DOC
EPA
United States
Department of
Commerce
National Oceanic and
Atmospheric Administration
Seattle WA 98115
United States
Environmental Protection
Agency
Office of Environmental
Engineering and Technology
Washington DC 20460
EPA-600/7:80-168
October 1980
            Research and Development
            A Comparison of the
            Mesa-Puget Sound
            Oil Spill Model with
            Wind and Current
            Observations  from
            August 1978

            Interagency
            Energy/Environment
            R&D Program
            Report

-------
                RESEARCH REPORTING SERIES

Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology.  Elimination of traditional grouping  was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:

      1.  Environmental Health Effects Research
      2.  Environmental Protection Technology
      3.  Ecological Research
      4.  Environmental Monitoring
      5.  Socioeconomic Environmental Studies
      6.  Scientific and Technical Assessment Reports (STAR)
      7.  Interagency Energy-Environment Research and Development
      8.  "Special" Reports
      9.  Miscellaneous Reports

This report has been assigned to the INTERAGENCY ENERGY-ENVIRONMENT
RESEARCH AND DEVELOPMENT series. Reports in this series result from the
effort funded  under the 17-agency Federal Energy/Environment Research and
Development Program. These studies relate to EPA's mission to protect the public
health and welfare from adverse effects of pollutants associated with energy sys-
tems. The goal of the Program is to assure the rapid development of domestic
energy supplies in an environmentally-compatible manner by providing the nec-
essary environmental data and control technology. Investigations include analy-
ses of the transport of energy-related pollutants and their health and ecological
effects; assessments of, and development of, control technologies for energy
systems; and integrated assessments of a wide range of energy-related environ-
mental issues.
 This document Js available to the public through the National Technical Informa-
 tion Service, Springfield, Virginia  22161.

-------
             A COMPARISON OF THE MESA-PUGET SOUND
               H   OIL SPILL MODEL WITH WIND
                   AND CURRENT OBSERVATIONS
                       FROM AUGUST 1978
                              by

                       Robert J. Stewart
                        Carol H. Pease
            Pacific Marine Environmental Laboratory
              Environmental Research Laboratories
        National Oceanic and Atmospheric Administration
                      3711 15th Ave. N.E.
                  Seattle, Washington  98105
Prepared for the MESA (Marine Ecosystems Analysis) Puget Sound
    Project, Seattle, Washington in partial fulfillment of

           EPA Interagency Agreement No. D6-E693-EN
                 Program Element No. EHE625-A
                   This study was conducted
                    as part of the Federal
                Interagency Energy/Environment
               Research and Development Program
                         Prepared for
           OFFICE OF ENERGY, MINERALS, AND INDUSTRY
              OFFICE OF RESEARCH AND DEVELOPMENT
             U.S. ENVIRONMENTAL PROTECTION AGENCY
                    WASHINGTON, D.C.  20460

                         AUGUST 1980

-------
                       Completion Report Submitted to
                PUGET SOUND ENERGY-RELATED RESEARCH PROJECT
                  OFFICE OF MARINE POLLUTION ASSESSMENT
              NATIONAL OCEANIC AND ATMOSPHERIC ADMINISTRATION
                                     by
                  Pacific Marine Environmental Laboratory
                    Environmental Research Laboratories
              National Oceanic and Atmospheric Administration
                            3711 15th Ave. N.E.
                         Seattle, Washington 98105
     This work is the result of research sponsored by the Environmental
Protection Agency and administered by the National Oceanic and Atmospheric
Administration.

     The National Oceanic and Atmospheric Administration does not approve,
recommend, or endorse any proprietary product or proprietary material men-
tioned in this publication.  No reference shall be made to the National
Oceanic and Atmospheric Administration or to this publication in any adver-
tising or sales promotion which would indicate or imply that the National
Oceanic and Atmospheric Administration approves, recommends, or endorses
any proprietary product or proprietary material mentioned herein, or which
has as its purpose to be used or purchased because of this publication.
                                     ii

-------
                                  CONTENTS

Figures	   iv
Tables  	    v
Abstract	   vi
Acknowledgements   	  vii

1.  Introduction   	    1

2.  Conclusions 	    2

3.  Recommendations  	    4

4.  Tidal Current  Simulation  	    6
          Functional Validation of Model  	    6
          Accuracy of the Model's Interpolation Scheme  	    7

5.  Significance of the Current Model Errors  	   13
          CASE 1:  No Knowledge of Tides	   14
          CASE 2:  No Knowledge of Nontidal Current	   19
          CASE 3:  Existing Model with Assumed Steady Current 	   22

6.  Wind Field Simulation	   23
          Significance of the Wind Model Errors	   26

7.  Drifter Response Studies  	   28

8.  References	   32


APPENDICES

    A.    Calculation of Tidal Phase and Amplitude from Model
          Coeficients	   34
    B.    Moments  for Randomly Oriented Constituent Displacements ....   36
                                      111

-------
                                  FIGURES

Number                                                                     Page

 1.       Location of Straits 11, 12, and 13, and the weighting
          factors applied to Parker's stations   	   38

 2.       Dispersion envelopes assuming no knowledge of tidal
          currents (Case 1)	   39

 3.       Tide model contribution to reducing position error
          relative to Case 1	   40

 4.       Dispersion envelopes associated with ignorance of
          high-frequency nontidal current oscillations (Case 2)  	   41

 5.       Observed net currents at various locations in the
          Strait of Juan de Fuca and the Central Basin	   42

 6.       Dispersion caused by randomly oriented net current
          of  .66 km/hr (Case 2)	   43

 7.       A comparison of Case 1 and Case 2 errors	   44

 8.       Difference-current dispersion ellipses at  1, 3,
          and 10 hours	   45

 9.       Wind Pattern No. 3	   46

 10.       Wind Pattern No. 4	   47

 11.       Wind Pattern No. 5	   48

 12.       Wind station locations	   49

 13.       Principal components of surface wind observations  	   50

 14.       Scatter plot of principal component weights  	   51

 15.       Scatter plot of wind observations, sorted by
          pattern type at four central basin stations  	   52

 16.       a.  Average residual currents on August 25,  1978
               (residual = drifter-model)  .....  	   53

          b.  Average residual currents on August 26,  1978
               (residual = drifter-model)  	   54
                                    iv

-------
                                   TABLES




Number                                                                    Page
1.
2.
3.
4.
5.
A comparison of a tidal analysis of model output with
an analytical prediction based on model coefficients 	
A comparison of simulated tidal currents with actual
A comparison of linear interpolation errors with

Parameters for the discrete frequency simulation of
the nontidal oscillatory currents 	
8
q
1?
18
20

-------
                                  ABSTRACT
     This  report  compares the  winds and currents observed  in  August  1978 in
the Strait  of  Juan de Fuca with  simulated  wind and current fields taken from
the MESA-Puget  Sound  oil spill model.  This model is described in a companion
report, Pease (in press).   A method is developed for relating these errors in
velocity to uncertainties in predicted position.  The tidal current subprogram
of the  oil  spill  model  is shown to reduce the uncertainty in trajectory posi-
tion by an amount that  is somewhere  in the range of 50%  to 90% of the total
uncertainty  that  can be  caused by ignorance of the tides.   It is also shown
that  the  uncertainty  in trajectory  position  is  strongly affected  by  our
inability  to predict the baroclinic motions in the region.  Over times less
than 10 hours,  the dispersion is  mainly  tidal,  and the tidal current subpro-
gram contributes significantly to  the prediction of position.  After 10 hours,
however,  the bulk  of  the  dispersion  is  due  to the  low-frequency (periods
longer  than a  week) baroclinic motions.  These  baroclinic motions are poorly
understood,  and a  program  of basic  research  directed at illuminating their
causes and  statistical properties  is called for, if predictions are to be made
over  periods  longer  than  10 hours.   The   regional  wind model  developed by
Overland,  Hitchman, and  Han  (1979)  and used as a subprogram in the model is
compared with wind  observations from a short period of time.  We conclude that
the  selection  of  a master station  for use in scaling the pattern strength
cannot  be  done  in an arbitrary fashion.  We also find that the repertoire of
patterns presently  available  in the program library is  not sufficiently com-
prehensive  to allow reliable modeling of the surface winds.
                                     vi

-------
                              ACKNOWLEDGEMENTS
     The authors are most grateful to Ms. Rita Chin who has labored diligently
over  the  computer analysis  of  the many diverse data sources incorporated  in
our study.  Ms. Chin also prepared the draft figures and was of general assis-
tance throughout the course of this project. Our thanks also to Ms. Sue Larsen
who prepared this manuscript and Ms. Virginia May who prepared the figures.

     We are indebted to Dr. James E. Overland, Mr. James R. Holbrook, Mr. Carl A.
Pearson, and Dr. Harold 0. Mofjeld, all  of  PMEL,  for  their contributions of
data, computer programs,  and meteorological and oceanographic  expertise.
Mr. Pearson was most helpful in providing his time and computer programs for
the tidal and spectral analysis of the various current velocity time series.
Mr. Holbrook provided us with the current meter observations from Straits 11,
12, 13 and further gave us his own analyses of these data which we have used
in several places.  Mr. Holbrook's interpretations of the data and his review
of portions of our manuscript were of great value to us.  Dr. Mofjeld has pro-
vided advice on a number of items, and his review of our manuscript resulted
in several changes which improved the report.
                                    vii

-------
                                  SECTION 1

                                INTRODUCTION


     In early 1978, an oil spill trajectory model was completed at the Pacific
Marine Environmental  Laboratory (PMEL)  for  the MESA-Puget  Sound project of-
fice.  This model  is  described in Pease  (in  press).   The model was developed
under the  direction of Dr.  Jerry A.  Gait, then of PMEL.  The model was devel-
oped for two  main  purposes.   In the event of an actual oil spill, it is to be
used to assist  in  the cleanup operations.  In  this  mode the model would pre-
dict locations  and times  of arrival of portions of  the spill.  This informa-
tion could  then be used to improve the  deployment  of cleanup equipment.  The
model is also intended for use in a  simulation capacity.   The site selection
of new  petroleum transshipment facilities, refineries,  or  storage facilities
involves the  consideration of a number of factors,  including probable environ-
mental  impacts.   It was  anticipated  that the  model predictions  would  be of
value in comparing alternative facility locations.

     Given  the   oceanographic  complexity of  the region  the creators of  the
model did  not aspire  to comprehensiveness.   However, they did wish to exploit
the substantial  literature on tidal currents that was available as a result of
a  long-term  program  of current meter  studies by  the National  Ocean  Survey
(NOS) (Parker,  1977;  NOS,  1973a;  NOS, 1973b; NOS,  1979;  and by Parker within
Cannon  (ed.), 1978).   It  was felt that a fine-scale model based on interpola-
tion  and  extrapolation of  the NOS tidal harmonic  analyses  would be a  good
start toward  the development  of a capability to predict  and simulate current
behavior.

     The details of the  surface  winds  were also little  understood.   But  the
concurrent  development  of  a  regional  wind model (Overland,  Hitchman,  and  Han
1979),  offered   hope  of simulating this  complicated variable  at a  level  of
detail commensurate with the intended resolution of  the model.

     Thus,   the  model   development  was  undertaken not  as the  final  step  in a
definitive  summary of  regional oceanography and meteorology, but  rather  as a
first step  at attempting  to integrate these phenomena.  The philosophy under-
lying this  effort  rested  mainly on the engineer's empiricism,  "let's do what
we can," rather  than a more scientific appraisal of  the possibilities.

     In late  August of 1978, a cooperative oceanographic experiment was under-
taken in  the region just  north of Dungeness Spit  (see  Frisch,  Holbrook,  and
Ages, 1980).   The  experiment  consisted  of drifter motion  studies,  CODAR  and
current meter  measurements,  CTD  casts,   and  surface wind  measurements.   The
results of  this  experiment provided a good opportunity to examine the wind and
current simulation portions of  the  oil  spill  trajectory model,  and so this
study was begun.

-------
                                  SECTION 2

                                 CONCLUSIONS
     Two questions  were  posed:   is the current  simulation  accurate enough to
be useful,  and  is  the wind properly  modelled  using the pattern method?  Both
questions  can be  addressed  from  two  viewpoints.   First,  is  the technique
properly implemented  in the  existing model,  and  secondly, is  the modelling
concept  valid?   The  implementation question  is simply an  independent check
that the model  operates  on its input data and coefficient arrays in a fashion
consistent  with  the equations that comprise the model.  The modelling concept
question is  deeper  and more philosophical.  The model was  created to exploit
the availability of both the tidal current  data and the regional wind model.
Little  thought  has been  given to the adequacy  of  these representations.   We
have attempted  to  quantify the contributions these submodels make to reducing
the uncertainty  of  a trajectory prediction, and we have made recommendations
for both the interpretation of existing model results and subsequent modelling
efforts.

     A  major difficulty  we encountered  in  addressing  the  modelling concept
question was  the determination of what constituted an acceptable error.  This
problem  included both the definition of  the  error  in  terms of  the various
parameters  that  enter the  problem and the  interpretation  of the error param-
eter.   We  have  attempted  a  novel  analysis  of  this  problem  for  the current
simulation  subprograms.   We  have  shown that  a  useful  measure  of trajectory
dispersion  is  the  variance  of  the  time-integrated  difference  between  the
actual  and  simulated velocities.   This  integrated difference-velocity  is  a
displacement, and  the variance of this displacement  is  a  measure of the area
surrounding  a  predicted position  in  which a drifting object  is likely to be
found.

     In  order to put into perspective the  tidal current subprogram's contri-
bution  to  trajectory prediction calculations, we have analyzed  the displace-
ment  errors for three  illustrative cases.  Case  1 considers trajectory pre-
dictions  made  with  perfect  knowledge  of  nontidal  currents but completely
ignoring  tidal  currents.   The  area of  the  resultant dispersion  is shown in
Figure  2.   Figure  3 depicts the reduction in the tidal  dispersion that should
be  achieved through  use of the  model,  based on our  estimates of the model's
errors.  Figures 4, 5, and 6  treat the converse case,  case  2,  in which tra-
jectory predictions are  made with perfect knowledge  of  tidal currents and no
knowledge  of the  nontidal  currents.   Figure  6 graphically  illustrates  the
importance  of the  nontidal,  slowly varying currents  caused by  density vari-
ations  within the region.

     These  slowly  varying  currents appear as net  flows  in the  15- and 29-day
current meter observations that are used  to estimate the  amplitude and phase

-------
of  the  tidal-current  constituents.   These  baroclinic motions  are  not  well
understood,  and  modelling  them appears  to  be  a  topic  of  state-of-the-art
research.   There seems  to  be no way of  dealing  with these currents except to
hypothesize  a  particular steady-current  field.  Figure 8 depicts  the trajec-
tory dispersion ellipses we can anticipate assuming it is  possible to select
the correct steady-current field.

     The study plan  for the wind field analysis  was developed around the  idea
of  taking  a detailed  look at  the  wind  field during  a short,  five-day  time
interval.   This approach provided us with good insights regarding the qualita-
tive performance of the model.  We found that for the particular conditions of
the test,  the model did not reproduce the fine-scale features seen in the  wind
observations.  This was  particularly true  in the region immediately  off  Port
Angeles.  We also  examined a method of  simulating  a  time  history of the  wind
and found  that  the arbitrary selection of Race  Rocks  as  a  master station was
not supported by the data.   A principal-component analysis  of a portion of the
data was performed.

     An attempt  was also  made  to reconcile  the Evans-Hamilton drifter  data
(Cox, Ebbesmeyer,  and  Helseth, 1979)  using  the  combined  wind and  current
fields.   The effort  was unsuccessful mainly because of the  effects of model-
ling errors.   It was  possible to detect a  current  reversal  off New Dungeness
on  26 August  1978  by  calculating the  average of the  difference  between the
observed  drifter  velocities  and the  model's predicted tidal-current  veloc-
ities.

-------
                                  SECTION 3

                               RECOMMENDATIONS
     The procedural  necessity  for making a best-effort analysis of the poten-
tial environmental  effects  of  a proposed development  has  led to the creation
of oil-spill models  not just for Puget Sound, but for a wide variety of loca-
tions.  The model  examined  here is representative, and we believe our conclu-
sions regarding  the  accuracy of the wind and current simulations have general
application.   Our  primary  recommendation  is thus that  the  users  of an oil-
spill model  should  be  alert to the possible inaccuracies.   If  the model has
been developed  for  a hydrodynamically complex region  such as the one studied
here, then  one  must remain  skeptical of any simplistic  interpretation of the
model's results.   In the Strait of Juan de Fuca, for example, the growth over
time of the position error exhibits a strong dependency on the slowly varying,
baroclinic  components  of  the   current;  and  these  current phenomena  are not
simulated  by  the  model.   Thus,  the model  contributes  beneficially to  the
prediction of a drifter's location only over brief time intervals.

     We found  that  for prediction intervals of of 1 to 5 hours, the model can
reduce position uncertainty  due to tides by a factor of about two, as compared
to predictions  using no tidal-current model.  During  this time, the position
uncertainty caused  by unknown  baroclinic currents is  small,  but growing, and
in  the  time  interval  from 5  to 10 hours,  uncertainties due  to baroclinic
currents  grow  to   overwhelming importance.   These  results  are,  of course,
specific  to the region studied, but the  principle can  be  applied  to  other
locations.

     Techniques for  using the model in an actual spill need to be developed to
accommodate these  short-comings if the  model is  to  be of any assistance.  If
the  tide  model had  only very  small  errors,  then in the  event  of a spill it
would be  feasible  to  simply subtract  the  simulated  current from an instan-
taneously  measured   current, and  so to estimate  the  important slowly varying
component.  However, as shown  in Table 4,  the  amplitudes  of the model errors
are  in  the range of 10  cm/sec  to  30 cm/sec for the important M2 and Kx con-
stituents,  thus even  a 20-cm/sec  steady current will be  masked by the model
error.  Nor can we  avoid this difficulty by using some other simple scheme to
interpolate between the tidal-current  reference  stations, as demonstrated in
Table 3.   Therefore, our second recommendation is that additional research be
devoted to either improving  the accuracy of the interpolation scheme or devel-
oping a  practicable observational scheme that will allow  the determination of
the  slowly varying currents over short time  intervals,  such as  1 to 6 hours.
Figures  I6a and  I6b suggest that  it might  be  possible   to  develop a hybrid
system  for  this purpose based on the use of 20 to 40 drifters coupled with the
existing  tidal-current model.    These  drifters should be  released  on  a 2-km
grid in the  subject area  and  positions  would have to be determined hourly,

-------
irrespective  of  the visibility.  Alternatively, a mobile  CODAR with complete
data reduction and analysis systems might be developed.  A research program of
this type is essential if we are to develop any long-term predictive capabili-
ties for use in a spill situation in the Strait of Juan de Fuca or the central
basin.

     The  current  subprogram's  shortcomings  are, of course,  caused by  its
failure to model  or otherwise account for  the  slowly varying baroclinic cur-
rents.  These  currents have  only recently been appreciated  for  their impor-
tance to pollutant  transport in the region.  Topics  that  might profitably be
explored  include baroclinic  effects  on  the  tides;  analytical  or  numerical
simulation of the coastal intrusion phenomena; and further field studies using
CODAR and  current meter  arrays.   It would be premature to expect  that these
studies would  immediately lead  to an improved  modelling  capability, but  it
might be helpful to use modelling as a focus for the  work.

     The wind  field simulation was also found to  be  rather inaccurate.   How-
ever,  this problem  continues  to be  studied  at PMEL as  part of the  Marine
Services program,  and progress  has  been made  over  the past  two years.   The
wind  field simulation  model  should be updated in  the  near future to  take
advantage  of  this  work.   There are  some  difficulties,  however,  that  will
require special attention.  Foremost is the problem of establishing a suitable
statistical framework on which  to perform the  comparison  of observation  and
simulation.   Recent  work in  nonparametric tests of multivariate  processes
should  be  examined to  determine whether they can help  solve this  difficulty
(cf.  Friedman and Rafsky,  1979).  An automated procedure  for pattern selec-
tion  is also required so that  statistical comparisons can be made using suffi-
ciently large numbers of samples.

     Finally,  we must  point  out that  this  study has  dealt with winds  and
currents.  These parameters, in  conjunction with topographical considerations,
are  undoubtedly  the  factors  that cause oil  spill  transport.   However,  the
equations that combine these phenomena and produce an oil transport prediction
are  still little  known.    It is  of vital importance  that the  direction  of
future  research  be  compatible with this appreciation of the problem.   Specif-
ically,  studies  of winds  and currents should be  limited to  wind and current
phenomena.  Combining them in  an "oil  spill" model seems to be pointless.   The
requirement  that the wind  and  current  models  be computerized and  of  known
accuracy should suffice to ensure their ultimate compatibility in an oil spill
model.

-------
                                  SECTION 4

                          TIDAL CURRENT SIMULATION
     Three current  meter  arrays were deployed in  the  central basin region on
July  16,  1978.  These  stations are  referred  to  as Straits 11,  12,  and 13.
Each station included a VACM current meter at a depth of 4 m, and each current
meter acquired enough data to allow at least a 29-day tidal harmonic analysis.
We have used this current meter data as the principal standard for judging the
accuracy of  our  tidal-current simulation routine.  The current meter data was
obtained from  Mr.   James  Holbrook of the PMEL Coastal  Physics  Group.   It was
subsequently analyzed using the R2SPEC computer program which is maintained by
Mr. Carl Pearson also of the Coastal Physics Group.

     The output  of the  R2SPEC analysis  includes  the  amplitude,  direction of
flood (ebb), and phase lag of the various tidal constituents.  The analysis is
done  both  in  east-west,  north-south components,  and  in terms  of components
oriented along the  major and minor axis of the tidal ellipse.  The major axis
of this  ellipse  determines the flood/ebb direction.  The major-axis represen-
tation of  the  motion was most  suitable  for  a  comparison with the tide model,
since the  tide model assumes all  motions lie  along the major  axis  of the M2
constituent  (Pease, in press).

FUNCTIONAL VALIDATION OF MODEL

     Although  the model  had been extensively debugged  and  tested in previous
projects,  the  availability of  R2SPEC gave us another  means of verifying the
proper operation of the  model.  Further, the  R2SPEC  test  was not redundant
with  earlier efforts to  validate the  correct  functioning  of  the model.  We
therefore proceeded with the test.

     The model generates  its  simulated currents  by summing as  many as three
reference station currents, each weighted by a factor lying between 0 and 1.0.
The  current  at the  reference station is in turn generated by the summation of
the  five  tidal  constituents  analyzed  and  tabulated   by  Parker  (1977).   To
perform  the  test we generated (U,V), (east-west, north-south), time series for
a  29-day period  at  the three current meter  station locations.   The simulated
time  series at  Strait  11 was  created with  the  sum  of  a .6  weighting on
Parker's Station 21 and a  .4 weighting  on  Station 30.   The Strait 12 current
was  generated  with  the sum of  a  .3  weighting  on  Parker's Stations 28 and 42,
and  a .4 weighting on Station  29.   Strait  13  was created with a weighting of
 .2 on Parker's  Stations  31 and  38,  and .6 on Station  30.   These synthetic
currents were  then  analyzed by R2SPEC, and the amplitude and phase of the five
tidal  constituents  of  the simulated tidal currents were determined.  Figure 1
shows the  location  of Straits 11, 12, and 13, and  the locations and weights of
the  reference  stations used in the summation.

-------
     An  alternative method  for  determining  the  amplitude  and phase  at the
three stations is, using trigonometry, to determine the amplitude and phase as
analytic functions of the weighting factors and the reference stations' ampli-
tudes and phases.  This derivation is outlined in appendix A.

     The comparison for the 5 tidal constituents used in the model is shown in
Table 1.  The  calculated phase and amplitude, equations (A4a), (A4b), and the
equivalent R2SPEC estimates are very similar for the M2 and Oa constituents at
all  three  locations,  and  they are  reasonably  close for  the S2, N2,  and Kx
constituents.  The  reason  for the slight disagreement in the latter constitu-
ent  estimates  is that  R2SPEC includes corrections  for  errors normally caused
by   the  presence  of  the  host  of  other  tidal   constituents.   These  other
constituents  are  not  present  in  the  model's  simulated  tidal  velocity,
equation  (A5),   and  so  these "corrections"  are  deleterious.   If  they  were
nullified,  the  analysis would  compare  even more  closely  with  the calculated
phase and amplitudes.   In  any event, the comparison  is  sufficiently close to
state that  the  model's  simulation of tidal  currents  accurately  reflects the
data taken  from  Parker  (1977) and stored in  the  various arrays.   There is no
time base error  and no indication of transformation  errors  in going from the
major-axis  coordinates  to  east-west,  north-south  coordinates.   We note  in
passing that the minor-axis velocity estimates from R2SPEC were uniformly zero
for  all  constituents  and locations,  reflecting exactly the model's simulation
process.
ACCURACY OF THE MODEL'S INTERPOLATION SCHEME

     We  know  that the model  operates  consistently on the data  stored  in its
coefficient arrays.   It  is  not certain,  however,  whether  these coefficients
have been  selected to create an accurate simulation of tidal  currents  at an
arbitrary point.   In  order to judge the  merits  of the model from this second
viewpoint, we  compared  the three model time  series  with  currents observed at
Straits  11,  12,  and  13.   Through a minor  communication  failure uncovered in
the  final  writing of this  report,  the model locations for  Straits  11  and 12
were taken to  be 48°13'N, 126°6'W and 48°20'N, 122°58'W,  respectively.   These
locations are  in error by about 2  km  from the sites of the July 16th deploy-
ment.  It was judged that the comparison was still valid,  however, since tidal
phase, amplitude,  and  direction of flood change very little over distances as
small as 2 km in this area.

     Table 2  shows the flood direction,  major-axis  amplitude,  GMT  phase, and
minor-axis amplitude for the M2, N2, S2, Kt, and Ox constituents, as estimated
from  the 29-day records  obtained  at  Straits  11,  12,  and  13.   Immediately
beneath  these  values  are  the  equivalent  parameters  as  calculated  from the
simulated time  series obtained from the model.

     In  general,  we can  see that  the model's  simulated tidal currents cor-
respond  reasonably well  with  the  observed  currents.   The  errors  in  flood
direction range  from 1° or 2° to 42°, with a typical value being approximately
10°. The simulated major-axis amplitudes are uniformly low, with the discrep-
ancy ranging  from a few  cm/sec, to  15 cm/sec in the K! constituent at Strait
11.  The percentage error  of this  deficiency ranges from 15%  or 20% for the

-------
                                                                                          TABLE  1
                                                      A COMPARISON OF A TIDAL ANALYSIS OF MODEL OUTPUT WITH AN ANALYTICAL  PREDICTION
                                                                                BASED ON MODEL COEFFICIENTS
00



H2

FLOOD
LOCATION
STRAIT
**
11
STRAIT
***
12
STRAIT
13

TYPICAL

CAL'C'D
R2SPEC
CAL'C'D
R2SPEC
CAL'C'D

R2SPEC
DIFFERENCE
DIR
OT
80°
80°
42°
42°
95°

95°
0°
amp.
cm/sec
36.7
36.8
36.5
36.1
33.5

33.6

-------
                                          TABLE 2
A COHPARISON OF SIMULATED TIDAL CURRENTS WITH ACTUAL OBSERVATIONS AT STRAITS 11, 12,  AND 13
Station Location
48°14'I123°6' Observed
Strait
11 48°13',U3°6' Simulated
4802r,122057' Observed
Strait
12 48°20',122<>58' Simulated
48° 19' ,122059' Parker 29
48°14' ,122°57' Observed
Strait
13 48°14, 122°S7' Simulated
Major Epoch Minor
°T cm/sec GMT cm/sec
92 43.9 296 1.4
80 36.8 249 0
65 45.6 319 21.9
42 36.6 311 0
60 31.2 312 .7
96 41.2 302 8.5
95 33.6 281 0
Major Epoch Minor
°T cm/sec GMT cm/sec
89 10.9 265 .6
80 6.3 249 0
75 10.6 294 4.4
42 6.9 307 0
62 6.0 298 1.1
96 8.1 269 .3
95 5.4 274 0
Major Kpoth Minor
°T cm/ sec GMT cm/sec
82 11.5 231 3.8
80 9.9 252 0
64 11.0 352 4.2
A2 10.8 323 0
74 9.3 14.1 3.3
109 17.9 252 3.H
95 8.8 298 0
Major Epoch Minor
°T cm/ sec GMT cm/sec
78 30.3 219 3.3
80 15.5 120 0
84 15.4 '238 4.1
42 13.7 188 0
4li 16.3 222 3.7
109 26.5 225 8.7
<)r, 13.5 170 0
Major Epoch Minor
"T cm/si-c GMT cm/sec
85 18.2 186 3.3
80 15.4 164 0
56 11.9 199 1.7
42 8.9 185 0
52 11.6 187 .3
105 15.0 178 2.7
95 9.3 161 0

-------
M2,  S2,  and Oj  constituents, to  40% or 50% for  the  N2 and Kt constituents.
The error in the GtfT Epoch value ranges from 5° or 10° to 40° or 50° (with one
error of  100°  in the Kx  component of Strait 11), but there  appears  to be no
pattern  in  this error  except that the Kj  constituent  is uniformly bad.  The
neglect  of  the  minor-axis  velocity in  the model  leads to errors  that are
typically 5 cm/sec  or less, with  the exception of the M2 component of Strait
12.


     A  portion  of  the  discrepancy between  the observed  currents  and the
model's  simulated  currents  may be due to  the differing depths  of Parker's
reference stations  and  the  current  meters  we  are  using for the comparison.
Parker's  stations  vary  in depth from 5 m to 22 m, whereas Straits 11, 12, and
13  are  at 4  m.   Over  such  a  range of depths we  know  that  the direction and
amplitude of  a tidal constituent  can change as a  result  of baroclinic effects
on the tide.   Directional changes  of  10°-15° and amplitude changes of 5 cm/sec
are  common in  this  region in the top  20 m to 30 m.   Because there is as yet no
known simple,  proven method for predicting  this  discrepancy, we shall ignore
it for the present  analysis.


     It  seemed possible  that the  particular  selection of weights  is solely
accountable for  the errors described  above.  It can  be seen from Figure 1 that
the  Strait  12 position and  Parker's  Station 29 are close together. Thus, one
would think  it possible to  use Parker's  Station 29 as a basis for predicting
Strait 12 currents.  The  consequence  of this selection may be seen in Table 2.
The  major-axis amplitude deficiency is exacerbated in the  M2  and S2 compo-
nents,  and  partially corrected  in  the  remaining  constituents.   On the other
hand, errors in  direction of flood are uniformly reduced, and the Epoch values
are  significantly  better for the  Kj,  S2,  and N2 components.  However, errors
still remain,  and it appears that we cannot expect  to achieve perfection from
a  simple  revision of the  coefficent arrays.


     The  determination  of the weighting factors and the  station indexing were
done manually by  the  authors of  the model  in a somewhat  subjective way. A
simple  objective approximation to  the process behind the station indexing and
weighting is  that of a  linear interpolation.   Are the errors seen at Straits
11,  12,  and 13  consistent with this  interpretation? If  not, can we expect to
diminish  the errors seen  in  Table  2 by using a more  objective weighting scheme
based on  linear  interpolation between stations?


     Table  3  shows the results of a simple analysis of  Parker's station data
which  suggests  an  answer  to  these  questions.   Parker's  station locations
were studied  and  eight station triplets were  found where all three stations
fell nearly on a straight line.  The  outermost stations were  then used to pre-
dict the M2 amplitude  and phase of the middle station.  The weighting factor
applied  to  a  predictor station was  simply the  fractional  distance from the
middle  station.   The  distance  between  the outermost  stations  varied from
12.6 to  39.6 kilometers,  and the weighting  factors  were  typically in the
range  of .45  to  .55, reflecting the fact that the  middle station was usually
about  halfway between  the predictor stations.   Because the predicted current
exhibits  an  error  in  both  phase  and magnitude,  we require  two  measures of
                                     10

-------
the error.   The least  ambiguous way to write  this error is as  the  sum of a
component that is in phase with the reference signal and a component that lags
by 90°   Thus:
                cos at - a cos (at + 6) = bl cos at + b2 sin at
(1)
where a is the linear interpolation estimate of a, and 6 is the phase error of
the  estimate.   We do  not consider the effect  of  the statistical uncertainty
that  clouds  our  knowledge  of the  amplitude and  the phase of  the  reference
signal, as well as our knowledge of the amplitude and phase of the predictors.
Equation (1) may be solved for bl and b2 with the result:
                                  = a cos (6) - a
(2a)
                               b2 = a sin (6)
(2b)
These bj  and b2 values for the eight triplets are shown in Table 3 in the two
far-right columns.

     The  bj  and b2  values for the  Straits  11,  12,  and 13 are  also  shown in
Table 3,  and we can see that their  range of values is not unlike that seen in
the  linear  interpolation  scheme.   There is  not  a sufficient number of data
points to allow us the luxury of a statistical test of the hypothesis that the
errors of both the model and the linear interpolation scheme are from the same
distribution.   However,  as  a  practical  matter  it  appears very  likely that
there is no significant difference between the methods.
                                     11

-------
                                                                                TABLE 3




                                                A COMPARISON OF LINEAR INTERPOLATION ERRORS WITH OBSERVED MODEL ERRORS
ro
Predictor Stations
Reference St. St.
Station No. Wt. No. Wt.
4 2 .47 5 .53
5 4 .51 8 .49
7 6 .50 8 .50
12 11 .49 14 .51
20 16 .39 30 .61
21 22 .44 20 .56
22 17 .54 26 .46
25 26 .52 78 .48
Strait 11
Strait 12
Strait 13
Reference St
Dist. Amp
(Km) cm/sec
38.0 41.8
39.6 42.7
13.0 47.3
12.6 72.7
28.2 47.3
16.7 48.9
20.0 63.7
13.9 67.0
43.9
45.6
41.2
. Analysis
M2
Phase
(LT)
39°
61°
54°
70°
74°
67°
84°
67°
296°*
319°*
302°*
Predicted Coefficients
M2 M2
Amp . Phase
cm/sec (LT)
31.4
43.2
43.0
80.5
41.8
54.3
64.5
68.6
36.8
36.6
33.5
63.5
49.8°
62.3°
59.0°
54.1°
79.1°
69.8°
90.4
249°*
311°*
281°*
Error
bl b2
(inphase) (90° Lag)
cm/sec cm/sec
-13.2
-.3
-4.8
6.3
-7.8
4.2
-1.2
-4.0
-17.9
-9-5
-9.5
13.0
-8.4
6.2
-15.3
-13.7
11.4
-15.8
27.2
25.9
5.1
10.9
t
                   *GMT PHASE LAGS

-------
                                  SECTION 5

                 SIGNIFICANCE OF THE CURRENT MODEL ERRORS


     As we  have noted,  tidal  currents created by the model  have  errors that
are a  sizable  fraction of the observed tidal currents.  It is not clear, how-
ever, whether  these  errors are important when viewed  in light of the model's
intended use.   In  fact,  the problem of what  is  an "acceptable" error has not
been discussed previously.   This  is a critical  deficiency  in our  thinking on
the  subject,  and  to  remedy this deficiency, we  offer the  following analysis
both  to help  critique  the  existing model and  to guide subsequent efforts.

     We have found that  one parameter useful as an error criterion is the mean
squared error  in predicted position.  This parameter is  an indication of the
area  that ought to  be  searched  if the  model is used  to  locate  a drifting
object  some  period of time after  it  has been  released.   If  this  number is
small,  say on  the  order of  1  (km)2,  then we can go to the location predicted
by the  model  and expect the drifter to be nearby, perhaps within sight of the
search  vessel.   Alternatively,  if the number is large, say 100 (km)2, then we
should  anticipate  substantial  difficulties  in  locating the  object.   Inter-
preted  another way,  if we run the model in a simulation mode, then the growth
of the mean squared  error with  time  makes it more and more  unlikely that a
model  prediction is  a truly representative trajectory.   This is particularly
true if the trajectory lies near a beach or some other geographic feature that
might catch the  drifter  or channel it into a particular region.

     Let us consider three illustrative cases.

 Case 1.  Trajectory predictions are made with no knowledge of tidal
          currents, but  with perfect knowledge of all nontidal currents.

 Case 2.  Trajectory predictions are made with no knowledge of the non-
          tidal  currents, but with perfect knowledge of the tides.

 Case 3.  Trajectory predictions are made with the present model with
          the  addition that we have perfect knowledge of the very-low-
          frequency (weekly and lower) currents.

     The first case is unreasonable in any practical sense because it supposes
that  we have  a perfect understanding  of wind-driven and  baroclinic current
phenomena that we  are  only now learning to identify and categorize and are far
from describing in a  statistically useful sense.  However, this case offers a
simplified  view of  the problem that  isolates  and helps  define the  role a
tidal-current  model  plays  in  trajectory calculations.   Specifically,  we can
use  this  case  to  relate errors in tidal-current  amplitude  and phase to mean
squared position errors  in locating drifting  objects.
                                     13

-------
CASE 1;  No Knowledge of Tides
     In addition to perfect knowledge of nontidal currents, we assume for case
1 that  amplitude and  phase functions are  independent  of position.  We assume
also that the trajectory calculation is to begin at a random time.  With these
qualifications, the  displacement due to one tidal constituent is given by the
time integration of the velocity.  This has a particularly simple form when we
are dealing with sinusoidal tidal constituents:
                               Udt =
                                            cos
                                                       9£) dt,
 which can also be written as:

            *• v   A *
_* sin(a0t
or-      *
where

and
                                   T2 _
                                      "
                                       • a*
                                TO = t2 -
                                                                         (3)
 Because  TI  is random, £_ is also random.  With trigonometry, we can transform

 equation  (3)  into the following:
                            =  £ V2 (1-cos
 where
 We let
                         = cos

                      sina^Tp
                                   V2  (1- cos
                                                 cos(|
                                                       (4)
                                                                          (5)
and  note that this parameter  is  simply the maximum excursion possible in time
period  T  for a particle being  advected  by a sinusoidal  current.  The  random
phase variate, |.,  is uniformly  distributed over  (0,2n).

     Based  on equation  (4),  the  distribution  of D.do)  is:

                             1                ,  -6«(tJ S D  S 60(T  )       (6)
 The first two  moments  of this  distribution may be  shown to be:

                                        1  = 0
                       og
                                               = %6J(T0)
                                                        (7)

                                                        (8)

-------
The first  result  (7)  is a  consequence  of the symmetry of  equation (6); the
second  result  (8) is  the  parameter we use to  characterize  the dispersion of
the case 1 assumption.

     These results are for the major-axis excursions of one tidal constituent
only.    If  we neglect  all  minor-axis motions,  the  displacement of a particle
that is  acted  on by a number  of  tidal constituents will be given by a summa-
tion of the form
                                      M
                              D(T0) = I  D (T0,4 )                       (9)
                                     £=1  *     *

The random phase  variates,  £„,  are  related  to  one  another by  the starting
time,  Tj, via the modulo operation,

                          4k  = mod  (akTi  +  6k>  2n)                      (10)

     There is some uncertainty on our part as to the nature of  the statistical
relationships  that exist between the random phase variates  defined by (10).
On one  hand,  equation (10) is  somewhat  analogous  to  computer  algorithms that
generate random  numbers.   On this basis, we believe that an argument could be
made that the phases are statistically independent provided the frequencies in
(10) are  incommensurate.   Unfortunately we know that the frequencies selected
to represent  tidal phenomena are not independent, and that there are a number
of simple relationships that link them.  For  example, the sum of the 0} and Kx
frequencies exactly equals the M2 frequency.  These linkages result in charac-
teristic tidal waveforms.   In Seattle, for example,  the  M2,  Oj, and Kj tides
often combine so that  consecutive diurnal high  waters are of nearly equal mag-
nitude  while  the intervening low waters are  of much different size.  Because
of the  linkage  in frequencies, the opposite pattern  occurs  only rarely.  So
from this  standpoint,  the  phase variates are to be expected to have some very
complicated,  multidimensional, statistical  dependencies.   We  have  not been
able to resolve  this problem, so we have opted to assume that  the phase vari-
ates for the different tidal constituents are independent.

     The importance  of this assumed independence  is  that we may use the cen-
tral limit theorem (CLT) to deduce the approximate  form of the  distribution of
D(to)-   Tne  simplest  form  of the  CLT  says that  the sum of  n independent,
identically  distributed, random  variates  with finite  mean  and variance con-
verges  to a normal (or Gaussian)  distribution in the limit as n goes to infin-
ity.  The  theorem also  holds for  variates  distributed  according  to a scale
factor  transformation  of a common distrubution function.  Equation (6) is of
this type with 6(T0) being the scale factor.

     It is  well  known that the percentile ranges  of a normal distribution are
related to  the square root of the variance,  the  standard deviation.  Thus, if
a2  is  the  variance of the  distribution on x,  then 68% of the distribution
Ills within ± a    and  95% lies between ± 2o   .  In analogy, and with reference
to the  CLT, wexshould expect  the variance ^f  our summed  variate to be func-
tionally related to  the percentiles  of that distribution.  Because the distri-
butions of  the  summands  have square-root  singularities at the  extremes  of
their range,  and because the  number of  summands  is small  (only five), the CLT
                                     15

-------
provides only  a  rough approximation to the  actual distribution of the summed
variate D(TQ).   Thus,  we cannot be certain that precisely 68% of our observa-
tions  will  fall within  one standard deviation of the mean.   However, exper-
ience with the CLT suggests that the bulk of the observations will fall within
one standard deviation,  and it is in this loose sense that we assert that the
standard deviation  is  related to a "characteristic" error of position.  Reit-
erating, we  assume  that the summands are independent, thus the summed variate
D(TO)  will be  distributed in an approximately  Gaussian form,  and this allows
us to  put  a  frequency interpretation on  the standard deviation that supports
our contention that the standard deviation of D(TQ) is an appropriate measure
of dispersion.  Given the  standard  deviation as  the  radius of the circle of
dispersion,  the  variance is thus proportional to  the  a-rea in which the drift-
ing object is  likely to be found.

     Also  because  of  the  assumed  independence  of the  summed  variates,  the
variance of  the sum may be related to the variances of the summands, equation
(8).   This follows  because of the well-known  result that the variance of the
distribution of the  sum of two independent random  variables  is  given by the
sum of the variances of  the summands,
                              a2  = a2  + a2   ,
                               ss    xx    yy
where s = x + y.
This result is exact and not dependent on the CLT approximation.  Thus, extend-
ing this result to five summands and applying it to our problem,


                      a2  = I %6|(T0) = Z/Vf (1 - cosa-to)
                       DD       £                       £
This  completes our  argument that the  standard  deviation (the square root of
the  variance)  of D(TQ)  provides a  useful and  readily calculated geometric
measure of the  dispersion.

      Case  1  assumes  we make our  predictions with  no knowledge of the tidal
currents.   We  interpret this  to mean  that  the major  axis  of tidal-current
motion has no  favored angular orientation.  Under the  assumption that all  five
tidal constituents  share a  common  major  axis,  the  radius  of dispersion is
determined from the  square root of (lla):

                                                       V
                                                                         (lib)


      Figure  2 shows this R(TQ) variate at Straits  11,  12, and 13. The  ampli-
tude  values  used  in the presentation  are taken from the July 16th analysis
 (see  Table  2).   The circumference  of  the circle  of dispersion  is  shown at
hourly  intervals  for  TO varying from  1 to  12  hours.  This half-day limita-
tion  on the  use  of  equation  (lib)  has been  chosen so  as to  accommodate the
restrictions  that are implicit in our  assumption that the phase and amplitude
                                     16

-------
parameters are  constant with respect to  displacement.   In an actual realiza-
tion, as  a particle is transported away  from the release point, the amplitude
and  phase parameters will  change.  To  illustrate  this advection  effect,  we
have sketched the dispersion circle about mean-flow streamlines taken from the
CODAR data over the period 23  to  25  August 1978.  Notice that the dispersion
curve from  Strait  13  encompasses the  Strait  11 site  at  hours  5  through 12.
Particles advected  into the vicinity of  Strait  11  would no longer have tidal
amplitudes and  phases  like  those at Strait 13  which  was  their release site,
but  rather would behave like particles  released at Strait 11.  These effects
would play an important and perhaps dominant role in long-term tidal displace-
ment calculations.

     Figures 19  and 20 of Parker  (1977) show that the M2 and Kj tidal ellipses
are nearly always in close alignment in this region.  Thus the assumption of a
shared major axis is not a bad  one.  It is possible, however, to calculate the
variance  of  the distribution of a sum  of random variates in which each tidal
current variate  can assume an  arbitrary  spatial orientation with respect to a
reference axis.  That is,

                              5
                      D(TO) = 2 60(T0)cos0 cos(40 + d)
                             £=1 s,        a      a


where <().  is the  angle between the reference axis and the variate's major axis.
We assume that  
-------
                                                                           TABLE  4
                                                           TIDAL CONSTITUENT ANALYSIS OF MODEL ERRORS
00
STATION ,
STRAIT 11
STRAIT 12
STRAIT 13
Constituent
M2
N2
S2
Kl
°1
M2
N2
S2
Kl
°1
M2
N2
S2
Kl
0,
Major Amp. (cm/sec)
30.5
9.7
6.3
4.6
28.1
9.6
1.8
3.8
1.8
6.8
28.9
13.7
2.1
8.8
17.5
Minor Amp. (cm/sec)
1.9
1.0
4.4
2.6
.9
1.1
.9
1.0
.3
1.8
.3
2.2
.3
.9
5.8
Flood Dir. (°T) Model Dir (°T)
84
111
18 80
58
79
90
148
112 42
84
119
98
125
155 95
98
116

-------
     Although the  assumptions  of case 1 are too coarse to provide a practical
guide to  the  dispersion problem, they are simple to understand, and therefore
they serve  a  useful conceptual purpose.  Moreover, case 1 lays the foundation
for an  interesting  comparison between the no-knowledge conditions  of case 1
and the  uncertain  knowledge  situation we face  when using  the tidal-current
model.    In  this real-life  situation, errors  in trajectory position  will be
caused by inaccuracies  in the model's tidal current simulation, as well as by
phenomena not  included  in the model.  The analysis above may be used to esti-
mate the  relationship  between the tidal-current errors and  the error in pre-
dicted position.    Table  4  shows the  R2SPEC analysis  of  the tidal-current
signal formed  from the difference between the observed tidal currents and the
simulated currents.  We consider  these difference currents to be random errors
associated  with the model's  predictions.   The  source of  these error signals
can be traced  back to the phase  and amplitude errors of Table 2 already dis-
cussed,  and to the error  in  the  alignment  of  the  major, or  flood,  axis.
Further,  these  errors are  augmented by the  model's explicit  neglect of the
minor-axis tidal velocities.

     The  last column in Table 4 shows the tidal-current flood direction speci-
fied in  the model.  Notice  in the Strait 12  simulation that both the major-
axis amplitude  errors are  small, and the direction  of  the  difference veloc-
ity's flood is randomly  scattered  relative  to  the model's  flood direction.
This scattering in the difference velocity flood  direction  is  what we expect
if  the  model had  no systematic  errors.   As  we  can readily see  from either
Table 2  or  the correlation of the difference velocity flood direction and the
simulated velocity flood  direction in Table 4,  some  sort  of systematic error
is  present  in  the model's  simulation  of  the Straits  11  and  13  velocities.
Nevertheless,  the  uniform  scattering assumption  provides  a useful  tool for
analyzing errors associated with  the model.

     Figure  3   shows the  dispersion associated  with tidal-constituent errors
of  Table 4 under  the assumption that the major  axis of motion  is randomly
oriented.   The diagram  is  rigorously correct  only for.  Strait  12,  but the
features  of the dispersion  envelopes of Straits  11  and  13  are qualitatively
correct  if  we  allow for  a  moderate (up to 33%) elongation  of  the circles in
the flood direction, and an equally sizeable  reduction  in the  direction per-
pendicular  to  the  flood.   This diagram illustrates  the  contribution that the
model makes to reducing  uncertainty in the predicted position of a drifting
object,  assuming  all  nontidal  currents  are  known exactly.   At best,  and
after  a  few hours, the model  predicts the location to within about  1  km at
Strait  12.   At Strait  13,   the  radius of the circle  of uncertainty is about
3 km after  5 hours,  and at Strait 11 the radius  is about 4.5 km after 5 hours.
This corresponds   to a  reduction in search area of  90%-95%  at  Strait 12, and
(perhaps) 50% at Straits  11  and  13.

CASE 2

     Case 2 is the converse of  case 1.  We  now assume  that drifter position
predictions  are made in ignorance  of all  nontidal motions,  but with complete
knowledge of the tidal  currents.   Again we are forced to limit  the duration of
the trajectory  prediction to small times, say  12 hours, to prevent out-running
our localized knowledge.
                                     19

-------
     Unlike  the  tidal currents,  the physical  mechanisms  behind the nontidal
currents are diverse and  poorly documented.   The  simplest  way to categorize
these motions  is in terms of  the manner in which  they  appear  in our current
observations.  One class of motion has periods  shorter than one hundred hours,
and these motions  can be resolved as frequency components in Fourier analyses
of the  current  records  now  available.   As a general  rule,  these motions are
weak; a  typical  root-mean-square amplitude for all the motions falling in the
frequency band between the daily and semidaily tides is  only  3  or 4 cm/sec.
The  other  class of  motion is  comprised of the long  period  oscillations that
appear  as  mean  values  in our  15-  and  29-day current  observations.   These
motions  are  not truely  steady   currents.   In all likelihood,  these  slowly
varying  currents are associated  with changes in the  density structure of the
region.   The  density structure  is  principally  dependent  upon fresh-water
runoff  from  the Fraser  River  and  also from  the  onshore advection  of low-
density  surface  waters by winds  acting on  the  region lying west of the coast
of the  state of Washington.   These motions  are generally of large amplitude,
and the  intrusions of low-density surface lenses have a sporadic appearance in
the current  records.

     The  estuarine  runoff  is perhaps  the  most predictable  feature  of  the
region,  and   in the  Strait  of  Juan  de  Fuca  at  midchannel,  the  average
outflowing current varies from  a minimum value of 10 cm/sec to  30 cm/sec in
late winter  to a high of 30 cm/sec to AO cm/sec in late summer (Cannon, 1978,
Figure  27).  This  flow is also  seen  in  CTD sections across the Strait and in
satellite images of  surface-water properties.   Superimposed on  this outward
flow are intrusions  of  low-density lenses of  water  advected to  the mouth of
the  Strait by  winds acting offshore.  These  intrusions  may  occur rather fre-
quently, and they  can result in eastwardly  (up-channel)  flow along the south
shore of the Strait of Juan de  Fuca with speeds of up to 50 cm/sec and dura-
tions of many  days.   The CODAR  images of the Port  Angeles region on 26 August
1978  document  the complex  behavior of  the upstream front  of  one  of these
intrusions (Fris-ch, Holbrook, and Ages, 1980).

     The dispersion caused by the weak, high-frequency motions  that are re-
solved  in our  current records may be estimated using a procedure that is en-
tirely  analogous to that used on the tides in case  1.  We  group the current
speed energy in  the various frequency bands into five discrete frequencies and
model these  motions  as though each discrete frequency was the equivalent of a
tidal  constituent  in case 1.  Table 5 shows the parameters  we have used for
this representation of the nontidal motions.  The values in this table reflect
a Fourier analysis of the kinetic energy at Straits  11, 12, and  13 as seen in
the  16  July 1978  deployment  (Holbrook,  personal  communication).*  We assume
for  this case  that the orientation  of this  motion is random and so use equa-
tion  (12b)  to calculate  the variance of the drift position.   Fiqure A shows
the  dispersion  envelopes  for these  representations.  Notice  that  the error
associated with  the  neglect of  these motions  is  a very small fraction of the
case 1  error,  and that it is  comparable with  the  tide model dispersion enve-
lope  at Strait  12  (Figure  3).   These  currents  are probably caused  by the
oscillating  wind  stress  that  acts on the region.   These stresses result in
highly   localized  surface boundary  layers,  and in  weak barotropic current

* Mr. J. R.  Holbrook, NOAA/PMEL,  Seattle, Washington.
                                    20

-------
                                  TABLE 5

             PARAMETERS FOR THE DISCRETE FREQUENCY SIMULATION OF
                      THE NONTIDAL OSCILLATORY CURRENTS
Freq. Band
(cyc/hr)
.008 - .016
.016 - .024
.024 - .032
.05 - .07
.09 - .2
Center Freq.
rad/hr
.075
.125
.176
.377
.911
Amplitude
Strait 11
2.2
2.2
2.2
3.5
8.1
of Discrete Sinusoid (cm/sec)
Strait 12 Strait 13
2.2 2.0
2.2 2.0
2.2 2.0
3.5 3.2
8.1 7.4
fields that  extend over the whole region.  It is reasonable to consider these
motions omnidirectional.
     Figure  5 depicts  the  magnitude and  direction of  the  average  current
observed at  a number of locations within the central basin and Strait of Juan
de Fuca  regions.  These  vectors are  taken from Figure  38  of Cannon (1978).
With  the  exception of  six  vectors  taken from the mouths  of Rosario  and Haro
Straits, which show a uniform out-strait flow, the remainder are widely scat-
tered  and  show  no obvious  channeling.  These averages are  the  net flow seen
over  various periods ranging from  10 to 30 days, and we reiterate that they
are not  good  estimates of  the true average because  of  this relatively short
record length.

     Although these  vectors are not good estimates  of  the true, statistical,
average current,  it is reasonable to  presume  that  they are representative of
the net current one might encounter in any given  drifter  experiment.   The
vectors themselves come  from  many different locations,  and  it  would make no
sense  to  try  to combine these values in an  elaborate  statistical model.   We
therefore limit  our analysis to the  simple determination of the average mag-
nitudes of  the observed net current  and let this value  characterize the ran-
dom,  net  drift  we should  expect  in  a  drifter  experiment.   This average was
found to be  .66  km/hour.

     Figure  6 shows three groups of concentric circles centered on Straits 11,
12, and  13  respectively.   These circles depict the hourly transport caused by
an  average  drift  of .66 km/hour.  The  circular  geometry reflects our uncer-
tainty in determining the direction of the  net drift.  These circles are to be
interpreted  as a  characteristic transport distance.  If there  were no tidal
currents  or  other  motions, and  if  the  streamlines were  straight,  then  a
drifter would, on the  average, be located somewhere on the circumference of
one of these circles.  Figure 4 and  Figure 6 are the counterpart to Figure 2.
A  comparison of Figure 4 and Figure  6 shows  that the very-low-frequency, den-
sity-related motions depicted  in Figure 6 are the  critical  elements in the
case  2 problem.

      A comparison of the  area of the  12-hour  envelope of  the net  drift portion
of  case-2  to the  case-1  envelope  (which encompasses all position errors from
the release to hour  12), shows  them to be of  comparable size.   A better com-
                                     21

-------
parison is  to  plot the radius of the position error as a function of time for
the  two  cases.  This  is  done  in Figure  7,  where we can  see  that the error
associated with  complete  ignorance of the tides  (case  1) is somewhat greater
than  the  error  caused  by ignorance of  the  nontidal  currents,  for trajectory
durations  less  than  10 hours.   After  that time,  the  case 2  current errors
become dominant.  The  contribution of  the  high-frequency,  nontidal oscilla-
tions in this  figure can be seen  to be of only modest importance.

CASE 3:  Existing Model with Assumed Steady  Current

     The uncertainty in position due to  our ignorance  of  the very-low-fre-
quency currents  (Figure 6)  is so  great as  to  require  the adoption of some
hypothesis  regarding the  behavior of  these currents.    The  alternative,  the
treatment of these currents as a  simple  random process, results in a region of
uncertainty that is so  large as to  obscure whatever insights the tidal-current
model might provide.  This is not a novel observation.  In a previous study of
oil  spill  trajectories  in the region (Stewart, 1978), the results were framed
in light of the  contrast between  a  no-steady-current hypothesis and a uniform,
ebb-directed,  steady-current hypothesis.  The necessity for such an assumption
that is provided by Figure 6 is simply another way of arguing the same point.

     The case  3  assumptions are  that we know these long-period currents. The
purpose  of this analysis  is to  outline the region of uncertainty associated
with the use of  our model in such a situation.  There are two ways we can pro-
ceed.  We  can  combine our previous analytical  results,  principally Figures 3
and  4,  or we  can  integrate  the  differences between   the observed current
velocities  of  the  simulated current velocities to  estimate the time behavior
of the error.   We have adopted the latter method because it is less dependent
on the many assumptions required  by the  analytical method, and because it will
serve as a check on our previous  calculations.

     Three  difference-current  time series were created  and then divided into
48-hour  intervals.   An initial time was randomly  selected  within each inter-
val.  We  then  integrated the difference velocity for durations of  1, 3, or 10
hours  to determine  one realization of  the  error  growth.  Using distinct 48-
hour  intervals,  we  acquired  between 20 and 30 independent  realizations for
each  of the three durations,  at all three stations.    The resultant  (x,  y)
positions  were then analyzed to  determine the variance-covariance matrix, and
the  coordinates  of  the ellipse  containing  68% of  the  observations were then
calculated  assuming the  (x,  y)  variates  were drawn from  a bivariate normal
distribution.   This  ellipse  was  compared  with  the scatter  plot  to ensure
qualitative agreement with the data.

     Figure 8  depicts these ellipses for  Straits  11,  12, and  13 at durations
of  1, 3,  and  10  hours.   The  steady flow used in  creating these figures was
taken from the net flow in the  current meter observation. Notice that Straits
11  and 13 have  flows consistent with the CODAR streamlines used in our pre-
vious  representations,  but the net flow at  Strait  12 is  oriented in the oppo-
site direction.   This  graphically illustrates  the importance  of  the slowly
varying  currents,  and it gives warning of  the dangers in  assuming knowledge of
this phenomenon.
                                     22

-------
                                  SECTION 6

                            WIND FIELD SIMULATION
     The  wind field  in  the region  is  modeled using a  time  series  of hourly
wind velocity observations at a master station and a large-scale wind pattern
that extrapolates  the wind velocity to areas that are distant from the master
station.  The wind pattern is determined using a  numerical model that calcu-
lates  the winds that would result from the application of a steady pressure
gradient across the region (Overland, Hitchman, and Han, 1980).  This submodel
accounts  for the  topographic  channeling of the dense  marine boundary layer.
The spatial  resolution of the program is such  that the major features of the
Olympic Mountain Range are discernable, but hills such as those that are seen
in  the city  of Seattle  do not appear.   The cost  of  computing these regional
wind patterns is  great, and so  the  model uses a  library  of  8 patterns,  each
selected  to  represent a  characteristic  pressure  gradient condition.  A se-
quence of pattern  types is selected for the period of the master station wind
velocity  observations, based  on the evolution of the  regional pressure  gra-
dient.   This  is presently a tedious and somewhat subjective manual task.   The
wind simulation  is thus  based  on empirical observations,  but the interpola-
tion/extrapolation junction i-s determined objectively from a physically based
numerical  model.   In this  respect  the  wind  model  differs  from the tidal-
current model in which the interpolation methods were based on manually deter-
mined  coefficients of uncertain  underpinning.

     Some numerical  algorithm is  required  to  combine  the wind  patterns and
master station observations.   Stewart (1978),  in an oil-spill study using the
MESA-Puget  Sound  model, developed  a technique wherein a  scaled  and rotated
hourly difference  velocity was applied throughout the region.  This technique
caused  the  simulated velocity  at  the  master  station  to  equal  the observed
velocity, and at  surrounding points the  simulated velocity changed  in ampli-
tude and  direction in a  fashion  determined by the  wind pattern.   A charac-
teristic  of  this method of extrapolating the observed velocity is that if the
pattern velocity  equals the  observed velocity, then the pattern velocity is
applied unperturbed throughout the region.

     The purpose of this algorithm was twofold.  First it caused the simulated
wind  field  to  vary  hourly.   These  high-frequency  variations  would not  be
present in the simulation if  the  wind  field had been  created by using  a se-
quence  of patterns.   Secondly,  the  algorithm assured  agreement  between the
simulated and observed winds at  the master station.

     A complete evaluation of the wind field simulation should be based on an
examination of both the patterns and the  algorithm used to turn these patterns
into extrapolation and interpolation functions.   However,  for this  study,  we
have judged it sufficient  to examine just the characteristics  of the patterns.
                                    23

-------
We have  taken  this course because Stewart's algorithm was selected largely on
subjective grounds,  and  whatever merit the algorithm might have, it is predi-
cated  on the  assumed agreement between  the model's  wind patterns  and the
observed winds.   If the  patterns are  a poor  fit to the data, then any agree-
ment between the  algorithm's velocities and the  observations  would be purely
fortuitous.  Little  would be learned  if the  fit  was less than perfect.  Fur-
ther,  an examination  of  the pattern  errors should  uncover  any simple method
for perturbing  the wind  patterns so as to achieve the purposes of the extra-
polation algorithm.

     Because the  determination of the pattern sequence for an extended period
is a  laborious  task, we  decided  to  closely examine the wind field simulation
over  a  short  time interval.  To  be  compatible  with the analysis of the joint
experiment  data,   the  period  22-26  August  1978  was  selected.  Large scale
isobaric  weather   maps and  the  hourly wind  observations  at Race  Rocks  were
obtained and given to  the wind model's creator, Dr. James E. Overland of PMEL,
who  determined a pattern  sequence  based  on 6-hour pattern  intervals.   The
period  selected for study turned out  to have rather anomalous weather for the
summer  months,  but we judged it  to  be as fair a  test as any other that might
be devised around  a  five  day period.   The pattern  sequence for this period was
determined to be:

                   6,5,5,5,5,2,2,3,3,3,3,4,4,5,5,3,3,4,4,4

where the integer  serves  as an index to the pattern type.  It can be seen that
the  bulk of the  interval was  accounted  for by patterns 3,  4,  and 5.  These
patterns  correspond to  pressure gradients such  that  the  large scale, extra-
regional flow is  from  the south,  southwest, and west respectively.  Figures 9,
10, and  11 show the  regional wind fields for these cases.

     To  see whether  there  is  a basis for  separating the  time  series  into
segments, we performed a  principal component analysis of the wind observations
from  Smith Island,  New  Dungeness,  Race  Rocks,  and  Tatoosh  Island.   These
locations  are  shown in   Figure  12.   This analysis  revealed that  42% of the
variance observed at the  four  stations could be accounted for by perturbations
to  the mean wind  speed  that would simultaneously  lie to the east-northeast at
Tatoosh,  south at  Race  Rocks,   west-northwest  at New Dungeness,  and north-
northwest at Smith Island.  The  mean wind velocity and these perturbations ere
shown  in Figure 13a.  This  result may  be interpreted as follows.  Under condi-
tions  that would  cause  the wind at Tatoosh to  become  more southerly (in the
meteorological  sense), the  wind at  Smith Island  would  swing a little to the
east,  the wind at Race  Rocks  would  swing to the north  from its mean north-
westerly direction,  and  the wind at New Dungeness  would swing to the southeast
from its mean southerly  value.

      The second principal component  is depicted  in Figure 13b, and  it  accounts
for  18% of the observed  variance.  Notice  that positive perturbations  to this
component  will  partially cancel  positive  perturbations to the  first principal
component  at  Tatoosh and  Smith Islands,  while  enhancing  the northerly and
easterly swings at  Race  Rocks and New Dungeness  respectively.  Alternatively,
if the  second component is  given  a positive  perturbation  while the  first
component  is  negatively perturbed,  then  the  Race Rocks  and  New Dungeness
                                     24

-------
velocities would  tend to remain constant while the wind at Smith Island would
drop  and  swing south,  and the  wind at Tatoosh  would drop  and  swing to the
east.  The third principal component accounted for 14% of the variance, and it
is the last  significant descriptor of the data, the remaining, smaller compo-
nents  accounting for  9%,  7%, 6%,  3%,  and  1% of the  variance respectively.
This component is shown in Figure 13c.  Whereas the first two principal compo-
nents  caused variations of  approximately equal  size  at all  four  locations,
this  third  component  is  effective mostly at  the two  eastern-most stations,
Smith  Island and New  Dungeness.   It  can be  seen that negative  amplitudes
applied to this  component will  result in  a  westerly rotation and increase in
wind  speed at Smith Island, and a calming at New Dungeness.  The wind at Race
Rocks  simultaneously  increases and swings to the north.  This third component
thus represents some kind of eddy in the lee of the Olympic Mountains.

     An obvious  question is whether these components  are related to the pat-
tern  sequences.   Figure 14 shows the amplitudes  of  these  three components at
three  hourly intervals.   The three plots, W2  vs  Wj,  W3 vs Wj, and  W3 vs W2,
are  two-dimensional plan views  of the three dimensional (Wj,  W2,  W3) space.
That  is,  any given wind velocity observation  can be approximated by a summa-
tion  of  the  three principal components with weights Wlf W2 and W3.  The (Wx,
W2>  ws)  space thus  provides  a  means  of  representing  the  observations.   The
bulk  of  the  observations form a cluster in the vicinity of the origin, with a
slightly  negative  average  value  for  the W2  component.   However,  the large
amplitude observations  tend to fall into  two  groups.   Observations  that were
made  during  Pattern 3 intervals have positive weights  on  the first component
and  negative weights  on the second  and third.  This  results  in an increasing
southerly wind at Tatoosh and  Smith Islands, a southerly wind at New Dungeness
that  drops  and swings  to  the  east,  and a north-northwest wind at  Race Rocks
that  increases  slightly.   Behavior of this type  would  seem to fall somewhere
in  between  Patterns  3  and 4 and  is  not  well  represented  by  either  (see
Figures 9 and 10).

     Observations  made  during Pattern 4  and Pattern  5  intervals,  that had
large  negative weights on the first component, had  large positive  weights on
the  second  component  and negative weights on  the third.  This corresponds to
an easterly  swing at Tatoosh,  a westerly swing at Race Rocks,  a slight drop at
New  Dungeness,  and a  slight westerly  swing  at  Smith Island.  The Tatoosh
behavior  does not  correspond to either  Pattern  4 or  5.   The New  Dungeness
behavior  seems more characteristic of Pattern 4  than  Pattern 5,  and yet Pat-
tern 5 intervals  had the extreme weights in this cluster.

     Another way of  looking  at the data  is to  draw a wind  velocity scatter
diagram  for  each pattern type at several  stations.  Focusing now on the cen-
tral  basin,  Figure 15 shows the wind velocity clusters at Smith Island, Point
Wilson, New  Dungeness, and Race Rocks.  Notice that the Pattern  3 data clus-
ters  nicely  at Smith Island and Point Wilson.  It is rather scattered at Race
Rocks  and strongly bimodal at New Dungeness, with the predicted wind rather
too  easterly for  the  large amplitude cluster.  Wind Pattern 4 correctly places
the  wind  velocity at New Dungeness  at  calm. Other stations tend to be rather
scattered,  although the  model predictions are indicative  of the predominant
quadrant.  Wind  Pattern 5  does  a poor job of sorting  or predicting the ten-
dency  of  the Smith Island, New  Dungeness,  and Point Wilson observations, but
                                     25

-------
it picks  up one  mode of  what appears to  be a  bimodal distribution at Race
Rocks.

     These  observations  are  too limited in number  to warrant further statis-
tical  analysis.   However, they do suggest  some  general conclusions.  First,
there  appears  to  be  a  strong  eddylike  motion  between  Race Rocks  and New
Dungeness.  This  motion  apparently results from  extraregional flows that fall
somewhere  between the Pattern 4 and  Pattern 5  boundary condition.  Further,
the bimodal behavior  of  the wind at Race Rocks and New Dungeness suggests that
there  is   some  threshold  value of  pressure  gradient  below  which light and
variable winds prevail.   Once this threshold  is  exceeded, the model's predic-
tions are  somewhat more  reliable, but  still subject to relatively large errors
in direction.  The  behavior of the wind at Tatoosh under Pattern 4 and 5 con-
ditions is paradoxical,  but  perhaps  this  simply reflects the  importance of
sub-grid-scale topography  on local winds.

     The fundamental  precept  of the wind model is that  a substantial fraction
of the  regional  wind patterns  can be  represented using  a library containing a
fairly small number of patterns based  on time-steady pressure  gradients.  This
assumption is  critical,  and  it requires  a  great  deal more study.   In the
five-day  period   we  examined,  the principal  component  analysis of  the wind
observations at  four  stations determined that  74% of  the  variance could be
accounted  for by using  three  wind patterns.   Unfortunately the three patterns
determined from   the  model  did not  show much   agreement with  the principal
component  patterns.

     Another important point  that can be made regarding the wind  field simu-
lation  is  the importance  of  matching the  master  station  selection with the
wind field behavior.  None of  the stations that we examined had uniformly good
agreement  with  the three  pattern types that occurred during the  observation
interval.  For example,  only one Race  Rocks observation  in Pattern  3 intervals
fell  within ten degrees  of   the predicted  wind.  By using Race  Rocks  as a
master  station for  Pattern 3  winds, we essentially negated the good fit seen
at the other three stations (Figure 15), since a  large error velocity would be
determined from  the  difference between the  Race Rocks  observations  and the
model predictions, and this error would then  be applied  to the other stations.

     More  generally,   the  selection  of an  algorithm for combining observed
winds  and  the regional  patterns appears  to be  a  difficult problem.  If the
wind field actually has  a threshold for its  response to the large-scale pres-
sure gradient, as suggested by the histograms  at  Race Rocks and New Dungeness,
then the algorithm will  have to  create very small-scale, weak  winds for condi-
tions  below the  threshold, coupled with patternlike  winds once the threshold
is exceeded.  The statistical  properties of the sub-threshold  local winds will
have to be studied to determine spatial coherence scales and  the characteris-
tics of the time  variability.

SIGNIFICANCE OF  WIND  MODEL ERRORS

     Because  of   the  limited   amount of data  and the subsequently  qualitative
nature of the  analysis,  there  is  no reasonable method for  estimating in a
 statistical sense the magnitude of the simulated wind velocity errors.  Thus,
                                     26

-------
even if we postulated a formula relating wind and drifter velocities, it would
not be possible to make quantitative estimates of the dispersion caused by the
model errors.  However,  it is possible to draw some conclusions regarding the
adequacy of the simulation method.

     To use  the model  in its present state, with  its  limited  pattern reper-
toire,  is to  chance misestimating  the wind by 45° to  180° in certain  key
regions.  The  area  off  Port Angeles is one such region when the extraregional
wind is from the south to southwest.  The model predictions could be in error
by  180°;  and with a 3% wind drift factor, the resultant drift could easily be
in  error  by  15  cm/sec  to  30 cm/sec, or  an amount equivalent  to  the steady
baroclinic currents.
                                     27

-------
                                  SECTION 7

                          DRIFTER RESPONSE STUDIES

     When we  first  proposed this work, we  envisioned  an analysis of the Cox,
Ebbesmeyer  and Helseth  (1979)  drifter  data.   This study was to  reveal the
response  of  drifters  to surface winds,  and it  was to be  the  subject  of a
separate  report.  Because  of the relatively large errors in the tidal-current
and wind models, the proposed analysis was not successful.  Although this work
may  seem  somewhat  tangential  to the present report,  it does  provide further
insight into  the importance  of the model  errors.   Certainly,  any model that
cannot make good hindcasts  must be of questionable validity  in  the forecast
mode.

     The  results  of this failed investigation are important also because they
bear on the concept of the oil-spill-of-opportunity research program, one task
of  which  was  to   infer the  transport  equations  for  oil  slicks  from  data
acquired  in actual  oil spills.  In contrast to most oil spill investigations,
we had the  benefit  of many  current meter  and  anemometer records, we had good
data on the positions of easily recognized  drifters,  and we had our computer
model with  which  to perform the various calculations and interpolations.  The
fact that our analysis was inconclusive despite all these advantages ought to
be  considered in the  design of any  future spill-of-opportunity experiments.

     The  drifter data  were  taken  from  the Plates  in Cox,  Ebbesmeyer, and
Helseth   (1979).  This  data  consisted  of the  position  of  97  drift  sheets
deployed  over the period 22-26 August 1978.  Positions were determined by air-
craft overflying  the  sheets, and the error  in  position was estimated by Cox
et  al., at 31 meters.   The  data were  acquired  only during daylight hours and
the  time  between  fixes was nominally about 30 minutes.  Using the known posi-
tion, we  determined the average velocity by dividing the distance traveled by
the  time  between fixes.  This velocity was applied at the midpoint between the
known positions.  If  the time interval exceeded two hours, the point was cast
out.

     We then  sorted the data by day and by location.    The latter sorting was
based  on  a nearly  square  grid in which the north-south cell  dimension was 2
minutes of latitude and the east-west cell  dimension  was 3 minutes of longi-
tude.  This corresponds  to a square about 3.6 km on a side.  This data set was
then examined, and grid squares that over  the  five-day experiment had  fewer
than ten  distinct drifters were combined with adjacent  squares to form a cell
with sufficient data  to  allow statistical treatment.

      Because  the wind model  had  proven itself  unreliable  in this  area,  we
decided  to use the observed winds at Race Rocks and at New Dungeness for the
estimate  of the surface  wind.  The wind data from Race Rocks and New Dungeness
                                    28

-------
were stored  in our data arrays  at  hourly intervals.   The hourly value at New
Dungeness  was  the   average   of  three   readings   taken  at  twenty-minute
intervals centered on the hour.  The drifter velocities, however, were deter-
mined at  irregular times as dictated by  the times of the position fixes.  To
accommodate  this  problem,  the wind was linearly  interpolated from its hourly
value to  the time of  the drifter velocity determination.   In this interpola-
tion, we  might have  done better at  New Dungeness had  we  used the original,
twenty-minute  wind records.

     Several  data  files  were then created containing  the  interpolated wind
velocities  and a residual velocity  equal to  the difference of  the drifter
velocity  and the  tidal-current  velocity as  determined  by the  model.   These
data  files  were  then analyzed  to  see  if there was  any statistical  basis to
support a drift equation of this general form:

                                                                          (13)
             cose
                             n lu.i +  Kl *  fu  1 *  F£ll
                             j (v*J +  [v*J +  [vjj *  UJ
where  (uj»vj) is  the  drifter velocity,  (un>v/>)  *s tne  l°cal wind velocity,
(u  ,v )  is the  tidal  current velocity,  (u ,v )  is the  slowly varying baro-
  w.  t                                      S  S
clinic   current  velocity,  0'  is  an  angle  of  rotation between  the  local
surface  wind  and  the motion it induces at the surface, and e. and  e»  are
random  measurement errors.   Since  we used  a  shore-based station to estimate
the  local winds,  we  must also  include a possible  rotation  and amplification
of the shore  station wind.  That is, in any given grid cell,
      cose" -
                     " -sine"!  fu

                     "  cos«"J  Lv
                                                                    (14)
 Thus the complete model is:

      =  or [~cose
         a [sine
             "Sin6
              cose
                             l  Fuwl +  K
                             j  [v*J    LV^

                                                                         (15)
 where a = a'*a", 6=6' +6".


 Let (uf,v') be the tidal currents predicted by the model, so

 The (Up, Vp, u , v ) covariance matrix is then given by:
       J\   *»   W   W
=  a  cosB a2,    - 2 sinO cose a     + sin e a      +  a
            Vw                 wvw           w w
                                                                        (16a)
VR
    2     22

*  '   8in9Vw
                   +2 sin6 costr
                                 Vw
                                                       22
                                                     cosO a
                                                           Vw
                                                                °ee <16b>
                                    29

-------
 2              2

a   cos6 sine (a
                         u u
                          w w
                               -  a
                                         V V
                                          w w
                           222

                      ) + a    (cos 6 - sin 6)
                           U V
                            w w
                                                                          (16c)
                                                                          ^  »-y
  u
 K W
             COS0 a
  u u
   W W
              Sin6 a
                                           u v
                                            W W
                                                                          (16d)
n v
 K w
      =  a  cos6 a
 u v
  w w
       -  a sin6 a
                                         v v
                                          w w
                                                                          (16e)
                                                                          v   J
      v U
         W
           cos9 a
u v
 W W
           Sin6 CT
                                        u u
                                         W W
                                                                          (l6f)
      vv
         w
      =  a  cos6 a
                        vv
                         w w
          a sine a
                uv
                 w w
                                                                          (I6g)
                                                                          <•  &'
          -•    -
provided  u~ - u_  is neglible.  Here  (u  ,v ) is assumed the independent vari-


able, and we assume
         =  a
                                          =a
     In  addition to the  seven conditions of equation  (16),  we may also show
      22         222

that a >0, a      >  a    and a     >  a    are implied conditions.

                      ee       VRVR     8£
The  covariances  on the  left hand sides of  equations  (16b)  through (I6g) may
                                             22           2

be estimated from the data,  as well as the a     , o      , and a      terms.
                           '                  u u     u v   '      v v
                                             w w     w w         w w


Thus we  have  seven equations, three  constraints,  but  only three unknowns, Of,

6, a2  .   There  are as many  as  50 different ways  of formulating a solution in
     Co

this over-determined  system, and each should  yield  reasonably similar esti-

mates of a, 6, and a2 ,  if the model  is correct.
                    OO


     We  performed  this  analysis using  four  different  methods  for  evalua-

ting a,   6,  and  a2   for  the grid  cells lying  just  off New  Dungeness.   In
                   o£

no  case   did  we  find a  data set that  yielded  four  estimates  of a,  6,  and

a2   that  simply  satisfied  the  constraints,  let  alone  exhibit  any con-
 Co

sistency.   In short,  the model,  equation  (15),  did  not predict  the  proper

covariance structure.
                                    30

-------
     The reason  for  this failure may be  directly  attributed to the errors in
the tidal-current model.  The  variances  of  the  residual velocity components
were typically  in the range of  10   cm/sec  2  to 300  cm/sec 2.  The amplitude
errors in  the  tidal-current simulation have been  shown  to be in the range 10
 cm/sec 2  to  30   cm/sec 2,  which correspond  to  variances that  are equal in

size to  the residual  velocity variances.  Thus,  VL, - ul  is  far from negli-

gible, and, in  fact, we could argue that most, if not all, of the variability

seen in (u_, v ) was due to model error and not the wind.


     We also  performed  a  number of tests using the Race Rocks wind records,
and we  tried lagging  the  wind  one hour.  None of these  tests  resulted  in a
significant relationship.

     Figures I6a and I6b show the average residual currents calculated for the
days  of  25 and 26  August  1978, respectively.  These  figures show  the  same
intrusion  reported in Frisch et al.  (1980),  and  which  is also seen in Plates
3d and 3e  of Cox et  al.  (1979).  The tidal-current model error is smaller than

assuming ul is zero, and so averages formed using the model should be slightly

superior to those  made without the model.
                                     31

-------
                                  SECTION 8

                                 REFERENCES

1.   Cannon, G.  A.  (editor) 1978:  Circulation in the Strait of Juan de Fuca:
          Some Recent Oceanographic Observations, NOAA Technical Report  ERL-
          399-PMEL-29, Environmental Research Laboratories, Boulder,  Colorado,
          49 pp.

2.  Cox,  J.  M., C. C.  Ebbesmeyer,  J.  M. Helseth 1979:   "Surface Drift Sheet
          Movements Observed in  the Inner  Strait  of Juan  de Fuca"  Evans-
          Hamilton Inc. 6306 21st Ave.  NE, Seattle,  Washington 98115.
          Completion  Report submitted  to MESA  Puget Sound Project  Office,
          NOAA.

3.   Friedman,  J.  H.  and  L. C.  Rafsky  1979:   "Multivariate Generalizations
          of  the  Wald-Wolfowitz  and  Smirnov  Two-Sample  Tests"  The  Annals
          of Statistics Vol. 7,  New York, pp. 697-717.

4.   Frisch, A.  S., J.  E.  Holbrook, and  A.  B. Ages 1980:  "Observations of  a
          Summertime  Reversal  in  the  Circulation  in the  Strait  of Juan de
          Fuca" submitted to J G R.

5.   Godin, G.  1972:   The Analysis of Tides, Univ.  of Toronto  Press,  Toronto
          and Buffalo, 264 pp.

6.   Gradshteyn, I. S.  and I.  M.  Ryzhik 1965:  Table of Integrals,  Series and
          Products, Academic Press,  New York and London,  1086 pp.

7.   National  Ocean  Survey  1973a:   Tidal Current Charts, Puget Sound North
          Part,  NOAA, National Ocean  Survey, Rockville,  Maryland,  12 pp.  +
          endpapers.

8.   National  Ocean Survey 1973b:  Tidal Current Charts, Puget Sound Southern
          Part,  NOAA, National  Ocean  Survey,  Rockville,  Maryland,  12  pp  +
          endpapers.

9.   National  Ocean  Survey  1979:    Tidal Current Statistics for Puget Sound
          and Adjacent Waters  (unpublished data  sets),  NOAA,  National Ocean
          Survey, Rockville, Maryland,  38 pp.

10.  Overland, J. E., M. H. Hitchman, and Y.  J. Han 1979:  "A Regional Surface
          Wind  Model  for  Mountainous  Coastal Areas" NOAA  Technical  Report
          ERL-407-PMEL-32 Environmental Research Laboratories, Boulder,
          Colorado, 34 pp.
                                    32

-------
11.  Parker, B. B.  1977:   Tidal Hydrodynamics  in the Strait of Juan de  Fuca  -
          Strait of Georgia,  NOAA Technical  Report, NOS-69,  National  Ocean
          Survey,  Rockville, Maryland, 56 pp.

12.  Pease, C.  H.  1980:  An Empirical Model for Tidal Currents  in Puget  Sound,
          Strait  of Juan  de Fuca,  and Southern  Strait  of Georgia, DOC/EPA
          Interagency  Energy/Environment R  &  D  Program  Report,  in  press.

13.  Schureman, P. 1958-  Manual of Harmonic Analysis and  Prediction of  Tides,
          Coast and  Geodetic Survey Special Publication No. 98,  U.S. Govern-
          ment Printing Office,  Washington,  B.C., 317 pp.

14.  Stewart,  R.  J. 1978:   "Oil Spill Trajectory Predictions for  the  Strait
          of Juan de Fuca and San Juan Islands  for the Bureau of Land
          Managements Review of the Northern Tier Pipeline Company's
          Proposal,"   Pacific Marine Environmental  Laboratory,  NOAA Environ-
          mental Research Laboratories, Seattle, Washington.
                                    33

-------
                                 APPENDIX A

                  CALCULATION OF TIDAL PHASE AND AMPLITUDE
                           FROM MODEL COEFFICIENTS
     The tidal model is based on the 5 constituents found in Table A of
Parker (1977), M2, S2, N2, Klt and 0 .  Parker's Epochs  (or phase lag) are
measured with respect to local time (LT).  Since the model also uses LT at
the subprogram level, the phase for one of the constituents is given by:


                          Qla = (v° + u) ' Mls,

where p denotes Parker's station number, £ denotes the tidal constituent
(e.g., M2,N2, etc.)i t denotes local time, and the "equilibrium argument," (V0
+ u), is taken from Table 15 of Schureman (1958).  The time in the model is
measured relative to 0000 January 1, 1978, and, for example, the equilibrium
argument for the M2 constituent is 201.8°.

     The major axis velocity of one of the model's reference stations thus
assumes the following form:

                         V|(T) = A| cos (a^T + 6^)                     (A2)

where the superscript again denotes the reference station number, and the
subscript denotes the tidal constituent.  Note that the  model calculation is
based on arrays of 0^ . and not on the epoch, K, and equilibrium argument.
                    t ,*
     The simulated current for tidal constituent "£" at  a grid location (i,j)
is determined from the summation of as many as three, weighted, reference
station velocities.  Thus:
                      = I W(i,j,m)A^1'J»nuco.(oDT 1- eri^'^),         (A3)
                         ..         Jv-            Jt     X • J&-
                       m=l                              '

 where  the  reference  station index, P(i,j,m), and weighting  factor W(i,j,m)
 are  determined  from  arrays in which the east-west and north-south positions
 determine  the principal indices i and j.  The index  'm' is  just the  sumnation
 index  and  is of no further consequence.  Our present discussion is devoted
 to the demonstration that equation (A3) is properly implemented in the model.
                                     34

-------
     Dispensing with the cumbersome notation of  (A3) we  can  see that  our
simulated tidal current for constituent, K, is created numerically using
an equation of the general form:
                                        3             .
                      A, COS(CTT + , in terms of the b  and 6  parameters.
iiu-s is readily done by expanding tfie cosine function with its (at + 8  )
argument into sine and cosine constituents of argument.CTt, each multiplied
by sines and cosines of argument 8  , (or 8,) and the b  (or a.) amplitudes.
by sines and cosines of argument  .  ,  	
This results in two equations for ij>lw  and
                            = tan
                                              m
                                     mI1bmcosem
                                                                         (A4a)
                    a, =
                                                                         (A4b)
These are the phase and amplitude of the simulated tidal current assuming
the model is functioning properly.  They are to be compared with the
R2SPEC analysis of a time series generated by the model.

     The model creates its simulated time series through the summation of
the five tidal constituents mentioned above.  Thus,
                                    .  .
                      Vlt3(t) = Z Vj'j(t)
                                   *
                                                                         (A5)
This velocity lies along an axis determined by the flood-ebb direction.  It
is converted to u (east-west), v (north-south) components with a simple
rotation transformation.  It is this time series of u, v components that
comprise the data analyzed by R2SPEC.

     The only difficulty in this comparison is the conversion of the model
phase lag parameters  into forms suitable for comparison with R2SPEC's output.
R2SPEC lists the constituent Epochs in  a form consistent with Schureman
(1958), which is to say that it is referenced to GMT time, not local time.
The conversion is made as follows:
and
                                       u)*  ~ 8tl(see Eq-
                             ;j =:  00^j + of  (8 hrs.)
                                                                        (A6)
where T denotes GMT  time  and a.  is  constituent Jfc's frequency in degrees per
hour.
                                     35

-------
                                 APPENDIX  B

           MOMENTS FOR RANDOMLY ORIENTED CONSTITUENT DISPLACEMENTS


     If the major axis of the 2th constituent is randomly oriented with
respect to a reference axis, then the amplitude of the motion along  the
reference axis will be given by the product

                              x = 6cos<|>0 cos£_;                   (Bl)

where 6 is the amplitude of the motion,

     4>. is the angle between the reference  axis and the  axis of motion,  and

     4- is the phase of motion.

We assume <|>, and £p are independent, uniformly distributed  random variates
lying in the interval  (0,27r).

     We now define y = 6 cos<)>. and z =  coslj., and note that the distributions
on y and z are given by:

                  fz(z) =     1  i  ; fy(y)  =      1      ;                (B2a,b)

                          7t(l-z2)%           7i6(l-/y\2)%
where -l^z^+1, and -6^y^

     For XQ greater than zero, we  can  readily  derive  the probability that
x is less than or equal to XQ:

     P[xgx0] = i-^fW* dz  fy z(y>z>  =  l-2/*dy  fy(y)[l-Fz(^)];6>x>0  (|3)

                      y
where F  is the  cumulative distribution  on  z.
       z
     The marginal distribution is  obtained  by  taking  the derivative of the
cumulative distribution, P[X^XQ],  whence:
 Using similar arguments,  a  corresponding form can be found for -
 Upon inspection,  it will  be seen that the equation for -6£xo-0 *s  Just a
 transformation of (B4)  in which x0 = -x0.   Thus,  (B4) holds for the whole
 range of x,  -
                                     36

-------
     The integral (B.4) can be integrated in term's of elliptic integrals
of the first kind, with the result:
                        £ (x0) = 2  1 F( n,  l-(^)2 )                  (B5)
                         x       nz 6    2      °

 (See Gradshteyn and Ryzhik, 1965, §3.152(10) pg 246.)

     This form can then be used to find the higher moments of the dis-
tribution.  The function is symmetrical with -6^x^6, and so all odd moments
are zero.  The second moment is found as follows:
_/2\2 62 r1
' W    J o
 M2 62 1  jl +
"W     2   2
                       E(k)dk .           (G. and R. ,  §6.147, pg 637)
                     0   2
                            G  = .2865 62.  (G. and R. ,  §6.l48,pg 637)
where K(k) = F( ^,k), is the complete elliptic integral; and

     G =  .915 965 594.. .,

is Catalan's constant.  Higher moments can be readily deduced from §6.147.
                                    37

-------
co
00
©STRAIT 12

        p-fc,
                                                         P"31  ADMIRALTY
                                                                  INLET
                                                                              factors

-------
to
I2HRS
IOHRS
5HRS
IHR
                                                                   12 MRS
                                                                   IOHRS
                                                                   5HRS
                                                                    HR
                                                                  STRAITS 12
                                                                         SMITH
                                                                           G>
                                                                  I2HRS
                                                                  IOHRS
                                                                  5HRS
                                                                   HR
                                                                STRAITS 13
                                                              PROTECTION
                                                                               ADMIRALTY
                                                                                     INLET

                                                                                 PT. WILSON
                      Figure 2.  Dispersion envelopes  assuming no knowledge of tidal currents
                                (Case 1).

-------
12 MRS
10 MRS
5 MRS
IHR
                                                          I2HRS
                                                          IOHRS
                                                          5HRS
                                                          HR
                                                           I2HRS
                                                           10 MRS
                                                           5 MRS
                                                           IHR
                                                        STRAITS 12
                                                                  SMITH
                                                                    e>
STRAITS 13
                                                      PROTECTION I
                                                                        ADMIRALTY
                                                                              INLET

                                                                         PT. WILSON
                Figure 3.  Tide model contribution to reducing position error relative
                          to Case 1.

-------
Figure 4. Dispersion envelopes associated with ignorance of high
          frequency non-tidal current oscillations (Case 2).

-------
               20 cm/sec
               .72km/hr
Figure 5. Observed net currents at various locations in the Strait of
       Juan de Fuca and the Central Basin.
                         42

-------
CO
                          Figure 6.  Dispersion caused by randomly oriented net current of
                                     .66 km/hr (case 2).

-------
I
§
CC
I
2
9-
8-
7-
6-
5-
4-
3-
2-
 I-
0.
            O  STRAIT  II
            D  STRAIT  12
            •  STRAIT  13
CASE I ERRORS
   CASE 2 ERRORS
     01   23456789  10
     TRAJECTORY DURATION (HOURS)	
                       ERROR CAUSED BY NON-TIDAL
                       "HIGH" FREQUENCY OSCILLATIONS
                   D
                     Figure 7. A comparison of Case  1 and Case 2 errors.

-------
Figure 8. Difference - current dispersion ellipses at 1,  3,  and 10 hours.

-------
Figure 9. Wind Pattern No. 3
              46

-------
Figure 10. Wind Pattern No. 4
            47

-------
PATTERN  5
                     Figure 11. Wind Pattern No. 5

-------
      49C
    48.5£
VO
      48C
                TATOOSH
      47C
             125C
124.5°           124°            123.5°



     Figure 12.  Wind station locations.
123C
                                                                                           122.5C

-------
       1st COMPONENT
                            NEW DUNGENESS  SMITH I.
       2nd COMPONENT
              RACE ROCKS
                             NEW DUNGENESS  SMITH I
       3rd COMPONENT
              RACE ROCKS
                 TATOOSH    ONEW DUNGENESS & SMITH I
Figure  13. Principal  components of surface wind observations.



                          50

-------
a
5- o
41
j. 00
21 °

11
0
-I- 0 <
.2. o °
-3-
4 - c
-cj a
-5 -4 -3 -Z -i




*
a . .0

* *
°B W, W3 * o °
o
•
o
a • a ' •
*o
a
T 2 3 •> 5 -5 -4 -3 -2 -i



4 •
V °
o
2- «
I • •
	 : 	 .„ .. 	 	
a °
o r5
4
0 0 .3
o .2

1
o
	 0 	 o 	
0 O "1
o •
• "3
-4
-5
*
" 1 2 3 4 ~5
^3 PATTERN
• 3
o 4
a
•
•

D

oL W,
-I' a' ° %
-2. o o • f
                  -5-4-3-2-1   12345
Figure 14.  Scatter plot  of  principal  component weights.

-------
                                        5 m/sec
                    SMITH 1.
POINT WILSON
RACE ROCKS    NEW DUNGENESS
          WIND
       PATTERN

              3
ro
          WIND
       PATTERN
             4
          WIND
       PATTERN
              5
                Figure 15.  Scatter plot of wind observation, sorted by pattern type
                          at four central basin stations.

-------
en
                                     \
                              VELOCITY SCALE
                                  50 cm/sec

DECEPTION
  PASS
                                                                      SMITH I.
                                                                        O
                                                                            ADMIRALTY
                                                                                  INLET
                                                                             PT. WILSON
                                                           PROTECTION
                           Figure 16a.  Average residual  currents  on  August  25,  1978
                                        (residual  = drifter - model).

-------
 tn
                                                   CATTLE PT.
Cl

0
-J
                               VELOCITY SCALE
                                   50 cm/sec
                                                                             ADMIRALTY
                                                                                   INLET


                                                                               PT. WILSON
                                                            PROTECTION I
                          Figure 16b.  Average residual currents on August 26, 1978
                                       (residual = drifter - model).
                                                                                                                 ft
to ^,
                                                                                                             rv  t—

-------