samson, 1980; Draxler, 1982).    Thus  Q(x,t\x',t") is assumed to  be
expressed as:
                                                     • 2      ,,«2
                                                   2a/

where x" = X - x'(f) and y" = Y - y'(t")  with (X,Y) being the coordinates of  the grid and x'(0
and y'(0  being the coordinates of the  centerline of the trajectory.  If x" or y" is greater than
3ax or 3ay which are  defined  below,  then segment / is not counted in grid box (X, Y).  It is
assumed that cx and ay can be approximated by

                                   ax(f) = oy(t') = a t1                             (13)

with  a dispersion speed, a, equal to 5.4 km  hr"1 (Samson, 1980).  To determine  the overall
probability  field  (Q-field),  Q(x,^x',t") is integrated over time  and space  as in (10).  This
field shows where over time period, T,  the air was  most and least  likely to have originated;  i.e.
the probability of the  air over any point moving over the  receptor during the time period, -.  In
practice  this is  carried out  by  summing the values of  Q(x,t\x',t')  in each grid box over  all
trajectory segments and dividing by the  total number of segments.

      Various weighting schemes can be  applied in the determination of the Q-field to derive  an
"implied"  transport bias (Keeler and  Samson, 1989).   For example, trajectories  associated
with  large amounts of pollutant deposition or precipitation  can be given more weight  so that  the
resulting probability fields are skewed towards those regions more often associated with these
types of events.  The weighted Q-field, Ow , is determined by
                                            17

-------
                             nseg
                                     Wt        r  / x"2      v"2 N~l
                                                                                    (14)
where x", /", ax and oy are defined  as  in  (12) and (13).   The total  number  of  three-hour
trajectory endpoints (segments), nseg, depends upon the number  of trajectories considered.
Each one can contribute  up to 24 segments  (72-hour back-trajectories) depending  on the
availability of meteorological information.  The weighting factor applied is given by,
Dj is  the value of the factor used in the weighting associated  with trajectory, j, (eg. wet S04=
deposition) and Dtota) is the total value of that factor over all trajectories.  The weighting factor
is also scaled by nseg, the total number of non-missing three-hour segments, and by mseg, the
number of  non-missing segments  associated with trajectory j.  Trajectories arriving  at the
receptor that  are  associated with large weights (i.e.. during high deposition events)  are thus
counted more heavily than  those associated with smaller weights. For an individual  receptor the
Ow field produced  when weighted by S04" wet deposition indicates the probable pathway associ-
ated with high and low deposition amounts.

2.1.6.3.     The Potential  Source Contribution  Function  —
      The potential source  contribution function (PSCF) is described by Zeng et al. (1989}.  It
basically consists of a ratio  between weighted and non-weighted Q-fields.  The  main differences
between the approach of Zeng  era/. (1989) and the approach followed here are  that  in this study
uncertainties  in the  trajectories, which are accounted for by  Gaussian  "puff-like" spreading
(equation 12), are assumed to exist.  Zeng et al. simply  counted the  number of  trajectory
endpoints in each grid box when generating the probability fields.  In  addition, all trajectories
are included  and  weighted by deposition in this study, while  their  weighted transition  proba-
bilities were determined by only including trajectories associated with events exceeding  a pre-
set concentration (equivalent to the  average  concentration at the site in question).  The PSCF as
employed in this study is determined by:

                               ocrc   Q(M*'.OweiQhted
                               PSCFm _                                             (16)
                                      U(X,f|X -Onon-weighted

Values greater than one imply  an enhanced probability  of the air residing over that  region prior
to high wet SO4* deposition events.
                                            18

-------
2.2.       DESCRIPTION OF  DATA BASE

2.2.1.    Meteorological  Data

      Meteorological data  were  obtained from the  National Meteorological Center (NMC) global
analyses, available at the National Center for Atmospheric Research (NCAR). This data set was
selected because it  includes information over the oceans.  An alternative data set  based on
rawinsonde observations was used in a preliminary study (Fernau, 1988), but  was rejected for
this effort  because it failed to represent circulation over the western  Atlantic Ocean, crucial for
the development of storms affecting eastern  North America. In addition, the NMC data offered
better spatial resolution over the entire region of interest.  The NMC data also included infor-
mation from a larger variety of sources and had been subject  to more  extensive quality control
procedures to insure accuracy and consistency (Trenberth and Olson, 1988).

      Synoptic and  asynoptic  meteorological data were collected via  NMC's  Global Data
Assimilation System  (GDAS) twice daily from numerous sources. Included among these were:
surface observations, rawinsondes, aircraft, ships, satellites and buoys.  In compiling the final
analysis NMC permitted a data collection cut-off time  of up  to ten hours'beyond the actual
analysis time. The irregularly spaced data (time and space) were interpolated  to a regular 2.5°
grid across the entire globe for both OOZ and 12Z.  To calculate the data values  at the grid points
an optimal interpolation scheme was used and the first guess fields for the interpolation were
generated from six or occasionally twelve hour global model predictions.

      The Global Data Assimilation System at  NMC, has been  evolving since September 1978.
Changes  were largely implemented  to improve  the accuracy,  coverage and/or detail  of  the
analysis.   Other  modifications were  necessary to  resolve  operational  difficulties as they
occurred or were discovered. The details of the changes that  occurred, and their effect  on  the
analysis,  are  discussed in  Trenberth  and Olsen (1988).  During  the 1978-1987 period  the
majority of the changes had no effect on  the information used in this study.  However, from
March 1984 to March  1986 there  were difficulties that resulted in  up to  a 5  m-sec"1 error in
the 850 mb level  winds.   This problem was attributed to small scale noise  in the 1000  mb
temperature field and it is believed that the resulting errors were not systematic.

      NCAR  houses NMC  data from July  1976 to nearly the  present for  temperature, relative
humidity, geopotential height, zonal and meridional wind components and pressure at up to 14
different levels.  During this period, however,  NMC failed to  archive portions of the data and
thus there are periods  of missing information in the data set.  The amount  and type of missing
data were also  summarized by Trenberth  and  Olsen (1988).  In the years 1978-1986 an
average of 7.8% of the data were missing.  The minimum amount lost was 2.5% in 1986 and the
maximum  was 13.6%-in both 1982 and 1984.

                                           19

-------
      For this study we obtained the  information from NCAR for the years 1979-1985.  The
14-layer,  2.5°  global data were reduced to 5° and 5  layers for OOZ covering eastern  North
America.  The choice of 5° intervals is discussed in Section 2.1.1.  The extent of the horizontal
grid is shown Figure 2.  The OOZ data  were then arranged into overlapping three-day records in
order to correspond to the three day duration of RADM simulation periods.  The first record of
the year contained information  from days 1, 2 and 3; the second record contained days 2, 3, and
4 etc.. This process increased  the amount of missing cases since a three-day record was not
considered complete unless all  three days were available.  One isolated missing day from the
NMC data would become three  missing cases because it would be included in three separate cases.

2.2.2.     Precipitation  Chemistry   Data

      Precipitation chemistry  data  were obtained from event  precipitation  chemistry  moni-
toring locations.  The data in the United States were obtained from the Utility Acid Precipitation
Study Program (UAPSP)  which provided  a six-year (1982-1987), 19-site  precipitation
chemistry data set covering the eastern United States (Mueller and Allan, 1987).  The UAPSP
data were augmented by precipitation chemistry measurements from two Canadian sites.  Figure
3 shows the location of sampling  stations chosen for  this study.  Table 2 lists the number and
location of the  sites.  Since the focus  of this study is on eastern sites with high deposition the
precision of the aggregation technique is tested for sites located in the  northeast  and midwest,
including the southern border and Ontario sites but excluding sites from the deep  south and
Great  Plains.   The sites commonly included in  northeast/midwest analyses are  identified in
Figure 3 and Table  2.

      The UAPSP data provided extensive information including: the start day and hour of the
first precipitation event, the end day and  hour of the last precipitation event (occasionally the
timing of  multiple events were  specified), the precipitation amount,  the pH, and a thorough
chemical analysis that included  the  determination  of the major ionic species in solution.  The
only meteorological information included in the data set were in the form of sample codes  which
distinguished events as either (1) snow and ice events, (2)  mixed (liquid  and solid sample)
events or hail events, (3) rain (uncoded events), and  (4) events with  thunder.

      All chemistry records were  examined for incorrect data.   Some errors in the  dates and
times of  precipitation begin/end and sample removal date/time  had to be  corrected or set to
missing.   Midpoint times and  durations  were calculated from the  precipitation  begin/end
information.  Some chemical measurements were discarded, including  phosphate, total acidity,
aluminum,  sulfite. nitrite and  total  organic  carbon.   Hydronium  ion concentration  was
calculated using the lab pH or, if missing, the field pH  according to the formula
                                           20

-------
Figure  3.    Location of the precipitation chemistry sites in the United States and Canada used
              for analysis of transport patterns.  These sites represent locations  of daily
              precipitation monitoring locations.  Sites indicated by solid circles were used to
              test precision of aggregation technique.

-------
   TABLE 2. NAME AND LOCATION OF DAILY (EVENT) PRECIPITATION CHEMISTRY SITES.
INCLUDES SELECTED SITES FROM THE UTILITY ACID PRECIPITATION STUDY PROGRAM (UAPSP)
    AND FROM TWO CANADIAN NETWORKS. ASTERISKS (*) DENOTE NORTHEASTERN AND
    MIDWESTERN SITES USED TO TEST THE PRECISION OF THE AGGREGATION TECHNIQUE.
SITE

* 1
* 2
' 4
•5
* 6
• 7
* 8
*1 0
* 1 1
1 2
•1 3
1 4
15
1 6
1 7
1 8
1 9
*20
*21
24
25

•27
*28
LOCATION
UAPSP
Turners Falls, Massachusetts (> 8/80)
Tunkhannock, Pennsylvania
Zanesville, Ohio
Rockport, Indiana
Giles County, Tennessee
Fort Wayne, Indiana
Raleigh, North Carolina
Gaylord, Michigan
Clearfield, Kentucky
Alamo, Tennessee
Winterport, Maine
Uvalda, Georgia
Selma, Alabama
Clinton, Mississippi
Marshall, Texas
Lancaster, Kansas
Brooking s, South Dakota
Underhill, Vermont
Big Moose, New York
Shawano. Wisconsin
Round Lake, Wisconsin
CANADIAN
Dorset, Ontario
Longwoods, Ontario
LATITUDE

42°59'
41°35'
39°59'
37°53'
35°17'
41°03'
35°44'
44°57'
38°08'
35°AQ-
44°37'
32°03'
32°28'
32°21'
32°40'
39°34'
44°20f
44°32'
43°49I
44°42'
46°13'

45°13'
42°49'
LONGITUDE

72°55'
76°00'
82°01'
87°08'
86°54'
85°19'
78°4V
84°39'
83°27'
89°08'
ea'ss1
82°29'
87°05'
90°17'
94°25'
95°18'
96°50l
72°52'
74°54'
88°37'
91°56'

78°56'
81°24'
                                   22

-------
                                H* (nmoles I"1)  =  106'pH.                           (17)


      Quality flags were also included in the data set.  They were processed after the method of
Endlich et al.  (1986) and  the  major steps  in this process are described  as below.  All
measurements were set to missing if the sampler was inoperative, water was in the dry bucket,
or agricultural activity, dust or smoke were reported  in the vjcinity of the  sampler.  If lab  and
field pH differed by more than 0.7 below pH 5 or 1.0 above pH 5 both measurements were'set to
missing.  Events with sample precipitation volume  differing from gauge precipitation  volume by
more  than  30%  (a measure of collection  efficiency)  were flagged.  Ion balance  was checked
using  the  following formula:

                          I Total Anion - Total Cation |
                          (Total Anion  + Total  Cation)/2 ~ °'6                     (1 8)


      Events failing (value s 0.6) were coded as missing.  Anions and cations used  in the  test
were  nitrate, sulfate, chloride, hydronium ion,  ammonium, sodium,  potassium,  calcium  and
magnesium.  Measurements  below detection level were flagged and negative values were set to
zero.  After application of the percent test,  as described below,  multiple events on  the same  day
were  totalled  and events captured by co-located samplers  were combined using  arithmetic
averages of the  measurements.  If one of the  two  co-located samples was considered to be of
better  quality either  overall or for a given  analyte then  only  the  less suspect  record or
measurement was kept.

      In  an attempt  to detect outliers, selected  data were subjected to the "percent test"
(Endlich, et al., 1986). The test was only applied to events which were suspect due to leakage,
contamination, pH disagreement, ion balance  failure  or collection efficiency failure.   In
performing the percent test the data  for  all sites  are grouped together by  analyte. For each
analyte the 90th, 95th, 99th  and 99.9th percentile are found using  the pooled data.   Then the
analyte values at each site are compared  to these global percentiles.  For a given precipitation
event, if one analyte  exceeds  its 99.9th percentile, two analytes  exceed their 99th percentiles,
three  analytes exceed their  95th  percentiles, or four  analytes exceed  their 90th percentiies the
event is considered to have failed the percent test and is flagged as such.  Flagged events were
eventually treated as  missing.

      The UAPSP data only cover the  United States, however, southern Ontario also experiences
large  annual  SO4* wet deposition (Finkelstein and  Seilkop,  1986). Thus, for  completeness, two
Canadian  sites were added  to the data base (Figure 3, Table  2).  The sites are  from the Acid
Precipitation 1n Ontario Study  (APIOS) network and are also daily/event samples.  These data
were  subjected to the same procedures utilized for the UAPSP data,  allowing  for minor differ-
ences in content, and were placed in the same data format as the UAPSP measurements.
                                           23

-------
2.3.      TESTING

      Testing of the aggregation and selection process was accomplished by a variety of proce-
dures which investigate the accuracy of successive stages of the analysis.  The following steps
were included:

      (1 )  The procedure for selecting sample events and generating aggregated estimates for
           annual deposition was tested by comparing estimates for annual deposition derived
           from meteorologically stratified samples both against observed deposition patterns
           and against estimated deposition from samples selected at random. The discrepancy
           between observations and aggregated estimates will also be compared to the natural
           year-to-year variability in  observed wet  deposition rates.

      ( 2 )  The statistical significance  of the observed relations between deposition and
            meteorological categories will be tested using the techniques of statistical
            regression.  The regression model also allows testing relationships between
            deposition and other meteorological variables (i.e.  precipitation) that are used  in
            the aggregation process.

      (3 )   A  simplified model  identifying source-receptor relationships for observed wet
            deposition was developed and applied based on (a) the final ensemble of events
            selected for future  in-depth analysis, and  (b) all events over the 1982-19S5
            period. A comparison between the two estimates illustrates both the accuracy and
            the potential limitations involved in predicting  source-receptor  relationships  from
            a finite sample.

 2.3.1.     Aggregation  Versus Random  Selection

      Techniques for  deriving aggregated  estimates for annual  deposition based on a  finite
 sample of events have been described  above (Section 2.1.3) . These  aggregation techniques may
 be applied to  ensembles generated  entirely at random or to  ensembles generated based  en
 meteorological stratification.  In addition, the  aggregation  procedure itself can  include
 knowledge gained from meteorological strata, or the procedure can be constrained to specifically
 exclude such knowledge.  Referring to Equation (7),  the prior estimate for deposition per event
 (  6-, ) can be derived  either  with or without knowledge  of the meteorological stratum associated
 with each event.  The prior estimate can also make use of other information (e.g. time of  year,
 precipitation)  during each event.

      A preliminary estimate of the  usefulness of meteorological stratification  in selection cf
 events  can be obtained by comparing aggregated estimates for random ensembles derived with
                                             24

-------
and without knowledge of meteorological strata.  The accuracy of the aggregated  estimates is
estimated by the  root-mean-square difference over 13 eastern sites (RMSD, Equation 9).  The
advantage gained by the use of strata, both  in  the selection of individual events and in the
aggregation estimate, will  be demonstrated by an improvement in  the  aggregated estimates
derived with knowledge of meteorological strata.

      In an  associated test, the RMSD for aggregate estimates is compared to the  natural  year-
to-year variability in observed wet deposition.  Year-to-year variability provides a measure of
inherent uncertainty in estimates for annual deposition.  If the RMSD for the aggregate estimate
is  no larger than the RMSD between  annual and multi-year average wet  deposition, then the
aggregated  estimate may be judged equivalent  to an estimate derived from data for a complete
year.

2.3.2.     Statistical  Regression

      The significance of differences in the observed pattern of wet deposition for events from
different  meteorological  strata have been tested by means of multivariate linear and logarithmic
statistical regression models.  The specific advantage of regression models (e.g.  Johnson and
Wichern,  1982)  is that it  permits a  persistent correlation between variables to be distin-
guished from a correlation  that may be due to random chance.  Regression analysis is a neces-
sary step in demonstrating that the relation between deposition and meteorological strata which
form the basis of the aggregation procedure and selection process is based on long-term trends
in deposition rather than on  arbitrary correlations.

      Because wet deposition must  depend on precipitation, a common starting point  is to
assume  a relationship  between deposition and precipitation  amounts.   Wet deposition and
precipitation have  previously  been  correlated through  both  linear  (Wilson et  al.,  1SS2;
Buishand et al.,  1988;  Serge, 1988;  Hooper and Peter, 1988) and  log-linear (Stein and  Niu ,
1988)  regression models.   A linear regression model for wet deposition with precipitation that
also includes variation by meteorological category has the following form:

                                       V  Bcttd)*,                                  <19>

where £>j is the estimated deposition for event /, fl^) represents the least  square  estimate (i.e.
average  deposition-to-precipitation ratio)  over ail events within the  meteorological category
containing event /, and  P-t is the precipitation amount for the  event.  The regression parameters
have been determined independently at each site for which observed wet deposition and precip-
itation are available.

      The importance of meteorological  category  in explaining  deposition can be verified by
demonstrating  that the   deposition-to-precipitation  ratio  (Rcat(i)) varies  significantly for
events from different meteorological  categories.   Variation  in 6^) can be  evaluated with the
                                            25

-------
following test statistic (heterogeneity  of slopes, in Freund et al., 1986):

                              1(0, - B0Pj)2-(Dj-  f3cat(i)Pi)2]/(N-1)
                    ' s             n                                                 v *•u i
                                   £(0i- Gcat(i)^i)2/("'-1)

where Dj  and P-t  represent  deposition and precipitation during the ith three-day event at an
individual site, B0 represents the overall deposition to precipitation ratio for the site, N is the
number of categories and n'  is the number of events with non-zero deposition and precipitation.
Note  that events with zero precipitation  (and therefore zero wet deposition) are excluded from
the analysis to avoid introducing bias.  The test statistic F, has an F-distribution  with (A/,/?'-i)
degrees  of freedom.  When this statistic assumes a value greater than a pre-determined critical
level  (1.66 for most cases examined here), the null hypothesis that that observed variations  in
wet deposition by meteorological category  (i.e. variations  in  Bcat(i)) are due to chance may be
rejected  with  95% confidence.

      Further insight into the the variation  of deposition among meteorological categories  may
be gained by comparing slopes  (3cat(j))  for  individual  meteorological categories with the
average  slope (80) for all wet deposition events at a site. The difference between  B^,, arc  3,
for an individual category are tested by the following statistic
                   Ts = n-                              n-
                                           i 2 // _ « n \  \* in    0     \ 2
                           :0i" Scatd) P,)
-------
where  £)average  is average  deposition per event at the site.  The coefficient of determination
represents the fraction of observed variation in  deposition  that is explained by the regression
estimate (6j)

     The techniques that have been demonstrated here in a linear regression model may also be
used in a logarithmic model  with the form

                                    log 5j - Y + (J log Pj                               (23)

or in a linear model that relates wet  deposition  to the precipitation  quantity raised to a power
other than one:

                                      &!-  Pcat(i)  PjY                                 (24)

The advantages  and disadvantages of these models for use in estimating wet deposition will be
explored below.

2.3.3.    Source-Receptor  Relationships

     Variation  between  categories is important  because if they tend to favor different source-
receptor relationships a sample of episodes from  a number of categories should contain a better
overall  mix of wind flow  patterns.  This would insure  that the resulting  aggregate  estimates of
wet deposition are not biased towards  a  small number of potential source regions.

     The wet differential probabilities, AOwel, were generated as:

                                      AOwet - Qj - "Qj                                (25)

to demonstrate  that  some  of the categories exhibited  bias towards  transport  from various
regions in eastern North America.  The  wet differential probability is the difference between the
probability,  of},  for contribution from  grid  cell /,/ due tojhe transport associated with category
c on days with precipitation, and the  mean probability, O^,  for contribution from  from  cell /,/
for all  days with precipitation.

      The source-receptor  relationships estimated  using a small number  of episodes  (19-30)
were evaluated by comparing Q-fields.  RMS errors (RMSE) between the actual Q-fieid and its
estimates were calculated according to:
                                           27

-------
where Oj •  is the 19 or 30 day aggregate estimate of transition probability  and Oy is  the
climatological transition probability value in grid cell / for site ;'.  S  is the number  of sites
considered  and  N corresponds to the  number of grid cells  over which the differences  are
calculated.  To convert to percents the RMS error is divided by the mean transition probability
over NX S grid points.
                                            28

-------
                                   3.     RESULTS
3.1.       METEOROLOGICAL CATEGORIES

3.1.1.    Choice  of  Stratification  Parameters

      Attempts at the stratification of the three-day periods into meteorological categories were
made using the  four  different combinations of meteorological data listed in Table 3.  These
factors were thought to be most likely related to precipitation and/or S04° deposition based on
dynamical considerations. These variables were extracted or  derived from the NMC  data set.
These included  those related to wind patterns:  850 mb wind velocity;  850  mb  geopotential
height; sea level pressure; and those related to vertical motion and/or precipitation  potential:
1000  mb  relative humidity;  850 mb temperature advection,  850  mb  vorticity advection and
500 mb vorticity  advection.  The four combinations were generated  at each grid point for 1983.


         TABLE 3.  METEOROLOGICAL FACTORS USED IN INITIAL CLUSTER ANALYSIS

         1. Sea level pressure  and 850 mb geopotential height (TR),
         2. 850  mb u and v components (UV),
         3. 850  mb temperature and  500 mb vorticity  advection (AD),  and
   	4. 1000 mb relative humidity and 850  mb vonicity advection (RV)	

      The AD and RV factors were not as well suited to the goals of this study because, while they
consider vertical motions and/or moisture,  they do not  include  representation  of  transport
patterns directly. The categories resulting from the  clustering  of RV and AD factors were also
quite 'unbalanced in size with  a  couple large categories  and many small  "outlier" categories.

      The TR and  UV factors are similar in  physical meaning in that both are related to wind
flow.  The UV factors are based solely on three-day progressions  of 850 mb wind velocity (i.e.
one level at  approximately 1500 meters  above MSL).  The TR factors also  account for wind
velocities and, in addition, consider the strength and development of cyclones.  Consequently
statistical analysis of the  UV and TR categories yielded similar  results.  It was found that the UV
components  were more sensitive to changes  in the number of categories than the TR  factors.
This  provides a  better means of determining the optimum number  of categories.  Therefore the
                                           29

-------
UV components were selected as the parameters of choice for this project.

      Due to the computer memory constraints of the statistical software used in this project it
was not  possible to place all the cases  during 1979-1985 into a single  cluster analysis.  Only
one year's worth of running cases (365 cases) at 48 grid points could be  clustered at one time.
Therefore, in order to minimize  the effect of missing  u-v data  and to capture a broader cross-
section of cases, consecutive three-day periods (record 1 - days  1,2,3;  record 2 - days 4,5,6;
etc.) for  1979, 1981 and 1983 were  pooled for cluster analysis.

      Selection of  the number  of  meteorological categories was  based on  the ability of  the
categorization to separate events with observable differences in SO4" wet deposition. The value
of increases  in step number  (i.e. increasing number of categories) was evaluated by calculating
the Beale (P> and  Calinsky  (C) criteria for within-group variance  of S04*  deposition  (defined
in Section 2.1.2).  High values of Fand C indicate that significant improvements in the  ability to
stratify  SO4* deposition occur  as a  result of  increased step  number.   Results are shown in
Figure 4.  As expected,  the  highest values of F and C occur when the step number is  low, with
diminishing returns  as the number of categories is  increased.   Figure  5  shows total within-
group variance of SO4" deposition (defined in Sections 2.1.1 and  2.1.2) as a function of number
of categories.   Within-group variance decreases as the step  number is increased, reflecting
improved ability to  stratify SO4*. deposition with  increasing  number of categories.  The pattern
of diminishing  returns with increasing step number is.also apparent in Figure 5.

      Choice of the proper number of categories  necessitates a trade-off between the desire for
precision in  distinguishing  between  different  meteorological  patterns  (facilitated by  a  large
number of categories) and the need for robustness in statistical estimates and  ease of computa-
tion (facilitated by  a small number of categories).  Appropriate  number of categories can be
identified in Figures 4  and 5 by identifying step numbers that show relative peaks  in F and C
and significant reductions in cluster variance.   Critical points  (identified by changes  in slope;
appear to occur at steps 8, 14 and 19. Based on this guidance we initially proposed a choice cf
19 categories.

3.1.2.     Description of Categories

      As described above (Section 3.1.1) it was decided to base the meteorological stratification
on the 850 mb wind fields (U and V components of wind vectors as the  two independent mete-
orological parameters) and to base the stratification on 19 independent categories.  This section
presents a description  of the categories in terms of meteorology  and  in  terms  of deposition
(Section  3.1.2.1).   We  also describe the choice to divide each  category into subcategories
representing  "wet"  and "dry"  events, i.e.  events with greater than average or  less than  average
total sulfate  wet deposition  at the 13 eastern sites  (Section 3.1.2.2).   The  38 subcategories,
representing "wet"  and  "dry" halves of the original 19 categories determined by wind field,
were used as the basis for event selection and aggregation.

                                            30

-------
                                 F-Statistic  -• C-Statistic
                         10     15     20     25     30     35

                                    NUMBER OF STEPS
                                                     40     45
50
Figure  4.
Beale and Caiinsky criteria (Fand C  statistics) versus the number of-categories
(step number), based on 850 mb wind field clusters and observed SO4~
deposition for 1983.  High values  for Fand C  indicate significant reductions of
within-group variance of SO4= wet deposition with increasing  step number.
      The mean 850 mb wind flow patterns on each of the three days were plotted to ensure that
the 19  categories  made  sense meteorologically  (i.e. all major  storm  patterns were  repre-
sented).  The plots depicting the flow patterns for the 19 categories  are included as an  Appendix.
Each  category demonstrates some unique mixes of cyclonic and anticyclonic circulations.  Table
4 is a summary of the major features associated with each of the 19 categories.
3.1.2.1 .
Wind  Flow  Patterns—
      The 19 meteorological categories represent clusters of 3-day events  and can be charac-
terized by the  evolution  of the meteorology from  day #1  to  day #3,  averaged for  all  events
contained within the category.  The mean 850 mb wind flow patterns on each of the three days
have been plotted  and examined to ensure that the 19 categories made sense  meteorologically
(i.e. all major storm patterns  were represented).  The plots depicting the flow  patterns  for the
19 categories.are included as an Appendix.  Each category demonstrates some  unique mixes of

-------
cyclonic and anticyclonic circulations.  Table 4 is a summary of the major features associated
with each of the 19 categories.
    *

    i
    I

    T
I
r
I
a
UU 7»
once, .
QA*.
•«*«




%..


                             10    15    20    25    30    35
                                   Number of  Categories
                                                          40
45
50
Figure  5    Cluster variance (within-group sum  of squares) for SO4" wet deposition versus
             number of meteorological categories included in the stratification,  based on 850
             mb wind  field clusters and observed S04= deposition for 1983.
                                          32

-------
    TABLE 4. SYNOPTIC DESCRIPTION OF THE NINETEEN 850 MB WIND FLOW CATEGORIES.
      N CORRESPONDS TO THE NUMBER OF OCCURRENCES; SYR-1979,1981 AND 1983
 (CONSECUTIVE 3-DAY PERIODS); 7YR - 1979-1985 (RUNNING 3-DAY PERIODS); AND  4YR -
              /Jw.,                1982-1985  (RUNNING)

 Category    N      N      N    Description
	(3yr)  (7yr)  (4yr)	

     1        13      76      41    Trof from James Bay to KS with a mdt SW flow to E.  Hi well off
                                 FL coast.  Trof rotates and moves slowly Ewd to Wrn Que thru TN
                                 with W to NW fLow behind and W-SW ahead. Hi moves Ewd.
     2        12      56      38    Sharp rdg from Lk Sup SEwd.  Mdt Sly winds between the rdg a.-.d
                                 a wk trof in the Dakotas. Trof strengthens and moves across Ml
                                 ending up N to S thru OH.  Good cyclonic circulation around the
                                 trof.
     3        6      49      28    Well developed Lo moves from lA to  Lk Ont.  Strong, closed
                                 cyclonic circulation around the Lo.
     4        8      42      21    Stationary front W to E thru  Lk Erie.  Sly winds S of front and
                                 wavy Wly flow to the N.
     5        25      135    72    WkLo over Nrn MN with front curling S and W into TX.  Lo
                                 strengthens and moves E to  Ont-Que boarder with front thru  E TN.
                                 W-SW winds ahead of system and mdt NWly winds behind.
     6        29      249    157   Stg Hi over NC drifts very slowly E as a wk trof moves SEwd  into
                                 Lk Sup.  Mdt SW flow over the MS valley spreads Ewd to Atlantic
                                 coast during the period.
     7        32      201     112   Continental Hi over SErn US weakens as  Nly winds spread Swc
                                 over the Grt Lks and New England.  A rdg re-develops from the Hi
                                 across Lk Mi  as Nly winds move E.
     8        18      159    92    A cold front along  the NE Atlantic coast  moves slowly offshore.
                                 An extensive  area of Hi pressure moves over the Ern US and a mat
                                 Sly flow develops W of the  MS river.
     9        33      263    139   Bermuda  Hi remains nearly  stationary over TN/GA.  A  very wk
                                 trof moves quickly across the Grt Lks.  A Igt  anticyclonic circula-
                                 tion covers the Ern US throughout the period.
     10       8      108    76    Trof thru Lk Sup and  SErn KS moves E and rapidly weakens.  Met
                                 to  stg Wly winds persist W of the MS river and S of the Grt  Lks.

                                     ABBREVIATIONS KEY
  trof -  trough of low  pressure
  rdg - ridge of high pressure
  Lo - low pressure system
  Hi - high  pressure system
  stg - strong
  mdt - moderate
  wk - weak
  thru  - through
  Lk - lake
  Grt Lks -  Great Lakes
N, S, E. W - north,  south, east, west
NE, SW, SE  - northeast, southwest, southeast etc.
Nly - northerly (same format  with other  directions)
Nrn - northern (same format with  other directions)
Nwd - northward (same format with other directions)

All standard State abbreviations are applicable
                                              33

-------
   TABLE 4 (CONTINUED) (N CORRESPONDS TO THE NUMBER OF OCCURRENCES; 3YR -
        1979, 1981  AND 1983  (CONSECUTIVE 3-DAY  PERIODS); 7YR - 1979-1985
            (RUNNING 3-DAY  PERIODS);  AND  4YR - 1982-1985 (RUNNING) )
  	I	
Category
  N
(3yr)
       N
     (7yr)
  N
(4yr)
Description
   11


   12

   13


   14


   15


   1 6


   17 -


   18

   19
 27     223    124
15

 8
13


13

10
        90

        50


        27


        34


        23


        68


        96

        78
 48
  16
              19
              16
              1 4
  37
 30
 48
A cold front N-S thru OH weakens and moves just off E coast.  Stg
NW winds behind front also weaken and shift to the W as a Hi
develops over TX.
Well developed Lo NE of Georgian Bay with a trof extending Swd
moves just E of ME.  NW winds cover entire Ern  US  behind system
Nw flow to the W of the Grt Lks moves across Ern US and
strengthens over the Atlantic.  A Hi develops over TX  and builds
across  OH valley.
A sharp rdg from Lk Sup thru GA with stg NW winds  to E and stg S
winds to the W moves offshore. A Lo  with a stg cyclonic circula-
tion moves over NE IA from CO.
Stg Lo over New Brunswick drifts E.  A weak Hi over the Gulf is
replaced by a weak Lo over GA.  Entire Ern US ends  up  in a wea*
cyclonic flow around the Lo.
A well developed Lo over Georgian Bay moves quickly NE and
another stg  Lo moves  into MO.  Stg S flow develops  aheac of
second Lo and W of a wk rdg separating the two  Lows.
Strengthening trof over MN moves to Wrn Ml and extends Swd to
LA. SW flow increases E of trof as does the NW flow W of re
trof.
A Hi over the Gulf which extends to IA weakens allowing ccns.s-
tent NW winds to cover most of the Ern US.
Well developed coastal Lo near DE moves NE as a rdg aicng MS
river moves over Ern states.  Mdt Siy flew W cf rcg anc aneac c'
wk trof NE-SW thru MN.
                                x,
                                    ABBREVIATIONS KEY
 trof -  trougn of low  pressure
 rdg -  ridge  of high pressure
 Lo - low pressure system
 Hi • high pressure system
 stg -  strong
 mdt - moderate
 wk - weak
 thru - through
 Lk - lake
 Grt Lks  - Great Lakes
                    N, S. E, W - north, south, east, west
                    NE,  SW, SE - northeast, southwest, southeast etc.
                    Nly  - northerly (same format with other  directions)
                    Nrn  - northern (same format with  other directions)
                    Nwd • northward  (same format with other directions)

                    All standard State abbreviations are applicable
                                             34

-------

                                          79-81
82-85
      0.12
      0.00
              1   2   3   4   5   6   7   8   910111213141516171819

                                     Meteorological  Category
Figure  6.    Comparison of frequency of occurrence of the 19 meteorological categories
             defined in  this study between 1973-1981  and  1982-1S85.
      During the seven year period there were marked differences in the overall frequency and
seasonal frequency of each  category.  Some types of events were rare while others occurred
quite often. The number of occurrences of each category over seven years are listed in Table 4.
Categories 6, 7, 8, 9 and 11 were the most frequent and categories 4,  13, 14, 15, and 16 were
relatively  infrequent during 1979-1985.   The  change in category  frequency  between  the
1979-81  period and the  1982-85 is shown in Figure 6.  The  overall importance  of each
category did not change significantly.
                                          35

-------
      Initially it was thought that it may be necessary  to impose seasonal requirements on
category selection due to the seasonal variations known to exist for SO4* deposition.  This would
ensure that storms  at all times of the year were represented.  Figures 7a-g, however, demon-
strate that seasonality is already inherent in the categories.  For example, categories 1, 2 and 3
were basically cold season events.  Figure 7b shows that category 4 occurred in all seasons with
a tendency towards Fall.  Categories 5 and 6 also occurred year  round, but they are more
dominant in the Spring and Fall.  Categories 7, 8, and 9 (Figure  7c) were predominantly warm
season  situations while 10 and 12 (Figure 7d) were cold season type  events.  Category  11
occurred fairly often in all months but  were most frequent between December and June.  Cases
in categories 13-16 were all most frequent during  the cold season as were the cases in cate-
gories 17, 18 and 19.  However, cases in 17 and 18 seemed to" occur more in winter and cate-
gory 19 cases in  spring. Missing  events, which resulted whenever the 850 mb data for any of
the three days were missing were recorded in  every month,  however, there were more in
August and less in  May and June.  On average there were approximately 5 events missing per
month  during the four-year period.

3.1.2.2.     Deposition  Patterns—
      While the underlying  reason for using  the wind flow to stratify deposition episodes was to
insure  representation of all  major air flow patterns, it was hoped that the metecrclogical
stratification .would also produce distinct S04" deposition patterns associated with each category.
It was found that this was partially true.  The stratification separated relatively "wet" periods
from relatively dry periods and thus separated those periods of enhanced pollutant deposition as
well.

      The  ability  to explain  the  variability  in  wet deposition by meteorological stratification
was tested in a number of ways.   First, the spatial patterns of the 85th  percentiles of S04=  wet
deposition associated with each category, were studied.  All available data (up to 7 years) from
each separate location that  satisfied the quality assurance measures discussed earlier were used
to calculate the percentiles. The spatial patterns changed between categories but the differences
were  not always  great.  For certain  categories, the magnitudes and locations of the relative
maxima  and minima were  significantly  different compared to the other categories  and some
categories appeared to be rather unique at certain sites and not at others.

      The 85th percentiles of SO4" deposition for each meteorological category were compared at
four locations in Figures  8 a-d. Deposition appears  to vary between categories and at each of the
sites explored there were different categories or groups of categories that resulted in  high 85th
percentile depositions.  This suggests that the 850 mb flow patterns that were  important  for
S04* wet deposition at  some sites were  not necessarily important for other sites.  Nonetheless,
there appeared to be some consistency in rank of categories between sites.  Category 5 consis-
tently produced relatively large amounts of SO4* wet deposition at all four sites while categories
12 and 13 produced small amounts.
                                           36

-------
      The utility of stratification in estimating long term averages relied on the assumption that
the variance within the strata were less than the overall variance.  Figure 5 above  showed that
there was less variance within  each meteorological strata as compared to  the variance over all
the data.  The average variance within  the 19  categories was approximately 85% of the total
variance.  Thus, the 850 mb wind categories did reduce the variance of the aggregate  estimates.
Statistical significance of  the relation between wet deposition  and meteorological category will
be  discussed in detail in Section 3.1.3 below.  Although there  were statistically  significant
differences in wet deposition among  categories, it is obvious that  factors  other than wind flow
(i.e. precipitation amount and location) explains  most of the variance in wet deposition.

      One factor  that influenced  the relative importance of  a'category was  its frequency of
occurrence.  As a first  guess,  one would expect that the most commonly occurring  categories
would be relatively important.  However, this  was  not  always be the case.   High pressure
systems occurred fairly frequently but they generally produced significantly less precipitation.
Therefore, in order to better identify which categories were  the most important, the total
amount of SO4" deposition associated  with each category during 1982-1985  was  compared.
Percent of total deposition due to each  category  were calculated for every site. There  were a
group of categories that consistently contributed a large percentage of the deposition at each one.
Storms associated with  categories 11, 9, 6, 5 and 7 appear  to be the most important  for  the wet
deposition of SO4=.  Some categories were important because they were  frequently  associated
with precipitation.  Others  were important becaus'e the average deposition per  event  was large.
In Figure 9 the  ratio of the percent of SO4° deposition to the percent of precipitation has been
graphed for each category.  In general, one  can assume that changes in precipitation should
result in proportional changes  in deposition.  Therefore,  ratios above one in Figure  9 indicate
relatively "dirty" categories.  While categories 5, 6,  7, and 9 had  ratios  greater than  one (as
did category 8, which was also fairly important), category  11, was not among the "dirty",   its
importance was  apparently more  related to the fact  that it frequently resulted in precipitation
at most of the sites.  In fact, the category 11 storms were responsible for the largest percentage
of precipitation at many sites in eastern North America.
                                            37

-------
                        M
M   J    J    A

    MONTH
O    N
                                                                          D2

                                                                          •
Figure  7a   Monthly mean number of occurrences of meteorological categories  1, 2, and 3
             during  1982 to  1985  (Note scale  is 0-25).
     O   50
                        M
M    J    J    A   S

    MONTH
O   N   D
                                                                           D5

                                                                             |4
Figure 7b  Monthly mean number of occurrences of meteorological categories 4, 5, and 6
             during 1982  to 1985  (Note scale is 0-50).
                                          38

-------
Figure  7c   Monthly mean number of occurrences of meteorological categories 7, 8, and 9
             during 1982  to 1985  (Note scale is 0-75).
Figure 7d    Monthly mean number of occurrences of meteorological categories 10, 11, and
              12 during  1982 to  1985 (Note scale is  0-50).
                                          39

-------
                       M
                           M   J    J   A

                               MONTH
                                                   N
                                                                      • 15

                                                                      n 14
Figure  7e   Monthly mean number of occurrences of meteorological categories 13, 14. and
             15 during 1982 to 1985 (Note scale is  0-25).
c
c
c
u
F
R
E
N
c
E
S
        25
         20 -
         15 -
10 -•
          5 .
                                                                      D17

                                                                      •  16
                       M   A    M    J    J    AS

                                    MONTH
 Figure 7f   Monthly mean number of occurrences of meteorological categories 16, 17, 18,
             and 19 during  1982  to 1985 (Note scale is 0-25).
                                         40

-------
          1 00
                12345
 6  7  8  91011121314151617181920
   Meteorological Category
Figure  8a.   Variation in amount of SO4" deposition between categories at Big Moose, New
             York.
               12345
6  7  8   9 1011121314151617181920
   Meteorological Category
Figure  8b.  Variation in amount of SO4* deposition between categories at Dorset, Ontario

-------
              12345
                  6  7   8   9 1011121314151617181920
                     Meteorological Category
Figure  8c
Variation in amount of S04" deposition between categories at Raleigh, North
Carolina.
           120
    n
                 1   2  3  4  5  6   7   8   9 1011121314151617181920
                                  Meteorological Category

Figure  8d.   Variation in amount of S04" deposition between categories at Zanesville, Ohio.
                                         42

-------
       2.00
        0.00
               12345
                    6  7-  8   9-10111213141516171819
                      Meteorological Category
Figure  9.
The ratio of the percent of SO4= deposition to the percent of precipitation for each
meteorological category.  Values greater than one imply more  deposition than
expected due to variations in precipitation amount.
      To gain a better understanding of the actual weather associated with each category the sea
level analysis from the  NCAA Weekly Summaries were examined.  It was hoped that definite
similarities would  appear between the  cases within categories and that  we would  then be able to
select a set of "most representative" cases for each category. 'These cases would  then be used in
the aggregation.   Unfortunately, there  was still too  much variability within  categories and it
was next to impossible  to identify typical cases. This was because there were  many  meteoro-
logical and deposition related parameters still varying within each category.

      For the purpose of event selection, it is desirable to  establish a  pool of  events with above
average  SO4= wet deposition in eastern North America. The ultimate  goal of the  categorization
process  is  to represent  meteorological events that contribute  significantly to  deposition in
eastern North America,  using a  minimal number of sample events. This aim would be thwarted
by  a selection process  that includes large numbers of events with minimal total wet deposition.
                                            43

-------
In order to improve the selection process the original  19 wind flow categories were divided  into
"wet" and "dry" subcategories.  The split into "wet" and "dry" halves was based on actual SO4"
wet  deposition received at the 13 northeastern sites during the years  1982-1985, with  the
"wet" subcategories consisting of events with total SQ^~ wet deposition above the  median.  This
resulted in 38 new categories consisting of the "wet" and  "dry"  halves of the original  19 wind
flow categories.

      An  examination  of 850 mb wind flow for the 38 subcategories  revealed little difference
between  equivalent "wet" and "dry" halves.  The major difference between wet and dry subcat-
egories concerns total precipitation, which is related  to meteorological parameters (e.g. abso-
lute  and  relative humidity,  atmospheric stability  indices)  other--  than  850 mb wind flow.   The
split into wet and dry subcategories based on SO4* wet deposition corresponded closely to a  split
based on total precipitation.  We have used the 38 wet and dry subcategories as the basis for
event  selection and aggregation, while  analyses for statistical significance {Section 3.1.3
below) were  based on the 19  wind flow categories with  precipitation  as an independent
parameter.

      The percent of total deposition due to each of the 38 subcategories were calculated as for
the original 19 categories.  Table  5  is a matrix of percent of total  deposition by site and category
('CT'). The same information  is portrayed  graphically in  Figure 10.   The  "wet"  sub-categories
are  numbered 21-40 with  21'corresponding to 1, 22 to 2, 23  to 3 etc.  The categories  that
consistently accounted for a relatively large percentage of  the deposition at all of  the sites were
25,  26. 27, 29, and 31.   These  were all from  the "wet" subset  and corresponded tc  tne
important categories selected  earlier from  the undivided set  of 19.   These  categories alcr.e
accounted for  30 to 60% of the 1982-1985 SO4* and  N03" wet deposition at the sites for which
data was available.

      A crucial step in  the development of the aggregation scheme  was the selection of an array
of categories  from which to select  events.  It was decided that  the categories selected sncuid
account for close to 75% of the total SO4" wet deposition.  A significant number of events would
be selected  from  the major  categories  (25,26,27,29  and  31).   However selection  from
additional categories were necessary to bring the SO4  wet deposition for included categories up
to 75% of total  deposition at most sites.  The  relative importance of the remaining categories
varied significantly from  one  site to the  next,  as some  wind flow  patterns  contributed
significant wet deposition at some sites but little deposition at others.
                                            44

-------
TABLE 5: MATRIX OF PERCENT OF TOTAL DEPOSITION BY SITE AND CATEGORY

CT
1
2
3
4
5
6
7
8
9
1 0
1 1
1 2
13
1 4
1 5
1 6
1 7
1 8
1 9
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Tot
SITE NUMBER
1
1.3
0.6
0.5
0.3
0.4
1.2
0.9
2.1
2.5
1
2.4
1.8
0.1
0.3
0.1
0.3
0.5
0.2
1.5
4
1 .1
4
0.8
8.9
8.4
8.9
6.3
1 4
2.1
1 4
3.7
0
1.3
0.9
1.3
1.2
0
1.2
100
2
0.3
0.2
0.1
0.2
0.7
1.1
0.9
1.6
2.4
0.2
1
0.7
0
0.2
0.2
0
0.1
0.1
0.3
4.1
0.7
3.3
1.3
9.4
7.3
20
7.8
1 4
2.4
1 1
3.7
0
1.2
0.4
1.1
1.1
0.4
0
100
4
0.4
0.8
0.2
0.2
0.5
1.4
1 .4
1
1 .4
0.8
0.8
0.6
0.3
0.1
0.4
0.2
0.4
0.2
0.5
4.2
1 .1
4.3
2.6
1 1
1 5
1 3
6.5
1 3
3.3
8.6
1.3
0.2
2.4
0.1
0.4
1.3
0.4
0.7
1 00
5
0.33
1.71
0.01
0.87
0.61
2.86
1.78
0.7
2.1
1.69
0.67
0.22
0.34
0.39
0.65
0.62
0.8
0.28
.1.58
5.26
2.14
3.33
1.28
10.7
12.5
11.5
3.02
9.78
4.9
6.25
0.81
0.27
5.37
0.75
1.23
1.52
0.07
1.18
100
7
0.45
0.66
0.06
0.07
0.57
2.8
1.56
1.06
0.78
1.34
0.61
0.52
0.21
0.2
0.27
0.12
0.68
0.45
0.38
6.79
0.77
3.28
1.51
8.61
15.3
15.4
3.03
1 2
3.63
7.65
1.01
0
3.23
0.57
1.24
2.22
0.58
0.48
100
8
1.34
2.83
0.31
0.92
1.08
0.75
1.2
6.26
3.69
2.44
3.59
0.93
0.77
0.58
0.99
0.15
0.81
0.29
2.31
2!16
1.9
5.94
1.93
8.15
1.39
9.75
4.83
9.42
0.59
12.9
2.66
0
1.91
0.41
0.11
2.14
1.1
1.42
100
1 0
0.15
1
0.24
0.2
1.9
6.33
1.52
0.61
2.99
0.75
1.6
0.28
0.11
0.15
0.06
0.05
0.58
0.39
1.29
3.5
1.67
4.01
0.89
1 4
17.6
8.87
1.94
11.7
3.02
7.21
0.65
0
2.06
0
0.66
1.4
0
0.59
100
1 1
.88
0.9
0.25
0.37
0.85
1.3
' 1.81
1.93
1.07
1.89
2.04
1.31
0.34
0.17
0.91
0.1
0.92
0.42
0.53
2.4
0.32
3.75
1.05
9.37
6.06
1 1 .8
6.91
16.8
3.19
11 .9
2.06
0.23
1.46
0.5
0.9
1.74
0.84
0.71
1 00
1 2
1.86
1.72
0.35
0.97
1.11
8.1
0.15
1 .23
7.66
3.73
1.59
0
0.75
0.51
0.6
2.28
0.87
. 0.41
0.8.4
6,31
1.75
4.66
0.99
8.73
8.15
2.8
0.97
1 2
2.36
7.87
1.19
0
3.12
0.67
1.66
1.79
0
0.25
100
1 3
1.17
0.35
0.02
0.17
0.38
1.56
2
4.09
1.38
1.78
3.04
3.39
0.09
0
0
0.64
0.64
0.46
'3.12
4.7
0.53
2.93
0.6
10.9
10.6
6.41
3.26
8.22
1.84
1 6
3.79
0
0.51
1.03
1 .98
1 .49
0.15
0.8
1 00
                              45

-------
TABLE 5 (CONTINUED)

CT
1
2
3
4
5
6
7
8
9
10
1 1
1 2
13
1 4
1 5
1 6
1 7
1 8
1 9
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Tot

1 4
1.76
2.99
1.24
0.45
2.69
3.45
3.66
11 .4
8.68
1.32
3.89
1.32
2.86
0.73
3.18
0.29
1.9
0.19
4.15
3.3
1.12
4.17
0.7
1.89
1.66
11.9
2.55
4.41
0.58
7.44
0.83
0
1.43
0.26
0.27
0.73
0
0.56
100

1 5
2.47
3.94
0.18
0.91
1.79
3.93
2.07
4.93
8.5
3.03
5.84
0.71
1.61
0.95
1.58
0.81
1.63
0.69
1.1-3
3.62
1.75
4.75
2.23
6.66
3
4.27
2.12
7.03
1.42
9.35
0.48
0.17
1.86
0.8
1.49
2.1 1
0.03
0.15
100

1 6
4.64
2.73
0.98
2.08
1.48
3.66
7.05
3.67
5.68
3.82
4.31
0.96
1 .4
0.09
1.95
1.51
1.28
0.98
3.01
3.07
1.59
2.69
2.7
5.73
3.96
7.54
2.24
5.42
1.32
6.56
0.84
0
2.21
0.41
1.1
1.1 1
0.08
0.12
100

1 7
2.05
1.58
0.46
2.24
1.5
9.01
1.78
4.25
3.71
4.09
2.22
0.21
1.21
0.62
1.6
1.23
1.86
1.01
3.4
3.18
2.2
1.55
2.77
5.42
8.S2
8.75
2.36
6.17
2.87
6.1
0.54
0
2.54
0.72
1.16
0.42
0.16
0.18
100
SITE
1 8
0.4
1.16
0.01
1.05
0.76
7.8
2.91
6.18
5.06
1.34
0.33
0.07
0.74
1.26
0.69
1.38
0.61
0.25
5.65
1.93
0.48
1.66
1.67
6.65
8.82
13.4
3.36
12.9
1.56
2.42
0.46
0.08
2.41
0.64
0.78
0.7
0.92
NUMBER
1 9
0.3
1.1
0
1.8
1
9.7
2.4
8.4
8.1
0.5
0.4
0.0
0.1
0.0
0.1
0.3
0.5
0
4.1
0.8
0
0.8
0.1
6.0
1 8
12
2.2
1 5
0.8
1.5
0.3
0
1.6
0
0.4
0.2
0.2
1.47 1.8
100
100

20
0.3
0.5
0.1
0.2
0.2
1.7
1.6
1.7
0.7
0.9
1.6
1.6
0.2
0.0
0.4
0.5
0.5
0.3
2.0
3.9
0.5
1.2
1.0
8.0
1 3
1 5
5.8
1 1
4.0
15
2.6
0.1
0.5
0.1
0.9
1.2
0.1
1.1
100

21
1.1
0.5
0.1
0.4
0.4
1.7
"1.1
1.1
1.7
2
1.6
1.1
0.1
0.0
0.2
0.4
0.5 .
0.3.
•0.7
- 5.8
0.5
1.4
3.2
1 1
1 1
1 5
4.6
13
3.4
1 2
1.3
0
0.7
0.1
1.3
0.8
0.3
0.3
100

27
0.3
0.9
0
0.4
1.3
3.7
1.2
1.4
2.1
1.4
1.4
0.8
0.3
0.3
0.2
0.1
0.3 .
0.9
0.1
5.8
1.3
2.5
1.5
1 3
1 7
5.8
1.9
13
4.8
1 3
0.5
0
0.4
0.1
0.4
1.7
0
0.3
100

28
0.5
1.1
0.1
0.8
1.3
2.9
1.1
0.6
1.5
1.4
1.0
0.2
0.5
0.1
0.2
0.4
0.5
0.7
•o'.i
5.1
1.3
2.0
2.9
1 3
1 0
1 2
2.3
1 7
3.9
8.7
1.2
0
2.2
0
0.3
3
0.1
0.2
100
         46

-------
 6
 7
 8
 9
10

12-
35-
36-
:'
 • r
                                          -
                                       "-".!	^v*-1
                                 •-
                                                .
                                               V" '; -;
   "^
      >
               "DRY" CATEGORIES "(1-19)
                                i
"WET" CATEGORIES  (21-39)
       2   4   5   1   8   10 11  12 13  14 15  16 17  18 19  20  21  2~!  2
                               SITE NUMBER
  Figure  10.  The percentage of total SO/ wet deposition associated with each meteoroicc
             category by site.
                                      47

-------
      A total of 9 to 15 categories (from the 38) were needed to explain 75% of the wet SO4"
deposition at each  site during 1982-1985.  Rather than try  to produce  an aggregate estimate
which would be best for the entire domain shown in Figure 1 (which would require up to 15
categories)  it was decided to focus  on a smaller northeastern region encompassing 13 sites.
(The eastern sites have been identified in Figure 1 and Table 2).  With the domain restricted to
the eastern  sites a total of 9 categories were needed to approach the 75% goal.  These  9 cate-
gories represent a minority of the 850 mb wind flow patterns, it is likely that a modest number
of sample events will  represent the  meteorological processes that affect SO4" wet  deposition
across eastern  North America.  The  identified categories also make sense in terms of the phys-
ical processes  (i.e. storm tracks) represented.

      Table 6 shows that the 9 categories with greatest SO4* deposition (21, 23, 25,  26, 27,
28, 29. 30  and 31) accounted for about 72% of the total SO4"  deposition within the  13  site
region.   The curve in  Figure 10 indicates  the total amount  of deposition accounted for as the
number of categories selected increased.  To explain more and more of the deposition one  needed
more categories at an increasing rate. The cost-to-benefit  ratio changed drastically with more
stringent requirements.   As  a first  approximation the nine  categories  mentioned above were
selected to be represented in the aggregation scheme.

      While the 9 major categories  account for  the bulk of  SO4*  wet deposition, concern  has
arisen that  some  important physical processes  may be excluded by the  9-category selection
criterion.  Many of the important categories  occurred primarily  in the warm  season.  Cold-
weather processes  (which may be  highly  nonlinear in terms of source-receptor relationships)
might be neglected by a selection process limited to the major categories. In addition, it  was
decided to  include a number of 3-day events on  which  very little precipitation fell in the  final
ensemble.   Inclusion  of "dry" events  enable  the final analysis to  include dry  deposition
processes.

      It  was decided that due to external constraints (RADM  modeling budget) no more  than 30
cases could be included in the final  ensemble.  It was also decided that from the nine important
categories up to 20 cases would be selected to reproduce the annual SO4* deposition pattern and
that the  other ten cases would be selected  to account for other, less prevalent conditions (winter
precipitation, unusual  meteorology and dry deposition). The final mix of episodes included (1)
an array of episodes chosen from categories that account for the  bulk of the annual SO4* wet
deposition,  and (2)  an array of episodes chosen to represent potentially neglected  processes:
cold season events and events with little rainfall that may contribute significant dry deposition.
                                            48

-------
     C
     U  F
     M R
     U  E
     L  Q
     A  U
     T  E
     I  N
     V  C
     E  Y
100%
 90%
 80%
 70%
 60%
 50%
 40%
 30%
 20%
  10%
   0%
                                                       £? ?4f '•'•'V;i
                                                .•
                                       '»i^?^**?~~~~ ~
                   0
                6    9   12   15  18   21  24   27  30
                      NUMBER OF CATEGORIES
                                                                         36
Figure   11.  Cumulative explanation of total SO4' deposition using categories ranked from
             greatest to least deposition.
      To determine how the  cases should be partitioned among the 9 important categories the
percent of total wet SO4' deposition figures shown in Table 6 were considered.  Relative t
percent contribution,  a  proportionate number of samples were  selected  from each  of the  cate-
gories.  The frequency of occurrence was also taken into consideration.  Constrained
numbers, the following  breakdown of category sampling was subjectively determined: four cases
were to be selected from category  29, three from 27, 26 and 31,  two from 25 and one fi
23,  28 and 30.   Initial tests of the  aggregation procedure were then based upon <
selected at random but according  to the predetermined  sampling criteria.
                                           49

-------
TABLE 6. PERCENT OF TOTAL SULFATE WET DEPOSITION BY METEOROLOGICAL CATEGORY
Category
1
2
3
4
5
6
7
8
9
10
1 1
12
13
14
15
16
17
18
1 9
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Total Dep.
1,927
2,459
474
1,042
2,112
5,638
3,683
4,189
4,570
3,857
4,474
2,854
796
517
970
789
1,582
1,060
2.459
11,953
3.347
8.683
4.381
27.497
28.978
32.625
11,725
33.151
9,108
29.049
5.134
254
5,252
1,046
2.467
4.212
917
2,072
% of Tot.
0.72
0.92
0.18
0.39
0.79
2.11
1.38
1.57
i.n
1.44
1.67
1.07
0.3
0.19
0.36
0.3
0.59
0.4
0.92
4.47
1.25
3.25
1.64
10.29
10.84
12.21
4.39
12.4
3.41
10.87
1 .92
0.09
1.96
0.39
0.92
1.58
0.34
0.78
                                  50

-------
      Eleven  additional cases were selected to  represent  winter  situations,  which were
hypothesized to have a higher proportion of non-linear effects, and dry deposition events.  The
basic criteria for sampling these days were as  follows:  five relatively wet winter cases and six
dry cases (the sum of the 13 site S04" deposition  was less than 50 mgm"2) from  previously
unsampled categories.  Given the number of categories from which these cases could be selected
it was impossible to sample proportionately,  therefore, other criteria were developed.  The
winter cases were selected from the "wet"  (21-39)  categories that displayed a relatively high
frequency in winter.  This reduced the number  of possible categories to 22,  24, 32, 34 and 37.
From this set  the five cases were selected randomly, one from each.  For the dry situation the
first  step in determining the set of categories  from  which to sample  was to consider only the
"dry" half (categories 1-19) .   The second  step was to isolate the  more frequently occurring
categories which also had a low mean deposition. Categories 6,  7, 8,  9, 11, and 19 met these
requirements.  From  this subset, six cases, each with a total 13-site wet SO4" deposition of less
than 50 mgm"2 were  randomly selected, one from each category.


3.1.3.     Statistical   correlations  between  category and  observed deposition

      As described above, wet deposition amounts received at various sites show considerable
variation for events among different categories.  In this  section results of  statistical tests of the
correlation between deposition, precipitation  and meteorological category wilt be  presented.

3.1.3.1.     Three alternative  statistical approaches—
      The  statistical analysis  will be  based on the  techniques of regression discussed  above
which use  correlations between wet deposition and (a) observed  precipitation, (b)  time of year,
and  (c)  meteorological category. Of the various factors which may be correlated with deposition
precipitation is apparently the most important,  and results  of regression analyses may depend
critically on the treatment of the  relationship  between deposition  and  precipitation.   Three
alternative correlations  between wet deposition and precipitation were  investigated.  These
were:

      •      Linear correlation between  wet  deposition and precipitation (Equation 19);

            Logarithmic correlation between  wet deposition and  precipitation (Equation 23);
            and

            A hybrid model correlating deposition with precipitation raised to an exponential
            power (Equation  24).

      Each of the three approaches has advantages. A  straight linear regression has the advan-
tage of simplicity; and  the regression parameters (  8cat(j)  in Equation 19) simply correspond
to the summed deposition-to-precipitation ratio observed during  events  associated with each
meteorological category. The  logarithmic model, by contrast, has the advantage of accounting for

                                            51

-------
 the observed tendency for SO4" and N03" concentrations to decrease during high-precipitation
 events (Stein and Niu, 1988).  This  tendency is especially important at coastal sites.  However
 the logarithmic model has  two  serious disadvantages.  First, evaluations of the accuracy of the
 logarithmic regression  are based on  the estimate  for  log  Dj  rather than for  Dj  itself.
 Consequently the logarithmic regression gives undue weight to events  with very  small amounts
 of deposition, which have negligible  impact on total wet deposition received at a site. On low-
 deposition days, the recorded discrepancy between observed deposition  and deposition estimated
 from  a linear regression model (Dj •  6j  ) will be small, but  the discrepancy recorded with  a
 logarithmic model ( log  Dj -  tog 6j ) can be very large.  For this reason we evaluated the
 accuracy of logarithmic models based on  the  linear coefficient  of  variation  (i.e.  ability to
 predict D\ , Equation 22) rather than based on its ability to predict log Dj.

      A second disadvantage of the logarithmic model is that  it requires twice as many regres-
 sion variables as does a simple linear regression.  Referring to Equation 23, if both  regression
 variables ( |J and 7 ) are allowed to  be category-dependent, then  the regression  will contain 40
 independent variables.  Accounting for other factors (i.e. time  of year)  increases the number  of
 independent variables still further.  It becomes  increasingly difficult to demonstrate  statistical
 significance when a model  contains such a large number of independent variables.

       The combined model represents an attempt to preserve the advantages of linear as opposed
. to  logarithmic  models while at the  same time  allowing for the tendency for SO4* and  NO3"
 concentrations to decrease during events with  heavier precipitation.  The combined'model  is
 based  on a linear regression  of  wet deposition with  P\ ^ as the  independent  variable.   The
 exponent Y is  derived from logarithmic regression (Equation 23).   The exponent is  allowed  tc
 vary  for different sites and for different species  (i.e. S04* and N03") but is not allowed to vary
 with  time of year or meteorological category.  Category-based  variations in  deposition are
 represented by allowing the linear  regression parameter (Pcat(i) 'n Equation 24) to vary with
 category.

 3.1.3.2.    Time-of-Year  Term  —
       Wet deposition rates for both S04" and NO3" both show strong seasonal dependence, with
 SO4* wet deposition rates typically much larger in summer than in winter and NO3" showing the
 opposite trend.  It has been suggested by Stein  (1989) that the  observed correlation between
 wet deposition and meteorological category might result  from  the seasonal nature of wet depo-
 sition rather than from  intrinsic differences in circulation pattern among the categories.   In
 order to avoid this error we have introduced time of year as an additional independent parameter
 in  the  regression  analysis.  Inclusion of the time-of-year  term  serves two purposes:   (a) it
 improves the overall performance  of the regression model by including season-based variations
 in wet deposition within the model; and (b)  it  insures that  correlations between observed wet
 deposition and meteorological category are not just a reflection of seasonal variations.

       Season-based variability is  represented  in  the combined regression model by adding a
                                            52

-------
time-of-year term Yj  with a value based on the day of the year:

                               Yj  = Sin [2ji(ND-81)/365]                          (26)

where  NQ is the Julian  day. ranging from 1  (January 1) to 365 (December 31).  The time-of-
year term  is incorporated in  the  linear and  log-linear regression  models by  replacing the
regression  equation  (24) with  the  following:

                                   fy  - P

where  8 is a regression parameter which is regarded as  independent of meteorological category.
The combined  regression  model (27) was  developed  in  stages, with P0 (without  category-
dependence) and 5 initially determined from a linear regression independent of meteorological
category.  After 5 was determined, a second  linear regression was performed to identify the
category specific parameters Pcat(i)-  Tne second linear  regression used the following equation

                                 &i -  PC.KI, d+*Y,)P,*                           (28)

with the category-specific 3^) parameters determined by regression of D{ against the  linear
variable (I+SY,)?^ (with 7 and 8  fixed).  This indirect method was  chosen  to  minimize the
number of  independent  regression parameters and to insure that the the parameters represent
physically  meaningful quantities.   In the completed model (Equation 28) there  were 22 inde-
pendent variables,  including 20 category-specific  3^ parameters plus y and  8.
 3.1 .3.3.    Results  —
      The linear, logarithmic and combined models were evaluated as predictors of SO4= and
 N03" wet deposition based on linear coefficient of determination (R2)  (Equation 22).  Note that
 R2  associated with the logarithmic model is the  linear  R2 and tests model ability to  predict
 deposition, rather than log D.  Results (Table  7) show that the  logarithmic model achieves
 significantly higher R2 values than the linear model for some sites,  especially coastal sites (e.g.
 Turners  Falls, MA) where the correlation between  deposition and  precipitation  was  weak.
 However the linear model  achieves higher  R2 than the logarithmic model at other locations (e.g.
 Longwoods, ON).

       The combined model appears to represent a significant improvement over both the linear
 and logarithmic models.  The R2 coefficients were higher for the combined model  than for the
 logarithmic model at all sites, despite the fact that the combined model had  fewer  independent
 parameters (21 as opposed  to 40 in the  logarithmic model).  The combined model also  had
 higher R2 than the linear model at  most sites. However there were some sites where the linear
 model achieved an R2 equal to or greater than the combined model.

       Incorporating the time-of-year parameter in  the combined model resulted in  a  significant
 improvement at  many sites, with  reductions in the  residual uncertainty (1-R2) of up to 25%.
                                            53

-------
At other sites incorporation of the time-of-year term had negligible effect.  The time-of-year
term had greatest impact at locations with severe winters (e.g. Underbill,  VT) and at coastal
sites, and little impact at southern or western sites.

      Incorporation of category-dependent P parameters (as opposed to a single po independent
of meteorological category)  caused an improvement  in R2 in all models.  Inclusion of category
dependence caused an improvement even when the seasonal dependence of wet deposition had
already been included in the regression.  Residual  uncertainty was  reduced by -10%  when
category-dependent  P's were included (e.g. R2 increases from 0.55  to 0.59 at  Tunkhannock,
PA).

      Statistical  significance of category-dependent  parameters Pcat(i) was evaluated by the
heterogeneity-of-slopes  statistic Fs (Equation 20).   As described earlier  the null  hypothesis
that  meteorological category had no impact on deposition (31  - {J2 " •••• P]  for  all categories)
was  rejected with 95% confidence whenever f| assumed a value of 1.66 or greater.

      Also shown in  Table 7A. the  F,  statistic demonstrated that category-based variations  in
SO4"  wet deposition  were  significant at the  95%  confidence interval or  higher at all sites
included in  this  study.  For NO3" (Table 7B)  wet deposition category-based variations were
significant at 18 of the 20 sites.  The remaining two  sites, Selma, AL  and  Brookings, SD, were
both at  the geographic edge of the  region represented by  the meteorological  categorization
scheme.

      Coefficients of determination (R2)  in the combined model were greater  than 0.5 at most
sites, indicating  that the regression was  able to account for 50% or more of the variation  in
observed wet deposition. The R2  coefficient  was significantly lower  at coastal  sites (Turners
Fails, MA and Winterport. ME) and at  sites that may be influenced by a nearby emission source
(i.e.  Fort Wayne, IN., downwind of Chicago).  At these sites  significant variations in  wet  depo-
sition rates could have been  caused by  small changes in local wind direction.
                                            54

-------
TABLE 7 A. LINEAR R2 COEFFICIENTS ARE SHOWN FOR THE LINEAR REGRESSION MODEL FOR
   SO4" WET DEPOSITION (USING EQUATION 19), LOGARITHMIC MODEL (EQUATION 23),
   COMBINED MODEL (EQUATION 24) AND COMBINED MODEL WITH TIME-OF-YEAR TERM
       INCLUDED (EQUATION 29). ALL MODELS HAVE p PARAMETERS VARYING WITH
METEOROLOGICAL CATEGORY. THE HETEROGENEITY-OF-SLOPES STATISTIC FS (EQUATION 20)
        IS SHOWN FOR THE COMBINED MODEL (TIME OF YEAR TERM INCLUDED).
R2 (Sulfate
SFTE
Turners Falls, MA
Tunkhannock, PA
Zanesville, OH
Rockport, IN
Giles County, TN
Fort Wayne, IN
Raleigh, NC
Gaylord, Ml
Alamo, TN
Winterpdft, ME
Uvalda, GE
Selma, AL
Clinton, MS
Marshall, TX
Lancaster, KN
Brookings, SD
Underhill ,VT
Big Moose, NY
Dorset, ON
Longwoods, ON
LINEAR
.144
.357
.731
.631
.575
.305
.656
.609
.406
.347
.266
.379
.365
.591
.727
.726
.436
.418
.576
.780
103
.236
.438
.682
.599
.574
.313
.619
.547
.373
.384.
.214
.430
.403
.651
.707
.731
.396
.358
.516
.678
Wet Deposition)
COMBINED
.273
.448
.730
.641
.590
.373
.658
.605
.463
.422
.276
.467
.432
.666
.731
.744
.455
.428
.573
.746
COMBINED +
TIME-OF-YEAR
.373
.591
.790
.719
.653
.433
.683
.662
.542
.481
.355
.510
.471
.682
.735
.745
.603
.617
.639
.822
FS
2.33
3.29
7.62
4.96
1.91
1.78
2.95
5.08
3.32
5.79
1.97-
4.42
2.77
1.81
3.80
1.81
8.18
5.63
3.79
4.16
                                   55

-------
TABLE 78. UNEAR R2 COEFFICIENTS ARE SHOWN FOR THE LINEAR REGRESSION MODEL FOR
   NOa'WET DEPOSITION (USING EQUATION 19), LOGARITHMIC MODEL (EQUATION 23),
   COMBINED MODEL (EQUATION 24) AND COMBINED MODEL WITH TIME-OF-YEAR TERM
      INCLUDED (EQUATION 29). ALL MODELS HAVE p PARAMETERS VARYING WITH
METEOROLOGICAL CATEGORY. THE HETEROGENE1TY-OF-SLOPES STATISTIC FS (EQUATION 20)
        IS SHOWN FOR THE COMBINED MODEL (TIME OF YEAR TERM INCLUDED).
R2 (Nitrate
SITE
Turners Falls, MA
Tunkhannock, PA
Zanesville, OH
Rockport. IN
Giles County, TN
Fort Wayne, IN
Raleigh, NC
Gaylord, Ml
Alamo, TN
Winterport, ME
Uvalda,GE
Selma, AL
Clinton. MS
Marshall. TX
Lancaster, KN
Brookings, SD
Underhill ,VT
Big Moose, NY
Dorset, ON
Longwoods,ON
LINEAR
.064
.282
.623
.498
.480
.283
.496
.591
.340
.024
.360
.366
.444
.524
.752
.709
.401
.329
.426
.666
IDG
.184
.479
.613
.516
.555
.349
.496
.535
.462
.170
.336
.478
.507
.577
.737
.716
.439
.384
.419
.568
Wet Deposition)
COMBINED
.223
.490
.667
.552
.577
.404
.532
.599
.524
.226
.389
.507
.538
.597
.755
.740
.472
.425
.458
.629
COMBINED +
TIME-OF-YEAR
.284
.547
.695
.644
.589
.488
.532
.631
.612
.286
.504
.647
.667
.682
.780
.751
.498
.433
.464
.670
FS
1.60
2.37
5.36
4.89
2.34
1 .92
2.32
4.26
2. S3
6.62
1 .98
1 .04
1 .74
3.68
3.05
1.30
7.05
5.24
1.96
6.99
                                    56

-------
   s
   T
   A
   T
   I
   S
   T
   I
   C
       6 T
       4 --
      -4 •-
      -6 -L
                                        sulfate  D  nitrate
              Mass.
Vt.        N.Y.        Ohio         Ind.
   RANKED SIMULATION NUMBER
Term.
Figure  12.   T8 statistic for meteorological  category  6, indicating deposition-to-precipitation
              ratios significantly above or below the norm at each site. Values of greater than
              1.66 or less than -1.66 are statistically significant at the 95% confidence level

      Although much of the remaining variations in S04" and NO3" wet deposition may be due to
random fluctuations, there was evidence that  at least some of the remaining variations may be
related to meteorological factors other than  the ones  represented  here.  Evidence of further
meteorological influence was based on  the high correlation  (correlation  coefficients  >0.8)
between residual errors  in S04* wet deposition at a  given site and  residual errors in NO3" at a
given site.  In other  words, when the  amount of SO4" wet deposition observed during a specific
event  exceeded the amount  predicted from the statistical correlation, there was a high proba-
bility that the amount of  NO3' also  exceeded  the predicted value.   Residual errors in S04= wet
deposition at each site were also positively correlated  with residual errors at neighboring sites
(especially in the Northeast),  though there was  little  correlation  with errors  at distant sites.
These two results suggest that the residual errors were associated with aspects of atmospheric
circulation that were not  adequately represented in the  categorization.
                                            57

-------
              Mass.
VL      ,  N.Y.        Ohio         Ind.
   RANKED SIMULATION NUMBER
Tenn.
Figure  13.  T, statistic for meteorological category 7, indicating deposition-to-precipitation
              ratios significantly above or below the norm at each site.  Values of greater than
              1.66 or less than -1.66 are statistically significant at the 95%  confidence level.
      Additional insights  regarding  deposition rates for  various  categories  was obtained  by
comparing the statistical  significance of  deposition associated with individual meteorological
categories with the norm at individual  sites.  As described earlier, the statistic Ts  (Equation
21) indicated when deposition associated with individual categories was significantly  greater or
less than the norm. Some values for the Ts statistic are shown in  Figures 12 through  14, based
on  the initial linear model (Equation 19).
                                            58

-------
              Mass.
Vt.         N.Y.         Ohio         Ind.
    RANKED SIMULATION NUMBER
Tenn.
Figure   14.  T8 statistic  for meteorological category 11, indicating deposition-to-precipita-
              tion ratios significantly above or below the norm at each site.  Values of greater
              than 1.66 or less than -1.66 are statistically significant at the 95% confidence
              level.
      Results showed that some categories (most notably the stagnant "waffling high", category
7) had significantly greater than average deposition at sites across the eastern and midwestern
U.S.,  while categories marked by more vigorous circulation (e.g. category 11) have deposition
below average.  An interesting result was obtained for category 6, where deposition was signifi-
cantly greater than average at northeastern sites and significantly below average at sites in the
Ohio valley.   Category 6 was characterized by a cold  front  in the eastern U.S. with a weak
southwesterly flow ahead of the front and  more vigorous winds behind it.

      Figures 12 through 14  also illustrate the  similarity in patterns of wet deposition of S04=
and of NO3".  Greater than average wet deposition of N03" was consistently associated with above
                                            59

-------
average wet deposition  of SO4", despite the fact that SO4* and NO3' show different seasonal
variations.

3.1.4.    Interannual  Variability

      The use of either  analytical or statistical  techniques to predict future wet deposition rates
raises questions  concerning  year-to-year variations in both deposition and  meteorology that
might affect the accuracy of predictions.  Wet deposition can  vary from  year to year due to
meteorological  fluctuations, including changes  in precipitation and changes in atmospheric
circulation patterns; and some of this variance  may be the result of long-term climatic changes.
Although the meteorological categorization and aggregation methods presented here have been
based on observations made during the period from 1982 to 1985,  we believe the same tech-
niques may be applied to future years.

      This section  will  present a summary of  interannual variations  in  S04" and N03" wet
deposition observed  during the  period 1982-1987, and to possible  correlation  with meteor-
ology. The extent of interannual variations will be used to establish the level of uncertainty in
aggregated estimates for annual deposition. This uncertainty may be reduced somewhat if year-
to-year variations  are  correlated with  observable  (and  possibly predictable)  changes  in
meteorology.    Special attention will be  given to results from the years 1986 and 1987,  which
were "not included  in the development of the meteorological  categorization and aggregation
techniques.

      Figures  15 through 17  show annual fluctuations in wet SO4* wet deposition, precipitation
and SO4* wet deposition-to-precipitation ratios  at three sites:   Turners Falls, MA, Zanesville,
OH  and  Giles County. TN; respectively.  Two features of these  fluctuations are worth comment.
First, the year-by-year  pattern shows significant differences among  the  three sites.  Annual
fluctuations in both wet  deposition and precipitation  are much greater at the Massachusetts site
than in the two inland  locations, possibly reflecting the importance of local meteorology at -a
location  closer to the ocean.  Year-to-year fluctuations in wet deposition average -25%  at the
Massachusetts site, while  at  the Ohio site an annual variation of  ~12% was more common.
These differences are significant at a 95% confidence level. Second, the year 1987 stands out as
a  year of unusually low SO4* wet deposition,  although this may be partly due to errors  in the
data.2 At the Massachusetts and Tennessee sites observed S04* wet deposition rates are 30%-
40% lower during  1987 than for the 1982-1985 period.  Precipitation  at these sites  during
1987 is  only  slightly below normal, and the  observed deposition-to-precipitation  ratios are
much tower than  in the preceding years.  This pattern of lower SO4* wet deposition during 1987
was repeated at most other sites included in this study, with S04" wet deposition  in 1987 aver-
2The sulfate wet deposition data for the second half of 1987 has recently been identified as in
  error due to problems in data analysis.

                                            60

-------
aging 24% lower than the  1982-1985 rates.

      Nitrate  deposition  during 1987 is much  closer  to  the  1982-1985  norm,  although
reduced NO3" was observed  at the Massachusetts and Tennessee sites. At the Ohio site observed
SO4" wet deposition during 1987 close to the  1982-1985 average, and NO3" wet deposition is
actually higher than average.   The S04" wet deposition-to-precipitation  ratio  is also above
average at the Ohio site.  Statistical correlations (described below and in  Section 3.1.3) show
that nitrate deposition during  1987 was significantly below the 1982-1985  average  at 5 of the
11 sites in the northeastern and midwestern United States,3 and  significantly above the  1982-
1985 average  at just one site.

      The extent of year-to-year  variations  can be  estimated by calculating the  root-mean-
square difference (RMSD) between wet deposition  during individual years at  each site and the
6-year average deposition at the site, i.e.
- V  f
                                                  i l°H ' DJO'l2                   (29)

where DJJ  is the observed deposition  at site j  for the iln year and DJQ is  the observed 1982-
1985  average deposition at site j.  This interannual RMSD  is equivalent to the  RMSD used to
evaluate aggregate  estimates  (Equation  9) and will be used  both  as a measure of inherent
uncertainty and as a standard for evaluating aggregate  estimates.  The interannual RMSD for
sulfate wet deposition at the 13 northeastern sites, expressed as a percentage of average annual
deposition, is  0.13.  For nitrate the interannual RMSD is 0.14.

      Reasons for the observed year-to-year variations  in wet deposition  are not immediately
apparent.  There  is  some  evidence that year-to-year variations may be due to  local meteoro-
logical fluctuations, since there is a strong correlation between variations  in sulfate deposition
and variations in  nitrate at each site.  The tendency for sulfate  and nitrate to exhibit similar
variations is apparent in Figures 15-17.  There is little correlation between sulfate and nitrate
deposition  at  different locations, as would  be  expected if year-to-year variations  were due to
changed emission rates.  However there  is also little  correlation  between observed annual
deposition  and meteorological  parameters (i.e.  precipitation) -or circulation patterns.  Although
there  is some correlation between wet deposition to precipitation, the ratio of annual sulfate wet
deposition  to  annual precipitation,  also shown in Figures 15-17,  varies significantly  from one
year to the next.
    Figure  2 identifies 13  northeastern and midwestern sites.  The multi-site analysis  in this
  section is based on 11  sites rather than  13, since data for 1986  and 1987 were missing for
  the two Canadian sites.

                                            61

-------
         3000 -•

         2500 -•

         2000 ••
               •
         1500 -

         1000 -.

          500 ••

             0--
       SULFATE (mg/m2)
            1982
             1983
1984
 1985
   1986
     1987
                  NITRATE (mg/m2)
             1982
             1983
• 984
 -955
   1986
    1987
25 j

20 -:

15 -.

10 -•

 5-

 0--
                  SO4 Wet Deposition/Precipitation Ratio
                        1983
                        1984
           1985
            1986
              1987
                                           PRECIPITATION (cm)
                 '9E2
                1983
 1984
•953
1986
1987
Figure  15.  Year-by-year variation in SO4" wet deposition, NO3" wet deposition, SO4" wet
             deposition-to-precipitation ratio,  and precipitation  at  Turners  Falls,
             Massachusetts.  Dashed lines represent predicted annual deposition based on
             meteorology and  1982-1985 statistical  correlation.
                                          62

-------
        2000 -•
        1500 -•
        1000 -•
         500 -•
            0
      SULFATE (mg/m2)
            1982
           1983
1984
1985
1986
1987
            1982
           1983
1984
1985
1986
1987
           35-j-
           30
           25 -•
           20 -•
           15 ••
           10 ••
            5-
            1982
          150 j

          120 ••
     SO4 Wet Deposition/Precipitation Ratio
           1983
1984
1985
1986
1987
    PRECIPITATION (cm)
Figure   16.
    1982      1983      1984      1985      1986      1987
Year-by-year variation in S04* wet deposition,  NO3' wet deposition, S04= wet
deposition-to-precipitation  ratio, and precipitation at Zanesville, Ohio.  Dashed
lines represent predicted annual deposition based on meteorology and 1982-
1985  statistical  correlation.
                                          63

-------
            1982
        1983
1984
1985
1986
1987
        2000

        1500

        1000  ••

          500  ••
NITRATE (mg/m2)
            1982
        1983
1984
1985
1986
1987
           35 T
           30
           25 ••
           20 ••
           15 -•
           10 -
            5--
   SO4 Wet Deposition/Precipitation Ratio
            1982
        1983
1984
1985
1986
1987
          150

          120
   PRECIPITATION (cm)
                           1983
                     1984
           1985
         1986
       1987
Figure  17.  Year-by-year variation in S04* wet deposition, NO3" wet deposition, SO4= wet
             deposition-to-precipitation ratio, and precipitation at Giles County, Tennessee.
             Dashed lines represent predicted annual deposition based on meteorology and
             1982-1985  statistical correlation.
                                          64

-------
      Year-to-year changes in atmospheric circulation can be represented by the frequency of
occurrence of events corresponding to specific meteorological  categories.  Changes in frequency
of occurrence  have been plotted in  Figure 18 for three of the most frequently occurring cate-
gories ("weak trof-southerly flow", category  6;   "waffling high", category 7;   and "northeast
cold front", category 11).  Variations in category frequency are approximately the same order
of magnitude  as year-to-year variations in  precipitation or wet deposition  (-10%).  JBince
these categories can be associated with high or low wet deposition (see Figures 7 and 12-14), it
is possible that these variations may explain  some of the interannual variability.
         0.2 y


       0.15--

         0.1 •:.

       0.05--
                                    CATEGORY 6
                                    CATEGORY7
                                    CATEGORY 11
           1982
1 983
•  1 984
1985
1986
1987
 Figure  18.  Year-by-year frequency of occurrence of events of meteorological categories 6,
              7 and 11, expressed as a fraction of all events.
      The impact of precipitation  and meteorology on wet deposition for  specific years can be
 estimated by applying the statistical model  for event-specific correlations  (Section 3.1.3)  to
 the observed annual meteorological  patterns.  In the model deposition per event is correlated
 with  precipitation during the event,  meteorological  category  and time of year, derived from
 observations during 1982-1985.  An  estimate for deposition for a specific year can  be obtained
 by applying the statistical  correlation (Equation 28) to the meteorological record (day-to-day
 precipitation,  time of year  and meteorological category) for the year.  The resulting estimate
 provides  an indication  for the extent of year-to-year variations that may be induced by mete-
 orology.  The annual estimate also provides a test of the ability to predict deposition for future
 years.

       Projected deposition values from this statistical  correlation are  also  shown for 1982
 through 1987 at the Massachusetts, Ohio and Tennessee sites  in Figures 15 through 17  (dashed
                                             65

-------
lines).   Results of the statistical  model project year-to-year variations in wet deposition that
are nearly as large as the observed fluctuations, suggesting that changes in meteorology may  be
a plausible explanation for the observed changes.  However the predicted deposition values do not
always  correspond with the observed  annual fluctuations.   For 1987  the  statistical model
predicts lower than average deposition at the Massachusetts site but not for Tennessee. Tests  for
statistical significance at the 95% confidence level  (Ts,  Equation 21) indicate that sulfate
deposition during  1987 was significantly  lower than  in previous years, even  accounting  for
changes in precipitation and meteorological category, at 13  of the 18 sites for which 1987 data
is  currently  available.   Nitrate deposition  is  significantly lower at 6  of the  18  sites and
significantly  higher at  one site.  There is room  for considerable improvement in  models to
predict changes in future deposition associated with meteorology:

      Figure 19 shows a comparison  between observed SO4" wet deposition during 1986 and
1987 and statistically  projected deposition (based on 1982-1985  observations) for  all  sites
included in this study.  Results show that observed deposition tends to be lower than projected
values,  largely due to  the anomalous year 1987.   Projected deposition falls within 20% of the
observed values  at  most  locations.  Tests  for statistical  significance  show that the use of
meteorological categories in the statistical projections result  in  an  improved ability to predict
variations in deposition that is significant at 95% confidence.
             4000
                         500    1000    1500   2000   2500   3000   3500
                              OBSERVED SO4 DEPOSITION (rng/m2)
4000
Figure  19.  Comparison between annual SO4* wet deposition observed during  1986 and 1987
              at 13 northeastern and midwestern sites with projected deposition  based on
              1982-1985  observations and 1986-1987 meteorology.
                                           66

-------
3.1.5.    Evaluation of  the category-based  aggregation  method

      Evaluation of the aggregation  technique focused on comparing nineteen-day aggregate
estimates generated using five  different event selection and weighting techniques.  Using each
technique one-hundred independent ensembles, each containing nineteen three-day periods,
were  selected  randomly from  1982-1985 observations.  In the first scenario events.were
taken from the  entire pool of observations and no additional knowledge was used in estimating
annual depositions. In  each of the other cases more information was employed.  For the second
and third scenarios events were again selected with no guidance, but  aggregation was performed
with  prior knowledge of the amount of precipitation that fell lit'each  site during each  event
(second scenario), and with  prior knowledge of both precipitation  and the season of each event
(third  scenario).   In the fourth case  events were  selected  (randomly) with  a prescribed
distribution among the seasons, and aggregated using the precipitation and seasonal information.
For the fifth case  the events were selected (randomly) from prescribed categories and  annual
depositions were estimated with the  aid  of  category-based  scaling (as outlined  above).
Aggregation was  based on equations (7) and (8) with the  prior estimate for deposition (D,)
reflecting the extent of prior  knowledge available for each case.

      The results of the various scenarios are compared in Figure 20. Evaluation was based on
the root mean-square difference (RMSD, Equation 9) between estimated mean annuaj deposition
and  actual annual deposition for the 13 sites in the northeast and  midwest.   The RMSD was
calculated for  each ensemble.  Figure 20 shows that the deviations  between estimated  and
observed deposition are smallest for the ensembles chosen by meteorological category and  that
the inclusion of more and more information sequentially improved upon the ensembles selected
at random.  Table 8 clarifies the meanings of the labels shown in the legend of Figure 20.  For
the category-based selection the  average deviation  is 18% of  the  observed deposition (2380
mg m-2 over the  13 sites), while  the smallest deviation for  a 19-day ensemble is 11%.  Note
that the  11% deviation is lower than  the  RMSD  for year-to-year variations in wet deposition
 (Equation 27).   Deviation  for  ensembles selected at random are significantly higher than for
 ensembles selected by category.  Use of observed precipitation in deriving aggregate estimates
 results in greatly improved accuracy. Aggregation based on random selection,  even accounting
 for precipitation and season, results in an average deviation of  22% (2850 mg nr2 over the 13
 sites).  Category-based selection reduces this deviation by approximately 25%.
                                            67

-------
                        —  R      —  R-P    — R-P.S
S-P.S  — C-P.C
              100%
       N
               40% --
          N   30% --
               20% --
               10%
                 0%
                           10    20   30   40   50   60   70   80   90   100
                                       ENSEMBLE  NUMBER
Figure   20  Comparison of selection and aggregation techniques for S04  wet deposition in
             northeastern North America.using nineteen three-day episodes.  The selection
             techniques include ensembles of events selected purely at random (R), selected
             with prescribed  seasonal distribution (S)  and selection with prescribed
             distribution among meteorological categories (C). Aggregation techniques are
             based on no prior knowledge of meteorology (R); prior knowledge of precipita-
             tion for each event (-P); prior knowledge of both event precipitation and season
             (-P.S); and prior knowledge of  event precipitation and meteorological category
             (-P.C).
                                           68

-------
3.2.        SELECTION OF CASES

      A total of 30 three-day events have been chosen for initial use in the analysis of annual
deposition.   Selection was based on methods outlined  in Section 2.1.5 with selection from
meteorological categories described in Section 3.1.2.  Events were selected  from  prescribed
meteorological categories  (including  the  split into  'wet* and  'dry' subcategories), with the
categories chosen (1) to represent meteorological processes that account for the  majority of
total  wet deposition  in  eastern  North America, and (2) to include important meteorological
patterns (i.e. cold-season  events and events with little precipitation for  evaluation of dry
deposition) that might be overlooked in the  first criteria.  Selection of categories was based on
the following  specifications:

      (a)   19 3-day events were selected from the 9 'wet1 categories which accounted for the
            greatest deposition on an annual basis. We estimate that these 9 categories account
            for about 75% of total annual S04" wet deposition at all northeastern and
            midwestern  sites (see Section 3.1.2).

      (b)  6 additional  3-day  events were selected to represent dry deposition  processes.
            These days  were selected  according to the criteria that the sum of 3-day SO4~ wet
            deposition "for the  13  northeastern  and midwestern sites total less than 50 mgrrf .

      (c)  5  additional 3-day events  were  selected  from among winter days (November-
            March).  These days were selected from categories that were not included in  trie
            original 19 and which may offer unusual  conditions.

      Choice of  the individual events was designed to meet two additional criteria:  random
 selection to avoid the introduction of  bias,  and accurate  representation of deposition at the 13
 northeastern  and midwestern  sites.   Evaluation of the  accuracy of individual  ensembles was
 based  on the aggregated estimates for annual deposition at each site derived from observed
 deposition on the sample days. Selection was made according to the following procedure:

       (1 )   The  distribution of sample 3-day events among the 38 categories was  prescribed
             based on deposition attributable to  each category and frequency of occurrence (see
             Section  3.1.2).

       (2)    Once the distribution  among categories was selected, ensembles of  specific 3-day
             events were generated. Each ensemble consists of a series of  3-day events  with the
             pre-determined distribution among the 38 categories.  The individual  3-day  events
             representing each category were selected at random.
                                             69

-------
      (3)   For each ensemble of specific events, estimated annual deposition was determined
           for the 13 northeastern and midwestem sites, based on observed deposition during
           the ensemble events and the category-based scaling factors.  The standard deviation
           between estimated and actual annual deposition for the 13 sites was determined for
           the ensemble.

      (4)   Ensembles were ranked based on standard deviation, and the best ensemble (with
           the lowest standard deviation among 20 randomly generated  ensembles) was
           selected as the representative ensemble.

      The selection was performed  in two stages.  Initially, the ensemble of 19 3-day events
representing categories  with  the highest  deposition was selected.   The  additional 11 events
representing  dry and winter deposition were selected after the initial ensemble had been
selected.  In the selection of the additional 11 events, estimated annual deposition was based on
an ensemble  of 30 events, with the prior 19 added to the randomly selected 11  events. The 11
additional events were chosen to provide the most accurate 30-event ensemble. The selected 3-
day periods are shown in Table 8.
3.3.       COMPARISON OF AGGREGATE ESTIMATE  WITH  LONG-TERM DEPOSITION
                                 »
      This section explores the accuracy of aggregate estimates derived based on the selected
30-event ensemble (Table 8).   Aggregate  estimates have been derived for sulfate and  nitrate
wet  deposition  and for total acidity, including estimates  for annual and seasonal totals.  The
estimates are based on the observed wet  deposition for  the selected events,  and on observed
correlations between wet deposition, precipitation and meteorological category fcr the 1S82-
1S85 period (Equations 7 and  8).  Aggregate estimates are compared with observed wet depo-
sition  during   1982-1985.


3.3.1.     Annual  Wet Deposition

        The following section  examines the ability of the aggregation technique to reproduce
annual deposition patterns for SO4", NO3* and H* wet deposition. The observed deposition  values
for the 30 episodes recommended for RADM simulation were used to estimate deposition using
the proposed aggregation technique.  From these results,  plots of deposition  across eastern
North America were drawn.  Estimates of annual and seasonal SO4=,  NO3" and H* depositions
were determined for all 22 sites.  The spatial patterns were analyzed for agreement with the
actual mean deposition patterns for 1982-1985.
                                           70

-------
  TABLE 8.  DATES OF THREE-DAY EPISODES (MIDDLE DAY) SELECTED FOR AGGREGATION.
  CATEGORY NUMBERS CORRESPOND TO DESCRIPTIONS GIVEN IN TABLE 4, WITH W AND 'D'
    REFERRING TO THE SPLIT INTO -WET" AND "DRY" SUBCATEGORIES BASED ON SULFATE
                           DEPOSITION (SECTION 3.1.2)
        Date
Category
Date
Category
      (Original 19  Periods)

      April 15-17,  1982     1W
      November  2-4, 1985   3W
      May  29-31,  1983      5W
      March  17-19,  1984    5W
      July  12-14,  1985      6W
      September 9-11,  1983  6W
      October 11-13, 1985   6W
      November  1-3, 1983   7W
      August 21-23, 1984    7W
      June 12-14,  1984      7W

      (6  dry  periods)

      April 28-30.  1985    19D
      June 10-12,  1983      8D
      July  17-19,  1985      9D

      (5  winter periods)

      February 1-3, 1982    2W
      December  9-11,  1983  4W
      November  4-6, 1985  12W
                  May 2-4, 1985,
                  September 9-11,
                  September 14-16,
                  September 25-27,
                  August  4-6, 1983
                  November  16-18,
                  September 26-28,
                  July  16-18,  1984
                  December  15-17,
                  December  16-18, 1985
                  May  14-16,  1982
                  November  6-8, 1982
                  February  9-11,  1985
                  November 13-15,  1983
1984
1983
1982
1985
1985
I
1982
8W
9W
9W
9W
9W
10W
11W
11W
11W
                       11D
                        7D
                        6D
                      14W
                      17W
3.3.1.1.    Sulfate Wet Deposition-
     Figures 21 a and b show the observed  annual SO4* deposition over the four-year period
(1982-1985) and the estimated annual deposition based on the 30-day ensemble.  In general,
the estimated depositions and spatial patterns agreed well with observed values within the thir-
teen  site domain. The annual and seasonal  results are summarized Table 9.  The mean RMS
errors shown were determined across all sites and are expressed as a percent of the average
deposition across all of these sites. Annual aggregate estimates were within 18% of the actual
deposition at the thirteen northeastern sites and within 10% at eight of those sites. The mean
RMS error as a percent of the average 22 site SO4* wet deposition was 13%.
                                        71

-------
Figure  2la
The mean annual SO4" wet deposition pattern across eastern North
America during 1982-1985.  (Units-mgm'2).  Contours correspond to
1000, 2000 and 3000 mgm"2.
                                         72

-------
Figure  21b
The aggregate estimate of the mean annual SO4= wet deposition pattern
across eastern North America for 1982-1985.  {Units=mgm~2).
Contours correspond to  1000, 2000 and 3000 mgm  .
                                         73

-------
TABLE 9.  ROOT-MEAN-SQUARE DEVIATIONS BETWEEN ESTIMATED AND OBSERVED SUMMER,
               WINTER AND ANNUAL WET DEPOSITION AT 13 EASTERN SITES.
          DEVIATIONS ARE EXPRESSED AS A PERCENTAGE OF OBSERVED DEPOSITION
Species
Sulfate
Nitrate
Hydrogen
Annual
13 %
20 %
17 %
Summer
20 %
27 %
24 %
Winter
20 %
35 %
21 %
      It can be seen that throughout eastern  North America there was  no clear tendency for
under or over estimation of SO4* deposition.  However there was a tendency to overpredict or
underpredict deposition  in certain regions.  Overprediction occurred at sites in the southern
Midwest (  Kentucky, Ohio, Indiana).   Underprediction tended to occur at  northerly sites
(Wisconsin, Michigan, Ontario, New York,  Vermont,  Maine and Massachusetts).  Underesti-
mation  also  occurred North  Carolina, Tennessee  and Georgia, while predictions  at the
Mississippi  and Alabama sites tended to be high.  Summer and winter depositions showed similar
tendencies towards over- and underprediction in specific geographic regions.   This apparent
regional consistency in over-  and  underestimation was not considered unrealistic.   On  the
regionaJ scale .the spatial variation  in precipitation amounts due to a typical weather system
varies significantly.  Since only a small number of episodes are used in the aggregation, the
spatial variation  associated with them  is carried  over  to the long-term pattern.

3.3.1.2.     Nitrate  Wet Deposition—
      On an annual  basis the  actual and estimated plots of NO3" wet deposition are shown in
Figures 22a and b. The mean RMS error as a percent of the average 22 site annual deposition
was  20%.   The RMS error is greater for nitrate wet deposition than for sulfate. possibly
reflecting the greater importance  of local meteorology in controlling the rate of nitrate depo-
sition.  Also in contrast to sulfate.  the RMS error for nitrate was significantly  larger than the
year-to-year RMS deviation (13%,  from  Section 3.1.4)

      Qualitatively the  spatial variations of  sulfate  and  nitrate  wet deposition  were quite
similar.  The area of greatest  deposition extended from  the Midwest to the Northeast.  While
depositions were least in the northwest and south.  As with S04", whether or not the NO3' wet
deposition at a given  location was over and underestimated depended its geographical location.  In
the lower Midwest the estimates were somewhat overpredicted  in northern Indiana and Ohio.
Similar to  SO4",  the deposition  estimates were generally  low from  southern  Ontario and
northern Pennsylvania to the Northeast coast. The largest deviation was at the Tunkhannock. PA
receptor.  As with SO4", overall the was no bias towards under- or overprediction. The slight
changes in regional biases between N03" and SO4* wet deposition were likely due to the differ-
ences in the emission patterns and in the  differences in their chemistry and rain-out charac-
teristics.

                                           74

-------
                                        1470

                                        I/'
                                1390    1620
Figure  22a  The mean annual NO3' wet deposition pattern across eastern North America
             during 1982-1985.  (Units=mgm'2).  Contours correspond to 500, 1000,
             1500 and 2000 mgm  .
                                        75

-------
             r   m
859
                                                                      Nitrate
Figure  22b  The aggregate estimate ot the mean annual NO3" wet deposition pattern across
             eastern North America for 1982-1985.  (Units=mgm~2).  Contours correspond
             to 500, 1000. 1500 and  2000 mgm'2.
                                         76

-------
3.3.1.3.     Acidity  Wet  Deposition—
      As expected the pattern of annual H* deposition reflected the behavior of N03' and S04=
deposition with maximum values from the lower Midwest across the eastern Great Lakes to the
Northeast United States  (Figures 23a and b).  The mean RMS error of 17% is between those of
N03' and SO4=.   The pattern  of significant underestimation  in the  Northeast,  in particular in
northeastern Pennsylvania and  upstate New York, was again apparent.  Some of the episodes
selected for aggregation were either associated with small  amounts or precipitation or low
pollutant concentrations  (or both).  It is likely that only one or  two of  the episodes were biasing
the results.  With only 30 events, of which only 21  had precipitation at Big Moose for example,
the characteristics of each  one will significantly influence the- deposition approximations.  It
would be extremely difficult to  identify an ensemble of 30 events that did  not result in some
bias somewhere  in eastern North America.


3.3.2.    Seasonal  Wet   Deposition

3.3.2.1.     Sulfate Wet  Deposition-
      Figures  24  a-d show estimated and observed S04" deposition  for the  summer and winter
halves of the year. The mean seasonal results were also summarized in Table 9.  The errors
resulting from the estimation of wet deposition using the aggregation method  were less  than
20% for eleven of the sites during summer and nine  in winter.           •

3.3.2.2.     Nitrate  Wet  Deposition—
      The actual  and  aggregate patterns for wet NO3' deposition in~ the summer and winter are
included in Figures 25 a-d.  In  the summer the highest values were reported in Ontario. Using
the aggregation routine, the  episodes selected predicted large wet deposition values to occur in
that area.   The  deposition values were somewhat over-predicted  in  the  Midwest excluding
Rockport, IN  and were again under-predicted  in the Northeast.  At the New York, Pennsylvania
and Massachusetts sites  the errors were  fairly significant but well  within 50%.  Throughout
the northeastern  region the  average error was 19%.  In winter the errors were worse.  The
average over the same  region was  25% but at the  Michigan, Vermont and Pennsylvania sites it
was close to 50%. In upstate New York and northern Ontario the errors were around 40%.  One
reason  the estimates were worse  in the winter (for S04* and NO3") is that the actual depositions
were generally less.  Therefore,  a  relatively small  departure, that made little difference on an
annual  basis, appeared  significant.
                                           77

-------
Figure  23a  The mean annual wet acidity (H*) deposition pattern across eastern North
             America during 1982-1S85.  (Units=mgm"2).  Contours correspond  to 15, 30,
             45 and 60 mgm"2.
                                          78

-------
Figure  23b The aggregate estimate of the mean annual wet acidity (HT) deposition pattern
             across eastern North America for 1982-1985.  (Units-mgm'*).  Contours
             correspond to 15, 30, 45 and 60 mgm"  .
                                          79

-------
           UA
            V*\
             u
\
  si

                          ^
                        ^•-ft   Na«:
                                                        ^h
                          cT	
                            k_ /
Figure 24a  The mean summer SO4* wet deposition pattern across eastern North America for
            1982-1985.  Summer corresponds to the period from April through October.
            (Units=mgm~2).  Contours correspond to 1000 and 2000 mgm' .
                                     80

-------
 X  /
              /*
y*~-^wi!      •"'  \   \
         \   Jr
          v|
             s
                                                                 Sulfate
Figure  24b The aggregate estimate of the mean summer S04* wet deposition pattern across
            eastern North America for 1982-1985.  (Units=mgrrf2).  Contours correspond
            to 1000 and  2000 mgrrf2.
                                       51

-------
Figure  24c  The mean winter S04" wet deposition pattern across eastern North America for
             1982-1985.  Winter corresponds to the period  from November through March.
             (Units-mgm"2).  Contours  correspond to 500 and 1000 mgm'2.
                                         82

-------
Figure  24d The aggregate estimate of the mean winter S04' wet deposition pattern across
             eastern North America  for 1982-1985.  (Units-mgnV'').  Contours corresp<
             to 500 and 1000 mgm"2.
                                          83

-------
Figure  25a  The mean summer NO3' wet deposition pattern across eastern North America for
             1982-1985.  Summer corresponds to the period from April through  October.
             (Units=mgm~2).  Contours correspond to 500 and 1000 mgm"2.
                                         84

-------
                    \   L
                                                                      Nitrate
                                                             /•s
Figure  25b The aggregate estimate of the mean summer NO3" wet deposition pattern across
             eastern  North America for 1982-1985.  (Units=mgm~2).  Contours correspond
             to 500 and 1000 mgm'2.
                                         85

-------
Figure  25c  The mean winter NO3" wet deposition pattern across eastern North America for
             1982-1985.  Winter corresponds to the period from November through March.
             (Units-mgm"2).  Contours correspond to 500 and 1000 mgm"  .
                                        86

-------
Fiaure  25d  The aggregate estimale of the mean winter NO3' wet deposition pattern across
             eastern  North America for  1982-1985.  (Units=mgnY2).  Contours correspond
             to 500 and 1000 mgm" .
                                          87

-------
3.3.2.3.     Acidity Wet  Deposition—
      The  seasonal estimates  of H* wet deposition behaved similarly to those for  SO4" wet
deposition.  Maps of the actual and aggregate estimates are  included in  Figures 26 a  and b,
respectively.  In summer, the  errors ranged from 3%  to 34%  with the  average  over  the
northeastern region being 16%.  For the winter estimates the results were similar, ranging
from 0% to 45%.  The average was also 16%.  For all three analytes the tendencies for over and
underestimation  were not as consistent  for the seasonal depositions.  As  well there were a
number of  sites where the bias changed  direction between seasons.  This  further highlights  the
influence a few slightly atypical episodes had on the aggregate estimates of deposition.  In reality
these episodes were not atypical since the long-term patterns  that were observed and that  the
aggregate estimates were measured  against were a compilation of a large  set of highly variable
episodes.


3.3.3.     Uncertainties  In   the  Aggregate  Estimates-
      The estimates of seasonal and  annual wet deposition were based upon the categories iden-
tified for event selection and their mean frequency of occurrence and deposition to precipitation
ratio.  How well the specific 30 events represented the categories  they were selected from
ultimately governed  the accuracy of  the aggregate estimates.  Events that had  uncharacteristic
pollutant concentrations,  relative to their  categories, caused  the long-term  estimates  to deviate
from the actual value. This can be inferred from the aggregation equation (see equation 7}.  The
actual deposition (^nnual) is scaled by the ratio  of the total amount contributed by the  events
selected, weighted  by category  frequency, to the total  amount expected due to the categories
chosen,  weighted by their frequencies,  and assuming  the  same precipitation  amount  as  the
events selected.  Mis-representation  of the categories that occurred  more frequently and'or had
large mean deposition to precipitation ratios  had the  greatest effect on  the accuracy  of  the
estimates.  Thus, the real goal of the events selected was to  match the mean  behavior of  the
collection of categories chosen to represent the long-term deposition.
      The  certainty  of the estimates basically  rests with the  relationship between the actual
long-term depositions and the categories represented.  Since events were selected from those
categories  containing the meteorological situations contributing the largest  percentage of  the
annual deposition, the relationship should be relatively  strong (i.e. Changes  in the total  annual
deposition  should be similar to  changes  in the total amount deposited by the sample of cate-
gories).   Roughly  75 percent of the  mean  annual  13-site deposition during 1982-1985
occurred with events in the meteorological categories used.  On a site by site basis and from year
to year, however, this percentage varied.  Values below  50 would imply  that the relationship
between the actual long-term depositions and the categories represented was weak.  In this case
the certainty of the  annual deposition estimates would be diminished. During  1982-1985 this
situation did not arise, the mean percents accounted for varied from 60 to over 80. For three of
the sites (Big Moose, Gaylord and Zanesville) Figure 27 shows the amount of variation in  the
                                           88

-------
Figure  26a  The mean summer wet acid (H*) deposition pattern  across eastern North America
             for 1982-1985.  Summer corresponds to  the period from April through
             October.   (Units=mgm"2).  Contours correspond to  10,  20, 30 and 40 mgm"2.
                                          89

-------
Figure  26b The aggregate estimate of the mean summer wet acid (H*) deposition pattern
             across  eastern North America  for 1982-1S85.   (Units=mgm~2).  Contours
             correspond to 10, 20 and 30 mgm"2.
                                         90

-------
Figure  26c The mean winter wet acid (H*) deposition pattern across eastern North America
             for 1982-1985.  Winter corresponds to the  period from November through
             March.  (Units=mgm'2).  Contours correspond to 10, 20 and 30 mgm' .
                                          91

-------
Figure  26d  The aggregate estimate of the mean winter wet acid (H*) deposition pattern
             across eastern North America for 1982-1985.  (Units=mgm )  Contours
             correspond to 10 and 20 mgm"2.
                                         92

-------
percentages from year to year.  In general, the  meteorological  situations represented did not
contribute as much during the  earlier years as they did in the later years.  The reason for the
increase is not certain, however, one possible cause  is that during the earlier years there was
more meteorological data missing (850 mb u-v).  More of the  deposition  events were  therefore
classified as missing at the expense of the amount of deposition  falling  within the categories.
Across all sites and all years, however, the percent accounted for  rarely dropped below 50_and
consistently stayed  above 60.

      In Figures 28 and 29 the total amount of wet S04* deposition by year is compared  to the
total amount in each year associated with the set of categories represented.  Similar temporal
variations between the two would suggest that deposition from the categories is related  to the
annual total. Comparisons were made at two Midwestern sites and at two northeastern sites.  At
Zanesville, OH the  annual amount and the category totals were closely related.  At Rockport, IN
the  relationship is  masked  by the greater frequency of missing  categories during the earlier
years but the amounts do vary similarly.   In the northeast,  there  is also some agreement with
increases  tracking  similarly but  again the  larger number of missing  categories during the
earlier period caused some discrepancies.  The effect of the missing category frequency has not
been examined in terms of  these results  to conclusively say it was causing the problem.  This
will be studied in the future.  The deposition amounts due only  to the collection of categories was
correlated with the  total annual  amounts.   Thus,  when  the specific events (30)  properly
portrayed  the behavior of the categories  they were  selected from, the  aggregation  technique
produced estimates of long-term deposition with a relatively high  level of certainty.
               82  83  84  85  86  |   82  83  84  85  86  |   82 83  84  85  86
                  Zanesville,  OH
Gaylord,  Ml
  YEAR
Bigmoose,  NY
 Figure  27.  The spatial and temporal variation in the percent of annual S04* deposition
               accounted for by the meteorological categories selected for representation.
                                             93

-------
   c
   s
      m
   D
   S?
      ~
      e
                             OH
       S-IN
      OH-C    B3 S-IN-C
                     1982
1983
1984
1985
1986
Figure  28.  Comparison between the year to year variation in annual SO4" deposition and the
              year to year variation in the total amount of deposition accounted for by the
              categories selected for representation.  S-IN « annual total at Rockport and S-
              IN-C - category total at Rockport.  OH = annual  total at Zanesville and OH-C -
              category total at Zanesville.
      The aggregate estimates of deposition were sensitive to the frequency of occurrence of the
categories, the mean deposition to  precipitation ratio of the categories and to the concentrations
associated with each event.  As mentioned above in reference to equation 7, the estimates were
most  sensitive to events from the  frequently occurring, "wet" categories. To insure that these
situations were represented properly more than one event was chosen from these categories.
This measure reduced the possibility of one non-representative event being selected and dras-
tically altering the estimates.  As the number of events selected from each category is increased
the estimates will approach the actual deposition.   Using all four years of data to determine the
category frequencies and the deposition  to precipitation  ratios  insured that the  most reiia^ie
values were  used.  Since these factors and the 30 episodes  were constant, the  values of the
aggregate  estimates  at  each site did not vary.   Given  that the long-term  depositions were
correlated  with the amount of deposition resulting  from  events within  these categories, the
                                            94

-------
uncertainty ranges associated with the estimated values were expected to be relatively small.
The deviation between the estimated and observed depositions supplied an indication of the level
of certainty to which long-term deposition could be reproduced.
                     1982
1983
1984
1985
1986
 Figure  29.  Comparison between the year to year variation in annual S04* deposition and the
              year to year variation in the total amount of deposition accounted for by the
              categories selected for representation.  NY = annual total at Big Moose and NY-C =
              category total at Big Moose.  VT = annual total at Underhill and VT-C = category _
              total at Underhill.
      The actual and  aggregate  maps of annual and seasonal deposition were discussed in a
 previous  section.  The 30 events reproduced the depositions well enough at each site to yield
 spatial patterns similar  to  the  observed  depositions.  The magnitude of the differences
 (uncertainty) between estimated and actual were calculated as a percent of the actual deposition
 for all of  the sites.  These were  summarized above in Table 9.  Expressed as a percent of the
 mean deposition  across all sites, the RMS errors  were smallest  for the annual estimates.
 Overall, estimates were  closest  for wet  sulfate deposition  with  smaller annual and  seasons!
 percent errors as compared to the wet deposition of N03" and t-T.
                                            95

-------
      It was most important to estimate the wet deposition over the smaller thirteen site domain
accurately.  For each of these sites the  estimates were compared to the actual deposition.  As
with the mean  error across all sites, the aggregation technique usually yielded better results
for  the annual estimates at individual sites.  Reasons for this were suggested in the discussion
concerning the  spatial plots of wet deposition. The percent errors for wet SO4" deposition  are
shown in Figure 30.  Positive  values  indicate an over-estimate and negative  ones under-
estimation.  On an  annual basis  the error was within  20 percent at all thirteen sites^The
estimates were most uncertain at Big Moose, NY and Fort Wayne, IN  with errors of -17 and 18
percent respectively.  At the  other locations  uncertainty was less than 11 percent.  The most
uncertain aggregate estimate occurred at Tunkhannock, PA during the winter  half of the  year.
The percent error was 54 percent. In general, the seasonal estimates were within 30 percent
across  the  thirteen-site region with the exception of Clearfield,  KY in  both seasons and
Winterport,  ME in  the winter.   The uncertainty in the seasonal estimates at Clearfield and
Tunkhannock appear to balance during an annual time frame so that the longer-term estimates
are more certain.   The most consistent  bias in uncertainty  occurred  at Fort Wayne  where for
both seasons and  for the annual period the estimates were high.  The aggregate  estimates
contained the smallest amount of uncertainty in Ontario, Ohio, southern Indiana, Massachusetts
and Vermont.

      Figure 31 shows the level of uncertainty associated with the aggregate  estimates of  wet
nitrate deposition.   Errors appear to  be  somewhat more variable from region to region and the
amount of uncertainty is greater than it was  for sulfate.  Biases towards over  or  under predic-
tion were also more consistent across seasons.  At Tunkhannock and Turner Falls, MA  the
uncertainty  was skewed  towards under-estimation.  While at Zanesville and Fort  Wayne  depo-
sition was over-estimated by around 20 percent.   At  Big Moose the  uncertainty  changed sign
from winter to  summer as it  did  at Underhill, VT and Dorset, ON.  At these  sites the altering
direction in uncertainty resulted in more stable annual estimates but a Big Moose the annual
estimate  remained  relatively uncertain.  In  Kentucky,  Michigan, North Carolina. Maine  and
southern Indiana  and  southern Ontario the aggregate estimates contained relatively little
uncertainty  for wet nitrate deposition.

      Levels of uncertainty in estimates of total acidity deposition were generally between those
of nitrate and sulfate.  This was shown  in Table 9 and is also apparent in Figure 32.  In Ohio,
Indiana. North  Carolina. Michigan, Vermont and Ontario  the degree of uncertainty  was relatively
small (generally within 20 percent).  Confidence  in acidity deposition was  somewhat less  and
biased  low over Massachusetts, Pennsylvania, Maine and  New York.  Overall the  aggregate
estimates of wet acid deposition were most uncertain at Tunkhannock.

      Compared to the observed  depositions the level of certainty in the aggregate estimates
varied between sites and seasons. The  magnitude of the uncertainty  indicated how well the 30
episodes reproduced the mean behavior of the categories they were chosen to represent.  The
                                            96

-------
 larger values indicated that the concentration of pollutants in  the precipitation for some of the
 episodes selected were significantly different than the mean volume-weighted concentration of
 the categories used.  In general, it only required a couple of unusual events to noticeable effect
 the aggregate estimate.  Considering that the same 30 episodes were used to estimate deposition
 at all locations across eastern North America the results were quite good.  Certainly  it would be
 possible to produce better estimates if separate events were used for each site.   This would
 significantly increase  the number of RADM runs necessary  and is therefore not an option.   The
 real root of uncertainty lies not in the accuracy of the specific predictions but in the  choice of
 categories or meteorological situations represented.  To place confidence in the outcome of the
 aggregation  study  and  in the long-term  estimates  of deposition derived through  the  RADM
 project the episodes  selected  must vary meteorologically and chemically.  They must  also
 represent the types of  situations that contribute a reasonable  portion of the total deposition at
 any given location.  These aspects have been continually tested throughout this study to insure
 confidence in  the estimates and in the reaction of the  estimates  to altered emission scenarios.
                               D  Annual
                                         Summer  B Winter
p   60%
E
R   *0% +
C
E   2 0 % 4-
N
        E   -20% i
        R
        R   -40% ••
        O
        R   -60%

                                           4^
                                                            1
                    MA   PA   OH  S-IN N-N  NC   M   KY  ME  VT   NY  N-ON S-ON

                                  Precipitation Chemistry  Site
                                        (by  state/province)
Figure  30.   The uncertainty ranges for annual and seasonal wet SO4" deposition due to
              estimation using the aggregation technique.
                                           97

-------
Annual
                                          Summer  ES  Winter
              MA  PA  OH  S-IN  N-IN  NC   MJ   KY   ME   VT   NY  N-ON S-ON

                              Precipitation Chemistry  Site
                                   (by  state/province)


Figure  31.   The uncertainty ranges for annual and seasonal wet NO3" deposition due to
             estimation using the aggregation technique.
                              Annual
            Summer  5  Winter
              MA   PA   OH  S-IN  N-IN   NC   Ml    KY   ME   VT   NY  N-ON S-ON

                             Precipitation  Chemistry Site
                                  (by state/province)


Figure  32.  The uncertainty ranges for annual and seasonal wet H* deposition due to
             estimation using the aggregation technique.
                                        98

-------
3.4.        SOURCE-RECEPTOR RELATIONSHIPS

      The use of the meteorological categories  was a key step in the aggregation project.  In
addition to providing accurate estimates of deposition it was also assumed that sampling from the
dominant categories (in terms of total  amount of wet deposition contributed) would insure -that
the important source-receptor relationships  were represented.  The  category-based method of
event selection and the specific 30 selected episodes have thus far only been evaluated in terms
of observed wet deposition.  The use of these categories, the behavior of which were quantified in
terms of deposition and  seasonality, also allowed  for the application of scaling and weighting
factors during aggregation. The actual episodes  selected from the categories were also shown to
provide good estimates of annual  and  seasonal  wet deposition at a majority of the sites and to
reproduce the spatial patterns well.

      This section  focuses  on the differences  in source-receptor relationships between cate-
gories,  the  benefits of event selection by category and on  the actual source-receptor relation-
ships represented by the 30 episodes.  It was  initially anticipated that  because the categories
represent different 850  mb wind-flow patterns  they should also be associated with different
source-receptor relationships.   This implies that the winds at 850  mb (-1500 meters} were
related  to the winds  responsible for transporting and dispersing pollutants in  the atmospheric
boundary layer (surface  to  1500 meters).  Evidence  to support this  hypothesis was uncovered
in the testing of the wet  deposition behavior of  the categories. A possible explanation for their
differences in  terms of  wet deposition  is that they were  associated with  different source-
receptor relationships.  This should result in different depositions between categories because
the chemical composition of precipitation  depended upon  transport.  To investigate this, mixed-
layer back-trajectories and Q-fie!ds for three of the monitoring sites;  Big Moose, New Ycrk;
Gaylord, Michigan; and Raleigh, North  Carolina were  analyzed to  provide a more direct measure
of source-receptor  relationships.

3.4.1.    CUmatoloaical Wind Flow  Patterns during  Precipitation

      Over eastern North America the air generally moves from west to east advectir.g air-
bourne pollutants along with it.  At any  particular time and place,  however, the  flow deviates
markedly from  this pattern.  Over the course of a year or longer certain upwind areas will have
a stronger  influence upon each location.  It is these long-term tendencies  that play an key role
in the spatial pattern of seasonal and annual deposition of SO4", NO3' and  H+  and  in the specific
amount that any given receptor experiences.  In terms of long-term wet deposition it is the  wind
flow  during and shortly before  each event that is important.

      To identify which areas have the  greatest tendency to affect Big  Moose, Gaylord and Raleigh
 the  trajectories  associated  with  all of  the  precipitation  events during  1982-1986 were

                                            99

-------
combined into overall Q-fields for each site.  Figures 33 to 35 show the mean probability of
transport  to  Big  Moose,  Gaylord and  Raleigh,  respectively  (units are 10"8 km"2).  The air
arriving at each location had the highest probability of passing  over areas immediately adjacent
to the receptor and the probability decreases with distance.  The rate of decrease depends upon
direction and  as expected the gradient is somewhat sharper toward the east.  This is because the
winds from the east were less common than those from the west. At Raleigh, however, a part cf
the  reason for the rapid decrease towards that east was probably due to the lack of data over the
ocean.  Compared to Big Moose, the boundary layer winds during precipitation events at Gaylord
and Raleigh had a greater tendency to come from the north and south and did not favor a westerly
direction as much.  At Raleigh, northeast and southeast winds were more common than at the
other sites and in general, the wind appears to be lighter there.-' This behavior was indicated by
the  size of the area encompassed by the 25 x 10"8 contour.  The Q-field for Gaylord  shows that
northwest  and southwest winds  were relatively more  prominent.   Relatively close  to the
receptors (within  about 500 km) it appears that any sources to west  and east would contribute
more at Big Moose and Gaylord while at Raleigh no such tendency was apparent.  In general, the
probability fields  for all  events (with and without  precipitation - not shown) tended to be more
skewed towards the west because non-precipitating events were even less often  associated with
easterly and  southerly winds.

3.4.2.    Wind flow Associated with  Sulfate Wet  Deposition

      To obtain an  idea of where the air originated during events with higher levels  of SO4" in
precipitation  the Q-fields were  re-calculated using the  same set of trajectories  but  with
deposition-related weights attached.  The weighted Q-fields are shown in Figures 36 to 38.
Subtle  differences  between these fields and the non-weighted fields were  discernabie by
comparing the  figures.

      At Big  Moose the weighting appeared to have resulted in  increased probabilities towards
the south to southwest and locally.  The values decreased to the west, northwest and east.  The
effect of the  weighting for Gaylord showed  up as slight  increases to the south  and north and
decreases to the west to northwest. At Raleigh  the probabilities increased very slightly to the
north and it  is more noticeable  over Virginia.  There were decreases in probabilities to the
southwest, west and northeast.  These changes suggested a possible relationship to  the spatial
distribution of SO2 emissions across the Northeast.  To highlight the differences between the
weighted and  non-weighted fields the potential source contribution functions (PSCF)  were
determined.  These new fields clearly  show which regions were  most often affiliated with high
and low deposition amounts and can be more easily compared to the emissions field.
                                           100

-------
Flaure  33.  The  mean  probability (10'8 km'2) for transport on precipitation days to Big
             Moose. NY, during  the period 1982-1985.
                                          101

-------
Flgur»  34.  The mean probability (10~8 km"2) for transport on precipitation days to
             Gaylord. Ml. during  the period 1982-1985.
                                          102

-------
Figure  35.  The  mean  probability (10'8 km"2) for transport on precipitation  days  to
             Raleigh,  NC, during  the period 1982-1985.
                                          103

-------
Figure  36.  The SO4" wet deposition  weighted probability (10~8 km"2) for transport to Big
             Moose,  NY during the period  1982-1985.
                                         104

-------
Figure  37.  The SO4" wet deposition weighted probability (10"8 km'2) for transport to
             Gaylord,  Ml, during the period  1982-1985.
                                         105

-------
Figure  38.  The SO4" wet deposition  weighted probability (10~8 km'2) for transport to
             Raleigh. NC.  during the period 1S82-1985.
                                        106

-------
3.4.3.     Potential  Source  Contribution  Functions

      Figure 39 illustrates the  transport bias field  for Big Moose,  NY.  Values are positive
(more often associated with above average 804" wet deposition) to the south to southwest and to
the northwest to north-northwest. When airmasses  originated over Big Moose and  over areas
immediately to the  south extending across  Pennsylvania to  southeast Ohio the resulting
depositions were most likely to be large.  Large positive values close to Big Moose are indicative
of high depositions during stagnant conditions when the ambient S02 and SO4* levels were likely
to be rather homogeneous over large areas.  In general the regions  with positive PSCF  values
agreed with locations of known  SO2  emissions. The main exception is off the coast of  Virginia
where the PSCF is large.   It is possible  that this area was linked to above average depositions
because of synoptic situation  or  because of the  importance  of precipitation  amount in
determining wet deposition amount there.  In Figure 39 it can be seen that  when the  air was
from the east,  northeast, northwest and west depositions were more often  below  average at Big
Moose.

      Sulfate wet deposition amounts were used to  weight the Q-fields in the numerator o? the
PSCF's hence, one could  argue that the results were as much a function of  large  precipitation
amounts as they are enhanced  SO4* concentrations (i.e. more precipitation fell when the winds
were from the southwest).  To test for this  possibility the  PSCF  field  for precipitation (i.e.
numerator weighted  by precipitation amount) was  determined.  The result  for Big Moose is
shown in Figure 40.  In this case the  largest values were located to the east where values
approached a minimum in the S04* weighted PSCF field.  There was overlap between the positive
areas in the precipitation and SO4° weighted fields in the vicinity of Big Moose and to the north
and south. However, along the Ohio River valley and over most of Ohio and southeastern  Indiana
the  SO4" weighted field indicated above average contribution  (positive) and the  precipitation
weighted field showed below average (negative)  contribution. This discrepancy highlights the
importance of the large sources of anthropogenic SO2  in the Midwest to the wet deposition of
SO4" at Big Moose.  These two figures clearly indicate that SO4* wet deposition is not driven
solely by precipitation  amount.

      The results of the PSCF  analysis  for Gaylord, Ml  are shown in Figures 41 and 42.  Over
all wet  events (Figure 41), wind flow from  the southwest and south-southeast were linked with
relatively high SO4" deposition  events.  The area most often tied to  large events extended from
Gaylord  across  southeastern  Michigan,  Ohio,  Kentucky and  western West Virginia  and
encompassed much of the Ohio River valley.  Winds  from  the  north also  appeared  to  have
frequently resulted in above average  SO4* wet deposition.  Flow from the southwest to northwest
and the east to southeast did not seem to be systematically related to high deposition events.
                                           107

-------
Figure  39.  The bias for transport associated with large SO4* wet deposition at Big Moose. NY,
              during  the  period 1982-1985.  Positive values represent  regions more often
              associated with above average SO4* wet deposition amounts .
      There is better agreement between the SO4* weighted and precipitation weighted fields at
Gaylord than was seen at Big Moose.  As with S04", there was a greater tendency for relatively
large precipitation events when the air arrived from the south  and north (Figure  42).  The
magnitudes of the PSCF values over these areas were significantly smaller than those generated
with  SO4" weighting.  It is interesting to note that  fairly  strong winds from the southwest seem
to have been associated  with  high precipitation events more often.  As well,  winds from the
north-northwest  across Lake  Superior  and  the Straights of  Mackinac were associated with
relatively large amounts of water deposition.  Some  of these events could be  associated with cold
outbreaks during the  winter that  pick up  large amounts of  moisture over  the  relatively warm
lake  water.
                                          108

-------
Figure  40.   The bias for transport associated with large water deposition at Big Moose, NY,
              during the  period 1982-1985.  Positive values represent  regions more  often
              associated  with above average precipitation .
      Similar analyses for Raleigh, NC  were hampered by the fact that many of the trajectories
from the eastern quadrants ended abruptly over the ocean.  It is therefore difficult to assess the
relative  importance of airmasses approaching from these areas.  For all of the wet events during
1982  to 1986 the PSCF field is displayed in Figure 43.  The regions most often leading to
relatively high wet SO4* depositions are to the northwest and towards the east to southeast.  The
reason  for the  large  values  to the  east  is uncertain.   It could  be due to a  greater influx of
moisture and precipitation associated with easterly winds  and/or due to polluted airmasses  from
the northeast that move out over the ocean and then back inland south of anti-cyclones.  It could
also be  partly  due to the difficulty  discussed above.   The region above 0.05  (5%)  across
Virginia, West Virginia and Ohio is more certain.  Flow from this area correlates well with  high
S02 emission regions in the Ohio  River valley.  This area takes on more significance when
compared to the precipitation weighted PSCF  fields for Raleigh  (Figure  44).  Clearly, these
areas do not have a tendency to be associated with the large precipitation events.  These events
usually  result with flow from the east and southeast.
                                           109

-------
 Figure  41.  The bias for transport associated with large SO4" wet deposition at Gaylord, Ml,
              during the period  1982-1985.  Positive values represent  regions  more often
              associated with above average precipitation .
      To highlight the airmass source regions  that were most often associated with high SO4=
wet deposition events the PSCF fields for Big Moose, Gaylord and Raleigh were combined.  This
new -product' field is shown in  Figure 45.  With the individual  PSCF fields it is not possible to
determine where along the trajectories  the  S04= has originated. The intersection of the PSCF
fields values from these  three locations, however, is  able to indicate more  specifically which
regions were consistently associated with relatively large amounts  of SO4" wet  deposition.
Areas above 0.25 or below -0.25 are the most significant.

      In Figure 45 it can be seen that  when the airmasses originated over Ohio, northeastern
Kentucky, western Pennsylvania and northwestern West Virginia they lead to relatively large
S04" wet depositions at all three sites.  This pattern is rather good agreement with the spatial
distribution  of SO2  sources.  In contrast  airflow from the northwest and  northeast were  not
systematically leading to high depositions.  Another area of that shows up as being  important to
wet SO4" deposition  is located off the coasts of North Carolina and Virginia. This could have been
due to the availability of moisture over the  Atlantic and/or the easterly  flow on the south side of
anti-cyclones and north  side of cyclones.
                                           110

-------
Figure  42.   The bias for transport associated with water deposition at Gaylord, Ml, during the
              period 1982-1985.  Positive values  represent regions more often associated
              with above average precipitation .
      Every location in  eastern North America will experience its own unique relationship with
the  various source regions that contribute pollutants to the atmosphere.   From day to day  and
even year to year these  relationships are likely to  change. Over the  relatively long time period
of 5 years the significant short-term variations are essentially  smoothed over.  These clima-
tological Q-fields were generated and analyzed for three opposing receptors. At each of the three
locations  studied the upwind regions most often associated with precipitation  events were
different, however, throughout the five years it was possible for the flow to come from  almost
any direction regardless of location.

      The aggregate estimates of seasonal and annual deposition were compared to the observed
values during  1982-1985.  Similarly,  aggregate  Q-fields were  compared to the actual wind
flow over a longer period of time.  Before such comparisons could be made it was necessary to
decide which flow pattern should be  reproduced.  Since wet deposition is of primary interest  the
precipitation only Q-fields should be  the  main goal.  However, the event selection protocol  was
designed to identify  storms  with a high likelihood  of producing S04=  wet deposition at  multiple
locations across eastern North America.   This implies  that flows from areas most often leading
to relatively high SQ4" are desirable. Therefore, an aggregate of the wind flow from the selected
storms should  most  resemble the SO4* weighted Q-fields.
                                           111

-------
Figure   43.   The bias for transport associated with large S04" wet  deposition at Raleigh, NC,
              during the period 1982-1985.   Positive values represent regions more often
              associated with above average wet deposition.
3.5.       THE VARIATION  IN  SOURCE-RECEPTOR RELATIONSHIPS  BETWEEN
           CATEGORIES

      Over all events (wet and dry) some categories favored flow from the  north to  northwest
(categories 7. 8), others favored the south to southwest (1, 5, 11) and still others showed bras
towards the west (6,  10).  Category 9 was associated with low wind flow situations and category
10  with  higher wind speeds.   Wet  differential  probability fields for these categories  are
included in  Figures  46-54.   Figure 46  shows  the  wet  differential  field for  category  20
(unbiased set of missing  days) and illustrates the magnitudes one can expect due to random
fluctuations.  Close to the site, the differential was as high  as 30 x 10~8 km"2 but further away
the random  fluctuations resulted  in differentials less than 10"8 km"2.  Wet differential  proba-
bilities seen in Figures 47 to 54 are greater than these values and, hence, can be  significant.
                                          112

-------
Figure  44.   The bias for transport associated with large water deposition at Raleigh, NC,
              during the period 1982-1985.  Positive values represent regions  more often
              associated with above average precipitation .
      The wet differential probabilities for these eight categories were similar to those  calcu-
lated  over all events  (dry and wet) which  were summarized  above.   The  main differences
appearing in  the wet on.ly cases are outlined in Table  10.  In terms of where  the air was most
and  least  likely  to originate  just  before and during  precipitation there were  identifiable
differences between categories.   Selection from a collection of these  categories should he!p
insure that the  resulting  aggregate estimates of wet deposition  are not  biased towards a small
number of potential source regions.
                                            113

-------
Figure  45.  The ensemble bias for transport associated with large S04* wet deposition at Big
             Moose, NY; Gaylord, Ml; and Raleigh, NC, during the period 1982-1985.
             Positive values represent regions consistently associated with above average
             SO4° deposition amounts.
                                          114

-------
Figure   46.   Difference between  climatological Q field (1982-1985) and the Q field for
              category 20 (unbiased set of missing days).
                                           115

-------
Figure   47.   Difference between climatological  Q field (1982-1985)  and the Q field for
              category 1.
                                          116

-------
Figure  48.   Difference between climate-logical  Q field (1982-1985) and the Q  field for
              category  5.
                                           1 1 7

-------
Figure  49.  Difference  between climatological Q field (1982-1985) and the Q field for
             category 6.
                                          118

-------
Figure  50.   Difference between  climatological Q field (1982-1985) and the Q field for
              category 7.
                                           119

-------
Figure  51.   Difference between climatological Q field (1982-1985) and the Q field for
              category 8.
                                          120

-------
Figure  52.   Difference between climatological Q field (1982-1985) and the Q field for
              category 9.
                                           121

-------
Figure  53.   Difference between climatological Q field (1982-1985) and the Q field for
              category  10.
                                          122

-------
Figure  54.  Difference  between climatological Q field (1982-1985) and the Q field for
             category 11.
                                          123

-------
    TABLE 10. DESCRIPTION OF SOURCE-RECEPTOR RELATIONSHIPS FOR SELECTED
                            METEOROLOGICAL CATEGORIES.
CATEGORY
                            DESCRIPTION
Category  1:
Category  5:
 Category  6:
 Category 7:
 Category 8:
 Category 9:
There area that the air was more likely to originate from extended to the
southwest and south covering the Ohio River Basin and parts of Virginia and
North Carolina and did not seem to be associated with Indiana and Ohio.  The
low-probability region extends a good distance to the west.
This category is different than Category 1  in that the flow bias is stronger
and aligned toward the south of the receptor.  This category might be more
likely to be influenced by east coast sources than would category 1 The
region less likely to influence the airmass over Big Moose extends westward
to encompass Michigan.
The biases in this pattern are relatively weak with a slight enhancement in
probability of wind flow from the  west.  The  area less-likely-to-contribute
included a large area around the  receptor itself and extends to the south and
east slightly.  The wind flow was stronger for precipitation events than for
dry events.
There is a pronounced positive bias towards the northwest and a little to the
'west.  There is some likelihood.of light wind situations during wet events in
category 7.  This is indicated by a large, positive bias region immediately to
the north of Big Moose.
Category 8 has a relatively strong bias toward air flow from the north,
though the differences do  not extend far from the receptor indicating a  weak
flow pattern. Category 8 exhibited lower concentrations of SO4" than did
other predominantly summertime categories.
Category 9, which exhibited a relatively high SO4" wet deposition to
precipitation ratio, is associated with  a moderately pronounced  differential
probability field.  This field decreases fairly  fast with distance suggesting a
tow wind  flow associated with the category.  There is  bias toward transport
from  the southwest in category  9.  For precipitation events the  maximum
positive area moves further away from the receptor in the southwest.  There
may  therefore be slightly stronger winds  associated  with precipitation
events.  A negative bias towards the north is rather pronounced.
                                          124

-------
 TABLE 10 (CONT.). DESCRIPTION OF SOURCE-RECEPTOR RELATIONSHIPS FOR SELECTED
                             METEOROLOGICAL CATEGORIES.

  CATEGORY	DESCRIPTION	
Category  10:    The relatively large negative bias suggests the transport was less likely to
                 be associated with  low wind speed situations.  The weak positive bias to the
                 west suggests that  air flow to the receptor had a somewhat enhanced chance of
                 arriving from the upper midwestern United States and extreme southern
                 Ontario.
Category  11:    The differences between the overall wet probability field and the wet prob-
                 ability field associated with category 11 were small throughout eastern
                 North America.  This suggests  that the wind flow pattern common to this
                 category is similar to the wind flow associated with the typical precipitation
                 event at Big Moose.
3  6.      SELECTING EPISODES TO  ESTIMATE  SOURCE-RECEPTOR
           RELATIONSHIPS

      The differences in the wind flow between categories suggests that using them as event
selection guidelines  should result-in a better representation  of the long-term S04" weighted
source-receptor  relationships.  In the context of this  study better  implies relative  to selecting
events -at random.   For the  estimation of annual depositions category-based selection  and
weighting was  shown  to improve upon  random and  seasonal selection methods.  This  was
demonstrated by selecting and aggregating  a number of nineteen-events ensembles using a
variety of methods.  A similar approach was developed for testing how well  categories dc  in
helping to reproduce the expected source-receptor relationships.

      The episode selection criteria  outlined  for estimating deposition were followed. Nineteen
were selected from categories 1W, 3W. 5W, 6W, 7W, 8W, 9W and 11W.  The random versus
strata comparisons were not based upon 30  episodes because it was the original nineteen that
were selected to best  reproduce deposition and wind  flow. The other eleven  were included  to
cover more unusual  situations.  Random selection of  events was limited to these events in the
wet halves of the categories.  The same episodes were expected to estimate deposition at locations
 across eastern North America so each ensemble of episodes was tested in terms of its ability to
 reproduce the wet-weighted fields at Big Moose, Gayiord and Raleigh collectively.   Twenty-five
 individual ensembles  were sampled and  aggregated using  the two techniques (random and
 strata).  For the random  case it was assumed that  there was no way to weight the various
 episodes.  The only  option involved whether or not to use all days or just the  wet  days. When
 categories were employed, however, there were a number of potential approaches to weighting
 the episodes.

                                           125

-------
           Keep all days or days with reported precipitation only and combine with no weights.

           Keep ail days and weight them according to the frequency of occurrence of their
           respective categories.

     •     Employ only the days with precipitation and weight them with the amount of
           deposition on each day and category frequency.

     •     Employ only the days with precipitation and weight with expected deposition (mean
           category deposition) and frequency.
                                                         ^f
     The Q-fields against which  our selection and aggregation methods were tested were the
SO4" wet deposition weighted fields shown in Figures 36 to 38. These Q-fields served as the
base-line for all comparisons. Two of the above approaches were tested.  In the first case both
the precipitating and non-precipitating days were  retained and both the random and category-
based ensembles were aggregated  without weights. It was assumed that the  way in which the
category selections were prescribed implicitly included a  weighting  since  those  categories
known to produce  a  large percentage of the deposition across eastern North America  were
sampled according to importance  and frequency.  All days were  kept because  the occurrence  or
non-occurrence of precipitation  at the  actual receptor  is not indicative of  the  processes
occurring over the surrounding area. The movement of airmasses during the entire evolution  of
a  storm system will'effect the chemistry  of the resulting   precipitation.   The  results  of
comparing the random and category-based ensembles in this situation indicated which approach
better  represented the wind flow associated with  long-term regional deposition.  Figure 55
shows the percent  RMS errors, ranked from  smallest  to largest, for the random and category-
based ensembles.
                                                                              »j.    *^
      The G-fields  were compared  over all areas with  probabilities greater than  10   km'* and
over the Midwest.  Figure 55 indicates  that selection by category is  consistently  better than
random  selection in reproducing  the source-receptor relationships important  to regional SO4"
wet deposition.  The two methods  were compared over the Midwest because flow from this region
is frequently associated with significant  amounts of S04" wet deposition (Figure 45).  Errors
are smaller over the Midwest  for  both  selection routines  and range from  20  to  45% when
categories are used to help select the episodes.
                                           126

-------
                        — C      —R      —C-MW  —R-MW
       100
                                           I—i—\—I—I—i—I—I—I—i—I
            1  23456789 10111213141516171819202122232425

                                RANKED ENSEMBLES

Figure  55.  Percent RMS errors, ranked from smallest to largest, for the random and
            category-based ensembles using trajectories from all days in 30 episode sample.
        125
              H—I—I	1—I-
            12345678  910111213141516171819202122232425

                                  RANKED ENSEMBLES

Figure  56.   Percent RMS errors, ranked from smallest to largest, for the random and
             category-based ensembles using trajectories from  precipitating days only in 30
             episode sample.
                                      127

-------
      In the second case only days with precipitation at the receptors were included  when
generating the estimated Q-fieids (third  option listed  above).  In this approach  the comparisons
are more specific to each location and not as indicative of the situation over larger areas.  The
results are  also  more sensitive to the stochastic nature of precipitation.  Each ensemble was
selected at  random either with or without the guidance of categories and the number of wet days
available for source-receptor evaluation varied between ensembles and sites.  Weights were
applied when aggregating each ensemble to an estimated long-term Q-field.  For the  category-
based ensembles information on the expected deposition and on the frequency of occurrence of
the particular categories sampled were used determine the weights.  The randomly selected
ensembles were weighted  according to the amount of deposition occurring  with each  event
selected. The comparison between the  random and strata based approaches is shown in Figure
56.  Over  the high probability areas (>10~4 km"2)  the method of pure random  sampling  did
slightly better than the category guided selections.  Over the Midwest regions,  however,  the
outcome was just the opposite. In general, neglecting the days without precipitation resulted in
a decrease in the quality of the estimates of between 10 and 20%.  Considering the  results of
both of the cases tested, the use of the meteorological categories appears to offer improvement
when  attempting to  represent the  long-term source-receptor relationships  with a finite
number of episodes.
 3.7.       SOURCE-RECEPTOR  RELATIONSHIPS ASSOCIATED WITH THE  30
        .    EPISODES    '

      Improvement in source-receptor representation due to category-guided event selection  is
 a fundamental assumption of the aggregation technique developed during this project.   The
 previous  section  corroborated this assumption.  Most  important, in terms of  applying the
 outcome of this study to RADM. is  the ability of the actual 30 episodes to reproduce the ciima-
 tological source-receptor relationships. These episodes were recommended for RADM simula-
 tion.   They were carefully examined for  source-receptor  representativeness  at  Big  Moore,
 Gaytord and Raleigh.  Mixed-layer  trajectories were  combined to produce Q-fields for each  of
 these sites.  As in the  previous section, the  long-term wind flow patterns were approximated
 using ail of the days (non-weighted) and with only the wet days (weighted).  The resulting  fields
 were compared to the long-term wet sulfate deposition weighted Q-fields.

      Of the 30 events selected  for RADM simulation two  did not have trajectory information.
 The estimated  Q-fields discussed here were generated with  up to 28 events  instead of  30.  The
 effect of missing this information was not known but  it was  assumed that with more trajectories
 the Q-fields would be more accurate. The non-weighted Q-fields for Big Moose, Gaylord and
 Raleigh are shown in  Figures  57 to 59.  Qualitatively, they  appeared similar to  the base-line
 fields.  The probabilities decreased  away from the receptors and tended to fall-off quicker to the
 east.  Comparing  the actual field to the  estimated field for Big Moose it can  be seen that the


                                           128

-------
estimation was not as smooth. This behavior was expected since the number of days included was
 Fiaure  57   Source-receptor  analysis (Q fields) for Big Moose,  NY derived  from both
               precipitating and non-precipitating days  within the 30 selected  episodes.
                                            129

-------
Figure  58.  Source-receptor analysis (Q  fields) for Gaylord, Ml  derived  from  both
              precipitating and non-precipitating days within the 30 selected episodes.
                                           130

-------
131

-------
Figure   60.   Bias in source-receptor analysis resulting from use of 30-everrt sample instead
              of 4-year sample at Big Moose, NY. Bias is expressed as a percent of predicted
              Q-field from  the 4-year analysis.  Both  precipitating  and  non-precipitating days
            •  during the 30 selected episodes are included in the analysis.
relatively small  and it was common to all three sites.  To quantitatively compare the  actual to
the estimated fields the percent difference between the two was determined over all areas with
probabilities greater than  10"* km"2.  Throughout this area the mean percent  RMS error was
46% for Big Moose, 48% for Gaylord and 37% for Raleigh.  Contrasting these values to those in
Figure 55 we see that the actual episodes selected were reproducing the wind  flow patterns
about as well as  can be expected.

      Figure 60 shows the spatial distribution of errors for Big Moose. Positive values indicate
over-representation  and  negative  imply under-representation.   The wind  flow from  the
northeast was  significantly over-emphasized but  throughout  the  remainder of  the  region
analyzed the flow is represented to  within  -20%.  Along the Ohio River  valley the bias was
towards under-representation and was generally equal to or  less than 20%. Northeast of Big
Moose the main  source area was southwestern Quebec (Montreal).   This area will  likely be
over-emphasized by RADM simulations using the 30 days.  These sources were less important
than those  in the Midwest where the 30 day estimates were representing the flow well.
                                          132

-------
Figure  61.  Bias in source receptor analysis resulting from use of 30-event sample instead
             . of 4-year sample at Gaylord, fvH. Bias is expressed as a percent of predicted Q-
              field from the  4-year analysis.   Both  precipitating  and non-precipitating days
              during the 30 episodes are  included in the analysis.
      The same type of plot for Gaylord is included in Figure 61.  The pattern of over and under
representation was more uniform than for Big Moose.  The  southern  section of the  midwest
including the Ohio River valley was under-emphasized as was southern Ontario and  New York.
Flow from the north was given more  weight.  Throughout a  rather large area the differences
were within  20%.  Of the areas shown to be most often  related to higher than average deposi-
tions (Figure 41) the sources  along the western end of the Ohio River valley are  most likely to
be misrepresented.

      The flow resulting from  the combination of all 30 days  at Raleigh overpredicted towards
the north and east and underpredicied  to the  west and south (Figure 62). Basically, the percent
differences were within 20% over much of the continent with  the exception of the  east coast
from east of Raleigh to the north across Pennsylvania.  Over the midwest the representation was
quite good.   RADM simulations using these 30  days will likely give sources from this region
proper emphasis while those  associated with the eastern seaboard could be significantly over-
emphasized.
                                           133

-------
Figure  62.  Bias in source receptor analysis  resulting from use of 30-event  sample instead
              of 4-year sample at Raleigh, NC.  Bias is expressed as a percent of
              field from the 4-year analysis.  Both precipitating  and non-precipitating days
              during the 30 episodes are included in the analysis.
      The  long-term wind flow at each site  was also estimated for precipitation events only.
The results of these fields were fairly sensitive to the weighting scheme applied.  The estimated
fields evaluated here were generated  by  weighting each of the days with precipitation  by  the
actual amount of S04" deposition reported and by the frequency of occurrence of the category i
was selected from.  Use of weights may be redundant since they are indirectly accounted for by
the breakdown of categories sampled and in some cases neglecting weights and just averaging the
trajectories associated  with  precipitation resulted  in a  more accurate Q-field.   The  percent
difference fields generated for the wet-only estimates and are shown in Figures 63 to 65.   rhe
overall RMS errors  attributed the  the 30 days were in the same range as those shown in Figure
56.   It was 61% for Big Moose,  52% for Gaylord and  76%  for Raleigh.  As in the previous
section the estimates of the flow deteriorated  when only precipitation  days were included.  Of the
28 available episodes at Big  Moose 19 resulted  in deposition.  Thus, only 25 days were used to
                                           134

-------
produce the estimated Q-field as opposed to 84 days when non-precipitating days were included.
At Gaylord, 16 episodes had deposition and at Raleigh only 15.  This translated to 25 and 23 wet
days at Gaylord and Raleigh respectively.

      The estimated flow during precipitation events at Big Moose (Figure  63)  indicates that
sources to the north were under-represented by up to 60%.  Sources  in the  central Ohio  River
valley  were over-represented by  about the same amount.  Flows  from areas to the  west and
areas south of  Big  Moose  but north  of the Ohio River  were portrayed reasonably well  consid-
ering  the  number of days  included.   For Gaylord (Figure 64),  the  sources to  the south to
southeast were significantly over-emphasized.  Elsewhere  the differences were  basically less
than 30%.  At  Raleigh, the 23 days used to approximate the long-term  wind flow on  precipi-
tating days  were yielding large  positive  biases towards  the  north and northeast (Figure 65).
Sources  towards the  midwest  were being under-represented by anywhere from  10 to 60
percent while over the  eastern Ohio  River valley, an area shown  of major importance (Figures
43 and 45) the estimated flow was within 30%.
 figure   63.   Bias in source receptor analysis resulting from use of 30-event sample instead
              of 4-year sample at Big Moose, NY.  Bias is  expressed as a percent of predicted
              Q-field from  the 4-year analysis.  Only  days with precipitation reported at Big
              Moose are included in the analysis.
                                          135

-------
Figure  64.  Bias in source receptor analysis resulting from use of 30-event sample instead
              of 4-year sample at Gaylord, Ml. Bias is expressed as a percent of predicted Q-
              field from the 4-year analysis.  Only days with precipitation reported at  Gaylord
              are included  in the analysis.
                                            136

-------
Figure  65.  Bias in source receptor analysis resulting from use of 30-event sample  instead
              of 4-year sample at Raleigh, NC. Bias is expressed as a percent of predicted Q-
              field from the 4-year analysis.  Only days with precipitation reported  at Raleigh
              are included in the analysis.
                                            137

-------
                                    4.    SUMMARY
4.1.       ACCOMPLISHMENTS

      A method for aggregating episodic deposition  estimates of the Regional Acid  Deposition
Model (RADM) has been developed.  The  method is based upon the stratification of three-day
periods into categories  of similar wind flow at 850 mb (roughly 1500 meters  above mean sea
level).  The  stratification  has been used to  identify  which  meteorological  situations  merit
simulation by  RADM  based on their  likelihood of producing sulfur  deposition at multiple loca-
tions  across eastern North America, their frequency of occurrence, and their  seasonality.  The
aggregation technique has been simulated using four years of precipitation chemistry data to
demonstrate that uncertainties in estimates of seasonal and annual sulfur deposition using the
aggregation technique  is significantly less than those obtained from  random (non-stratified)
sampling of the data.

      The following specific tasks have been completed as part of this project:

      (1 )   An extensive data base with daily precipitation and  wet  deposition  of sulfate, nitrate
            and total H* was developed.  The data base covered the  years 1982-1985  and
            included 23 sites  in the northeastern, southern and  midwestern U.S. and  southern
            Canada.  Meteorological information to be used in subsequent analysis was also
            collected.

      (2)   Techniques of cluster analysis were  used to stratify 3-day events  over the 4-year.
            period into categories based on the meteorological data described above. Four
            independent stratification schemes were investigated: stratification by 850 mb
            wind field, stratification by 850 mb geopotential and mean surface pressure,
            stratification by 850  mb temperature advection and 500 mb  vorticity, and
            stratification by 500  mb vorticity advection  and 850 mb relative humidity.  These
            four stratification schemes were evaluated to find the combination  of meteorological
            parameters which best explained variations in sulfur wet deposition.

      ( 3 )   A set of 19 meteorological categories was compiled based on  similarity in wind flow
            at 850 mb.  These 19 categories were each subdivided into "wet" and "dry" cases,
            making 38 categories. These 38 categories were used as the basis for selection of
            individual days and derivation of aggregated estimates for annual deposition.
                                           138

-------
     ( 4 )  Accuracy and validity of the 38-category stratification was evaluated based on the
           ability to predict annual sulfate deposition  at eastern and midwestem sites from a
           finite sample of 3-day events. We compared estimates for annual deposition
           derived from a random sample of 3-day events with estimates from a cluster-based
           sample of events.  The standard deviation between estimated and actual deposition
           was lower by 25% with cluster-based selection.

     (5)  A preliminary selection of 30 individual 3-day events for in-depth study with
           RADM was made. Selection of individual days was designed to satisfy the following
           criteria:
           (a)   representation of the meteorological categories that account for the bulk of
                 dry deposition in the eastern U.S.

           (b)   Inclusion of days representing a wide variety of circulation patterns,
                 including relatively  rare  winter circulation patterns.

           (c)   Inclusion of a number of days with  very little wet deposition, to allow  for
                 simulation  of dry deposition processes.

           (d)   Selection of individual  events within  each category by a random process,
                 rather than prescribing specific events.

           (e)   Accurate representation of total annual deposition at all  northeastern and
                 midwestem sites from the  ensemble of  individual events.

      The selected ensemble allows for  accurate estimation of summer, winter, and total annual
deposition at all sites in the northeast and midwest.
4.2.       RECOMMENDATIONS FOR FUTURE WORK

4.2.1.     Evaluation  of  the 30  Episodes and  the Meteorological  Categories

      Meteorological categorization was employed to help select episodes for RADM  simulation
with the purpose of trying to represent the dominant wind flow situations.  Categorization based
on the 850 mb wind flow was used because it was assumed that the ensuing categories would best
segregate general  meteorological situations and the deposition events.  These assumptions were
tested. The categories were shown to explain a good degree of the variance in wet deposition (S5
% confidence level) and in terms of their  mean 850 mb wind flow  they were seen to differ
systematically from each other and resemble common meteorological patterns.  In the context of
offering improved aggregate estimates the categorization by 850 mb wind flow was  shown to be

                                          139

-------
warranted.  Nonetheless, within categories there exist large variations from the mean flow and
the mean deposition pattern.  The following tasks investigate the magnitude and causes of those
variations.

4.2.1.1.     Source-Receptor   Relationships

      The hypothesis that the categories would differ chemically (at any given site) was based
upon the assumption that the categories represented systematically different wind flow patterns
and hence  would differ with  respect to source-receptor relationships.  This concept is very
important to the  overall RADM project.  The collection of events selected from the various
categories should adequately represent the potential influence of all the major source regions
effecting eastern North America.   In  addition, the frequenc/'with which  a particular  source
region impacts upon any given receptor must be as is  in nature.  If the overall source-receptor
behavior represented by the collection of events is biased then the aggregate of RADM runs
resulting from various emission change scenarios will also be biased.

      The source-receptor relationships  associated with the categories and with  the 30  days
selected for simulation were tested during the past year of research.  Results indicated that  for
Big Moose,  NY;  Raleigh, NC; and Gaylord, MI the categories did favor different upwind regions
and the 30  three-day episodes also differed in terms of  where the  air was most likely to  have
originated.  How  well the wind flow over the 30 days represented -the expected wind flow was
also tested,  however, results of these comparisons are still  being investigated.

      During  the  next  year a new method for  evaluating  the various approaches (30 day
aggregate vs. actual vs. random 30 days) will be applied to  aid in this  aspect of the source-
receptor testing.  This new method will be based on a slightly modified version of the "threat
score" (Anthes 1983) and should supply a more meaningful gauge  for assessing the effective-
ness of the selected days.

      In terms of the source-receptor relationships and whether the specific source regions are
given proper weight, the categories and the 30 days have yet to be evaluated. This aspect, which
will require a more detailed look at the meteorology associated with the episodes and categories,
will be explored in the upcoming year.  In studying the 30 days the approach will be to examine
the winter and summer situations separately.  The flow patterns are more distinct  in the winter
and thus the source receptor relationships for the associated categories should be more clear-
cut.  As a result, it is hypothesized that the wind flow for the days selected to represent this
time of year should do a better job at  reproducing the  overall pattern (as compared to the
reproduction of the summer half).  If this is shown to be the case then it will only be necessary
to concentrate on improving the categorization and selection  of warm season events.

4.2.1.2.     Chemical  and  Meteorological  Behavior of the  Categories

      In general, the  categories  were  shown  to differ  chemically  and  meteorologically.

                                           140

-------
However, as it was  pointed out earlier there remains a  large amount of variability.  While the
current methodology  has been shown to improve upon a random selection of episodes, the ability
of our specific events to represent the  expected range of situations is still clouded by the within
category variability.   This variability  is great  enough  to  allow  some overlap.   Events  from
different  categories can be more similar than different and also not resemble the mean category
pattern at all.  These effects can lead to selecting events which do  not  possess the desired
pollutant  transport  paths  and  the  under  or  over representation of certain  meteorological
situations.

      The within category variability  is most notable in the summer when just a few categories
are predominant.  There are clearly other mechanisms previously not accounted for that govern
the precipitation and pollutant deposition behavior.  To understand and include these effects the
meteorology of these situations needs  to be examined in more detail.  The  goal will be to better
understand what differentiates the events in these more ubiquitous categories.  This knowledge
could be used to further stratify some  categories leading to better weighting factors for  some of
our selected days. It will also identify any key situations that are not presently being modeled.

      In  the process of testing the nineteen categories for their  ability to  explain the variation
in wet deposition a number of regression models were developed.  These models  are capable of
relating  depositions  and concentrations to specific parameters in  quantitative terms.   The
importance of a variety of  meteorological factors, in terms of their influence on  precipitation
chemistry, can be  quantitatively assessed in this way. This approach can be used in conjunction
with additional stratification to uncover  the other dominant mechanisms controlling deposition
and to improve the  aggregate estimates.  Using the results of the regression models alternate
aggregation schemes can also be explored and techniques that can better account for the effects of
meteorological  variability can be incorporated.  As discussed  above, specific knowledge of the
source-receptor relationships associated with the categories and the 30 events have  not  been
determined.   This information  is especially lacking for the  light wind categories which are
associated with high deposition to precipitation ratios and relatively large numbers of events.
The regression studies and  sub-stratification  along with  careful  meteorological analysis will
add more to our knowledge of the source receptor relationships under these types  of situations.


4.2.2.       Application  of  aggregation  to other  atmospheric  phenomena

4.2.2.1.     Dry  Deposition

      A  method for  aggregating dry deposition estimates  from the RADM simulations has not
been developed.  This problem is difficult to tackle because the amount of suitable dry deposition
data is very limited.   In addition, the mechanisms associated with dry deposition processes are
not completely understood.  A variety  of factors such as boundary layer turbulence and surface
characteristics are known to influence  the amount of pollutants that are dry deposited.  In  order
to apply  similar aggregation methods to dry deposition it  will be necessary  to study  the behavior
                                           141

-------
of the various  contributing mechanisms and the ambient levels of SO4", SO2 and NOX with
respect to the meteorological categories.

      The data gathered during the RADM intensive  evaluation period (Summer, 1988)  can
provide some of the information needed to address the dry deposition issue.  With  a better
understanding of the within versus between category variation in ambient concentrations it will
be possible to evaluate the suitability of  estimating the long-term patterns of dry deposition
using a small subset of episodes.

      In  the  present selection six of thirty episodes were selected to enhance dry deposition
representation.  The only criteria used in selecting these episodes was that they be associated
with  a small amount of wet deposition across the Northeast as aT whole (< 50 mgm"2).  However,
with  no dry deposition or ambient concentration data it has been very  difficult to  evaluate the
days chosen. The information gained through the ideas mentioned above will help test these days
and  also indicate whether or not more episodes need to be  selected to  estimate dry deposition
patterns.

4.2.2.2.     Visibility

      Visibility  in the eastern United States  is known to be  largely  governed by the concentra-
tions of SO4*.  Therefore, the methods developed thus far could be applied  to visibility.   Specific
aspects of the problem that need to be addressed include:

      •   '  Identification of the sources of visibility data across  eastern North America;

      •     Examination of methods for reducing  meteorological variability in visibility  data
            due to humidity;

      •     Examination of the behavior of visibility associated with the different meteorolog-
            ical categories;

      •     Evaluate need for additional meteorological categories based on  differences in
            visibility;

      •     Identification of some of the conditions (categories) that  are conducive to good poor
            visibilities. Study of the specific relationships existing between these conditions
            and  visibility.

      •     Test performance  of the current 30 days in predicting visibility behavior  over
            seasonal and annual time periods.

      •     Selection of specific episodes useful  in predicting long-term visibility behavior
            through an aggregate technique. This could include:
                                            142

-------
           —   Episodes to supplement the existing 30 days, or

           —   An independent set of episodes,

4.2.3.    Quantification  of Aggregation  Sensitivity

      There are a number of factors that influence the aggregate  estimates.  In particular, the
variability of  deposition-to-precipitation  ratios within  categories,  between categories and
between years can affect the estimate.  The variation  in the frequency and amount of precipita-
tion associated with  the  categories and variation in the frequency of occurrence of the  actual
categories also affects the estimates. The sensitivity of the aggregate  estimates to these types of
factors needs to  be quantified and the factors that have the most impact on the results need to be
highlighted.   This will  also enable  us to determine what time scale the aggregation  scheme
handles  best  (i.e. one year averages, four year averages)  and the amount of inter-annual
variability associated with the average  estimates.
                                           143

-------
                                  5.     REFERENCES


Anderberg, M.R.  1973. Cluster Analysis for Applications.  Academic Press, New York, 359 p.

Beale, E. M. L  1969.  Euclidean cluster analysis.  Bull. Intrnl.  Statist. Inst., 43, p. 92-
     94.Berge, E. 1988.  Time-trends of sulfate and nitrate in precipitation in Norway
     (1972-1982).   Atmospheric  Environment, 22:  333-338.

Buishand TA, G.T. Kempen, A.J. Frantzen, H.F.R. Reignders and A.J. Van Den Eshof. 1988.
     Trend and seasonal variation of precipitation chemistry in the Netherlands.  Atmospheric
     Environment,  22:339-348.

Calinski, T. and Harabasz, J.  1974. A dendral method for cluster analysis.  Communications in
     Statistics., 3,  p. 1-27.

Cass, G.R., 1981. Suifate air quality control design.  Atmospheric  Environment, 15, 1227-
     1249.

Chang. J. S.. Brost. R. A.. Isaksen.  I. S. A., Madronich, S., Middleton,  P.. Stockwell, W. R., and
     Walcek, C. J.  1987.  A three-dimensional Eulerian acid deposition model:  physical
     concepts and formulation.  J. Geophys.  Res.,  92, 14681-14700.

Davies.  T.D., G.  Farmer and R.J. Barthelmie.  1990. Use of simple daily atmospheric circulation
     types for the interpretation of precipitation  composition  at a site  (Eskdalemuir) in
     Scotalnd,  1978-1984.  Atmospheric  Environment, 24A: 63-72.

Draxler,  R.R., 1987,  Sensitivity of a trajectory  model  to the spatial and temporal resolution  of
     the meteorological data during CAPTEX.  J. Climate Appl. Meteor., 26. 1577-1588.

Endlich, R.M., B.P. Eynon, RJ. Ferek, A.D. Valdes and C. Maxwell. 1986.  Statistical Analysis oT
     Precipitation Chemistry Measurements over  the Eastern United  States. UAPSP 112,
     Utility Acid  Precipitation Study Program,  Washington,  D.C., Available from UAPSP  Report
     Center, P.O. Box 599. Damascus, MD, 20872.

Fernau,  M.E. 1988.  Use of Cluster Analysis to Define Periods of Similar Meteorology and
     Precipitation Chemistry in Eastern  North  America.  Ph.D. Thesis, University of Michigan,
     University Microfilms, Ann Arbor,  Ml, pp.  330,  1988.

Finkelstein, P.L and S.K. Seilkop.  1986.   Changing patterns in acid rain.  Extended Abstracts,
     5th Joint Conference on  Applications of Air Pollution Meteorology. Nov. 18-21, 1986
     Amer. Meteor. Society, Boston,  MA, p. 79-82.

Freund, R. J., R. C. Littell and P. C. Spector.  1986. S-4S System for Linear Models SAS
     Institute Inc., Gary, NC.
                                          144

-------
Heffter, J. L  1980.  Air Resources Laboratories Atmospheric Transport and Dispersion Model
     (ARL-ATAD).  NOAA Tech. Memo. ERL ARL-81; Rockville, MD, 1980; p. 17.

Hidy, G.M., Mueller, P.K. and E.Y. Tong, 1978,  Spatial and temporal distributions of airborne
     sulfate in parts  of the U.S.  Atmospheric  Environment, 12: 735-752.

Hooper, R. P. and N. E. Peters.  1989.  The applicability of principal components analysis for
     determining sources of wet deposition,  presented at the Joint International Association of
     Hydrological Sciences/American Geophysical Union Symposium  on Atmospheric
     Deposition, Baltimore,  MD., May 10-11, 1989.

Johnson, R. A. and D.  W. Wichern.  1982. Applied Multivariate Statistical Analysis,  Prentice-
     Hall,  Inc., Englewood Cliffs, NY.

Kahl J.D.  and Samson  P.J.  1986: Uncertainty in trajectory calculations due to low resolution
     meteorological  data, J.  Climate Appl.  Meteor. 25: 1816-1831.

Kahl J.D.  and Samson  P.J.  1988: Trajectory sensitivity to rawinsonde data resoltuion.
     Atmospheric Environment, 22:  1291-1299.

Kalkstein LS. and  P. Corrigan, 1986: A synoptic climatological approach for geographic
     analusis: Assessment of sulfur dioxide concentrations. Anal. Assoc. Amer. Geographers,
     76:   381-195.

Kalkstein L.S., G. Tan and J.A. Skindlov.  1987.  An evaluation of three clustering procedures for
     use  in synoptic climatological classification.  J. Climate Appl.  Meteor., 26: 717-730.

Keeler G.J., and  P.J.  Samson,  1989. Spatial  Representativeness of trace element ratios.
     Environ.  Sci. and Techno!., 23:  1358-1364.

Lamb R.G. and J.H.  Seinfeld, 1973, Mathematical modeling  of urban air pollution - general
     theory,  Environ.  Sci. and Technol.,  7: 253.

MAP3S/RAINE Research Community.  1982. The MAP3S/RAINE precipitation chemistry
     network: statistical  overview for the period  1976-1980.   Atmospheric Environment,
     16:1603-1632.

Moody J.L. and Galloway J.N. 1988: Quantifying the relationship between  atmospheric transport
     and  the chemical composition  of precipitation on Bermuda,  Te//us,  40B:  446-462.

Moody J.L. and Samson P.J. 1986: The Influence of Meteorological Factors on Precipitation
     Chemistry. UAPSP 111, Contract U101-4. University  of Michigan  final report for Utility
     Acid Precipitation Study Program, Washington D.C. and Electric Power Research Institute,
     Palo Alto, CA. Available from UAPSP Report Center, P.O. Box 599, Damascuc, MD, 2Q872.

Moody, J.L. and  P.J. Samson. 1989.  The influence of atmospheric transport on precipitation
     chemistry at two sites in  the midwestern United States.  Atmospheric Environment, 23:
     21 17-2132.
                                          145

-------
Mueller. P. K. and M. A. Allan. 1987.  Utility Acid Precipitation Study Program, Third
     Summary Report.  UAPSP 114, Utility Acid Precipitation Study Program, Washington,
     D.C., Available from UAPSP Report Center, P.O. Box 599, Damascus, MD, 20872.

Samson. P.J.. and M.J. Small. 1984. Atmospheric trajectory models for diagnosing the sources
     of acid  precipitation, in Modeling of Total Acid Precipitation Impacts, J. L. Schnoor, ed..
     Butterworth Publishers,  Boston,  1984.

Samson, PJ.  1980. Trajectory analysis of summertime sulfate concentrations in the north-
     eastern United States. J. Appl. Meteor.,19: 1382-1394.

Samson PJ.  1989a. Aggregation Pollutant Deposition Episodes into Seasonal and Annual
     Estimates. EPA Cooperative  Agreement CR-814854-01. Prepared for the United States
     Environmental  Protection Agency, Meteorology and Assessment Division,  Research
     Triangle Park,  NC.  January, 1989.

Stein, M. S.  and X Niu.  1989. "Analysis of Precipitation and Deposition Data", personnel
     communication.

Summers Peter W.  1987: Empirical source-receptor relationships in eastern  Canada deter-
     mined  from monitoring data and airmass trajectory climatologies: In EMEP - Workshop  in
     data  analysis and presentation, Cologne, FRG 15-17 June 1987, EMEP/CCC-Repon 7 87.

The  Sulfate  Regional Experiment: Report of Findings. EA-1901, Volume 1. Electric Power
     Research Institute, Palo Alto, CA, March, 1983.

Trenberth, K.E. and  J.G. Olson. 1988.  Evaluation of NMC Global Analyses: 1979-1987. NCAR
     Technical Note NCAR/TN-299+STR, National Center for Atmospheric Research, Bouider,
     CO.

Vukovich. P.M. and J. Fishman: 1986.  The climatology of summertime ozone  and sulfur dioxide
     (1977-1981).  Atmospheric  Environment, 20:

Ward,  J. H.  1963.  Hierarchical grouping to optimize an objective function. J. Am. Statist.
     ASSOC.,58: 236-244.

Wilson J.W., Mohnen, V.A. and J.A. Kadlecek, 1982, Wet deposition variability as observed by
     MAP3S. Atmospheric  Environment, IB: 1667-1676.

Zemba, S.G., D. Golomb and J.A. Fay.  1988  Wet sulfate and nitrate deposition patterns in
     eastern North America, Atmospheric Environment, 22:  2751-2761.

Zeng Y. and P.K. Hopke. 1989. A study of the sources of acid precipitation in Ontario, Canada,
     Atmospheric  Environment, 23:  1499-1509.
                                          146

-------
         APPENDIX A
Mean 850 mb Wind Flow Patterns
     for the  19  Categories

-------
EASTERN NORTH AUCRICAN MEAN  asows w»e  rr.o - a
                   TRAJECTORY O.USTER  1       J.ex. N-
     100      93      SO     33     10
                                                70     Si
  EASTERN NORTH  AUE.R;CAN  UEAN asous WM: FXLD -  19 CLUST
                     TRAJECTORY C.USTES  1        J.sr.. N-
      100     9S      gc     IS     10     73     70     63
                             A-2

-------
EASTERN  NORTH AMERICAN WEAN  JJOMB WIND
                 TRAJECTORY CLUSTER  1
                                       » O.USTERS
                                       3.6S. N- O
100     9i     90     ai     ao
         Ai  -  X«»n  850  sb  vind  flov
 for  t-i«  thr««  days  la Ci-star  1.    The
 clusters were  exirac'ei Ircnv a see of ;cr.aecutive
 three-day periods  during 1919, 1931 and  1933.   N
 ccrreSFcr.is ta t.w.e r.un^er of bccurrer.ces  of  clu.ite:
 1  durir.g that  period.   The  percer.t cf ti.-e  (out ci
 ail cases) the cluster occurred is also  showr.  in
 tlJ« upper right.   Wind vectors represent  six hours
 of motion at reported wind speed.
                       A-3

-------
EASTERN NORTH  AMERICAN MEAN  SiOUB WWC HEl^  - 4  CLUSTERS
                   TRAJECTORY  CLUSTER  2        JJJ. M-
    100     as      ao      as      *o
                                                70     85
 EASTES.H NOR'TH  AMERICAN WEAN  asoue WIND rE'.2  - a  CL-STESS
                    TRAJECTORY  CLUSTE.R  2        3.;-.  N- 12

-------
EASTERN NORTH AUERICAN MEAN sioue  WNO FELO  -  » CLUSTCRS
                TRAJECTORY O.USTEH  2      3,27.. W- 12
    100
                                    71
    rigur«  X2  -  M«an   850 =b  wind  flow  p«tr«r=a
    for tJs« thr««  days  in  Cluster 2 .   T.".e
    clusters were  ex^rac.cd frcr. a  se*  o£ ccr.jecutive
    t.s.ree-day periods during 1979,  1981 and 1983.  N
    corresponds to the nunoer  of occurrences of cluster
    2 during that  period.   The percent  of tine (out of
    all cases)  the cluster  occurred is  also shown in
    the upper right.  Wind  vectors  represent six hours
    of motion at reported wir.d speed.
                          A-5

-------
EASTERN NORTH  AMERICAN y£AN 850U8 WNC  FCLD - « CLUSTERS
                  TRAJECTORY CLUSTER   317S.  N-  i
     co     si      so      as     to                    a
EASTERN NORTH AMCRJCAN WEAN  sioue WIND F
                  TRAJECTORY  CLUSTER  2
                                                  is CLUSTE.RS
                                                 1.7-.  N-  6
      100
                                                 70     e:
                             A-6

-------
EASTERN NORTH AMERICAN MEAN tioua  wsfO ndfl  - » CLUSTERS
                 TRAJECTORY C1.USTER  3       t.7X.  N-  (
            X3  -  >t«»=  850  nis  wind  flow  p»ttar=s
    for  th«  thr««  
-------
EASTERN NORTH  AMERICAN  MEAN asoua wo FCLD  - »  CUSTESS
                   TRAJECTORY CLUSTER  «       2.2X. H-   a
                           13      iO      75      70     iS
 EASTERN  NORTH AUESICAN MEAN 850M8 WIND FiELD  - IS  CLUS*E?.S
                    TRAJECTORY CLUSTER  4       2.2". N-   8
      100     »3
                             A-8

-------
CASTESN NORTH AMERICAN MEAN aS
-------
EASTERN NORTH AMERICAN  MEAN (sous WMO FCLD -  » CLUSTERS
                  TRAJECTORY CLUSTER  3       i.9X. M-  25
 EASTERN NORTH AMERICAN WEAN aiouB WIND FIELD  - »  C.USTERS
                    TRAJECTORT CLUSTER  5       6.95. N-  25

      100     91
                             A-10

-------
EASTERN NORTH AMERICAN MEAN asous  WIND
                TRAJECTORY CLUSTER  5
 It CLUSTERS
«.«. M- 25
                                                      -123
         •3     90     85     «0     75    70     t-
          A3  -  M««n  850  mb  wind  flow  patttms
   for  th«  thr««  days  In  Cluster  5.   The
   clusters  were  extracted from a  set  of  consecutive
   three-day periods during 1979,  1981 and  1933.   N
   corresponds  to the nusvfcer  of occurrence; of cluster
   5  during  that  period.  The percent  of  tir.e (out of
   all  cases)  the cluster occurred is  also  showr.  in
   the  upper right.  Wi.-.d vectors  represent six hcur:
   of =otion at  reported wi.-.d speed.
                        A-ll

-------
EASTERN NORTH  AMERICAN utAN asoua w*c  FCLD - » CLUSTERS
                  TRAJECTORY CLUSTER   «       §.OX. N- 2»
    100     13      90             tO      7S      70      «1
 EASTERN NORTH  AMERICAN MEAN  asoua wwo  FELB  - «  c.us*ERS
                   TRAJECTORY  CLUSTER   6      j OI M_  29
     100     92
                             A-12

-------
EASTERN NORTH AMERICAN MEAN asous w»c PELD -  a
                TRAJECTORY CLUSTER  I      g.ox. M- 2S
   100
           AS  -  Haan  850  mb  »ind  flow  pattama
   for  tha  thraa  days  in Cluster  £.   T.w.e
   clusters were extracted £rom a set of ccnaecutive
   three-day periods during 1919,  1981 and 1983.-  N
   corres?cr.dj to the nirrier of occurre.-.ces of cluster
   6 during that period.   The  percer.t of tiae (out cf
   all cases) the cluster occurred is also shown i.-.
   the upper right.  wind vectors  represent six hours
   of motion at reported  wind  speed.
                         A-13

-------
EASTERN MOUTH  AUERICAN KAN »sous w»e mo  - »  CLUSTERS
                   TRAJECTORY CLUSTER  7       g.gx.  N- 32
                                         75      70      8S
  EASTERN NORTH AMERICAN MEAN  850U3  WIND  FIELD - IS CLUSTERS
                    TRAJECTORY C-'JSTER   7       g.8S.  N- 32
                                                               * 10
      100     »;     to     13      *o  '           70      e:
                             A-14

-------
EASTERN NORTH AMERICAN WEAN esous w»c fELD - » D.USTEKS
                TRAJECTORY CLUSTER  7      8.85.  N-  32
    100
          81     90
           AT  -  Maaa  850  ab wind  flow  ptt-«rti«
   for  ti«  tir««  d*y» In  Clustaz  7.   T^-.e
   clusters  were  exrracte'i frcn: a  set of consecutive
   three-day periods during 1979,  1931 ar.d 1932.   N
   corresponds  to the r.u-ii-er  of occurrences of cluster
   7 during  that  period.  The  percent of tiM (out of
   all cases) the cluster occurred is also shown in
   the upper right.  Wind vectors  represent six hours
   of motion at  reported wind  sp-eed.
                           A-15

-------
EASTERN NORTH AMERICAN MEAN  SSOUB WINS mo  - T»  CLUSTERS
                  TRAJECTORY  CLUSTER  8        s.ox. N- w
    100     K     10
                                 80     75
                                                      91
 EASTERN NORTH AUIR.-CAN MEAN  SSOUB WIND FIELD  - is  c^s'^s
                   TRAJECTORY  CLUSTER  8        s.os.  N-  18
     iOO     »S
                           •3      10     75     70     15
                             A-16

-------
EASTERN NORTH AFRICAN WEAN JSOUB WIND  rtto - « CLUSTERS
                TRAJECTORY CLUSTER   8      5.ox. M-
   100    »1     90     IS     10     7!     70
  ricrur« AS  -  Main  850  mb  wind  flow  patterns
  for ti«  t_hr««  days  In  Cluster  8 .    The
  clusters  were  exrrac-ed Irom a set  of  consec-.tive
  three-day periods  during 1979, 1931 ar.d  19S3.   N
  corresponds  to the p.usrer of occurrences  of  cluster
  8  during  that  period.   The percent  of  tijr.e (out cf
  •11 cases) the cluster occurred is  also  shown  i-
  the upper right.   Wind vectors represent  six hours
  of motion at  reported wind speed.
                        A-17

-------
EASTERN  NORTH AMERICAN MEAN asous w»« rtiB  - i»  CLUSTERS
                   TRAJECTORY CLUSTER  »        ».«.  M- 33
     100     »S
 EASTERN NORTH AUERXAN U£AN  850W8  W»C  FIELD - IS CLL'STESS

                    TRAJECTOSr  CLuSTER   9       9.:-,  N- 33
      100      »5
                                   »0     7S     70     6i
                              A-18

-------
EASTERN NORTH AMERICAN MEAN lious WIND ncLD - n  CLUSTERS
                 TRAXCTORY CLUSTER  i       « 1X. N- 33
   100     91     90     «J     13     75     70
   Fig^r*  A9  - M««n  8£0  ab  wind  flew  patterns
   2or th« thr««  diva ii  Cluat»r  9.   The
   cluster: were extracted free, a set ef  consecutive
   three-day periods during 1979, 1981  and  1983.   N
   corresponds to the nu.-r.reT of occurrences  of  cluster
   9  during that period.  T.-.e percent cf  tirr.e (cut cf
   • 11 cases)  the cluster occurred  is ilsc  showr.  in
   the upper right.  Wind vectors represent  six hcurs
   of inction at repcrtec wir.i speed.
                        A-19

-------
EASTERN NORTH AyERCAN HAN eSOuS WM) FTLD -  « CLUSTERS
                   TRAJECTORY O.USTE* X5       j.jr.. N-   a
    ioc
 EASTERN NORTH AMERICAN UEAN  850U6 WIND r£L2 -  IS CLUSTERS
                    TRAJECTO*- CLUSTER 10       2.2r..  N-  S

                                                               1 2=
                             A-20

-------
EASTERN NORTH AMERICAN UEAN esoua WIND FIELD  - a CLUSTERS
                 TRAJECTORY CLUSTER TO      j.2r..  N-   8
   IOC
                90
   rig-ur«  JUO  -  Maaa  850  nb  wind  «io»  p»ttar=3
   •or  th«  thr««  
-------
EASTERN NORTH AMERICAN  MEAN asoua wno  fits - 19 CLUSTERS
                   TRAJECTORY auSTER n       7.47.. N. 57
   ico     es      gc     i:     ic
                                               o     a:
EASTERN NORTH  AMERICAN  WEAN SSOUB WIND FIE.D -
                   TRAJECTCRr C.USTER  n        j t-  N_ 2-
                           A-22

-------
EASTESN NORTH AMERICAN uf AN 550U8  WIND HELD - is CLUSTERS
                 TRAJECTORY CLUSTER  11      7.*S.  N-  27
    rig-ur«  *•"  ~  H*»n  85°  °*>  wind  flo«  patterns
    for  ti«  tixraa  days  ia  Cluster  11.   The
    clusters vere  ext.racrei frcr. a jet  cf  ssr.sec-tive
    three-day pericis  durir.g 19"5, 155: ar.i 19 = 2.   N
    ccrresponds t=  the nuster cf occurrer.ces  of  cluster
    11 during that  pericc.   The percent of tinie  (cut of
    all cases) the  cluster occurred is  also shown  in
    the upper right.   Wir.d vectors represe.-.t  six hours
    of motion at  reported wir.i spesi.
                           A-23

-------
EASTERN NORTH AMERICAN  MEAN asous  WIND  rzio -  n CLUSTERS
                   TRAJECTORY CLUSTER 12         4.IS. H-  15
    100      93      1C      6S     10      73      70     l-
 £>ST£SN NORTH  AMERICAN  UCAN 850U8 WINS  FX^D - IS C»US*E»S
                    TRAJECTORr CLUSTER  12        4 •£. s. 1S
     100
                                                 70      e:
                             A-24

-------
EASTCRN NORTH AFRICAN  WEAN aicuB WIND FCJ - w CLUSTERS
                TRAJECTORY CLUSTCR  12       4 m  N.
                                               K
          A12  -  M«.n  850  ab  »ind  flow
  for  the  three  d*ya  ia Cluster  12.   T.le
  Clusters  were  extracted  from a set of c=r.ae;utive
  three-day period  during 19"9, 1951 a--.d 1983.  V
  corresponds  to the nuri«r of occurrences of cluster
  12 during that period.   The percent of tine  (out of
  all  cases)  the cluster occurred is also shown in
  the  upper right.   wind vectors represent six hours
  of motion at reported wind speed.
                         A-25

-------
EASTERN NORTH AUCRICAN MEAN  asous  WIND rcj  - »  CLUSTERS
                  TRAJECTORY CXUSTER  13       2.2X. N-   8
   100
 EASTERN NORTH AMERICAN MEAN esous WINO ncj -  is C^USTE=S
                   TRAJECTO=r CLUSTER  U       2.;-.  N-  a
                                                             - —
     100
                   •°     is     «o     ?:     70     t:
                           A-26

-------
EASTERN NOflTH AMERICAN MEAN JSOuS WHO fTLD - W CLUSTERS
                 TRAJECTORY CLUSTER  U      J.2S.  N-  1
   100
          it      BO
                              1C     7S     70    65
            i.13  -  Main  650 ai  vind  flow
   .for  th«  thr««  days  in  Cluster  13.
    clustera were  er^racted frc=
                                   a  set  of  C3
    three-day periois during 1979,  1931  a.-.d 1953.   N
    corresponds  to  the ni^rier of  occurrer.cei of  cluster
    13 during that  period.  The percent  of time  (out of
    •11 cases) the  cluster occurred is  also shown  in
    the upper right.   Kind vectors  represent six hcurs
    of notion at  reported wind sp«ed.
                         A-27

-------
EASTERN NORTH  AUCRJCAN WEAN tsoua WINO  rtLD - » CLUSTERS
                   TRAJECTORY CLUSTER H        V4X N.   5
   too     »:
                                              ro     a:
EASTERN NORTH  AMCRCAS  MEAN esous WINO FXLO - is CIUSTE^S
                   TRAJECTORY CLUSTER H       1-4-  N,   «
    100     91
                           A-28

-------
E.ASTESN  NORTH AULRICAN UCAN 850U8 WIND FIELD - 1J  CLUSTERS
                 TRAJtCTORY CLUSTER 14       t*-. M.  S
   ICO    IS     BO     12     10     73     70     IS
   riyur«  A14  - >U»n  BIO  ais  wind  fiov p»tt«rns
   for th« tfcz««  days  in  Cluster  1«.   The
   clusters were extracted frori a set of  consecutive
   three-day periods during 1579, 1981  and 1983.   N
   corresponds to the nusier of occurrences of  cluster
   14  during that period.  The percent  of ti-Tie  (out of
   all cases)  the cluster occurred is also shown  in
   the upper right.   Wind vectors represent six hours
   of  motion at reported wind speed.
                          A-29

-------
EASTERN NORTH  AMERICAN UEAN .«soue w>c FCLD
                   TRAJCCTORT C.USTES  B
                                                  CLUSTERS
                                                   N.   3
    100
EASTERN NORTH  AMERICAN  U£AN BiOug W5O  FIELD - ?9 CLUSTERS
                   TRAJECTORY O.USTLR 13        ..*-. N.   «
   100     IS      tO     13     tO
                           A-30

-------
EASTERN NORTH
                   MEAN »soua wwo  ncto - a CLUSTERS
               TRAJECTORY CLUSTER 13       ^z. N-  5
          A15  -  Haan  850  nb wind  flow p«tt«r=a
  for ti« tir«« itys in  Cluster  15 .   T.w.e
  clusters were ext.racte£  free a  se*  ci  rcr.iecu;iv«
  three-day periida duri.-.g  1979,  19B1  and 1963.   N
  correspor.da to the r.uz-ier  cf occurrence: cf cluster
.  15  during that period.   The  percent  of tirr.e (cut cf
  all cases)  the cluster occurred is  also shewn  in
  the upper right.  Wind vectors  represent six hours
  of  notion at reported wind speed.
                        A-31

-------
EASTERN NORTH AMERICAN MEAN tioue  w»c  FCLD - w CLUSTERS
                  TRAJECTORY O.USTER W        17* N.
    103     f.
EASTERN NORTH AMERICAN UEAN  SSOMB  WIND
                  TRAJECTORY CLUSTER  is
                                             - is  C.-JSTERS
                                                - N_  6
            /            ^    y  vx
     100     •!     tO            |0     71
                           A-32

-------
EASTERN NORTH AMERICAN MEAN 850UB WIND FELD - 1» CLUSTERS

                 TRAJECTORY C1.USTEK  tt       \7~, N-  (
    loo     is     sc
                              10    7S     70     61
   rig-ar«  A16  -  Haan  850  =b  wind  flow  p*--en:3
   for  »^»  rhr««  days i-  Cl--s-«r  16.    The
   clusters  were extracted  lrcr> a set of ccr.secutive
   three-<±ay p«rioda during  1979,  1981 ar.d 1982.  N
   correspcr.ds to the r.urier of occurrences of cluster
   16 durir-.g that period.   The  perce.-.t of fine  (cut cf
   all cases)  the cluster occurred is also showr. ir.
   the vrpp«r right.  Wind vectors represent six hour:
   of seticn at reported wind sp«ed.
                           A-33

-------
EASTERN NORTH  AMERICAN MEAN  ajous  WIND
                   TRAJECTORY CLUSTER 17
                                                it CLUSTERS
                                                j.j-.  s-  u
   100      e:
EASTERN NORTH  AMERICAN  MEAN esoua  WIND  FIELO -  is CLUSTERS
                   TRAJECTOR-r  CLUSTER 17        J.sr.. N.  '.:
    100

-------
EASTERN NORTH AMERICAN MEAN 850MB  WIND TELO - « CLUSTERS
                 TRAJECTORY CLUSTER  17       J.ss. N- O
    ioo     8:
            XI7  -  K..=  «£0  mb  wind  flow  p«««»a
    for  th«  thr.«  diys  in Cluat.r  17.
    eluacera were  ex-.r3c-.ed frcrr. a «'. of "r.secu--^
    -ree-d^y period  during 1979, 1981 ar.i  1983    >
    correapcnd, to the nusfcer o£ occurrence^  -  c.«.  .
    17 during t.-.« period.  The percent of t^-.e  (ou. of
    .11  ca^es)  the cluster occurred  is  .lao  shcwr. ir.
    the  upper  right.  Hind vectors ze?re3er.t 3-x ..cur:
    of action  .t  reported wind  speed.
                           A-35

-------
EASTERN NORTH AUtSiCAN UEAN eSOuB  WIND  FIELD -  IS CLUSTERS
                   TRAJECTORY  CLUSTER 18         2.6-.  N- U
    ICO     s:
   STERN  NCSTH AUE=:CAM MEAN  aicua WINS FXLD  - '.s  c^.S'E'.s
                     TSAJECTOPr CL'JSTE8  'S        J.6". N-  O
      :co  .
                              A-36

-------
EASTERN NORTH AMERICAN WEAN JSOUB WIND FIELD - 11  CLUSTERS
                 TRAJECTORY CLUSTER 18       3.S-. N-  U
    ICC
           SJ     10     IS     »0
    rig-ur«  A18  -  M«»n  850  mb  wind  flow  patterns
    for  tb«  t!ir««  days  is Clustar  IS.    The
    clusters were  ex-racied frsm a set of consecutive
    three-day periods  during 1979, 1981 ar.d  1982.  N
    corresponds to  the nu.-rier of occurrences of cluster
    18 during that  period.   The percent of ti--ne  (out cf
    all cases) the  cluster occurred is also  shewn  in
    the upper right.   Hind vectors represent six hours
    of motion at  reported wind speed.
                          A-37

-------
EASTERN NORTH  AMERICAN MEAN asoua wmo FCLO -  is CLUSTERS
                   TRAJECTORY CLUSTER  19        7.8^.  N- 10
    :so      s;      sc      i:      ic      7:      re     s:
 EASTERN NORTH  AMERICAN  UEAN 8icu3 WIMC rxL3  -  is  C^L'STLRS
                    TRAJECTORY CLUSTER 19        ;..-  N. -c
     100
                            A-38

-------
EASTERN NORTH AMERICAN I«*N ssoue WIND FIELD - u CLUSTERS

              TRAJEC'ORY CLUSTER  19      2.ST.. N- 10
  X19 -

i. thr..
                   «  850
                          Clu.t.r  19.
   - = r ti.  thr..  i*7»  ia  Cu.t.r    .
    L3-. = S  were extrac.ec fr« a ,e^  o£ c=r.« = u-.ive

   chree-d^y period, &rir.5  1S79. 196:   : -;e/c,./....


   S'LM.S"«K P;;-';,:^";"'--^  ';:;';;
     f notion a- repcr-.ed -ir.d spesd
                       A-39

-------
       APPENDIX  B
85th Percentile  Wet  Sulfate
 Deposition Plots for the
       19 Categories
            B

-------
85th  Percentile  S04= Deposition  (STEP  19)  -  CLUSTER 1
         Figure  Bl  -  85th  Percentile  Three-day  Total
         Deposition  Pattern  due to  Cluster  1  Cases.
         The 85th percentile at each  site  was derived fror.
         the distribution of all available three-day total
         deocsitior. cases within the  cluster.
                               B-2

-------
85th  Percentile  S04= Deposition  (STEP  19)  -  CLUSTER  2
           Figure  B2  -  85th  Percantile  Three-day  Total
           Deposition  Pattern  due  to  Cluster  2  Cases.
           The 65th percer.tile at each site  was derived from
           the distribution of all available three-day total
           deposition cases within the cluster.
                                B-3

-------
85th  Percentile  S04= Deposition  (STEP  19) -  CLUSTER  3
            Figure B3  -  8£th  Percentile  Three-day  Total
            Deposition Pattern  due  to  Cluster  3  Cases .
            The  85th percer.tile at each  site  was derived fro:?.
            the  distribution of all available three-day total
            deposition cases within the  cluster.

-------
85th  Percentile  S04= Deposition (STEP  19) - CLUSTER  4
          Figure  B4  -  85th  Percentile  Three-day  Total
          Deposition  Pattern  due to  Cluster  4  Cases.
          The 85th percentile at each  site  was derived from
          the distribution of all available three-day total
          deposition cases within the  cluster.
                                B-5

-------
85th  Percentile  S04= Deposition (STEP  19)  - CLUSTER 5
          Figure 35  -  85th  Percentile  Three-day  Total
          Deposition  Pattern  due  to  Cluster  5  Cases.
          The 85th percer.tiie at  each site was derived from
          the distribution of all available three-day total
          deposition cases within the cluster.

-------
85th  Percentile  S04= Deposition  (STEP  19)  -  CLUSTER 6
              Figure  B6  -  85th  Percentile  Three-day Total
              Deposition Pattern due  to  Cluster  6  Cases.
              The  85th percer.tile at  each site was derived from
              the  distribution of all available three-day total
              deposition cases within the cluster.
                                   B-7

-------
85th  Pcrcentile  S04= Deposition  (STEP 19) - CLUSTER  7
             Figure  B7   -  85th  Percentile  Three-day Total
             Deposition  Pattern  due  to  Cluster  7  Cases.
             The 85th percentile at each  site was derived from
             the distribution of all available three-day total
             deposition  cases within the  cluster.

-------
85th  Percentile  S04=  Deposition  (STEP  19)  - CLUSTER 8
      Figura  B8  -  85th  Parcentila  Three-day  Total
      Deposition  Pattern  due  to  Cluster  8  Cases.
      The 85th percer.tile at each site  was derived fror.
      the distribution of all available three-day total
      deposition cases within the cluster.

-------
85th  Percentile  S04= Deposition  (STEP  19)  - CLUSTER 9
           Figure  B9  -  85th Percentile  Three-day  Total
           Deposition  Pattern  due  to Cluster  9  Cases.
           The 85th percer.tile at each site  was  derived from
           the distribution  of all available three-day total
           deposition cases  within the cluster.

-------
85th  Percentile  504= Deposition  (STEP  19)  - CLUSTER 10
          Figure  BIO  -  85th  Percentile Three-day
          Total  Deposition  Pattern  due  to  Cluster  10
          The 85th percer.tile at  each site was  derived from
          the distribution of all available three-day total
          deposition cases within the cluster.
                                B-ll

-------
85th  Percentile  S04= Deposition  (STEP  19) -  CLUSTER  11
         Figure  Bll  -  85th  Percentile  Three-day
         Total  Deposition  Pattern  due  to Cluster  11,
         The 85th percer.tile  at each site was derived frcrr.
         the distribution of  ail available three-day  total
         deposition  cases within the cluster.
                               B-12

-------
85th  Percentile  S04= Deposition  (STEP  19)  - CLUSTER  12
            Figure  B12  -  85th  Percentile  Three-day
            Total  Deposition  Pattern due  to  Cluster  12
            The  85th percer.tile at each site was derived from
            the  distribution of all available three-day total
            deposition cases within the cluster.
                                  B-13

-------
85th  Percentile  S04= Deposition  (STEP  19)  -  CLUSTER 13
        Figure B13  -  85th  Parcentile  Three-day
        Total  Deposition  Pattern  due  to  Cluster  13
        The 85th percentile  at  each site was derived from
        the distribution of  all available three-day  total
        deposition cases within the cluster.
                            B-1A

-------
85th  Percentile  504= Deposition  (STEP  19)  -  CLUSTER  14
           Figure  B14  -  85th  Percentile  Three-day
           Total Deposition  Pattern  due  to  Cluster  14
           The  85th percentile at  each site was derived from
           the  distribution of ail available three-day total
           deposition cases within the cluster.
                               B-15

-------
85th  Percentile  S04= Deposition  (STEP  19)  -  CLUSTER  15
         Figure  B15  -  85th  Percentile  Three-day
         Total  Deposition  Pattern due  to  Cluster  15.
         The 85th percer.tile at each site was derived frcrr.
         the distribution of all available three-day total
         deposition cases within the cluster.

-------
85th  Percentile  S04= Deposition  (STEP 19) - CLUSTER  16
           Figure  B16  -  85th  Percentile  Three-day
           Total  Depoaition  Pattern  due  to Cluster  16
                 th percentile at  each site was derived f-om
           the distribution of all available three-day total
           deposition cases within the cluster.
                                 B-17

-------
85th  Percentile 504= Deposition  (STEP 19) - CLUSTER  17
        Figure  B17  -  85th  Percentile  Three-day
        Tot»l  Deposition  Pattern  due  to Cluster  17 .
        The 85th percer.tile  at  each site was derived from
        the distribution of  ail available three-day total
        deposition cases within the cluster.

-------
85th  Percentile  S04= Deposition  (STEP  19)  - CLUSTER  18
             Figure  B18  -  85th  Percentile  Three-day
             Total  Deposition  Pattern  dua  to Cluster  18.
             The  85th percentile at  each site was derived from
             the  distribution of ail available three-day total
             deposition cases within the cluster.
                                  B-19

-------
85th  Percentile  S04= Deposition  (STEP  19)  - CLUSTER  19
      Figure  B19  -   85th  Percentile  Three-day
      Total  Deposition  Pattern  due to  Cluster  19.
      The 85th percentile at each site was derived frcrr.
      the distribution of all available three-day total
      deposition cases within the cluster.

-------
85th  Percentile  S04=  Deposition (STEP  19)  -  CLUSTER 20
               Figure B20  -  85th  Percentile  Three-day
               Total Deposition  Pattern  due  to  Cluster  20
               The  85th percent-le at each  site was derived fror.
               the  distributicr. cf ail available three-day total
               deposition cases within the  cluster.
                                   B-21

-------
           APPENDIX  C
     Site  Specific Tables  of
   Total  and Percent  of Total
     Wet  Sulfate  Deposition
                 KEY

No. - Number of times cluster had precipitation
%ofDep - Percent of  total deposition
Total - Total wet sulfate deposition (mgro )
AVG - Average deposition of wet  events only  jmgm
Max - Maximum deposition value reported (mgm )
S.D. - Standard deviation of the wet events  (mgnf

-------
SITE 1
CL
9
11
7
6
5
8
12
1
3
10
19
2
17
16
14
i
15
18
13
20
No.
68
90
58
53
47
45
41
28
25
29
26
12
14
10
•
g
i
8
i
150
%ofDep
16.7
16.13
9.78
9.63
9.38
8.4
5.51
5.36
4.45
3.08
2.63
1.75
1.67
1.62
1.59
1.1
0.99
0.18
0.07
0
TOTAL
2612.81
2523.21
1529.61
1505.99
1467.35
1314.23
861.36
838.47
696.54
481.18
410.67
273.63
261.53
253.94
248.1
171.88
154.12
28.34
10.87
7160.86
AVG
38.424
28.036
26.373
28.415
31.22
29.205
21.009
29.945
27.862
16.592
15.795
22.802
18.681
25.394
35.443
21.485
38.53
3.543
2.717
47.739
MAX
230.09
150.78
192.73
117.5
150.78
192.73
80.43
11724
102.3
83.5
125.49
50.62
66.97
80.43
46.37
52.67
71.35
8.82
4.76
230.09
S.D.
45.458
24.801
35.594
29.668
27.58
37.178
20.564
28.932
26.133
16.24
23.99
14.462
18.981
23.95
10.67
20.802
36.021
2.692
1.838
44.174
SITE 2
CL
7
9
11
5
8
6
12
1
i
10
4
14
17
16
2
18
15
19
13
20
No.
58
55
67
40
35
53
23
19
24
24
12
12
12
6
12
13
c
12
•
160
%ofDep
20.66
16.76
12.43
10.06
9.42
8.42
4.43
4.38
3.43
2.55
1.45
1.39
1.18
1.08
0.91
0.55
0.53
0.35
0
0
TOTAL
3460.22
2807.97
2082.47
1685.34
1578.31
1410.03
741.24
734.31
574.71
426.5
243.19
232.36
197.68
181.06
153.12
92.29
89.26
58.67
0.5
6288.38
AVG
59.659
51.054
31.082
42.133
45.095
26.604
32.228
38.648
23.946
17.771
20.266
19.363
16.473
30.177
12.76
7.099
9.918
4.889
0.5
39.302
MAX
328.08
328.08
146.87
143.16
327.9
140.77
145.56
124.61
81.33
65.91
47.28
74.05
46.18
68.69
37.61
39.51
29.68
14.4
0.5
204.37
S.D.
73.2%
66.853
33.437
37.464
62.082
26.044
33.533
28.209
19.565
14.666
14.797
21.394
14.973
21.668
10.356
12.576
11.151
5.047
0
38.606
CM

-------
SITE 4
CL
6
7
9
5
11
8
1
3
10
4
14
12
2
17
19
16
18
13
15
20
No.
75
61
68
60
78
29
32
23
41
11
16
29
23
18
8
8
14
9
8
19?
%ofDep
15.9
14.15
13.91
11.7
9.44
7.51
4.51
4.47
4.11
2.77
2.49
1.93
1.91
1.62
1.25
0.66
0.62
0.53
0.51
0
TOTAL
3451.02
3071.06
3018.09
2538.01
2049.42
1629.42
978.58
969.31
892.98
601.1
541.44
417.97
414.33
351.88
271.91
144.18
135.39
113.97
111.29
7266.73
AVG
46.014
50.345
44.384
42J
26.275
56.187
30.581
42.144
21.78
54.645
33.84
14.413
18.014
19.549
33.989
18.023
9.671
12.663
13.911
37.651
MAX
180.65
164.53
40237
17036
152.91
44826
12321
14325
173.74
204.13
10821
50.52
53.99
68.09
87.3
58.71
51.49
43.51
56.64
230.92
S.D.
50.758
35.461
63.322
40.517
29.947
92.825
32.238
38.152
35.657
75.285
29.263
13.329
13.387
16.237
28.051
18.838
14.986
13.57
18.098
39.294
SITE 5
CL
6
7
9
5
11
10
14
1
2
8
3
19
17
4
16
15
12
13
18
20
No.
79
57
58
50
56
36
16
25
27
29
18
13
15
14
9
10
14
5
10
161
5ofDeo
15.36
13.25
11.88
11.26
6.91
6.59
5.76
5.59
3.85
3.72
3.35
2.76
2.32
2.15
1.86
1.4
1.03
0.61
0.35
0
TOTAL
3695.48
3187.9
2858.47
2710.61
1663.57
1584.97
1385.26
1346.05
927.13
895.62
805.53
663.97
559.47
518.18
447.4
337.53
248.24
146.07
84
9042.69
AVG
46.778
55.928
49.284
54.212
29.707
44.027
86.579
53.842
34.338
30.883
44.752
51.075
37.298
37.013
49.711
33.753
17.731
29.214
8.4
56.166
MAX
190.27
333.5
170.22
333.5
134.91
233.66
233.66
169.1
103.4
79.03
233.66
137.4}
164.68
103.3
175.01
92.98
38.87
64.08
17.09
280.62
S.D.
45.879
58.679
35.692
57.408
30.691
54.861
72.276
51.079
32.208
26.83
53.201
42.757
39.479
32.082
50.53
29.485
12.266
29.275
6.463
55.165
C-2

-------
SITE 7
CL
6
7
9
5
11
1
10
8
14
3
17
i
12
2
16
18
19
15
13
20
No.
71
57
61
43
48
28
33
20
15
17
17
11
15
13
1
8
i
5
<
152
%ofDc?
18.06
16.97
12.74
9.18
8.26
7.24
4.97
4.09
3.43
3.34
2.91
1.58
1.53
1.43
1.36
1.02
0.85
0.83
0.2
0
TOTAL
3466.08
3256.63
2445.27
1761.87
1584.62
1389.6
953.38
784.8
658.39
640.67
557.85
302.99
293.78
273.72
261
196.3
163.77
160.2
40.45
6801.45
AVG
48.818
57.134
40.086
40.974
33.013
49.629
28.89
39.24
43.893
37.686
32.815
27.545
19.585
21.055
37.286
24.539
40.943
32.04
10.113
44.746
MAX
234.88
186.23
160.3
226.18
128.68
160.55
13539
172.58
125.11
128.68
124.29
84.28
40.84
56.61
105.01
91.2
91.2
86.54
19.6
353.8°
S.D.
47.94
51.377
38.462
46.933
34.104
50.764
30.708
44.462
30.774
31.792
32.953
22.565
12.401
14.023
35.182
28.441
34.082
32.164
7.256
52.32
<;TTF 8
CL
11
9
8
7
<
3
^
19
12
i
10
17
i
14
i
15
1
1
1
2C
No.
75
66
39
30
31
22
25
16
19
23
20
16
11
10
20
i
6
i

14
%ofDeo
16.51
13.11
11.1
10.95
9.22
6.25
4.73
3.74
3.59
3,5
3.03
2.95
2.85
2.49
2.15
1.4
1.3
0.7
0.2

TOTAL
2037.28
1618.52
1369.53
1351.92
1138.2
771.6
584.13
461.26
443.45
432.04
373.63
364.34
352.28
307.79
264.98
172.63
171.63
95.06
32.76
4870.14
AVG
27.164
24.523
35.116
45.064
36.716
35.073
23.365
28.829
23.339
18.784
18.682
22.771
32.025
30.779
13.249
19.181
28.605
15.843
6.552
34.057
MAX
124.37
101.64
105.13
330.62
123.42
112.16
112.16
62.02
75.76
63.92
45.93
61.99
103.14
75.56
73.78
35.14
136.16
54.55
15
330.6

S.D.
26.377
24.459
27.159
59.368
29.691
29.844
28.098
20.264
21.268
15.175
' 13.447
16.072
37.131
22.365
18.247
12.736
52.909
19.654
4.992
44.688
C-3

-------
SITE 10
CL
6
5
9
7
11
3
10
1
2
8
14
17
19
4
12
16
18
13
15
20
No.
93
46
65
48
66
18
31
25
18
16
12
19
11
7
9
4
6
*
i
i
161
%ofDep
23.97
15.87
14.7
10.38
8.82
4.25
3.78
3.65
2.67
2.55
2.21
1.98
1.88
1.09
0.94
0.7
0.39
0.11
0.06
0
TOTAL
2674.48
1770.82
1640.54
1158.78
983.93
474.72
421.28
407.08
297.98
284.06
246.12
220.8
209.94
121.72
104.67
78.44
44.07
12.56
7.24
3949.21
AVG
28.758
38.496
25.239
24.141
14.908
26.373
13.59
16.283
16.554
17.754
20.51
11.621
19.085
17.389
11.63
19.61
7.345
6.28
3.62
24.529
MAX
202.95
215.16
117.38
96.74
151.13
65.66
48.57
75.59
99.25
64.03
58.59
44.05
31.49
70.31
47.83
46.9
31.49
6.28
6.28
105.14
S.D.
31.137
39.804
21.807
23.902
24.766
21.203
14.06
18.354
24.162
18.385
17.665
11.439
8.189
24.737
13.886
21.357
12.028
0
3.762
24.977
SITE 11
CL
9
11
7
t,
g
6
10
3
12
1
17
14
4
15
18
19
O
4.
16
13
20
No.
73
91
57
50
41
75
50
24
30
29
21
15
15
14
13
13
22
8
8
198
%ofDep
17.89
13.94
13.61
10.21
8.83
7.36
5.09
4.01
3.37
3.28
2.66
1.63
1.42
1.41
1.25
1.25
1.22
1
0.57
0
TOTAL
3886.21
3028.78
2957.89
2219.3
1918.94
1598.49
1105.56
870.54
732.01
712.52
578.51
354.81
307.72
306.08
272.34
271.42
265.21
218.25
123.28
9651.78
AVG
53.236
33.283
51.893
44.386
46.803
21.313
22.111
36.273
24.4
24.57
27.548
23.654
20.515
21.863
20.949
20.878
12.055
27.281
15.41
48.746
MAX
362.37
359.07
289.71
279.55
236.63
93.49
93.49
90.24
93.54
76.11
82.38
78.63
54.35
75.48
164.14
73.09
40.05
97.6
49.06
313.13
S.D.
75.088
50.239
71.324
46.443
56.984
18.214
21.848
30.889
24.753
19.603
21.964
19.039
14.127
20.465
43.535
23.808
10.7%
30.673
15.586
55.283

-------
SITE 12
CL
9
6
5
11
1
10
3
16
14
2
7
17
8
4
15
12
19
13
18
20
No.
36
51
22
32
26
25
15
8
11
13
12
8
11
10
6
t
4
j.
•
112
%ofDep
19.65
16.24
9.84
9.46
8.17
6.09
5.01
3.94
3.64
3.47
2.95
2.67
2.2
1.95
1.27
1.19
1.09
0.75
0.41
0
TOTAL
1912.42
1581.01
957.66
920.71
795.02
592.33
487.85
383.85
353.87
337.6
287.46
259.54
214.54
189.99
123.5
116.29
105.76
73.45
39.87
2898.22
AVG
53.123
31
43.53
28.772
30.578
23.693
32.523
47.981
32.17
25.969
23.955
32.443
19.504
18.999
20.583
23.258
26.44
18.363
13.29
25.877
MAX
214.56
214.56
227.57
73.69
84.84
84.84
75.55
93.06
74.14
74.14
64.49
75.55
33.13
44.03
40.1
46.81
30.25
28.6
18.67
206.37
S.D.
58.198
42.681
61.717
21.99
20.35
19.899
23.784
30.95
21.456
19.253
21.179
22.86
11.063
12.719
10.577
13.994
2.62
7.363
5.845
25.283
SITE 13
CL
11
6
5
9
7
8
12
1
19
10
3
16
17
15
A
i
18
14
13
20
No.
89
56
41
56
52
46
37
30
25
39
17
10
16
A
10
11
12
t
2
159
%ofDep
19.04
12.14
11.26
9.6
8.41
7.36
7.19
5.87
3.92
3.62
2.95
2.62
2.13
1.03
0.88
0.77
0.61
0.51
0.09
0
TOTAL
1917.71
1222.67
1133.82
966.99
847.03
740.93
723.62
590.96
394.69
364.35
297.27
263.89
214.96
103.33
88.78
77.51
61.07
51.12
9.37
3853.9
AVG"
21.547
21.833
27.654
17.268
16.289
16.107
19.557
19.699
15.788
9.342
17.486
26.389
13.435
51.665
8.878
7.046
5.089
10.224
4.685
24.238
MAX
107.33
112.77
104.32
94.55
122.59
65.45
80.42
127.76
38.82
47.87
54.33
85.68
50.05
67.85
29.01
16.35
17.7
24.77
8.9
112.46

S.D.
21.749
32.428
29.015
21.437
20.237
15.538
18.167
24.29
11.868
10.592
• 13.803
24.807
14.212
22.889
7.996
5.668
6.087
10.329
5.961
25.997
C-5

-------
SITE 14
CL
7
8
9
11
3
6
1
19
5
2
15
13
17
14
12
10
4
16
18
20
No.
49
46
57
71
20
29
28
10
26
27
11
6
18
10
14
15
9
4
4
145
%ofDcp
15.59
13.99
13.09
1132
5.42
5.1
5.06
4.71
4.58
4.12
3.44
2.86
2.63
2.16
2.14
1.89
1.15
0.56
0.19
0
TOTAL
1598.28
1434.15
1342.66
1160.97
555.53
523.48
518.41
482.62
469.58
422.42
352.46
293.4
269.72
221.58
219.61
194.26
118.1
57.42
19.9
3632.68
AVG
32.618
31.177
23.555
16.352
27.776
18.051
18.515
48.262
18.061
15.645
32.042
48.9
14.984
22.158
15.686
12.951
13.122
11.484
4.975
25.053
MAX
382.37
382.37
401.1
161.9
161.9
93.55
161.9
117.5
60.93
50.64
76.61
94.1
40.49
38.68
66.54
50.04
28.88
25.72
7.97
192.41
S.D.
66.725
56.898
52.593
21.579
39.317
23.803
29.578
43.327
16.052
11.834
26.095
34.743
11.928
14.754
15.969
14.347
8.866
9.526
2.451
35.95
SITE 15
CL
9
11
5
8
6
7
1
2
3
10
17
4
14
15
16
13
19
12
18
20
No.
55
71
39
41
48
45
27
30
19
29
17
16
13
12
9
9
11
9
7
163
<*ofDep
15.53
15.19
8.45
7.05
6.93
6.34
6.09
5.69
4.93
4.45
3.74
3.14
2.81
138
2.29
1.79
1.28
1.19
0.72
0
TOTAL
2178.93
2131.73
1185.32
989.22
972.52
889.8
854.55
798.5
691.84
624.59
525.45
440.56
394.43
334.41
321.8
250.59
179.86
167.03
100.4
4457.37
AVG
39.617
30.024
30.393
24.127
20.261
19.773
31.65
26.617
36.413
21.538
30.909
27.535
30.341
27.868
35.756
27.843
16.351
18.559
14.343
27.346
MAX
195.3
97.09
104
191.73
113.1
138.24
97.09
66.67
97.09
51.98
77.28
113.1
76.88
46.53
58.85
67.05
47.86
57.18
51.98
195.3
S.D.
44.448
24.506
21.955
32.844
21.555
28.245
21.717
17.771
26.502
17.231
20.933
29.693
21.75
17.51
14.836
20.709
13.35
17.234
18.208
29.072
C-6

-------
SITE 16
CL
7
9
11
1
6
5
8
10
4
2
3
19
16
17
15
14
12
13
18
20
No.
58
61
68
31
64
47
37
36
17
29
17
17
10
13
9
11
16
7
7
165
%ofDep
14.59
11.1
10.87
7.71
7.62
7.22
5.92
5.14
4.79
4.32
3.67
3.13
2.61
2.39
2.36
2.3
1.81
1.4
1.05
0
TOTAL
2236.6
1701.24
1665.93
1181.38
1168.07
1105.85
907.18
788.44
733.61
662.34
563.11
480.13
399.33
366.08
362.03
352.22
276.97
213.98
161.42
5373.42
AVG
38.562
27.889
24.499
38.109
18.251
23.529
24.518
21.901
43.154
22.839
33.124
28.243
39.933
28.16
40.226
32.02
17.311
30.569
23.06
32.566
MAX
254.14
216.22
123.19
155.75
12733
123.19
159.77
103.09
107.48
59.58
94.51
79.63
94.51
78.89
78.27
94.51
77.96
61.08
61.08
254.14
S.D.
55.72
43.099
27.446
38.107
20.374
33.177
28.676
20.814
33.154
14.161
31.03
26.77
26.184
26.887
26.732
26.869
21.343
22.702
19.623
41.579
SITE 17
CL
6
7
9
11
10
5
8
1
4
2
19
14
16
15
17
-
13
18
12
20
No.
74
47
52
52
30
38
35
27
14
17
21
12
11
12
12
12
8
9
10
151
%ofDeo
17.93
10.53
9.88
8.32
6.96
6.92
6.61
5.23
5.01
3.78
3.57
3.15
2.39
2.31
2.28
2.01
1.21
1.17
0.75
0
TOTAL
2403.38
1411.58
1324.33
1114.81
932.88
927.35
885.78
700.9
670.91
506.34
479.05
422.86
320.54
310.12
305.41
269.08
162.12
156.65
100.12
3893.04
AVG
32.478
30.034
25.468
21.439
31.0%
24.404
25.308
25.959
47.922
29.785
22.812
35.238
29.14
25.843
25.451
22.423
20.265
17.406
10.012
25.782
MAX
135.2
88.33
84.72
76.86
113.02
97.46
80.94
85
126.04
94.17
81.57
68.46
59.97
59.97
97.46
47.9
46.63
30.33
29.57
96.17
S.D.
28.669
29.137
21.505
17.428
27.888
22.599
23.077
17.977
32.045
26.456
18.538
19.949
14.742
14.612
30.809
13.533
12.927
8.119
9.429
22.375
C-7

-------
SITE
CL
9
6
7
8
5
19
14
10
11
4
1
16
3
2
15
17
18
13
12
20
18
No.
68
74
57
35
38
26
16
28
29
10
19
10
16
13
8
12
4
7
5
160

%ofDep
17.95
16.62
16.35
9.54
7.42
7.11
3.68
2.9
2.75
2.72
2.33
2.16
1.67
1.63
1.33
1.31
1.18
0.82
0.53
0

TOTAL
2504.82
2319.08
2281.78
1331.78
1035.05
992.43
512.95
404.95
384.26
379.44
324.57
301.18
233.3
227.7
185.1
183.29
164.09
114.54
73.43
5132.71

AVG
36.836
31.339
40.031
38.051
27.238
38.17
32.059
14.463
13.25
37.944
17.083
30.118
14.581
17.515
23.138
15.274
41.022
16.363
14.686
32.079

MAX
145.26
126.02
135.46
156.28
106.03
130.02
87.08
51.92
75.89
72.39
40.79
73.88
31.03
53.45
44.68
33.01
111.19
56.06
57.11
159.37

S.D.
35.551
27.616
34.781
39.16
26.236
30.397
23.176
13.294
15.53
27.4
13.605
20.043
9.34
12.952
12.481
10.35
47.472
18.794
23.794
33.882
SITE 19
CL
6
9
;
8
(
19
11
4
14
10
1
2
3
17
16
12
18
13
15
20
No.
63
57
44
29
21
18
11
6
3
1)
4
7
7
6
2
2
2
2
1
99
%ofDg>
27.16
23.11
14.57
10.61
7
5.86
1.96
1.88
1.56
1.27
1.1
1.06
0.86
0.74
0.71
0.26
0.18
0.06
0.05
0
TOTAL
1999.28
1701.4
1073.04
781.02
515.66
431.7
144.45
138.05
115.15
93.14
81.24
78.12
62.95
54.73
52.2
19.29
12.99
4.33
3.65
1859.52
AVG
31.735
29.849
24.387
26.932
24.555
23.983
13.132
23.008
38.383
8.467
20.31
11.16
8.993
9.122
26.1
9.645
6.495
2.165
3.65
18.783
MAX
106.62
145.26
135.46
120.2S
69.57
74.16
75.89
57.03
85.03
24.41
26.5
28.29
32.17
24.41
27.98
17.38
11.38
3.51
3.65
121.66
S.D.
27.301
33.868
29.394
35.05
22.898
20.769
21.185
23.856
42.413
8.068
11.031
11.871
11.389
8.495
2.659
10.939
6.908
1.902
0
21.779
C-8

-------
SITE 20
CL
7
11
6
9
5
8
10
1
12
19
17
16
3
4
2
14
18
15
13
20
No.
71
86
74
66
43
49
44
34
35
21
17
10
21
14
12
9
12
8
8
181
%ofDep
16.44
16.3
15.08
11.92
8.17
7.52
4.91
4.21
4.19
3.1
1.67
1.34
1.29
1.22
0.98
0.56
0.47
0.42
0.21
0
TOTAL
2670.1
2647.88
2448.57
1935.92
1326.61
1221.64
796.83
683.04
680.22
504.25
270.79
217.95
209.74
198.46
159.63
90.68
75.75
' 68.73
33.62
4810.84
AVG
37.607
30.789
33.089
29.332
30.851
24.931
18.11
20.089
19.435
24.012
15.929
21.795
9.988
14.176
13.303
10.076
6.313
8.591
4.203
26.579
MAX
151.18
233.2
264.74
15027
227.74
12536
93.31
10627
67.84
110.01
61.67
93.31
63.72
69.39
40.12
19.16
22.72
14.44
9.53
151.18
S.D.
39.701
40.89
54.4
28.83
42.302
29.064
20.992
26.842
21.604
27.256
20.752
28.287
13.247
19.714
11.309
6.289
6.762
3.954
3.26
28.728
SITE 21
CL
7
9
11
6
5
1
8
10
4
12
16
3
17
2
19
14
18
15
13
20
No.
65
74
79
78
45
33
44
38
13
29
9
21
14
17
17
9
12
5
5
185
<7cofDer
16.18
14.25
13.26
12.78
11.41
6.91
5.71
5.43
3.55
2.38
1.63
1.52
1.27
1.07
0.92
0.69
0.64
0.28
0.11
0
TOTAL
3248.21
2861.77
2662.01
2566.08
2289.95
1386.54
1145.87
1089.85
713.23
478.23
328.14
305.31
254.97
215.81
184.28
139.14
128.25
56.02
22.93
8019.24
AVG
49.972
38.673
33.6%
32.898
50.888
42.016
26.043
28.68
54.864
16.491
36.46
14.539
18.212
12.695
10.84
15.46
10.687
11.204
4.586
43.347
MAX
255.09
255.09
216.03
255.09
230.9
174.88
124.7
174.88
295.54
51.43
174.88
52.3
63.81
40.27
28.88
42.07
62.94
26.95
9.24
419.5
S.D.
60.174
42.0%
32.821
42.433
52.327
42!662
32.513
30.082
83.919
14.245
52.703
13.961
18.442
12.085
9.371
13.354
16.918
10.719
3.384
61.703

-------
SITE 27
CL
6
9
11
5
7
10
1
8
3
2
17
4
12
18
14
16
19
13
15
20
No.
83
67
63
47
53
38
25
32
20
15
13
14
16
11
9
4
i
6
t
147
%ofDep
21.07
14.94
14.22
14.21
6.98
6.19
6.14
3.23
2.47
2.21
1.92
1.91
1.35
0.85
0.75
0.56
0.45
0.29
0.25
0
TOTAL
3502.05
2483.48
2363.91
2361.19
1160.56
1029.34
1020.29
536.45
410.71
367.11
319.38
317.73
225.07
140.45
125.06
92.95
75.51
48.35
41.09
4906.59
AVG
42.193
37.067
37.522
50.238
21.897
27.088
40.812
16.764
20.536
24.474
24.568
22.695
14.067
12.768
13.8%
18.59
15.102
8.058
10.273
33.378
MAX
257.77
265.19
265.19
178.57
90
142.58
142.58
78.21
69.39
58.46
72.1
68.71
58.87
38.91
24.1
56.63
31.5
13.58
13.84
169.36
S.D.
42.839
40.573
46.135
45.409
18.766
29.25
38.687
18.639
17.61
19.959
26.25
21.186
15.713
14.1
9.052
21.37
12.733
5.453
6.233
28.888
SITE 28
CL
9
5
6
7
11
1
10
4
17
8
2
14
3
12
18
16
13
19
15
20
No.
65
48
78
41
68
28
43
14
18
19
15
14
12
8
9
6
8
4
5
138
<5k>fDep
18.9
14.12
12.93
12.55
9.74
5.61
5.31
3.69
3.52
2.88
2.43
2.25
2.16
1.38
0.8
0.63
0.53
0.35
0.23
0
TOTAL
3940.1
2943.62
2696.28
2616.8
2031.65
1169.49
1106.47
769.66
733.6
599.67
507.3
469.98
451.01
286.76
166.97
130.32
110.64
73.38
47.08
6274.74
AVG
60.617
61.325
34.568
63.824
29.877
41.767
25.732
54.976
40.756
31.562
33.82
33.57
37.584
35.845
18.552
21.72
13.83
18.345
9.416
45.469
MAX
458.42
391.45
117.02
193.36
180.94
157.2
88.17
141.18
183.09
. 103.26
73.43
78.65
125.6
86.81
61.57
33.65
32.66
32.55
15.73
458.42
S.D.
75.112
76.565
26.926
58.354
30.52
42.594
22.646
41.308
47.878
32.445
19.221
20.795
32.21
35.788
18.304
11.754
9.201
9.569
6.67
60.823
C-10

-------
           APPENDIX  D
    Site  Specific  Tables of
   Total and Percent  of  Total
           Precipitation
                  KEY

No. - Number of times cluster had precipitation
%of Pre -  Percent of total precipitation
Total - Total precipitation  (cm)
AVG - Average precipitation  for wet  events^ only
Max - Maximum precipitation  reported (mgm *)
S.D. - Standard deviation of precipitation
       the wet events only(mgm" )

-------
SITE 1
CL
11
9
5
12
6
3
7
8
1
10
19
2
16
17
14
15
4
18
13
20
No.
104
75
55
43
66
27
69
47
33
41
28
18
11
18
9
5
12
12
6
179
%ofPre
18.95
12.02
8.88
8.21
6.78
6.46
6.31
5.89
5.06
3.92
3.77
3.52
2.69
2.57
1.88
1.02
0.74
0.74
0.59
0
TOTAL
179.41
1 13.74
84.06
77.67
64.19
61.17
59.74
55.78
47.87
37.06
35.73
33.28
25.47
24.35
17.83
9.61
7
6.99
5.62
428.41
AVG
1.725
1.517
1.528
1.806
0.973
2.266
0.866
1.187
1.451
0.904
1.276
1.849
2.315
1.353
1.981
1.922
0.583
0.582
0.937
2.393
MAX
16.13
5.58
6.09
4.55
4.06
4.58
5.05
5
4.39
3.55
8.38
8.25
3.81
4.11
3.35
4.19
1.52
2.39
2.74
19.82
S.D.
1.986
1.411
1.393
1.411
1.027
1.265
0.897
1.217
0.975
0.994
1.776
1.973
1.35
1.322
0.741
1.609
0.584
0.783
0.92
2.478
SITE 2
CL
11
9
;
6
t
8
3
12
1
10
2
14
17
16
4
18
19
13
15
20
No.
79
80
70
67
48
43
27
30
28
37
16
14
14
10
13
16
23
4
9
207
%ofPre
14.34
12.98
11.6
9.31
8.79
6.78
6.23
5.94
5.93
3.94
3.32
2.48
1.85
1.76
1.4]
1.04
0.98
0.77
0.56
0
TOTAL
105.79
95.81
85.59
68.72
64.84
50.02
45.96
43.81
43.76
29.13
24.5
18.29
13.67
13.02
10.38
7.64
7.26
5.65
4.16
321.48
AVG
1.339
1.198
1.223
1.026
1.351
1.163
1.702
1.46
1.563
0.787
1.531
1.306
0.976
1.302
0.798
0.478
0.316
1.413
0.462
1.553
MAX
11.63
11.63
5.07
11.63
4.01
5.07
5.08
4.06
6.38
4.27
7.49
4.83
3.13
3.43
2.49
1.52
1.48
3.76
1.37
13.28
S.D.
1.633
1.693
1.293
1.588
1.183
1.223
1.178
1.162
1.556
0.905
1.835
1.401
0.941
1.262
0.708
0.53
0.423
1.66
0.393
2.195
D-l

-------
SITE 4
CL
6
5
7
11
9
3
8
1
10
12
4
14
17
2
18
15
19
13
16
20
No.
83
66
73
92
77
28
38
35
47
34
16
18
23
25
18
10
10
12
9
205
%ofPre
14.63
11.01
10.8
9.88
9.42
6.79
6.28
5.97
5.69
3.35
3.27
3.06
2.77
2.57
1.11
1.01
0.95
0.77
0.63
0
TOTAL
117.48
88.43
86.71
79.35
75.66
54.49
50.46
47.96
45.72
26.91
26.28
24.6
22.25
20.65
8.93
8.1
7.66
6.17
5.08
264.87
AVG
1.415
1.34
1.188
0.863
0.983
1.946
1.328
1.37
0.973
0.791
1.642
1.367
0.967
0.826
0.4%
0.81
0.766
0.514
0.564
1.292
MAX
7.69
4.65
8.24
5.16
8.44
4.79
8.85
4.54
7.95
3.16
4.5
4.43
3.16
1.91
2.62
2.34
2.16
2.29
1.5
6
S.D.
1.94
1.308
1.248
1.013
1.331
1.333
1.708
1.221
1.567
0.828
1.382
1.311
0.915
0.557
0.606
0.613
0.746
0.593
0,492
1.249
SITE 5
CL
6
5
7
9
11
10
1
14
2
4
3
19
8
16
17
15
12
18
13
20
No.
91
57
63
70
68
43
31
19
30
18
22
15
31
10
20
11
15
11
I
195
%ofPre
17.88
10.11
9.13
7.86
7.61
7.41
7.15
6.49
4.86
3.8
3.58
3.19
3.09
2.26
2.06
1.38
0.78
0.77
0.58
0
TOTAL
185.49
104.84
94.75
81.51
78.98
76.86
74.16
67.28
50.39
39.45
37.16
33.13
32.01
23.45
21.38
14.34
8.1
8
6.03
383.06
AVG
2.038
1.839
1.504
1.164
1.161
1.787
2.392
3.541
1.68
2.192
1.689
2.209
1.033
2.345
1.069
1.304
0.54
0.727
0.861
1.964
MAX
11.23
9.07
9.07
5.92
4.83
7.29
8.58
7.38
4.78
9.19
4.62
4.32
3.81
5.86
5.16
2.59
2.39
4.09
2.08
18.56
S.D.
1.947
1.838
1.452
1.112
1.235
1.778
. 2.025
1.807
1.509
2.201
1.251
1.576
1.081
1.628
1.177
0.954
0.573
1.153
0.686
2.564

-------
SITE 7
CL
6
7
11
5
1
9
10
3
14
17
g
2
12
4
16
18
19
15
13
20
No.
95
72
82
61
36
73
43
26
17
28
31
24
27
12
11
17
8
7
9
195
%ofPre
19.08
13.36
10.54
9.93
8.15
7.97
5.27
4.23
3.87
3.18
2.99
2.35
2
1.98
1.4
1.15
1.03
0.94
0.56
0
TOTAL
161.69
113.27
89.38
84.18
69.07
67.52
44.69
35.89
32.79
26.99
25.35
19.95
16.98
16.76
11.88
9.79
8.74
7.94
4.77
258.24
AVG
1.702
1.573
1.09
1.38
1.919
0.925
1.039
1.38
1.929
0.964
0.818
0.831
0.629
1.397
1.08
0.576
1.092
1.134
0.53
1.324
MAX
8.94
6.35
6.48
11.91
7.87
5.08
3.4
4.17
4.32
3.81
8.05
3.17
3.28
3.56
3.68
2.67
3.35
2.49
1.65
7.8
S.D.
1.82
1.632
1.25
2.091
2.104
0.962
0.96
1.257
1.182
1.108
1.646
0.947
0.721
1.233
1.447
0.703
1.248
0.765
0.549
1.201
SITE 8
CL
11
9
8
2
5
7
3
1
12
17
10
6
19
4
15
14
13
18
16
20
No.
91
80
57
31
42
47
25
35
30
21
26
30
19
11
13
13
12
9
6
188
%ofPre
17.75
11.85
10.94
7.62
6.96
6.16
6.09
5.37
4.06
3.84
3.71
2.98
2.97
2.61
2.36
1.83
1.67
0.61
0.6
0
TOTAL
181.88
121.43
112.04
78.04
71.33
63.15
62.42
55.05
41.63
39.36
38
30.48
30.39
26.75
24.2
18.71
17.14
6.24
6.17
344.87
AVG
1.999
1.518
1.966
2.517
1.698
1.344
2.497
1.573
1.388
1.874
1.4$2
1.016
1.599
2.432
1.862
1.439
1.428
0.693
1.028
1.834
MAX
13.76
9.04
5.03
11.66
9.3
4.17
11.66
5.74
4.42
6.1
8.79
6.96
3.94
7.62
7.29
2.92
7.29
3.15
3.28
18.71
S.D.
2.126
1.69
1.533
2.492
1.704
1.204
2.556
1.359
1.243
1.552
1.858
1.596
1.254
2.794
1.983
0.961
1.945
0.982
1.16
2.14
D-3

-------
SITE 10
CL
6
5
9
11
7
10
1
3
8
17
14
2
12
4
19
16
18
15
13
20
No.
99
53
74
91
53
46
31
22
22
23
15
22
19
10
15
6
10
4
i
185
%ofPre
20.57
12.9
11.95
11.84
8.S6
6.23
5.47
4.16
3.25
2.75
2.37
2.3
2.1
2.04
1.08
0.87
0.76
0.28
0.23
0
TOTAL
133.98
84.02
77.8
77.09
57.72
40.6
35.62
27.08
21.14
17.93
15.44
14.95
13.67
13.29
7.02
5.64
4.93
1.8
1.47
238.42
AVG
1.353
1.585
1.051
0.847
1.089
0.883
1.149
1.231
0.961
0.78
1.029
0.68
0.719
1.329
0.468
0.94
0.493
0.45
0.49
1.289
MAX
8.38
8.13
3.99
5.59
4.09
4.15
5.16
3.6
4.24
3.1
2.54
1.88
2.51
4.08
1.45
1.98
1.45
0.66
0.66
5-26
S.D.
1.683
1.402
0.942
0.908
1.066
0.836
1.092
1.207
1.182
0.693
0.768
0.507
0.569
1.166
0.361
0.859
0.474
0.203
0.294
1.168
SITE 1 1
CL
n
9
5
6
7
1
8
30
3
12
14
17
2
4
15
19
16
18
13
20
No.
108
82
63
81
72
37
52
60
28
38
18
22
30
16
14
17
10
15
13
242
%ofPre
13.42
12.25
11.05
9.03
9.03
7.03
6.91
6.69
5.18
3.82
2.84
2.58
2.54
1.92
1.49
1.32
1.21
1.07
0.62
0
TOTAL
130.73
119.37
107.59
87.97
87.92
68.49
67.3
65.2
50.42
37.22
27.65
25.13
24.72
18.71
14.51
12.9
11.77
10.46
6.01
379.43
AVG
1.21
1.456
1.708
1.086
1.221
1.851
1.294
1.087
1.801
0.979
1.536
1.142
0.824
1.169
1.036
0.759
1.177
0.697
0.462
1.568
MAX
7.49
7.67
6.91
5.97
6.89
7.31
6.38
6.79
5.26
5.82
6.91
3.83
1.9
3.05
3.81
3.81
4.57
4.5
2.18
7.85
S.D.
1.396
1.738
1.61
1.141
1.359
1.534
1.577
1.236
1.503
1.176
' 1.74
0.978
0.558
0.913
1.044
0.939
1.318
1.097
0.576
1.717
D-4

-------
SITE 12
CL
6
9
11
1
5
10
14
3
2
4
8
7
16
19
17
15
12
18
13
20
No.
70
43
52
31
33
32
16
17
17
14
16
21
9
6
10
9
10
3
4
132
%ofPre
21.95
11.85
11.42
9.1
6.97
6.77
5.1
3.83
3.69
3.67
3.37
3.03
3.01
1.73
1.63
1.6
0.9
0.21
0.19
0
TOTAL
222.48
120.17
115.74
92.22
70.61
68.64
51.65
38.81
37.36
37.18
34.19
30.68
30.55
17.53
16.48
16.27
9.12
2.16
1.88
239.07
AVG
3.178
2.795
2.226
2.975
2.14
2.145
3.228
2.283
2.198
2.656
2.137
1.461
3.394
2.922
1.648
1.808
0.912
0.72
_ 0.47
1.811
MAX
33.7
33.7
7.11
9.45
8.26
9.45
9.35
9.35
8.81
8.81
6.07
6.1
9.35
6.14
3.3
6.1
2.24
1.02
0.63
7.09
S.D.
5.688
5.053
1.955
2.186
2.271
2.365
2.577
2.367
2.145
2.342
1.91
1.586
3.222
2.094
1.089
1.75
0.893
0.267
0.17
1.698
SITE 13
CL
11
12
9
5
7
8
6
3
19
10
1
16
17
2
14
15
4
18
13
20
No.
110
44
69
48
70
55
70
22
31
55
36
12
21
12
7
5
14
15
6
199
%ofPre
20.3
10.39
8.33
6.93
6.89
6.43
6.39
5.86
5.22
5.07
4.38
3.72
3.29
1.5
1.5
1.33
1.13
1.06
0.26
0
TOTAL
185.43
94.91
76.07
63.33
62.92
58.69
58.39
53.5
47.7
46.32
39.97
34
30.09
13.69
13.67
12.19
10.29
9.72
2.42
412.02
AVG
1.686
2.157
1.102
1.319
0.899
1.067
0.834
2.432
1.539
0.842
1.11
2.833
1.433
1.141
1.953
2.438
0.735
0.648
0.403
2.07
MAX
11.84
6.43
6.35
6.09
4.24
5.75
4.01
6.12
7.39
5.21
4.25
10.5
6.23
5
6.12
7.19
3.32
3.89
1.22
11.94
S.D.
1.964
1.903
1.37
1.366
0.953
1.222
0.9
1.632
1.899
1.126
0.919
2.959
1.738
1.374
2.571
2.796
0.941
1.053
0.483
2.694
D-5

-------
SITE 14
CL
11
9
7
8
2
1
3
5
6
14
10
19
17
15
4
13
12
16
18
20
No.
79
68
53
55
32
30
24
34
32
11
19
13
20
12
10
7
15
5
4
164
%ofPre
15.72
12.16
10.87
10.75
7.19
6.76
6.47
5.12
4.61
3.8
3.46
3.09
2.24
1.99
1.98
1.61
1.54
0.44
0.18
0
TOTAL
156.6
121.16
108.3
107.12
71.65
67.35
64.47
51.04
45.88
37.82
34.49
30.76
22.28
19.82
19.72
16
15.38
4.42
1.81
330.5
AVG
1.982
1.782
2.043
1.948
2.239
2.245
2.686
1.501
1.434
3.438
1.815
2.366
1.114
1.652
1.972
2.286
1.025
0.884
0.452
2.015
MAX
10.03
12.3
8.1
8.1
6.99
10
10.03
6.6
6.81
9.14
12.3
6.99
6.17
5.03
4.69
5.86
5.03
1.27
1.02
8.74
S.D.
2.236
2.348
2.214
1.806
1.994
2.29
2.849
1.742
1.962
2.711
3.336
2.204
1.339
1.269
1.523
2.187
1.327
0.245
0.386
2.06
SITE. 15
CL
11
9
6
1
2
8
5
10
3
7
14
17
4
16
15
19
12
13
18
20
No.
75
62
56
33
33
55
39
33
21
48
13
17
18
9
14
13
11
10
7
183
%ofPre
15.51
8.77
8.74
8.11
7.84
7.66
7.28
5.34
5.16
4.32
3.99
3.89
3.26
3.16
1.9
1.68
1.33
1.27
0.8
0
TOTAL
175.13
98.99
98.68
91.63
88.54
86.51
82.26
60.29
58.33
48.83
45.04
43.91
36.8
35.64
21.46
18.94
15.04
14.37
8.98
383.42
AVG
2.335
1.597
1.762
2.777
2.683
1.573
2.109
1.827
2.778
1.017
3.465
2.583
2.044
3.96
1.533
1.457
1.367
1.437
1.283
2.095
MAX
11.76
5.62
6.35
7.44
11.76
10.32
6.5
5.94
11.59
5.87
11.59
7.7
6.3
10.08
4.14
8
5.51
4.6
4.29
13.36
S.D.
2.118
1.505
1.741
2.121
2.542
1.985
1.913
1.997
2.622
1.202
' 2.974
2.114
1.569
3.534
1.459
2.229
1.804
1.572
1.563
2.156

-------
SITE 16
CL
6
11
9
7
1
5
4
10
8
2
14
3
19
16
17
15
12
18
13
20
No.
66
70
63
60
34
50
17
43
42
29
15
21
18
10
16
10
16
9
8
186
%ofPre
11.84
11.04
7.72
7.71
7.68
7.55
6.83
6.29
6.29
4.68
4.44
3.67
3.2
2.72
2.72
1.99
1.34
1.27
1.02
0
TOTAL
157.48
146.9
102.65
102.56
102.21
100.42
90.82
83.71
83.61
62.2
59.08
48.8
42.62
36.24
36.19
26.46
17.76
16.93
13.54
462.86
AVG
2.386
2.099
1.629
1.709
3.006
2.008
5.342
1.947
1.991
2.145
3.939
2.324
2.368
3.624
2.262
2.646
1.11
1.881
1.692
2.488
MAX
12.56
10.61
6.07
11.61
8.15
15.12
12.56
8.15
7.82
6.35
9.66
8.84
7.87
9.02
5.79
6.45
6.35
4.9
3.79
17.25
S.D.
2.749
2.38
1.738
2.161
2.446
2.965
4.024
2.283
2.143
1.634
3.585
2.531
2.565
2.767
1.901
2.438
1.773
1.322
1.564
3.145
SITE 17
CL
6
10
11
9
4
7
5
1
8
2
19
14
17
3
15
16
13
18
12
20
No.
87
37
58
61
17
49
44
30
41
20
21
13
15
13
15
11
9
9
10
162
%ofPre
23.11
7.96
7.82
7.46
7.24
7.13
6.47
6.42
5.08
4.42
3.19
2.71
2.46
2.22
2.07
1.55
1.1
0.9
0.68
0
TOTAL
274.05
94.39
92.67
88.42
85.86
84.5
76.75
76.07
60.26
52.37
37.81
32.13
29.2
26.35
24.59
18.4
13
10.71
8.08
286.63
AVG
3.15
2.551
1.598
1.45
5.051
1.724
1.744
2.536
1.47
2.619
1.8
2.472
1.947
2.027
1.639
1.673
1.444
1.19
0.808
1.769
MAX
13.09
16.7
5.33
7.72
16.7
8.35
5.82
11.68
7.62
16.7
5.39
5.54
5.46
4.52
4.11
4.11
3.76
2.79
3.1
9.45
S.D.
2.862
3.497
1.242
1.564
4.314
1.91
1.638
2.818
1.843
3.868
1.7
1.456
1.846
1.378
1.218
1.225
1.389
0.743
1.11
1.913
D-7

-------
SITE
CL
6
9
7
8
5
19
11
1
4
10
14
17
16
A
-
18
15
13
12
20
18
No.
96
71
63
40
47
29
41
27
11
36
17
21
12
19
18
•
10
•7
g
188

%ofPre
19.34
17.74
14.76
9.23
6.55
5.95
4.19
3.54
3.13
3.08
2.81
2.14
1.71
1.63
1.6
1.09
0.71
0.5
0.29
0

TOTAL
208.89
191.62
159.44
99.71
70.78
64.27
45.3
38.2
33.81
33.29
30.3
23.1
18.46
17.59
17.31
11.76
7.65
5.38
3.15
305.3

AVG
2.176
2.699
2.531
2.493
1.506
2.216
1.105
1.415
3.074
0.925
1.782
1.1
1.538
0.926
0.962
1.68
0.765
0.769
0.394
1.624

MAX
10.08
13.49
14.12
9.78
5.33
9.65
6.86
5/76
9.08
2.7
4.63
4.7
4.45
3.59
4.62
8.89
1.9
2.67
2.39
9.91

S.D.
2.321
2.763
2.543
2.384
1345
2.413
1.314
1.487
2.862
0.856
1.427
1.527
1.322
0.951
1.101
3.198
0.607
0.912
0.807
1.672
SITE 19
CL
6
9
7
8
5
19
11
14
10
4
1
4.
3
17
16
18
12
15
13
20
No.
90
82
73
41
39
24
33
10
29
8
15
14
11
17
4
6
8
-
<
149
%ofPre
26.43
19.93
14.93
8.24
7.1
4.36
3.05
2.64
2.62
2.6
2.52
1.35
1.35
1.19
0.63
0.39
0.36
0.17
0.14
0
TOTAL
217.13
163.74
122.71
67.73
58.35
35.84
25.07
21.69
21.56
21.4
20.67
11.11
11.09
9.75
5.16
3.18
2.93
1.38
1.15
238.39
AVG
2.413
1.997
1.681
1.652
1.4%
1.493
0.76
2.169
0.743
2.675
1.378
0.794
1.008
0.574
1.29
0.53
0.366
0.46
0.23
1.6
MAX
9.12
13.49
14.12
7.22
4.85
5.46
6.86
4.77
2.46
9.08
5.76
3.59
2.84
2.59
2.64
1.65
1.19
1.1
0.51
8.15
S.D.
2.314
2.755
2.285
2.033
1.31
1.253
1.306
1.598
0.814
3.705
1.917
0.922
0.857
0.685
0.914
0.606
0.375
0.557
0.221
1.894
D-8

-------
SITE
CL
11
7
9
6
12
8
5
10
1
19
3
17
2
16
4
18
14
15
13
20
20
No.
114
87
83
94
45
59
51
50
39
27
26
22
17
12
16
18
10
12
12
232

%ofPre
19.17
11.64
9.82
9.5
8.17
6.86
5.61
5.46
5.22
3.87
3.53
2.57
2.01
1.65
1.51
1.24
0.94
0.7
0.52
0

TOTAL
188.25
114.28
96.47
93.3
80.23
67.41
55.07
53.64
51.24
37.99
34.7
25.24
19.77
16.25
14.82
12.18
9.25
6.87
5.15
368.04

AVG
1.651
1.314
1.162
0.993
1.783
1.143
1.08
.073
.314
.407
.335
.147
1.163
1.354
0.926
0.677
0.925
0.572
0.429
1.586

MAX
8.28
8.77
8.13
4.8
7.11
9.02
4.95
5.36
3.83
3.49
3.71
5.63
4.38
3.49
5.31
2.59
1.93
1.72
1.85
9.02

S.D.
1.475
1.278
1.316
1.203
.459
.576
.071
.187
.142
.037
0.849
1.301
1.319
1.026
1.298
0.654
0.665
0.409
0.516
1.801
SITE 21
CL
11
6
7
9
5
10
1
12
8
3
2
17
4
18
19
16
14
13
15
20
No.
122
112
92
91
62
69
39
47
57
27
24
25
15
24
29
13
14
11
12
256
•fcofPre
16.6
10.92
9.78
9.35
9.17
9.11
6.8
5.86
4.26
3.89
2.31
2.26
2.05
1.75
1.7
1.62
1.23
0.73
0.62
0
TOTAL
190.93
125.59
112.51
107.51
105.5
104.82
78.21
67.39
49
44.7
26.62
25.97
23.6
20.14
19.53
18.68
14.17
8.36
7.12
416.18
AVG
1.565
1.121
1.223
1.181
1.702
1.519
2.005
1.434
0.86
1.656
1.109
1.039
1.573
0.839
0.673
1.437
1.012
0.76
0.593
1.626
MAX
7.66
7.29
6.45
7.29
7.66
14.45
6.17
5.45
3.99
5.1
6.07
5.24
6.37
3.62
1.99
3.73
3.4
1.6
1.65
14.4
S.D.
1.433
1.259
1.153
1.136
1.478
1.965
1.371
1.156
0.926
1.118
1.456
1.215
1.584
0.954
0.609
1.287
0.899
0.515
0.476
1.88
D-9

-------
SITE
CL
11
6
9
5
10
1
7
3
12
2
17
g
4
14
18
16
19
15
13
20
27
No.
Ill
121
106
65
61
37
81
27
43
27
27
51
17
16
21
12
17
6
8
248
%ofPre
16.24
14.77
11.77
9.94
8.3
6.9
6.57
3.84
3.78
3.49
2.84
2.84
2.37
1.82
1.58
0.99
0.82
0.65
0.48
0
TOTAL
159.66
145.19
115.73
97.69
81.6
67.84
64.62
37.78
37.2
34.29
27.91
27.88
23.32
17.91
15.54
9.77
8.05
6.41
4.71
281.27
AVG
1.438
1.2
1.092
1.503
1.338
1.834
0.798
1.399
0.865
1.27
1.034
0.547
1.372
1.119
0.74
0.814
0.474
1.068
0.589
1.134
MAX
6.05
5.84
5.82
5.84
5.52
5.36
4.19
4.11
3.27
3.82
4.06
3.13
3.83
4.51
2.82
1.71
2.44
3.21
1.53
5.86
S.D.
1.277
1.291
1.173
1.142
1.142
1.539
0.854
0.996
0.761
1.252
0.983
0.62
1.161
1.251
0.75
0.611
0.652
1.167
0.44
1.053
SITE 28
CL
6
9
11
5
7
10
1
3
17
4
2
12
8
14
19
18
13
16
15
20
No.
Ill
78
90
62
63
54
34
24
28
15
23
34
30
15
15
13
12
10
10
198
%ofPre
15.33
13
11.17
10.74
9.82
6.92
6.09
4.75
3.52
3.27
3.2
2.77
2.67
2.27
1.12
1.11
1.01
0.7
0.56
0
TOTAL
130.86
111
95.38
91.66
83.79
59.04
51.97
40.54
30.07
27.89
27.34
23.67
22.79
19.34
9.52
9.44
8.64
5.94
4.78
261.51
AVG
1.179
1.423
1.06
1.478
1.33
1.093
1.529
1.689
1.074
1.859
1.189
0.6%
0.76
1.289
0.635
0.726
0.72
0.594
0.478
1.321
MAX
5.17
9.42
5.36
8.36
6.56
4.29
6.6
5.16
5.46
4.18
3.32
2.12
3.22
2.86
2.54
1.63
1.76
1.25
2.3
9.42
S.D.
1.143
1.622
0.95
1.505
1.465
0.964
1.622
1.198
1.131
1.245
'0.853
0.626
0.933
0.823
0.802
0.422
0.569
0.547
0.675
1.391
D-10

-------