Professional Documents
Culture Documents
2008-Validating and Improving The Canadian Coast Guard SAR Planning Program - CANSARP - Ocean Drift Theory
2008-Validating and Improving The Canadian Coast Guard SAR Planning Program - CANSARP - Ocean Drift Theory
\
by
© Lindsay E. Hillier
December 2008
Abstract
The Canadian Coast Guard Search and Rescue Coordinator uses a software
system to estimate the drift of targets in the ocean, and consequently determine a search
area. Existing software applies a simple drift algorithm (MiniMax) that has been in use
since World War II (Canadian Coast Guard/Department ofFisheries and Oceans Canada
[CCG/DFO], 2000).
The Coast Guard must be aware of the effectiveness of the drift prediction
algorithm, and the efficiency of the environmental inputs used. This thesis determines
the practicality of the available methods of MiniMax and the stochastic Monte Carlo
approach. In addition, we explore the implementation of higher resolution ocean and sea
current inputs. This both improves the current MiniMax algorithm and allows
Using an assembled database of drifting buoys in the North Atlantic Ocean, the
the Monte Carlo methods are evaluated. Results from the assessment indicate that
actual length. These results are used to determine where improvements must be made to
the current algorithms and environmental inputs for eventual application to the search
system.
11
Acknowledgements
I wish to thank Dr. Fraser Davidson and Dr. Entcho Demirov for both co-
supervising my project and for their guidance and assistance whenever needed.
I would also like to extend a sincere thanks to Dr. Andry Ratsimandresy, Debbie
Power, and Adam Lundrigan at DFO for their patience and assistance whenever needed,
and Oceans National Search and Rescue Secretariat New Initiatives Fund project in
personnel at the Canadian Coast Guard College (Westmount, NS) for their ongoing
support and positive attitude toward such a large project being undertaken in such a short
time.
Physics and Physical Oceanography Department also deserve thanks. Their smiles,
antics, and words of encouragement saw me through some long days of research.
lll
Contents
Abstract............................................................................................................................... ii
AcknowledgemeiJts............................................................................................................ iii
Contents ............................................................................................................................. iv
List ofTabies..................................................................................................................... vi
. oJ,rv·
L 1st ··
-~.· 1gures .................................................................................................................. Vll
Appendix B Cha11ges Made to the Monte Carlo Method i11 CANSARP Sciet~tific.. B-1
Appendix C Ge11eral Cha11ges Made to CANSARP Scientific......•........................... C-1
v
List of Tables
Table I Available Configurations of Output Data from Mercator Mode/... ............................................ 23
Table 2 Drifter Data and Sources with Contact Names ...................................................... .................... 30
Table 3 Mean and Standard Deviation ofDrift Length for I 2 SLDMBs in August 2007 Using Current
Estimation Methods Compared to Actual SLDMB Drift Length ................... .......................... ... 5 I
Table 4 Data from Monte Carlo Simulations over 48 Hours with SLDMB 17324 with Mercator
Currents and Varying Numbers of Particles .............................................................................. 77
VI
List of Figures
Figure 1.1 Sample data transfer from a drifting buoy to CANSARP from deployment to the point that it
is usable in CANSARP for search planning. ........................................................................... 8
Figure I.2 Delineations ofmodel boundaries used in CA NSARP a) Bathymetry of the Gulf ofSaint
Lawrence and a subset ofthe IML Model Boundaries in upper left panel. b) The Grand
Banks model geographical boundries delineated by the blue line. ......................................... 9
Figure I.3 Example ofa convex hull plotted around points ranging between 2 and 8 in both the x andy
axis. Note that the figure is formed by outlining the outermost points from the cluster of
particles.......................................... ....................................................................................... 18
Figure 2.1 Climatological sea currents for a) Summer and b) Winter as used in CANSA RP Scientific.
Coastline and bathymetry displayed to provide reference. Quivers shown are reduced such
that 119 ofall quivers on grid are displayed.......................................................................... 22
Figure 2.2 The air or sea deployable Self Locating Datum Marker Buoy (SLDMB) is composed ofan
air-deployable buoy, a GPS receiver, and an ARGOS satellite data transmitter
(http://www.seimac. com/) ..................... ....................................... .......................................... 2 7
Figure 2.3 Comparison ofamount ofavailable drifter data between -90° and -30° west, 25° and 80°
north. From this plot it is apparent that the GTS data from MEDS has a vast supply of
drifters, while the SLDMBs are lacking. ........................................................................... .... 31
Figure 2.4 Number ofobserved drifter data per squared kilometer for available timeframes for: a)
MEDS GTS 2001-2007 b) MEDS Raw 1994-2002 c) MEDS P&S 1989-2003 and d)
SLDMBs 1999-2007. Note varying density scales for plots. ................................................ 33
Figure 2.5 a) SLDMBs on Newfoundland shelf deployed in August of 2007. Each SLDMB is a
represented by a different color and entire life spans are illustrated. Bathymetry lines are
labeled and numbered in meters. b) All available MEDS GTS drifters in the Northewest
Atlantic for August 2007........................................................................................................ 35
Figure 2.6 Mean velocity comparison between a) SLDMB and b) GTS drifters over their respective
lifespans. Each dot represents one drifting buoy.................................................................. 36
Figure 2.7 a) Distances oftotal drifts of 12 SLDMBsfrom August 2007 for 48 hours on the
Newfoundland Shelf (blue) with mean length of70 km along with predicted CNOOFS
trajectories (red) with mean length 22 km and predicted Mercator trajectories (blue) with
mean length 26 km. b) Both current types under calculate the total distance of SLDMB
trajectories. The difference in SLDMB and CNOOFS (magenta) plot has a mean of48 km
and the difference between SLDMB and Mercator (cyan) plot has a mean of 44 km. These
lines are always positive indicating that the SLDMB trajectory is always longer................ 37
Figure 2.8 CANSARP Scientific simulations with MinMax Method using both CNOOFS (top panels)
and Mercator (bottom panels) currents at levels closest to surface (left) and 15 m (right) on
the Newfoundland Shelf over 48 hours for SLDMB 17324. Blue lines represent the drifter
trajectory while the black line is the predicted trajectory. Black circles indicate the search
area. While some improvement is seen in total distance, resolution ofdrift is lost and
Vll
- - - - - - -- - ---- -
distance is still far under calculated for al/12 cases ofSLDMBs in August 2007 (by about 3
times at the surface and about 5 times at 15 m for CNOOFS currents; 2. 7 times for
Mercator currents at the surface, and about 2.6 times at 15m.) This figure illustrates a
unique case in the Labrador Current where the Mercator currents almost match the SLDMB
trajectory. .......... .................................................................................................................... 38
Figure 2.9 CANSARP Scientific simulations with MiniMax Method using both CNOOFS and Mercator
Currents at levels closest to surface and 15 m offthe Newfoundland shelfover 48 hours.
While some improvement is seen in total distance, resolution ofdrift is lost and distance is
still far under calculated, as specified in Figure 2.8............................................................. 39
Figure 2.10 Correlation between distance ofCANSARP Scientific output using Mercator Currents at
surface level and 15 m versus actual drift trajectories. No obvious patterns can be observed
geographically................... .................................................................................................... 4 0
Figure 2.11 Continuous and discontinuous CNOOFS model output data used in CANSARP Scientific
simulations. Top images show simulations with continuous data while bottom images show
simulations with discontinuous data ...................................................................................... 42
Figure 3.2 Example Monte Carlo Simulation in CANSARP Scientific with CNOOFS currents for 48
hours using 250 particles (green dots). The dashed green line is the predicted search area
by the Monte Carlo method, the solid pink lines are the predicted search trajectory by
CANSARP Scientific (mean of left and right clouds ofparticles), and the solid blue line is
the actual SLDMB trajectory. The numbered line is a line ofbathymetry at 200m depth .. .48
Figure 3.3 Rule of Thumb (black circles), Ekman (red circles), and Madsen (green circles) current
estimation algorithm outputs as they exist in CANSARP Scientific on 12 SLDMBs from
August 2007 over 48 hours. Blue lines are SLDMB tracks punctuated by a green dot at the
start and a red dot at the end ofthe drift. Actual positions relative to landmasses and
bathymetry can be seen in Figure 2.5a.................................................................................. 52
Figure 3.4 Sample Pollard Millard Simulations for 48 hours with SLDMBs 17316 and 17326. Blue
lines are SLDMB tracks while black lines and circles represent the predicted Pollard
Millard trajectory and search area, respectively. Predicted trajectories are 2 to 3 x shorter
than actual trajectories .................................................................................. ... ..................... 54
Figure 4. 1 Rule of Thumb, Ekman, and Madsen current estimation methods with wind V = 10 mls and
U = 0 mls. Methods illustrate calculated trajectory velocities of0.328 m/s, 0.2148 mls and
0.2662 m/s, respectively. ....................................................................................................... 56
Figure 4.2 a) CNOOFS manually extracted currents and b) calculated current output from CANSARP
Scientific for a simulation at (-47.829, 46.995) starting on August 4, 2007 for 48 hours.
Note that north is up .............................................................................................................. 51
Figure 4.3 Simulation ofSLDMB 17316 on starting on August 4 at 0500Zfor 96 hours at 6-hour
intervals (displaying first 67 hours). The blue line represents the SLDMB Trajectory, the
red line represents CANSARP Scientific's calculated trajectory and the grey quivers are
CNOOFS current vectors.. .................................................................................................... 58
Figure 4.4 a) Mercator Currents and b) calculated current output from CANSARP Scientific for a
simulation at (-47.829, 46.995) starting on August 4, 2007 for 48 hours. Note that north is
up........................................................................................................................................... 59
Vlll
Figure 4.5 Simulation ofSLDMB 17316 on starting on August 4 at 0500Zfor 96 hours at 6-hour
intervals (displaying first 67 hours). The blue line represents the SLDMB Trajectory, the
red line represents CANSARP Scientific's calculated trajectory and the grey quivers are
Mercator current vectors............................................................................................. .......... 60
Figure 4.6 Mean position of250 particles after 24 hours ofdrift with SLDMB 17316 using the Monte
Carlo method for the a) CNOOFS and b) Mercator Currents . ............................................. 62
Figure 4. 7 SLDMBs from August 2007 simulated with 24-hours of CNOOFS currents using the Monte
Carlo Method with 250 particles........................................................................................... 63
Figure 4.8 SLDMBsfrom August 2007 simulated with 24-hours ofCNOOFS currents using the
MiniMax Method. See Figure 2.5afor location with respect to bathymetry..... .................... 64
Figure 4.9 SLDMBs from August 2007 simulated with 24-hours ofMercator currents using the Monte
Carlo Method with 250 particles........................................................................................... 65
Figure 4.10 SLDMBs from August 2007 simulated with 24-hours ofMercator currents using the
MiniMax Method. See Figure 2.5afor location with respect to bathymetry......................... 66
Figure 5.1 Search trajectories and velocity profiles using CNOOFS currents for SLDMB 17303, 17316,
and 17347. Column 1: Drifter trajectory and CANSARP Scientific Search trajectory over
48 hours Column 2: Calculated velocity profile and CNOOFS current velocity profile
comparison at start location ofdrift .. ..... ............................................... ........................ ........ 71
Figure 5.2 Velocity magnitude profile using CNOOFS data for a simulation at point (40, -65) ............ 73
Figure 5.3 Monte Carlo search areas with 1000 (green dashed line), 500 (red dashed line) and 250
(blue dashed line) particles after 24 hours for SLDMB 17303. Areas increase in size with
the number ofparticles applied with values of0.0183o2, 0.0258o2, and 0.0328o2,
respectively. The solid blue line is the SLDMB actual trajectory for 24 hours while the
green dot represents the start of the drift and the red dot represents the end ofthe drift. .... 78
Figure 5.4 Drift trajectory of lost mooring starting from 53° 10' N, 50° 54 ' W at 18:00Z on May 17,
2008. Reported locations illustrated from ARGO satellite fixes ................................... ........ 80
Figure 5.5 Predicted trajectory ofmooring from 18:12 on May 17, 2008for 94.3 hours using CNOOFS
currents at 10m depth. Actual drift trajectory is in blue with start location at location at
94.3 hours noted. Search trajecto1y and area are plotted in black Bathy metry is contoured
by multi-colored lines and labeled according to their depth in meters. ................................ 81
Figure 5.6 CANSARP Scientific simulation on May 17, 2008for 94.3 hours starting at 18: 12:00 UTC
run with CNOOFS currents with surface velocity field as quivers from AVISO altimetry
data. Contours represent mean sea surface height in em. Quivers represent sea surface
velocity for the 11h ofMay. Black trajectory is the predicted drift trajectory by CANSA RP
Scientific and the circle is the proposed search area. .................................................. ......... 83
Figure 5. 7 Daily progression of mooring trajectory (black) with corresponding daily averaged sea
surface height altimetry data on (L-R): a) May 17 , b) May 18, c) May 19, d) May 20, e)
May 21. Drift trajectories are plotted from May 17 to end ofday altimetry is plotted. The
green dot indicates the mooring start location and the red dot indicates its last position on
each day. Value ofsea surface height is indicated on colored contours in units ofem ......... 85
IX
Figure 5.8 Displacement calculations in the x andy directions where the current velocity is projected
parallel and perpendicular to the local current in the region over a 48 hour period. Each of
the 12 drifters explored are indicated by a different color dot in the plot. The mean
displacement on this plot is (656.6991, 74.4230) . ................................................................. 88
Figure 5.9 Normalized error calculations in the x andy directions where the current velocity is
projected parallel and perpendicular to the local current in the region over 48 hours. Each
color dot represents the error produced by a different drifter. The mean error position is (-
0.273, 1.52) ......................................................... ................................................. .................. 89
Figure B- I Time improvements made in Mercator Monte Carlo script by number ofhours ofsimulation
run and number ofparticles used.................................................................... .................... B-2
Figure C-1 Regular Mercator grid and irregular CNOOFS grid.......................................................... C-1
X
List of Abbreviations and Symbols
E: Total Error
Xl
GDP: Global Drifter Program
LW: Leeway
Xll
.-------------------------~-------------- --~~--------
Xlll
-----------
Chapter 1 Introduction
Search and Rescue is used all over the world to locate missing people and objects,
or to assist those in danger on both land and water. The focus ofthis thesis is ocean drift
calculations applied in maritime Search and Rescue. In all Search and Rescue cases,
notification or a distress signal must be received by the Search and Rescue coordination
centre in order to begin the process of creating a search plan. This notification may come
from the persons or vessel at risk, an observing vessel, or from individuals awaiting an
creating a search plan using the most up-to-date and accurate information known about
the case. Important information includes the Last Known Position (LKP) of the object,
how long the object has been missing, and an accurate description of the object of
interest, called the search object. These inputs are used, along with the corresponding
environmental inputs (winds, currents, etc.) for the search, to run a computer program
that applies search theory algorithms to help locate the search object. Each country has
its own computer system that applies different calculations to predict the search object's
drift in the ocean, but all aim to fmd the object in a reasonable period of time.
Once the search plan is prepared, Search and Rescue Units (vessels reserved
specifically for finding search objects) are deployed. The type and number are dependent
on the location of the search, and the vehicles available in the region. Fixed-wing
aircraft, helicopters, and seagoing vessels may be deployed for the search. Each of these
vehicles are requested to search a given area using a prescribed pattern from the search
1
plan. It is the Search and Rescue Coordinator's job to provide the most efficient pattern
In addition to these Search and Rescue Units, any vessel in the area may be
requested to assist with the search. These craft may be asked for information regarding
the incident or may be requested to assist in the actual search process, depending on a
Once the search plan has been carried out, and every reasonable effort has been
made to find the search object and the search is not successful, a call must be made as to
when to reduce or terminate the search. This is done based on the environmental
conditions in the area, available resources, and time lapsed. Each country has regulations
While every country has variations in procedures and guidelines, the above is a
general Search and Rescue process. Details and comparisons of each nation's
methodologies follow.
The Canadian Coast Guard is currently using search planning methods developed
for use in WWII. The original search theory's purpose was to determine an area in which
to search for enemy vessels. Following the war, the United States Coast Guard took
ownership of the algorithm and adapted it such that it was useful in Search and Rescue
operations.
2
In the Simplified Search Planning Method (SSPM) - the manual method of
plotting a search - a number of assumptions are made about the search object's probable
location, the nature of visual detection, and the way in which searches are conducted.
These include:
1) The possible search object locations are distributed around a datum position in a
3) The inverse cube model ofvisual detection 1 is sufficiently accurate under all
search conditions
4) Searches are performed as series of equally spaced parallel sweeps relative to the
search object
5) Specific levels of coverage (search effort) are used for each case in a series of
automate the approach to searching. This automated approach allows the incorporation
of more data, and more complex inputs to generate search scenarios in little time. The
search planner is able to evaluate several possibilities using various times, positions,
1
Inverse Cube Model: "The instantaneous probability that the search object will be detected is inversely
proportional to the cube of the range from the observer of the object (Soza and Company, 1996)."
3
1.3 Inputs for Search and Rescue Planning
CANSARP requires several factors in order to produce a search area, the most
vital of these being the Last Known Position (LKP). This position is used to compute a
datum (the most probable area of a search object corrected for drift over time, that
increases with subsequent searches). Four possible scenarios generally exist for
1) Single Position Known: The last observed position of the search object is of
2) Multiple Positions Known: This situation involves the reporting of more than
one location such that the actual last known position of the object is
questionable.
locations along the track have been reported, but certainty may still be
questionable.
4) General Area Known: If nothing more than a general region is known for the
search object, then a search area is established based on fuel endurance of the
4
1.3.2 Leeway
Movement of the search object through water, caused by the direct action of the
wind on the exposed surfaces of the object is called Leeway (LW). The shape, size and
orientation of the search object cause the LW term to vary making it difficult to
determine impact on object direction and speed (CCG/DFO, 2000). Leeway is applied
type/orientation of drift object in the wind) exists, and is applied to the left and right of
the downwind direction, should the object diverge. Leeway is applied to the search
3) Use leeway rates tables from National SAR manual and extract appropriate
taken, and the divergence angle as per the National SAR manual is both added
and subtracted to the downwind direction to produce the minimum and maximum
expected divergence.
5
Leeway rates and directions are implemented from tables in the National
SAR Manual Chapter 7, Section 7.31, that were developed through observations
of common drift objects and can be used to calculate leeway speed and
divergence for various objects at wind speeds of 5 to 40 knots using wind spends
measured at the 10m standard reference height (Allen & Plourde, 1999).
The fmal product that CANSARP uses is the total water current (TWC) to
indicate the datum point from which the search will be based. This current is defined as
the vectorial sum of all applicable currents (sea current or climatological current, tidal
current, wind driven current, etc) in a particular drift plot (CCG/DFO, 2000). In
discussed in greater detail in Section 1.6), the total water current is computed based on
whether ocean model currents are applied. If model currents are used, all current and
wind forces are considered a part of the current field and no calculation is required. If
simple measurements of current speed are applied, then winds and any other suspected
current forces must be vectorially summed to produce the total water current vector.
There is an order by which currents are applied in the Canadian Coast Guard' s
Self Locating Datum Marker Buoy (SLDMB) data, then 2 model output currents; the
Grand Banks Model and the Institute Maurice Lamontagne (IML) GulfModel, followed
6
by subjective currents, wind driven currents, tidal currents, and finally sea currents
(Canadian Coast Guard College [CCGC], 2005). The first three current types are
preemptive in CANSARP such that just one of the selections are used (in the order listed)
regardless ofhow many other current types are selected (no calculation of total water
current required) while the final four current types can be vectorially added (CCGC,
2005).
In situ observed currents are estimated from surface drifters released by the on-
scene search and rescue unit. These currents are important to a search since they are
aircraft or ocean vessel deploys a surface buoy into the water at the Search and Rescue
Scene nearest the last known position ofthe search object as possible. Location data is
collected by the drifter's internal GPS and is transferred via ARGOS satellite to the LUT
(local computing station receiving transmitted data) for transfer to one of the three
Canadian Coast Guard's Joint Rescue Coordination Centre (JRCC) and then to the
7
Argos Satellite
Transmitter
Data sent to
CANSARP terminals
~
for use in SAR
Figure 1.1 Sample data transfer from a drifting buoy to CANSARP from deployment to the point that
it is usable in CANSARP for search planning.
Presently, CANSARP uses ocean forecast systems from the Grand Banks and
IML models that take real time data and project currents for hours to days in advance
(CCGC, 2005). The data is automatically downloaded to the CANSARP workstation for
predicting drift (CCGC, 2005). Each of these models has different boundaries. The IML
model encompasses the St. Lawrence River, the Gulf of St. Lawrence, and the southwest
coast ofNewfoundland and Labrador/the northeast coast ofNova Scotia (Saucier et al.,
2003), as in Figure 1.2 (a). The Grand Banks model has boundaries encompassing the
entire Labrador Sea as per Figure 1.2 (b) (Tang et al., 2008).
8
52.
5 1.
50
72 70.
The IML model has a horizontal resolution of 5 km, and a vertical resolution of 5
m from the surface to 300 m depth. Below 300 m, the resolution is at 10 m increments
(Saucier et al., 2003). This model is hydrostatic and provides solutions to the mass,
momentum, heat, and salinity conservation equations. Details and equations can be
The Grand Banks model is based on the Princeton Ocean Model (POM). It has a
free surface and applies sigma coordinates in the vertical direction. The model
determines velocity, potential temperature, salinity, and turbulence. The model grid is of
116° x 116°, and has 16 sigma levels in the vertical. Y ao & Peterson (2000) discusses this
9
1.3.6 Subjective Currents
Subjective currents are estimated from the SAR scene or from other data sources
and can be combined vectorially with other current sources to produce a resultant current
(CCGC, 2005). They differ from measured currents as they are estimated from the scene
CANSARP calculated wind currents are local currents generated by the effect of
wind on surface water calculated by CANSARP using observed and forecast data, and
should be used in conjunction with tidal currents and sea currents, if available (CCGC,
2005).
There are two types of calculations for wind driven currents presently used in
CANSARP; the empirical or Rule of Thumb method and the Ekman method. The
Empirical method uses 3.3% of the wind speed offset 20° to the right ofthe wind
direction (CCGC, 2005). The Ekman method is based on the Ekman boundary theory
(Madsen, 1977) and is generally used when time permits as it is computationally more
Tidal currents are only available for select geographical regions in CANSARP.
These currents are static models that do not incorporate external inputs, but do change in
10
time, and are calculated in 15-minute time steps (CCGC, 2005). These currents account
The last type of current is the sea current. In CANSARP, "sea currents" refer to
steady state ocean currents such as climatology, but these change seasonally. One-hour
time steps are used in CANSARP to calculate sea current drifts (CCGC, 2005).
There are five basic steps involved in search planning in Canada (CCG/DFO,
2000):
1) Estimating the datum (most probable position of a search object corrected for
3) Selecting the appropriate search pattern considering the size of the area and
11
1.5 United States Coast Guard (USCG) Model
The United States Coast Guard's search model has origins dating back to 1942
responded to a German submarine threat in the Atlantic (Frost & Stone, 2001 ). Bernard
Koopman, who joined ASWORG in 1943, has been credited with the statistical
foundation of the search theory (Frost & Stone, 2001). Koopman defined the elements
of an optimal search as having the following four properties (Frost & Stone, 2001 ):
The optimal search problem is defined as "finding the allocation over some
subsets of the possibility area for the limited amount of available search effort that
maximizes the probability of success." Solving this problem indicates how search effort
should be distributed.
for locating both stationary and moving targets, and document a few cases of successful
12
In the early 1960s the Americans tried out the MiniMax theory, similar to the one
the Canadian Coast Guard uses today, which was intended to allow different drift
scenarios and handle a maximum and minimum drift parameter calculation. Leeway was
the most common ofthese drift parameters, and tables were created with leeway
The Search and Rescue Program (SARP) that followed was the first attempt at
computer based search plan approach around 1970 (Frost & Stone, 2001). It was
simplicity to the user in mind, requiring just 4 input variables; incident date and time, last
known position of the distressed craft, probable position error of the distressed craft, and
probable error ofthe search craft (Frost & Stone, 2001). SARP calculated drift
trajectories based on hourly time steps, and accessed wind and current data using nearest-
approach to search planning and evaluation (Frost & Stone, 2001). This was a semi-
random approach using the Monte Carlo method, which will be discussed in greater detail
in section 3.2. The CASP program applies 500 points centered on the "head" of the mean
sea current vector using a stochastic approach to determine locations independent of one
another, producing a region of normally distributed points (Frost & Stone, 2001). When
13
error is summed to the mean sea current vector, the resultant vector is used to compute a
sample drift velocity. This sample drift velocity is found by taking samples from the
environmental forces (winds, currents, etc). The drift velocity is then multiplied by a
time interval to obtain sample drift distance (Frost & Stone, 2001). This procedure
repeated several times will provide a reasonable probability distribution for search.
According to the number of replications requested, the search is then updated using the
Monte Carlo approach. The solution does not resemble a circular pattern at all like the
CSPM provided. Probability densities are represented by color in the CASP output (Frost
By 2001, the United States Coast Guard was using more advanced techniques to
determine the trajectories of oil spills than they were for people in distress, and the
Search and Rescue Optimal Planning System (SAROPS) system came into development.
The SAROPS 1.0 alpha version was released in March of2005, with an operational
Kit (CJMTK)-based graphical user interface (GUI), Environmental Data Server (EDS),
and simulator engine that performs the particle motion and search optimization
(O'Donnell, J.D. et al., 2005). The GUI applies a wizard-based interface, supports vector
or raster plots, displays environmental data, and displays recommended search patterns
and probability maps (Spaulding, 2008). The EDS requires surface current and wind
data, and any other available factors including visibility, cloud cover, sea state, etc. to run
(Spaulding, 2008). It automatically selects the best data available to run, and
14
accommodates for varying spatial data resolutions (Spaulding, 2008). Sources of this
data are National Oceanic and Atmospheric Administration (NOAA), Navy, regional
associations, universities and commercial providers. This data is passed to the simulator
engine which computes the Monte Carlo particle simulation and simulates distress
There are identified needs and a plan for the development of open-sourced coding
and more readily available versions of the software for the scientific community to
review and use, as SAROPS is not currently a product available for public use (Allen &
Howlett, 2008).
The Norwegian model called LEEWAY also employs the Monte Carlo method.
It is a part of a suite of oceanic models including a ship drift model and a 3-dimensional
oil spill model. It was developed and implemented by the Norwegian Meteorological
Institute for the operational community. The program has the following features:
current at a certain depth comparable to the draft of the object. In the absence
15
3) The search object of interest is assumed to adjust its motion instantaneously
4) Surface wind fields are developed from an operational 3-D baroclinic ocean
version of the Princeton Ocean Model (POM) which solves the primitive
5) Stokes drift is assumed in leeway drift (Breivik & Allen, 2008). That is, the
where POC is the probability of containment, and POD is the probability of detection. A
Monte Carlo approach was thus decided upon to produce a probability distribution for
both the latitude and longitude uncertainty values. This is because Markov processes are
of random evolution and are "memoryless" depending only on the current state, and not
on how an object behaved in the past to arrive in the current state (Kom & Kom, 2000),
and in the search model, concern is placed on the Last Known Position. Throughout the
integration of the members involved in the Monte Carlo problem, once movement left or
16
right of downwind has begun, the object cannot cross paths and change direction. In
The initial drift distribution for the Norwegian drift model is prescribed on the
accuracy of the LKP. If the LKP is well known, the initial drifter locations in the
ensemble are tightly concentrated. In LEEWAY, the ensemble size is set to 500 (see
Breivik, 2008 for further detail). The search area is determined using the convex hull
polygon derived from the particle distribution. The convex hull of a set of points where
the smallest convex polygon that encompasses all points of the set (Brown, 1979). Figure
1.3 illustrates a convex hull surrounding 225 points with values between 2 and 8 in both
17
. ..~.~- ...~ .....~.- ~~ --· '· ..
.
..' :·.. .... . ~
. ~··· ... . - .~
·~ •
.. •: tit :
.. : .
- ~~' ........ •. ..... .. . .
··:······· .
: .
· · ·~ .
:
11'1 :
.•. . ;
II)
~ ' . ~ .. · :
> :•
>- . ~. .
0;0000 0: .. ~·:· · 0000000~00.0 ~00000
.: . . .. .. . ...
" 0 "'''· ; · : O O O OOOO O O O R OOO
: . . •..• ' ..
. :.
. '.· ... .. ~
.: :
·····~· ·:.· .. ·~ · · ·~·· ·· .... .;... . ..
~ :. .-
0 0~--L---~--~--~---5~--L---~--~--~--~10
X Values
Figure 1.3 Example of a convex hull plotted around points ranging between 2 and 8 in both the x
and y axis. Note that the figure is form ed by outlining the outermost points from the
cluster ofparticles.
It has been noted that validation of the Norwegian model is lacking. Small drift
experiments have been undertaken, but no evaluation of the model has been formally
completed. Of major concern in this method is the lack of account for jibbing, capsizing
or swamping of the search object (Breivik & Allen, 2008). A further improvement
suggested by Breivik (2008) is for a higher resolution model, particularly near shore
18
1.7 Thesis Objective
The Canadian Coast Guard (CCG) Search and Rescue (SAR) objective is to
"prevent loss of life and injury through search and rescue alerting, responding, and aiding
activities using public and private resources (CCG/DFO, 2000)." The purpose of this
project is to validate and improve the Canadian Coast Guard Search and Rescue Planning
Centre. The ultimate goal is to improve the efficiency of search theory given present day
computing capacity and available environmental inputs in efforts to fulfill the CCG
mandate.
This project addresses fine tuning search and rescue theory as well as numerical
simulations to demonstrate the impact of applying updated search and rescue theory.
Statistical success of the existing algorithm called "MiniMax" are determined and
compared to the Norwegian SAR version of the Monte Carlo method. All computation
CANSARP program called CANSARP Scientific. Using various sea and wind current
inputs, the most effective combination of algorithms and inputs will be determined
mathematically and proposed for eventual implementation into the CANSARP program.
Current inputs being used include historical seasonal currents, and outputs from
Mercator model forecasts, as discussed in sections 2.2 and 2.4. Wind inputs used in
CANSARP Scientific can be manually input as constant values, although option exists to
19
read the Canadian Meteorological Center's (CMC) Generalized Equation-of-State Model
20
Chapter 2 Environmental Data Sets
Environmental data sets include current and wind data, as well as ground truth
data. The currents used in Cansarp and Cansarp Scientific are either climatological
currents or model output currents, and the winds are CMC Winds from Environment
Canada. Ground truth data includes a variety of drifting buoys that simulate drifting
seasons; winter and summer. The sources of these currents include geostrophic
current map of the Atlantic Ocean by the International Ice Patrol, and a collection of
American and British pilot charts of the North Atlantic Ocean combined with gridded
data from CANSARP V2.0 (Seaconsult Marine Research Ltd., 1993; Murphy & Hanson,
1989; El-Sabh, 1976). The coverage ofthese current grids is limited to the Northwest
Atlantic Ocean as seen in Figure 2.1, and the use of these currents for Search and Rescue
has some obvious limitations. Due to geographical and temporal resolution, these
currents are applied in the original CANS.ARP program only when no other data is
available for a particular case, or when little is known about the conditions at the drift
target's LK.P.
21
60 60
58 58
56 56
54 54
.
..,
52 ..
..,
52
,.
i!
...J
50
48
,.
i!
...J
50
48
0
-2000 -1 800 -1600 -1400 -1200 -1000 -800 -600 -400 -200 -2000 -1800 -1600 -1400 -1200 -1000 -800 -600 -400 -200
Bathymetric Depth (m) Bathymetric Depth (m)
Figure 2.1 Climatological sea currents for a) Summer and b) Winter as used in CANSARP Scientific.
Coastline and bathymetry displayed to provide reference. Quivers shown are reduced
such that 1/9 ofall quivers on grid are displayed.
Current files obtained from the French operational ocean forecasting organization,
model, forced by atmospheric wind stress, evaporation, precipitation, sensible heat flux,
latent heat flux, infrared flux, cloud coverage, surface humidity, air temperature at 2 m,
and winds at 10 m (Modeling the Ocean at Mercator, 2007) . The Mercator system
22
. - - - - - . , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -- - - -- ------------
Each of these data sets provides daily outputs of a number of state variables, sea
Scientific are sea current velocity readings from the output state variables.
for applications like search and rescue, navigation through ice, the offshore oil industry,
weather forecasts, and marine habitat management in the Northwest Atlantic Ocean. Its
goal is to provide a coupled ice ocean forecasting system enabling users to make better
The CNOOFS model produces a daily forecast using the Nucleus for European
Modeling ofthe Ocean (NEMO) to determine the ocean state (CNOOFS, 2007). It
applies 1-way nesting within a Global Ocean Forecast System (Mercator's PSY3V2
currents with global, native grid at 1/4° horizontal resolution and 43 vertical levels) and
temperature, salinity, and ocean currents at hourly intervals (CNOOFS, 2007). Its
23
domain ranges 103.12 W to 27.23 W longitude and 26.69 N to 83.68 N latitude with
closed boundary conditions on the eastern and southern boundaries (CNOOFS, 2007). At
present, this model applies the free-surface and uses tidal forcing with one main
component and outputs at 1/4° in the horizontal and 46levels in the vertical (CNOOFS,
2007).
Winds in the GRIB (GRidded Binary) format are available from two different forecast
systems; a GEM regional and a GEM global format available in both low and high spatial
resolution files that provide a forecast for twice a day; OZ and 12Z.
The low resolution files in regional model output offers 0 - 48 hour forecasts with
surface fields available at 3-hour intervals (Environment Canada, 2007). At 60°N, the
resolution is 60 km. The lower resolution global model output offers 0 - 120 hour
forecasts at 6-hour intervals with 2.0° degree resolution (Environment Canada, 2007).
The high resolution files are global at 0.6 x 0.6° and 30 km resolution, while the
regional files are at 15 km resolution (hour forecasts at 6-hour intervals with 2.0° degree
resolution (Environment Canada, 2007). Again the regional forecasts are 0-48 hours
while the global forecasts offer 0 - 144 hours (hour forecasts at 6-hour intervals with 2.0°
degree resolution (Environment Canada, 2007). The time resolution of the files is 3-
24
In CANSARP Scientific, it is possible to apply any of the above formats, but most
commonly used are high resolution files at 1-hourly resolution in efforts to provide the
highest definition possible for a given simulation. At present, the official CANSARP
program does read GRIB winds in a similar manner, but reads lower resolution winds at
programmed to select the best available GRIB wind file for the scenario. Selection is
based on a pivot date of April4, 2005. Simulations for dates prior to the pivot date use
the regional 3-hourly winds whereas those following the pivot date apply the regional 1-
The CANSARP program downloads environmental inputs twice daily from the
IML and Grand Banks Ocean Model as well as the CMC winds. IML Gulf data is
downloaded once a day from the server at IML in Rimouski, Quebec. The current data
output from the Grand Banks model is updated twice daily by the Bedford Institute of
Oceanography and passed to a server at the Canadian Coast Guard College for
implementation into CANSARP. The CMC winds are downloaded hourly (the wind data
includes hourly wind observations in addition to the GEM model output, unlike
CANSARP Scientific) from a server at the Canada Ice Center though the data is produced
by the group at CMC in Dorval. All environmental data is obtained at 1 hour resolution
25
GRIB is a standard compact data format for gridded meteorological data. GRIB
Self-Locating Datum Marker Buoys (SLDMBs) determine and transmit their own
position via satellite transmission (CCG/DFO, 2000). They are used to determine the
environmental effects (currents, winds, etc) on drifting objects, and to track debris of a
SAR incident in the water during a search procedure, in efforts to reduce the size of the
air deployable buoy, GPS positioning system, an ARGOS satellite transmission system,
and a sea surface temperature sensor. They can be set-up to simulate either a person in
water (PIW) or a 4-person liferaft, and have a lifespan of about 5 days once deployed.
The data transmitted from the SLMDBs are received by the Rescue Coordination
Centre (RCC) that deployed the drifter in a SAR case, or in many cases, by a research
agency Local User Terminal (LUT), for analysis and testing. This project uses SLDMB
track data from the Canadian Coast Guard College archival ftp site:
26
-------- - -- - - - - - - - - - - -- - · - - - - - - - - · -
Figure 2.2 The air or sea deployable SelfLocating Datum Marker Buoy (SLDMB) is composed ofan
air-deployable buoy, a GPS receiver, and an ARGOS satellite data transmitter
(http://www.seimac. com/).
In addition to the SLDMB drifters, WOCE drifters were used in this project for
both validation and experimental purposes. While these drifters serve a similar·purpose
to the SLDMBs, they have slightly different characteristics and were created for different
reasons.
The World Ocean Circulation Experiment (WOCE) was a large experiment under
the World Climate Research Programme (WCRP). As part of the goal to observe poorly
projects continued partly post 2002 through the CLimate VARiability program
(CLIV AR). Drifters in the WOCE program provided validation data. They are usually
27
~------------------------- ------------------
drogued at 10 or 15 m and transmit their positions via the ARGOS satellite transmission
system.
Drifting buoys provide information about surface drift which can be used to
compare to models and satellite data, and can contribute to circulation models. The
Marine Environmental Data Service (MEDS) in Canada is part of the Surface Velocity
Program Data Assembly Centre (DAC) that combined with the United States' Atlantic
Atmospheric and Oceanographic Laboratory, assembled and made data from WOCE
intervals is available through this program (Department of Fisheries and Oceans Canada
[DFO], 2007).
The MEDS database is a part of the Integrated Science Data Management (ISDM)
branch of the Department ofFisheries and Oceans (DFO) Canada. ISDM receives,
processes, filters, and archives drifting buoy data over the Global Telecommunications
System (GTS) and also archives non-real time data from other sources. This data is
attainable by making a formal request to the DFO with the specifics of data required.
This data includes more than 22 million drifter position records ofGTS data from 1978
onward, and Global Drifter Program (GDP)/ formerly Surface Velocity Program (SVP)
WOCE data available from 1979 to present with drogued buoys available up to just 2003
(DFO, 2007). The Raw and Pressure and Sensor (P & S) data come from the Atlantic
Oceanographic and Meteorological Laboratory (AOML), also under the GDP. The Raw
files are data received by the AOML that are placed in an archive. The P & S files are
28
merged pressure and sensor files containing edited pressure data with buoy ID, date time
and position and sensor output data containing buoy ID, date time and sensor values of
In the case of CANSARP Scientific, these drifters are used as search targets.
They are input according to their identification number and simulations are run based on
their LKP position and time. Simulations can be run based on whether the drifter is
In an effort to determine the best dataset for use with CANSARP Scientific, a
29
---,- - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - - - -- ------- ------
Of the above drifter data, the SLDMBs and MEDS drifters are used here since
SLDMB data is currently used by the Coast Guard for marking and studying SAR cases
and the MEDS dataset is comprehensive and vast with both drogued and undrogued
cases. The total number of drifters from 1995 to 2007 in the Northwest Atlantic is seen
in Figure 2.3:
30
1800.-------.--------------,,------------ - - . - - - - - - - - - - - - - - - . - - - - - - - .
1600
!::!
CD
~ 1000
....
0
a;
.tl
E
:I
z
Figure 2.3 Comparison ofamount ofavailable drifter data between -90° and -30° west, 25° and 80°
north. From this plot it is apparent that the GTS data from MEDS has a vast supply of
drifters, while the SLDMBs are lacking.
As per Figure 2.3, the MEDS Global Telecommunications System (GTS) data
contains approximately 4 times more drifter position data as MEDS Pressure and Sensor
(P & S). It is also quite evident that the SLDMB data contains very few drifters, which is
Each of the data sets had a different range of availability; GTS runs from 1978 to
present, P & S runs from 1989 to 2006, Raw runs from 1993 to 2003, and SLDMBs from
1997 to 2007. There is also a kriged2 data set available from 1989 to 2006 though it was
not explored as interpolated data is not of as much interest for this project. From each of
the above sets, several years of data were obtained for exploration according to
2
Kriging is a geostatistical optimal interpolation technique used to interpolate unknown values at known
locations by applying a semivariogram. It incorporates measures of uncertainty and error, and optimal
weighting can be applied based on the semivariogram used (Cressie, 1990).
31
- -- - - - - - - - - - - - - - - - - - - -- - - - - - - -- - - - - -
availability and size of files at the time of download. Figure 2.3 below illustrates the
datasets obtained for this project. A useful period oftime for analysis, largely dependent
upon environmental inputs, was selected for more consistency, once data analysis began.
32
a h
5000 200
4500 180
4000 160
N
3500 ~ 140 e
15 "'~
3000 :;
15 '0
.. 60· 120 c.
.. ~
500 35 20
0 -
-70 -60 ..so 40 -30 -20
Longitude
c d
200
180
70 160
N N
E
140 ~
"'..c..... 65
.,....
5
.. ..
~
'0
120 :;
;
~55 100 ~
...c ....10 0
00
......
0
.c
E
50 80
0
;
.c
:I
60 E
z :I
z
40 40
35 20
Figure 2.4 Number ofobserved drifter data per squared kilometer for available timefram es for: a)
MEDS GTS 2001-2007 b) MEDS Raw 1994-2002 c) MEDS P&S 1989-2003 and d)
SLDMBs 1999-2007. Note varying density scales for plots.
One possible solution to the inconsistent time and shortage of SLDMB data issues
would have been to obtain older data (pre 1998), but corresponding environmental inputs
for earlier dates was difficult to obtain. The focus in this thesis will thus be on 12
SLDMB drifters from August 2007 that were released on the Newfoundland shelf. This
33
was decided because the most up-to-date model data was available for use in this period,
and because there were enough drifters in our region of peak interest. Findings in this
area would be expanded and tested with various other drifters in locations off the shelf
following primary analysis. Figure 2.5a shows the available SLDMB drifters from the
August 2007 dataset, while Figure 2.5b illustrates the available MEDS GTS drifters for
the same month. Figure 2.6 compares the velocities at which the SLDMBs and MEDS
GTS drifters travel. Their mean velocities indicate that their drift characteristics are
similar.
34
···· :c~ ··· ....... ·:·· · .... :·········\· ..............., ..................... ·························.················:
• Start Locabon (LKP)
........ti... .: . !>~Dt.!!117m r._. • End Location
· SLDMB 17330
...
:y :
;
~- .... .
~
~-. . . · · ··~
:
~
~= : .
...... ~.. Aoo ... _it ............... .. / ....... -
Gl I~ :;
"t:J
/: i
,g ........... .... . (
:. . . . .
47 .5 ~~~~~ ·17P~ ~
1ij
.~.~~;~r
...J
'~ \: :
''''2
46.5
46 ..... . .
. . /!: ~ -~~ -~·-·······
45.5 ...... . . . . . . . . . . . . . . ;. i.7 T. . . . r. i
4
-~ -53 -52 -51 -50 -49 ~8 ~7 ~6 ~5
Longitude
Figure 2.5 a) SLDMBs on Newfoundland shelf deployed in August of 2007. Each SLDMB is a
represented by a different color and entire life spans are illustrated. Bathymetry lines are
labeled and numbered in meters.
b) All available MEDS GTS drifters in the Northewest Atlantic for August 2007.
35
,.--..,.-,---------------------------------------~ - · ------
1.5 ······-:·
Figure 2.6 Mean velocity comparison between a) SLDMB and b) GTS drifters over their respective
lifespans. Each dot represents one drifting buoy.
various depths (uppermost available level and 15m in each model) influenced drift
of the trajectory of the drifter as Figure 2.7 illustrates. As per Figures 2.8 and 2.9,
minimal change is seen when currents at depths are applied whether on or off the shelf.
To further this study, it was tested to see whether there is a distinctive spatial pattern
between where CANSARP Scientific predicts the drifters to exist and where they actually
travel in a given timeframe (48 hours in this study). It seemed that there was no pattern
36
- - SLDMB
- CANSARP Sciennfic CNOOFS
17316 17317 17319 17323 17324 17325 17326 17329 17330 17747 17748
SLDMB Number
10
: :
E8
., I
/:======~=========1\---l'~,>·~·-·~~~"\
I \ :0 \
~ 6 HoO / ./ :., oo;o \ •ooooO \
~ I/ \ I
i5 I / I1 \\ : __ ,.__ _ __ _, \ I
4
I.,./.... OOOOOOOOO OO 000000000
)..---
000000: 00 00000000
; "'"HH" .. ,H!\
''" HHO\
I
I I \ I .
I \ I ;
I \ \
\ \ :
\ I :
··• ~ ~ 1·
\ \:
\ ,: / /
\ ~ / :I
\r- - ---- .,L---~
,/
Figure 2.7 a) Distances of total drifts of 12 SLDMBsfrom August 2007 for 48 hours on the
Newfoundland Shelf (blue) with mean length of70 km along with predicted CNOOFS
trajectories (red) with mean length 22 km and predicted Mercator trajectories (blue)
with mean length 26 km.
b) Both current types under calculate the total distance ofSLDMB traj ectories. The
difference in SLDMB and CNOOFS (magenta) plot has a mean of48 km and the
difference between SLDMB and Mercator (cy an) plot has a mean of 44 km. These lines
are always positive indicating that the SLDMB trajectory is always longer.
37
47.1 •··· 47.1
47 47
46.9 46.9
468 ...... , .. .
,.. 46.7
46.6 . ii" 466
.J .
46.3 .
~~(:)
462 L-~~--~~--~~7~2·~--~~7~--~~~6~8-
\
LongHude
·::. ..•.
46.9 ., ...
46.8
46.6 .
46.5
464
46 3
LongHude
Figure 2.8 CANSARP Scientific simulations with MinMax Method using both CNOOFS (top panels)
and Mercator (bottom panels) currents at levels closest to surface (left) and 15 m {right)
on the Newfoundland Shelf over 48 hours for SLDMB 17324. Blue lines represent the
drifter trajectory while the black line is the predicted trajectory. Black circles indicate
the search area. While some improvement is seen in total distance, resolution ofdrift is
lost and distance is still far under calculated for al/12 cases ofSLDMBs in August 2007
(by about 3 times at the surface and about 5 times at 15 m for CNOOFS currents; 2. 7
times for Mercator currents at the surface, and about 2.6 times at 15m.) This figure
illustrates a unique case in the Labrador Current where the Mercator currents almost
match the SLDMB trajectory.
38
"' "'
" "
428 42.8
14"
~ "'
424 424
422 .,,
"' "'
" "
42.8
428
; <42.6 1426
:!!
424
.,,
"' '"
., 42
Figure 2.9 CANSARP Scientific simulations with MiniMax Method using both CNOOFS and
Mercator Currents at levels closest to surface and 15 m off the Newfoundland shelfover
48 hours. While some improvement is seen in total distance, resolution ofdrift is lost and
distance is still far under calculated, as specified in Figure 2.8.
39
.,-----------~--------------------------------- ------
~
,....,
• sCD
u
c:
CD
"'0 • • ....
CD
~ 50
• • ~
'tV 0
•• •• •
...J CD
u
.
c:
•• • • t'O
iii
.-.. • : •• i5
40
••
•
• • • •• • •• • • -. ·-
,•• • • • •• •
• • •
• •
• •••
30
• • • ••
•
•• • • •
•
-70 -60 -50 -40 -30 -20
Longitude
~
,....,
• sCD
u
c:
~
"'0
Ql
• • ~
••• •• • •
:;]
:t=
'tV • 0
..
...J CD
u
c:
••
•• • ... • • t'O
iii
• • ••
',
i5
40
•
••• •••
• • • • • • • : #• •• ·---
•' · • • • • •
•• • ••
30
••
•• • •
• • • ••
•
•
-70 -60 -50 -40 -30 -20
Longitude
Figure 2.10 Correlation between distance ofCANSARP Scientific output using M ercator Currents at
surface level and 15 m versus actual drift trajectories. No obvious patterns can be
observed geographically.
40
Tests were run with "continuous" versus "discontinuous" output currents from the
CNOOFS model. In this context, continuous refers to a special model run producing
hourly output data for the entire length of the simulation in one run. Discontinuous refers
to separate model runs, re-initialized daily, producing data output files containing hourly
data; the CNOOFS present standard. Again, each case underestimated the drift length of
the SLDMBs, but in all cases, the discontinuous data produced longer search trajectories
than the continuous data an average of 150 km. In the discontinuous data, the SLDMB
tracks were about 3 times longer than the predicted trajectories while in the continuous
data, they were about 4 times longer. Examples of these simulations can be seen in
Figure 2.11.
41
47.1
······\ · · · · •
•
Start Location
48 Hours 48 6
47 - - SLDMB Track
46.9
484
468
48 2
.. 46.7 Cl)
"0 "0
iii"' 466
....1 . .:~% ............;......... "'
""
iii
....1
. ~· ...
465
464
463
46 ·2 L---~::------:~:....._--4---:t-7.7
2----4~7:-----4
7 6.::-8-
-524 -52 2 -52 -51 8
Long~ude Longitude
47.1
• Start Location
• 48 Hours
47 ..... · ....... - - SLDMBTrack 486
·····:········
.. 46.7
"0
"'
.
"0
48 2
""
iii
....1 46.6
"'
""
iii
....1
465 ....
46.4
Figure 2.11 Continuous and discontinuous CNOOFS model output data used in CANSARP Scientific
simulations. Top images show simulations with continuous data while bottom images
show simulations with discontinuous data.
42
Chapter 3 Drift Algorithms
The concept of the MiniMax Method used by the Canadian Coast Guard was
developed by Bernard Koopman in 1949 with the intention of defining a search area with
100% containment (Koopman, 1980). As its name implies, this method calculates the
minimum and maximum possible locations of the target drift object based on a number of
(CCG/DFO, 2000). The midpoint between these two extremes is the datum point for the
search. Datum is the most probable location for a search object at a given time
(CCG/DFO, 2000).
Following the "datum" determination, the sum of the squares of all possible errors
is calculated to represent the square of the total error (CCG/DFO, 2000). The search area
size is determined based on the total error, E. There are three components that are
considered in calculating the total error; the total drift error, the initial position error, and
The total drift error, denoted De, is either the combination of all individual drift
errors (de), or is calculated as a function of the distance between the minimum and
maximum locations (de minimax). Individual drift errors are cumulative throughout the drift
due to assumptions that simplify calculations. The Canadian Coast Guard assigns a
43
constant value to de of 1/8 of the determined drift (default) or 1/3 the determined drift if
(CCGIDFO, 2000). The distance between the minimum and maximum locations, also
where Distance is the distance between dmin and dmax, de_min is 1/8 (or 1/3) dmin and de_max is
1/8 (or 1/3) dmax. The value of D e will usually be equal to the value of d e_minimax unless
(3.2)
LetXbe the error (in meters) ofthe initial position based on the source reporting
the Last Known Position (LKP) ofthe search target (CCG/DFO, 2000). When Xis
reported as a fixed position, it is known as the "fix error" and the position error is
attributed to navigational systems indicating a position, and the dead reckoning error,
based on a percentage of the distance the search target has drifted since its last reported
position (CCG/DFO, 2000). The fix errors are based on a table with error values
associated with respective navigation systems, and the dead reckoning errors are also
indicated in a table based on the type of aircraft or vessel searching for the target
(CCG/DFO, 2000). Dead reckoning values are only applied if the source of the LKP
indicates they should be. Similarly, a search craft error called Y is to be applied when
appropriate, and is equal to the fix error of the search craft. Total probable error is thus
(CCG/DFO, 2000):
44
(3.3)
Of course, the value of the total errorE must be re-computed in time as drift
changes, impacting the datum, or if the search unit or initial position changes.
Once this is complete, the search radius is implemented with radius of E around
Last
Known
Position
(LKP)
Total Drift Ve~tor······...
Figure 3.1 shows that the search is based from the Last Known Position (LKP).
From here, the sea current (water forces acting on the object) vectorially add with the
wind driven current (water forces driven by winds) to produce a total water current.
Once this new location is established, leeway divergence is applied. A lookup table for
45
leeway divergence based on the object type search and produces leeway angles ranging
from 0 to 60° to either side of downwind (CCG/DFO, 2000). Once a minimum and
if the total drift time is greater than 4 hours (CCG/DFO, 2000). The leeway uncertainty
value is simply the distance between the minimum and maximum locations, summed
with their respective confidence values and averaged which then represents the radius of
each error circle. Once small error circles are established, the total error as described
above is plotted around the search area, centered on the midpoint between the minimum
and maximum positions from leeway divergence is produced, and a search plan, often
based on a square of rectangular shape, can be decided upon by the search planner based
on available resources. Of course, the search area will increase as time goes on as errors
mcrease.
probability distribution that evolves with time forms a search area of particles. Presently,
although planned, the Monte Carlo Method is not used in CANSARP. In CANSARP
Scientific, the Monte Carlo Method option is based on the Norwegian model entitled
is determined in the same manner as in the MiniMax method. The difference is that there
are many randomly perturbed particles with the convex hull determining the search area.
At each time step, the wind and current are perturbed in magnitude and direction
46
randomly with a component specific standard deviation. The contributed random noise is
additive to the trajectory calculation. Next, the leeway coefficients have a perturbation
applied to it, and leeway vectors for both crosswind and downwind are computed using
linear regression formulae produced by Allen and Plourde (1999) according to the type of
object drifting. The advected particle results in a component of the particle cloud for
each iteration and particle required. The number of particles computed is user dependent,
all particles and distributes them based on the Last Known Position of the drifter of
interest. The first two particles are undisturbed, and each of the others is randomly
perturbed, constrained by a radius determined from the initial position error and number
of particles in the simulation. With each time step, the radius expands by an amount dr
defined as the initial position error divided by the number of particles in the simulation
(adjustable by user) with default value 0.5, multiplied by the radius of search, as
determined above. This is all computed in a polar coordinate system prior to determining
This method varies considerably from that of the Norwegian SAR that initially
equal to half of a user-specified radius. In Norway, two radii of uncertainty are applied,
47
..-~--------------------------------------------------------------- ------- ----
since the search with leeway is usually bimodal3 . Each radius is determined in a similar
way to which the Canadian search areas based on error are found (Breivik, n.d).
In CANSARP Scientific, all options for running the Monte Carlo Method are the
Below is an example simulation using the Monte Carlo Method with CNOOFS currents
for 48 hours and 250 particles, with a constant standard deviation of 0 applied to the
: :
:r-----..: _______ [__
:
47.05
'
47 ··•·.····
46.9
-48.2 -48.1 -48 -47.9 -47 .8 -47.7 -47 .6 -47.5
Longitude
Figure 3.2 Example Monte Carlo Simulation in CANSARP Scientific with CNOOFS currents for 48
hours using 250 particles (green dots). The dashed green line is the predicted search
area by the Monte Carlo method, the solid pink lines are the predicted search trajectory
by CANSARP Scientific (mean of left and right clouds ofparticles), and the solid blue
line is the actual SLDMB trajectory. The numbered line is a line of bathymetry at 200 m
depth.
3
In this case, bimodal refers to having two areas of equal probability and equal size.
48
3.3 Drift Method Applications in CANSARP Scientific
either of the aforementioned methods. The total water current can be determined from
ocean forecast inputs or if background climatology is used, wind currents must be taken
into account. In CANSARP Scientific, the CNOOFS and Mercator ocean forecast
currents can be selected. If the climatological sea currents are used, the variable wind
driven component of the current can be added through three alternatives; the Rule of
The Coriolis force pushes moving objects right (left) in the Northern (Southern)
Hemisphere. The faster the object moves, the more strongly it is affected by this force.
the Coriolis and wind force, must balance. To achieve this "geostrophic balance," the
ocean current must move the water mass from the surface to about 500 m depth to the
right of the wind direction while the Coriolis force pushes the water 90° to the right
explaining how surface current flows, including internal forces within the water that
49
direction (Lenn & Chereskin, 2008). This became known as the Ekman spiral. Each of
the three current estimation methods available in CANSARP Scientific are based on the
Ekman theory.
The Rule ofThumb method is the simplest form of accounting for wind stress
forcing and the Coriolis affect through a constant term. Here the Coriolis force can be
approximated with the current magnitude at 3.3% of the wind speed and current direction
approximation holds for values of approximately 45° latitude in the Northern Hemisphere
(Stewart, 2005).
The Ekman Method uses lookup table for winds, based on Ekman theory and
eight 6-hr periods (0, 6, 12, 18, and 24 hours). Wind speed and direction are considered
to be those which were valid at the end of the period. Based on each period's location
(latitude and longitude), direction and speed, a directional offset and speed factor are
extracted from tables and applied to the current direction and speed of the wind to
This method uses a 19-hour history of winds at the given location to calculate,
based on Ekman Boundary Layer Theory (Madsen, 1976), the drift (direction and speed)
due to winds. When this method was founded, its improvements on Ekman's method
were that it assumed that the vertical eddy viscosity increased linearly with vertical
distance from a sheared boundary, and that is could be applied in both shallow and deep
50
waters (Madsen, 1976) - a characteristic uncommon for models of its time. Madsen
Table 3 contains the mean and standard deviation in the length of each of these 3
current estimation methods as compared to the actual length ofthe SLDMB's drift.
The SLMDB trajectory is 2.3 to 3 times as long as the predicted trajectory in each
of the above cases, on average. This is a considerable discrepancy and the results are
51
"'1;---t.~~--""
- .
.,~._....,.,.,~_,~
47.1
-411
..."
...
• 4117
!. e.e
·~~,~~~-~
...~~ ,_...
..~.~..~,-...rr,-..~,--7..,,
..."·'
'"
'" ...7.6
•••
.,
<11
.., ..,
... ... '"
... ... ..,
i '" !467
11-' <166 •
~ ... ..
... ..,
.... ... "'
,., 45.3 4Q 1 ·~
"'
-48.6
"'
...
~·
... ~ ..... .... ;
'"
" ~-~;----.t;;----,~oi;,,--:rt;;-----,t;-"'7,
,_,., •.-
,_..
.. 1
. ., ....... -526
'"""""'
-51 <1 .52'1 -5'1 ..ei18
Figure3.3 Rule of Thumb (black circles), Ekman (red circles), and Madsen (green circles) current
estimation algorithm outputs as they exist in CANSARP Scientific on 12 SLDMBs from
August 2007 over 48 hours. Blue lines are SLDMB tracks punctuated by a green dot at
the start and a red dot at the end of the drift. Actual positions relative to landmasses and
bathymetry can be seen in Figure 2.5a.
52
3.4 Additional Current Approximation Algorithm: The Pollard-
Millard Method
do not consider inertial oscillations4 • To test the impact this change may make, the
Millard (1970) proposed that most inertial oscillations at the sea surface could be
attributed to winds, and applied the linear momentum equations of the mixed layer to
explain wind-stress induced currents caused by wind stress considered as a body force.
au
- - Jv=F-cu
at
av
- - fu = G - cv
at
where au and av are acceleration terms,fis the inertial frequency, -cu is the linear
at at
damping or friction term, and F and G are wind stress terms solved iteratively as follows:
(3.4)
4
A periodic motion in which the fluid inertia is balanced by the Coriolis Force. Inertial oscillations are
dependent upon their latitude position and travel clockwise in the northern hemisphere and
counterclockwise in the southern hemisphere. They are the most common currents in the ocean, and are
caused by fast changes in winds at the sea surface (Stewart, 2005).
53
where (U; sinB;.U; cosB;) are hourly averages ofwind velocity, Co is the drag
coefficient, Pais the density of air, Pwis the density of sea water, and Zo is the depth ofthe
oscillatory pattern is generated similar to observed drifter track behavior. Winds were
CMC GEM, updated hourly, from which wind stress is calculated. The resulting effect is
projected onto the climatological currents. These oscillations are not accounted in the
steady state methods of Rule of Thumb, Ekman, and Madsen. While oscillations are
4 7.1
47
..
..,
~ 46.9
1ii
--'
Figure 3.4 Sample Pollard Millard Simulations for 48 hours with SLDMBs 17316 and 17326. Blue
lines are SLDMB tracks while black lines and circles repres ent the predicted Pollard
Millard trajectory and search area, respectively. Predicted traj ectories are 2 to 3 x
shorter than actual trajectories.
54
Chapter 4 Validation of Environmental Inputs and
Current Estimation Algorithms
inputs and existing algorithms. Reviewed first were the algorithms for estimating the drift
including the Rule of Thumb, Ekman and Madsen methods. The approach used was to
apply constant wind current values of 0 m!s for u and v to observe the direction and
distance each drift by applying a constant current value. For example, the v component
ofthe current was held at 0 while the u value was given a value ofO.l m/s. Ifthe drift
traveled directly east, its direction was validated. The distance was calculated using
d = VI t where dis the distance, Vis the velocity (0.1 m/s in this case), and tis the time.
For a one hour period, the expected distance is 360.00 m. By running a simulation in
CANSARP Scientific, it was observed that the direction is directly east with a magnitude
of 361.32 m. The results were accurate in this case, and the Matlab functions used to
Wind currents were validated in a similar manner. Currents were held constant at
0 m!s and one component of the wind vector was given a speed. In this case the direction
should appear opposite to that of when currents are applied. Originally, it was found that
wind currents were being offset by 180° as a result ofbeing read in the same way as sea
currents. This issue was rectified by altering the code to account for the different
conventions.
55
Next each wind current estimation algorithm was reviewed with the corrected
wind convention and constant (zero) sea currents. Manual calculations were completed
to verify the Rule of Thumb and Ekman methods, according to the National SAR
Manual's worksheet. Since the Madsen method has no accompanying calculations in the
manual and uses look-up tables in CANSARP Scientific, any results in relative
agreement with the Ekman method were taken to be acceptable for validation purposes.
35 35 35
..
34.5 34.5 34.5
Figure 4.1 Rule of Thumb, Ekman, and Madsen current estimation methods with wind V = 10 mls
and U = 0 m/s. Methods illustrate calculated trajectory velocities of 0. 328 m/s, 0.2148
m/s and 0.2662 m/s, respectively.
Once the direction and distance were verified in these basic plots, the application
of model currents from CNOOFS and Mercator were tested. Comparisons were made by
feather plots and frame-by-frame trajectory progressions in an effort to gauge whether the
56
CANSARP Scientific calculations and the actual drift trajectories matched the input
currents. In Figure 4.2, the top panel illustrates manually extracted currents from the
CNOOFS current data file over 48 hours. These values were selected based on the
CANSARP Scientific calculated trajectory ofSLDMB 17316. For each point in the
trajectory, the nearest U and V components were extracted from the data file and plotted.
neighbor interpolation is illustrated. These currents are the ones applied to predict the
Figure 4.2 a and b is 42.41° with a mean velocity magnitude difference of0.035 rnfs .
-0.3 ........................................... .
0 5 10 15 20 25 30 35 40 45
Time (Hours)
~ -0.2 ............. i............... "i .... .. .... .. ... ;................:......... ....... ~.......... . .... ~ .. .... ...... ...;.... ...... .... ..[ .... ....... .... ;......... .
u
-0.3 ..... . ...... .. .....·.. .,... ................. ·····'···· ·· ····· .... , ..... .. .................. .
5 10 15 20 25 30 35
Time (Hours)
Figure 4.2 a) CNOOFS manually extracted currents and b) calculated current output from
CANSARP Scientific for a simulation at (-47. 829, 46.995) starting on August 4, 2007 fo r
48 hours. Note that north is up.
57
"" ""
"
....
I ...
.... ....
~.J8.1 -4805 -48
.._...
-tH15 -478 -4715 -478 --4715 -477 -4785 -178 48
~, -4105 ...a
-
-t705 _., , -4785 -c1a -4715 -c11 -47 85 -478
""
.....
-
-47115 -478 -4785 -478 -4715 -e77 -47 ~ ....7.8
""
"
....
~,
48
-4805 """
-
-4705 ""'" -4785 -478 -4715 -477 -47115 -478
Figure 4.3 Simulation ofSLDMB 17316 on starting on August 4 at 0500Zfor 96 hours at 6-hour
intervals (displaying first 67 hours). The blue line represents the SLDMB Trajectory, the
red line represents CANSARP Scientific's calculated trajectory and the grey quivers are
CNOOFS current vectors ..
In Figure 4.2, the feather plots shows minor differences that may be accounted for
in the different methods of data extraction, however, the direction and magnitude are in
58
reasonable agreement. In Figure 4.3, while the calculated trajectory by CANSARP
Scientific was not identical to the actual SLDMB data, the direction of the output
trajectory does agree with the applied CNOOFS current directions in time. The average
predicted trajectory is 100.50° while the average difference in distance is 1.66 x 103 m.
Similarly, the Mercator currents were verified graphically as in Figure 4.4 and
Figure 4.5 where the mean directional difference between the two cases is 64.69° and the
-s
'iii' 0 .2 : ...... ·· ····i ........ ..
z:.
0 .1 :
.
. .
··~
:
............ ··~······
... ... ..... ...... ····· · · ··~· ·· ········· ·· · ·.····· · ···· · ·····.·
: . . .
.... ........ .
: . .
J ~ -~~~//~~/'/~/~~~f.~~~~~~~~~~~~~--y~~~~j~~~~/~~~..\\\\\
-0 :
. . . . .. . . . . . . . . . . . . . . . .
.
0 5 10 15 20 25 30 35 40 45
Time (Hours)
~ 0.2
E
; : 0 .1 ......................... ...... . . . ................. ...... .... ... .
.
:=
~ 0
\. l .\"l · l · \· l · ~ · \ - lll · \ · lll · \ ·l ·l · J · ri ·I · t · ri ·l · J · lr ·r · l · "l · r ·rt · ·,~'''~~
> -0 .1 .... .. . ... .. .... . .. ... . .. . . .. . ···· . ·.. . .
"E
...~ -0 .2
::l
u -0.3
25 30 35 45
Time (Hours)
Figure 4.4 a) Mercator Currents and b) calculated current output from CANSARP Scientific for a
simulation at (-4 7.829, 46.995) starting on August 4, 2007 for 48 hours. Note that north
is up.
59
'\
I
t
\
...., ... -'10 ..na
Lcngi1ude
-47.7
472
47.1
47
~
~
...
....
...7
'"18 2
'"18.2
472
472
47.1
•7 1 \
47
47 I )
j ... j ...
1... /
~
....
~
... /'
/
... /' -
/
...7 '6.7
<07
/ / /
48
18.2 '"182 ...., '"182 ...., ..
472 47.2 47 2
471' 47.1
I
47
I I;
47
/ /'
<0.7
I
'"18.2 ...., ... -4711 -'10
longitude
_., -'6152 -48.1 -48 -4711 .... 7 8
Long<tude
-477
_.,. '"182 ... ,
Figure 4.5 Simulation ofSLDMB 17316 on starting on August 4 at 0500Zfor 96 hours at 6-hour
intervals (displaying first 67 hours). The blue line represents the SLDMB Trajectory, the
red line represents CANSARP Scientific's calculated trajectory and the grey quivers are
Mercator current vectors.
60
Like the CNOOFS currents, differences are evident in Figure 4.4 due to extraction
methods. In the frrst panel, the Mercator currents are manually extracted based on the
positions of the trajectory of SLDMB 17316; the nearest U and V components were
extracted from the data file. In the second panel, extraction by nearest neighbor
interpolation is seen. Quivers illustrated in Figure 4.5 are more sparse than in Figure 4.3
as a result of the time intervals in the data; Mercator data varies every 6-hours whereas
the CNOOFS data varies hourly, producing a more dynamic search trajectory. Here the
average difference in direction in the SLDMB trajectory versus the CANSARP Scientific
predicted trajectory is 82.60° while the average difference in distance is 1.13 x 103 m.
The next verification was of the Monte Carlo method. This process was fairly
straightforward since the Monte Carlo trajectory is calculated in the same way as
MiniMax. In the CANSARP Scientific Monte Carlo method, the search area is
determined by randomly dispersing a defined number of particles such that the mean of
all the particles indicates the most likely location for the drifting object being sought
after. Accordingly, the mean of the cloud of dispersed particles was determined to ensure
that the search area was appropriate to the calculated drift. The mean point calculated by
the CNOOFS currents is at (46.98, -47.83) while the mean point produced by the
Mercator currents is at (46.95, -47.83). The calculated Root Mean Squared Error of the
distributed points to the mean for CNOOFS is 334.56 m while for the Mercator currents it
is 459.28 m, thus the Mercator simulation produces a slightly larger search area. The
mean search radius for the CNOOFS currents was 1.69 x 103 m and 2.25 x 103 m for
Mercator. This would produce a maximum search circle of about 16 km 2 . The feasibility
61
of searching this area will depend on a nwnber of factors including the nwnber and type
of search units available, and their location, but this size is generally a non-issue. The
MiniMax search area for the same scenario averages 24.5 km2
Figure 4.6 illustrates the mean particle location and the distributed particles using both
current sets:
47 11 47, ,-~.-------,,-----,~r=========il
--:~--:~--.~r=
. =;Mo=:=nt==e~C=;arl=;oP::"'llltl::::c::::les===i] • Monte Carlo Pe11icles
• Mean Particle Location • Mean Pellicle Location
- CANSARP Calculated Trajectory - CANSARP Calculated Trajectory
e LKP POSition • LKPPOSitiOO
47.05
.. 47.05 .
47
4695 .
46.9 469 .
46
~8 -47.95 -47.9 -47.85 -478 -4775 -477 -4765 -476 46
~8 -47 95 -47 9 -47 85 -47 8 -47 75 -47.7 -47 65 -47 6
LongHude
LongHude
Figure 4.6 Mean position of250 particles after 24 hours of drift with SLDMB I 73 16 using the
Monte Carlo method for the a) CNOOFS and b) Mercator Currents.
Analysis (Figure 4.6) shows that the mean particle location is indeed at the end of
the CANSARP Scientific calculated drift trajectory, and is situated in the middle of all
randomly generated particles indicating that the determined search area is centered
Finally, the drift trajectories were verified using 12 SLDMB drifters from August
2007 by comparing the MiniMax Method output to the Monte Carlo Method output.
62
....
~-,
~~:>~,
i" Sl}·~ ··:'\
.... ~~:t~~;:J
....
- .•.
-~~,--~~---=----.~ ---7,~--.h----7,
·~,
"'
·-...
.,..
.,
....
....
: . ···..
............:...
!~ ....
...
48 15
.
.... ...,
·~~,--~~--~----~~.--~~--~.-~
·- ·-
..~b,.-~~--~.---.~.~.--~~--~.---7. ....
""Ji• _..,. _..,. """ _.., _..,
longll...
_..,
"'
., .. 471 ·
.,..
490
....." ....
47 4955
___)
.... ... 495
I....
~·
i .....
... ~
.... a~ 4945
:!!
4015
4015
' ·1
... , ... ,
I
I .,~',,
' '.. _...__...
., ·..
494
..... ....
\ /
4935 ·!·.
••;;. -415 -41
""Ji• _..,. ..474 -47.3 A72 A71
_,.,
Longitude
4955
49.5
! 4945
~
4 9.4
4935
.,,
49
~6 -5055 -505 -5045 -504 -5035 -503
LongiUdO
Figure 4.7 SLDMBsfrom August 2007 simulated with 24-hours ofCNOOFS currents using the
Monte Carlo Method with 250 particles.
63
64
~~ ,) '- . -~--~ .......
<705
....
...
i .....
I
~ ....
4675
.,
,_
-~~.-.~~~,~.~~-.,r,,.,.,~,~~,~ ~~~~~-7, ••• "'J! 1 ..oi80!J _.. -"'M ..o~tt -47t5
l.._,..
""" .aJ~ -477 _.785 ..ne
4665
46J;h,~~..*,,~--~,~.--~,~,~~.. .. ..*,,.---..~,~,--~..,
<705
47.05
,_...
"'
48Jil5
"
... "'"
...
• 48.85 . "'
4675
~
~ ...
.. , 4875 .
., ...
.... 48.05
,..,......
..~,_•.---..~,••---...,~.~~..
46J!~.--~ .
;t,,. ---..~,.,--~,,
'"J!o .A7.5 ..,
. ,. ,.._ A)J . , ..,
4935
... 4
~-lOs -50.55 -50.5 -so•s ,_...,.
-504 -50!5 -503 -5025 -502
....
~
4tl5 .
4945
~ '\
....
., ...,..,
... i
+· . ..
.....
,_.... ..,,
49
las -Sl5.5 .505 .sueS ~• -50315 -503 -5:125 -502
- ,!o,----._.,t._,-.---._,;t,.---_,.;, "
"'.1-~....---...,;t,. ---._.,.t,. -,- --_
..
Figure 4.9 SLDMBs from August 2007 simulated with 24-hours ofMercator currents using the
Monte Carlo Method with 250 particles.
65
~
I ·· i " I"
•.jj,
-
.,~
..,, •.~;,
·-
~·
•.~; ,
""
.,.
,_..
.,
""
...
~
§ ....
I" ~ "
'
..J 48.8
-4875
..,
... ...
"'' ·- •.~; ,
- ....
..;;. ~15
,..,.....
~13
~" ~1
..
47 05
47
4705
47
...
46 05 4605
""
469 ·····: 469 ...
• 4685 ; • 46 85
-g ~ 14945
li ~
....j 46 8
46 75
46.8
4675
...
467
467 ..,.
46 65
..... .... .... ..., ..., ...,
4665
"""'"' .. -50.25
longitude
Longitude
...
...
411
.l0e -50.55 -505 -5045 -50 4 -50.!5 -503 -5025 -502
""""""'
Figure 4.10 SLDMBs from August 2007 simulated with 24-hours ofMercator currents using the
MiniMax Method. See Figure 2.5afor location with respect to bathymetry.
66
In each of the above figures, the calculated trajectories agree for both the
MiniMax method and the Monte Carlo method, and thus the implementation Monte Carlo
In 2006, C-CORE worked with Oceans Ltd. and the Canadian Coast Guard to
these tests, all uncertainty parameters were set to zero (downwind and crosswind leeway
standard deviation, sea current components standard deviation and wind components
standard deviation), and the leeway divergence coefficient was changed to 1 to produce
the largest spread of particles possible (Choisnard, 2006). Winds were taken from the
CMC GEM Regional model and sea currents from the climatological grid, with wind
driven current calculated according to the Rule of Thumb Method (Choisnard, 2006).
From the 18 tests conducted, conclusions indicated that the difference between two
simulation positions is usually less than 2 km when the drift is less than 24 hours
(Choisnard, 2006). It was concluded that the Monte Carlo code in CANSARP Scientific
67
Chapter 5 Exploring Issues and Possible Change for
CANSARP Scientific
quite obvious that the searches generated were too short in length to reasonably predict
the drifter trajectories. The following sections will examine the underestimation of
length using a number of approaches, and will discuss possible ways to improve this
ISSUe.
Throughout this project, Y4° CNOOFS model output currents were being used in
point that these experiments were run, a higher resolution version of the model was being
whether these higher resolution sea-surface data files would improve the search trajectory
produced by CANSARP Scientific. While the new model output data was not ready to
test in this thesis, an attempt to explore the impact of physical processes that govern the
surface layer dynamics and the role of the surface layer Ekman transport was taken into
account. This was done by calculating the Ekman velocity components according to
Kundu (1990) equations 5.1 and 5.2, and comparing these values to those extrapolated in
68
(5.1)
(5.2)
In equations 5.1 and 5 .2, u and v are the horizontal and vertical components of current
velocity from the CNOOFS model, 7 represents the wind stress calculated from the CMC
GEM Winds data at the time and location of the simulation, z is the depth from 0 to 100
m, and pis the density of water, taken to be 1027 kg/m3 . The Coriolis Parameter,!, is
observation, and 8, the thickness of the Ekman Layer is determined as 15 ~ ~ 2;• where
Vv is the eddy coefficient taken to be constant at 0.01 m 2/s, as per Kundu (1990).
The Ekman velocities were computed for 1 - 100 m depth from 7 (the wind
stress). If the calculated wind speeds are high, then the surface current is probably
affected substantially by winds, thus accounting for the large (approximately 3x larger)
drift produced by CANSARP Scientific versus the SLDMB trajectory. If this were the
case, then the CNOOFS model, which only has a surface layer resolution of 5 m may not
Plots were generated for each of the 12 SLDMB LKP positions and times in
August 2007 for 1-hour intervals in the CNOOFS data. There were also sample plots
69
done at (38, -55), (40, -45), (40, -55), (40, -65), (55, -55), for both CNOOFS and
Mercator data on August 7, 2007 at 01:00. These samples represent locations in the
Labrador Current and within the sub-polar and sub-tropical gyres. A sub-sample of these
70
48.6
-10
4 8.4 .65
;... , . ;. .
.
-15
-20
"'
15.
,ll -30
l •. • -35
-522 -52
-5
-10
........
. 5urllocaot11
St.C»oeT....
-15
-20
~ -25
"'
15.
~ -30
-35
..... f--
-40
C81c1J8le<1E~Velocity
-45 U.Componert OfCalclial:ed Ekman V'*>aty
.... -
- V-~ItoiCBicliatedEkmanVelooly
486
-5
48.5
-10
48.4 ·i-
-15
48.3 -20
482 .
E-25
1-30 ... i
48.1 .j ...
-35 -~·······
-40 .j.
48
- Cak:lAate<l Beman Velocity
-45 "''•:·· ··· - ~C~oiCBic""'edEkmanVelooOy
- V-~ ol ColcUoled Ekman Velooly
47.9
.. .( -e-CNOOFS Velocll'f
-50 -+- lJ.C~d~S
-+- V-~olCNOCIFS
47.g26 -52.4 -52.2 -52 -51 8 -0.3 -02 -0 1 0 0.1 02 03 04
Longlude Volocily (mls)
Figure 5.1 Search trajectories and velocity profiles using CNOOFS currents for SLDMB 17303,
17316, and 17347.
Column 1: Drifter trajectory and CANSARP Scientific Search trajectory over 48 hours
Column 2: Calculated velocity profile and CNOOFS current velocity profile comparison
at start location ofdrift
71
The calculated Ekman velocities were small compared to the CNOOFS averaged
velocities at 3m depth. This indicated that even with data extrapolation, surface velocity
values for CNOOFS currents would change minimally from the velocities at the 3 m
depth - certainly not significantly enough to account for the discrepancy between the
Of interest in this study was that not all velocity magnitude profiles behaved as
anticipated (speed greater at surface than at depth). For example; at (40, -65), the 3m
speed of the CNOOFS currents were less near the surface than at a depth of ~ 95 m
(about 0.03 rn/s difference). Generally, current magnitude decreases with depth, and the
Figure 5.2 shows this profile, with the CNOOFS current velocity magnitudes
offset such that they dissipate to 0 rn/s at their last recorded point.
72
0
Figure 5.2 Velocity magnitude profile using CNOOFS data for a simulation at point (40, -65).
It should be noted that this experiment assumed constant vertical eddy diffusivity,
and this may make the calculations inapplicable since this component becomes smaller
Overall, while this is not a conclusive study, it can be said that higher resolution
in the depth of the current data may not necessarily improve the overall drift calculations
in CANSARP Scientific.
73
5.1.2 Radius Determination
Another observation made upon examining drift and search trajectories of drifters
was that drifts react with varying geographical location, and particularly similar were the
It was observed that the predictions made by CANSARP Scientific were of reasonable
direction in both of the above cases, but the search trajectory was being underestimated.
Drifters that did not fall into either of the above 2 categories were classified in a grouping
CANSARP and CANSARP Scientific both apply a "CANSARP Safety Factor" that
increases with the iteration ofthe search attempt. This safety factor's purpose is to
ensure the containment of the search object, based on elapsed time since primary search
efforts were taken. For example, the first search's radius is multiplied by a factor of 1.1 ,
the second by 1.6, the third by 2.0, etc. Each of these factors was determined
These factors are independent of the oceanographic environment the drift occurs
in. Here we consider revamping the use of a constant safety factor to make it dependent
74
the 12 SLDMB drifters from August 2007 off the coast ofNewfoundland, it was found
that the radii of the drifters on or near the shelf (based on cumulative error of
safety factor of 8. 7 in order to incorporate the search object. The radius of the search
areas affected by eddies however, would only have to increase by 2.1 times overall, or
apply a factor of2.3 to successfully encompass these drifters. Of the three drifters in the
encompassing of the drifter required anywhere from 3.2 to 7 times as large a radius. The
possibility of further examining this group with a larger subset may lead to more detailed
results.
While expanding a search radius by 7.9 times (producing an average search area of
nearly 12000 km2 and still only successfully retrieving an average of 40% of objects) is
not feasible in most search cases (due to limited search unit availability/resources and
time), implementing a larger radius combined with improved environmental inputs may
lead to a higher search success rate. On the contrary, implementing a radius that is 2.1
times larger in regions exhibiting mesoscale eddies may be reasonable as the average
search radius using this factor is 11 km, providing a total area to be searched of almost
400 km 2 • As mentioned in Chapter 4, search time and the ability to cover an area in some
reasonable time will depend on the available search units, their location at the time of the
incident relative to the search area, their type, and the number available to complete the
search plan.
5
In this context, "overall average" refers to an average including the presently incorporated CANSARP
Safety Factor
75
5.2 Optimizing the Number of Particles used in the Monte
Carlo Method
Once validated, the Monte Carlo Method was explored in terms ofboth efficiency
and productivity. One way to find a balance in both is to alter the number of particles
used for simulations, since this is a user-controlled variable and is simple to change. In
an effort to quantify the optimum number of particles for use in the Monte Carlo Method,
It was established that both Norway (Breivik & Allen, 2008) and the United
States' (Frost & Stone, 2001) outdated CASP model apply 500 particles in their search
algorithms. The United States now allows an option of 3 modes entitled "fast,"
"normal," and "comprehensive." These options apply 2500, 5000, or 10,000 particles per
scenario, respectively to allow the user to choose between speed of search planning and
statistical validity. When cases have more unknowns, rapidly changing environmental
carried out whereas most cases apply the "fast" mode in order to reduce processing time
76
At present, though it is modeled after the Norwegian version of Monte Carlo, the
As a benchmark, one drifter was used to test the length of time and the coverage
ofvarying numbers of particles. Table 4 contains the results ofthese simulations over a
Table 4 Data from Monte Carlo Simulations over 48 Hours with SLDMB
17324 with Mercator Currents and Varying Numbers of Particles
These simulations can be compared to the same simulation using the MiniMax
Method for the same amount of time with the same drifter. This experiment yields a
To determine the maximum radius ofthe particles, the distance from end of the
search trajectory to the furthest distributed particle was found. As the software functions
now, it requires less time to run few than to run several. The radius increases somewhat
77
47.1
47.05
~~~~,
47 /JL .... L ..... ,~~ ---
' II : : II
I I ; II
..
.., .
I I
I
: II
II
I I
..... Ir ·l' .
~ 4695 .. ... It·
iij II . . I I
:~
...J
t., - : :
~, :~ -~~-------------;~~/
46.9 .. >~:;::::::::::::::1::;:;:~~ .. ;
46.85
8------~~----~~~7.8~-----~~7~.7~------4~77
46 ~~ 6------~~~75~-----~~74
Longnude
Figure 5.3 Monte Carlo search areas with 1000 (green dashed line), 500 (red dashed line) and 250
(blue dashed line) particles after 24 hours f or SLDMB 17303. Areas increase in size with
the number ofparticles applied with values of0.0183o2, 0.0258o2, and 0. 0328o2,
respectively . The solid blue line is the SLDMB actual traj ectory fo r 24 hours while the
green dot represents the start of the drift and the red dot represents the end of the drift.
While this method already provides a larger radius of search than the MiniMax
Method, selecting the ideal number of points for use depends on the amount of time the
search planner has, and the available resources. In time and produced search area, it
appears from Figure 5.4 that running a simulation of 1000 particles is reasonable as the
area is searchable and not drastically different from an area produced by 250 particles.
This would require about 19 minutes to process, in addition to the search planner's task
of assigning the search plan with tracks for the search and rescue units to follow.
Ideally, basing the number of particles run on the resources and time available for
planning would be implemented in a manner similar to the USCG's approach where the
weights of the statistical significance and time available are weighted in determining how
78
many particles to apply. Selecting the number of particles based on the current behavior
in a given region may be of use. For example, it would be useful to be able to select a
higher number of particles for a simulation in a region where the currents are non-
uniform and have small-scale detail, whereas in regions of more uniform motion, fewer
particles could depict an equally useful search area. One proposal for future development
in a region could be determined to converge or diverge based on its Root Mean Squared
Error. If the value were larger than the benchmark value, the particles would be
diverging and more particles could be applied to the search simulation. Similarly, if the
RMS value were smaller than the benchmark, the particles would be converging and
set adrift from the slope east ofthe Strait ofBelle Isle. It drifted from 53° 10' N, 50° 54'
W on this day at approximately 18:00Z and moved slowly northeastward, then seaward,
79
2811
311
2500
a
--
3000
7.5 l<m
Figure 5.4 Drift trajectory of lost mooring starting from 53° I 0 ' N, 50° 54 ' W at / 8:00Z on May 17,
2008. Reported locations illustrated fro m ARGO satellite f zxes.
On May 20, 2008, a notice was sent out from the Bedford Institute of
requesting that if it were not a difficult task, to try to predict the trajectory of the mooring
for retrieval on May 21 by the Canadian Coast Guard vessel, the CCG Hudson.
In an effort to assist this team, and to use this opportunity as validation for
Figure 5.5 at 10m depth. Simulations were also run at 50 m and 100m depth, and
80
53.2
··1········ ····· ·······...........................
- - Mooring Trajectory
• Start Point
• End Po1nt
524
52.2 ....
52
51 .8
-51 .5 -51 -50 -49.5 -49
Longitude
Figure 5.5 Predicted trajectory ofmooring from 18:12 on May 17, 2008for 94.3 hours using
CNOOFS currents at 10m depth. Actual drift trajectory is in blue with start location at
location at 94.3 hours noted. Search trajectory and area are plotted in black.
Bathy metry is contoured by multi-colored lines and labeled according to their depth in
meters.
The predicted search trajectory follows the shelfbreak for the duration of the
simulation whereas the mooring traveled almost perpendicular to this prediction. Since
the LKP given for this case is in the Labrador Sea, it would be expected that the current
in this region be strong enough to carry the mooring in the direction of the prevailing
current, as the predicted trajectory goes. However, this was not the case, and further
81
Altimetry data obtained from the Aviso website (http://www.aviso.oceanobs.com)
illustrated the daily averaged surface velocity fields (called gridded absolute dynamic
velocities) and sea surface height (called gridded absolute geostrophic velocities). This
Envisat), and is represented on a 1/3° by 1/3° on the Mercator grid, available as daily
averages. These velocities are computed using geostrophic method (Picaut and
McPhadden, 1989). It is based on the assumption that the surface pressure gradient is
balanced by Corio lis acceleration due to the surface flow. This method is reliable outside
velocities plotted as quivers, and the overall surface velocity field as contours.
82
54 .. . ....... "f ........I...
·~ · ~· ..·;/._ ....·;
\ \ \: ,
15
\ 10 15
'
, : '
53.5 .. .. ··r ·:···· \······· , ....... l ... ·· ··1······· ·i ········v ·· . . . ~ . ...... .... .. .\ ... .... ······ ~
. .
~ 20
53
·······<··]····\········\·
' · ~
I · ······\··· ·
····~ ··· · · · ··:······· ..... ·· ··········· ·!····· ·· ····· ·:··············:
Q)
"0
\ \
~ 52.5 ... ... T ·:····r ······s ·· ····y· ....~
...J 10 _;,
I .' \ \ \
52 ........................ . \ \
\ \
15
, \ \ :
' \ \ '
51 .5 .. ····; ....:.......... .
\.
....... \ .. \ :
\' :
: 1
\ : \ ~ \
I i l \ "
-52.5 -52 -51.5 -51 -50.5 -50 -49.5 -49
Longitude
Figure 5.6 CANSA RP Scientific simulation on May 17, 2008 for 94.3 hours starting at 18:12:00
UTC run with CNOOFS currents with swface velocity field as quivers from AVISO
altimetry data. Contours represent mean sea surface height in em. Quivers represent sea
surface velocity for the 1 l'h ofMay. Black trajectory is the predicted drift trajectory by
CANSA RP Scientific and the circle is the proposed search area.
While this figure verifies that the CANSARP Scientific search trajectory traveled
with velocity of the currents for that simulation, no further insight is gained regarding the
83
A daily plot of the mooring and its respective daily averaged sea surface height
84
~O r---~-------.--~---.----r---.---.----,---,
~2
(-60.8881. ~.1!16) "(53.1.SS.. -60.811}
-8
~ 53
528 528
526
o_
520 "o ,
,., -51 8 -515 -51.-4 -50.8 -50 15 -50.4 -50 2 .5(1 -52 -518 -516 -514 -512 -51
longlude
-508 -5015 -504 -502 -50
53•
/"'~,13>4, -40.861)
153.o8TI, 41.1300) :
528
526 520
..52 -51 8 -516 -5 1 4 -512 -51 -508 -505 -504 -502 -50 -52 -518 -516 -514 -512 -51 -508 -50 6 -504 ..502 -50
longl:udt longlude
~~-1334,-60.811)
tl53.odll-61.1172) f .
528
526
-52 -51.8 -51 .6 -51 .4 -51 2 -51 -508 -506 -50<4 -50 2 .so
longitude
Figure 5.7 Daily progression ofmooring trajectory (black) with corresponding daily averaged sea
surface height altimetry data on (L-R): a) May 17, b) May 18, c) May 19, d) May 20,
e) May 21. Drift trajectories are plotted from May 17 to end ofday altimetry is plotted.
The green dot indicates the mooring start location and the red dot indicates its last
position on each day Value ofsea surface height is indicated on colored contours in
units ofcm.
85
Figure 5.7 shows the mooring following movement of a water mass with a relatively
low velocity west - southwestward. The sequence of sea surface height observations
suggest that during this time, a meander may form and grow along the Labrador Current.
The southwestward drift ofthe mooring follows the direction of fastest growth in this
meander. Mesoscale meanders and eddies are energetic elements of ocean circulation
and have spatial scales close to the internal Rossby Radius of Deformation. This radius
changes as a function ofthe latitude and vertical stratification (Gill, 1982). In the
Labrador Sea, the Internal Rossby Radius is about 20 km, whereas this value increases
toward the equator to values of about 240 km (Chelton et al., 1997). While the effect of
the mesoscale meander is seen in the altimetry data, neither geostrophic current (averaged
over 1/3° by 1/3° squares nor the CNOOFS model output data are able to resolve motion
under~ 20 km diameter, as seen in the mooring's track (found to have a Rossby Radius
al. 's (1997) mapped gravity wave phase speed value at 47°) that is probably responsible
for the tiny fluctuations in the actual trajectory as well as the overall direction of drift.
After the first day's drift, both the altimetry and CNOOFS data were reasonably
able to explain the mooring's path, but because the drift simulation began in a poorly
resolved region, CANSARP Scientific predicted day 1 in the incorrect direction, and so
the following days were also predicted incorrectly, as a result of the initial error.
86
5.4 Analysis of Errors Produced by CANSARP Scientific
The observed mooring's behavior in the preceding case demonstrates that the
CNOOFS model in CANSARP Scientific was not sufficiently accounting for certain
small-scale behaviors in the ocean. These types of model errors are unavoidable in
forecasting models. One way to account for this error in search predictions from
CANSARP Scientific without more refined model data is to quantify the model error and
model predictive skills and to use this information to optimize the Search and Rescue
procedure.
predictive skills ofSLDMB position. The purpose ofthis experiment was to evaluate the
error ofCANSARP scientific and CNOOFS predictions ofthe SLDMB positions. The
model error for each case was quantified in an orthogonal local coordinate system that
had one axis parallel to the local direction of the current and a second perpendicular axis.
CANSARP Scientific was run for each of the 12 SLDMB drifters from August 2007 for
each hour over a 48-hour period starting from the actual drifter location at each respective
time. This was done (rather than starting from the computed location after each hour) so
that the error calculated was not cumulative, but rather individual for each hour, in order
Figure 5.8 shows this result with a different color representing each drifter. The
87
(5.3)
where s 11 is the error in the position in direction parallel of the surface model current, and
& ..L is the error in perpendicular direction, and fJ values represent points of the parallel
and perpendicular vectors actual and computed (denoted by subscript 'C'), respectively.
. ' ....... ..1.......... ' ... 't ' .... ..+........ - ~ ...........:........... !........... :........ .
2500 ······· · ·-···················· ..••1.
... 1 ••• :_ .. :
. .: .
. . . • • • • ••• 1 ' ' ' '
:
. ........ ........ t. :
"E
..
..
•
:.
.:. .. : ~ ·~
15001- ....
. . .. ..• . ··:···········~········· . .... . ...... . ~·· \:t..... ~.· ........ · · ····· · · ·· ····.····· ............. .
.. .
.. :...
....
::I
u
m
:
1000 r- .. ....... ; .. ......... ~ .....
. .~ .
.. :. ..• ..... ;.. .. ;- ...........:........... .. ~
~ ..
. ..• : : .•.,. .
u
.3
~
~ 500
•:
•
.......... ,.. ........ . -.....
... •·r,... . ·.. :...........:.
.
~
··:..... ... ~- .. :-· .. ~ ·-····. ·· ... -......:.....; ...
:
~
:
~ :.
.
·~:·::-:· ·· ;.~ ~>
·-
u
"0
c
·~
• : • · · ~ ••• • :• •
0 ~--~----+---~--~~~~~~-+~-r~----~--~--~~~--~--~
.. . :. •
. . .. ".. · . : • • ••• •
. •• .:
:
• •
~
~ . .. ~ ~· . ..,!!•• ;: · • +
...... :+ • ' •: • : ~· : •. : !•
Cl) : • • : • :'t • • • .:. ~ : : ' : •
(L : • •~· ' ·. : · .. • • • -. · . : : .: • ~ ·.- ·. :
"E -500 ······ ···.;·. ···· ·· · ··~--~··· • ·· ! ' ' '~ ~--···.·~. ······ ·: · ·····.··•···:_··'- ········ ·:······ .. ~ - ~ - ...... ~-- ~ · ~( ·· ~·t········· -
Q)
:• • · ~ · . t ~: .
• : : •
: : •
· .
~. · . ! .· .~ .
E
Q)
u . .. • •. :' ! • • • ~: • • • • •• • .
•...• .. =....... .. ........ :........... -.:...........
• :•
.!!
c. -1000 ~ ......... :; ........... ~ .... ... ... ~ . ..... ~ ..· ..
:· .... ~ .. ·:·. ~~... , •.. -~ ·; · ~ . • • ~.
.~
0
-1500 ~ .
·. : : :
.
.. ... ·.
ft- :
~
• --. • • ; · •• • .
.. " ................ ...... ,: .. ......... :. .. ~. ~ .... ~ . .......... .. .... .... ·:......... ...:............ :............:........ ..
.
Figure 5.8 Displacement calculations in the x andy directions where the current velocity is
projected parallel and perpendicular to the local current in the region over a 48 hour
period. Each of the 12 drifters explored are indicated by a different color dot in the p lot.
The mean displacement on this plot is (656.6991, 74.4230).
Figure 5.9 shows the error in computed drifter positions. The predicted drifter
along the direction parallel to the local current. This in particular implies that the model
underestimates the surface current velocity or that the predominant current pattern from
88
the current model (CNOOFS) may be shifted left or right of the actual current pattern. It
can also be said that they-component of the error is usually positive, explaining that the
floats tend to propagate right of the computed velocity. Figure 5.9 shows the error
calculated as:
(5.4)
20.---------or---------.----------r---------~---------.----------.
... ....
15 ...... ......... . .......... ......................... ................. .... ... .
"E
: ..
0
~
3 10 .. ..•• ~ .....
:
. . ....... ........ ·-
ii
u
0
. .... ...
... .
-1
.2 5- ................ . ,... . .... , ....... . .. .. ........ v .. . . ..... ~ ............ ' .' ............ ·:- . ......... -
.... )
~ ..) ·. ~ 1
...,. • • •. ••
•• _, ••
u •• • • \ • · ·~.a ·
=o : •• .._
... ~
~ 0 ~--------~--------~~----~~~~~~~~--------~--------~
~ ·~ ~A~~: ·
~ • t. ~~. .
~ -5 ........ ........... , ) .... .... .................. :.................... ·.:~-. ~;--;~-~- . -~ ... • .......! ........... ~- ....................... .
'0
Ql
!::!
ii -10
E
0
z
-15 ............. . . ... -
- 2~L
0----------~2~
0 ----------~10~--------L
i
0 __________1L0--------~20~------~
30
i
Normalized Error Parallel to Local Current
Figure 5.9 Normalized error calculations in the x and y directions where the current velocity is
projected parallel and perpendicular to the local current in the region over 48 hours.
Each color dot represents the error produced by a diffe rent drifter. The mean error
position is (-0. 273, 1.52).
89
..----,--- - - - - - - - - - - - - - - - - - - - -- -- - - -- - - - - - - - - ---- - - -----------
The cloud of points in the plot of error is extended in the direction of predicted
current velocity, suggesting that the errors related to the underestimation of current
incorporate this error analysis in the planning of a Search and Rescue algorithm is the
90
Chapter 6 Conclusions
In this thesis, the CANSARP Scientific software was first tested and validated,
with a number of minor changes made such that it would function properly. Once this
task was completed, search methodologies in the United States and Norway were studied
and compared to the existing MiniMax Method in Canada. Environmental datasets used
Based on the available data, the present MiniMax drift algorithms in both
CANSARP and CANSARP Scientific were explored. In doing so, it was determined that
all of the current estimation methods (Rule of Thumb, Ekman, and Madsen)
underestimate the length of actual drifting buoys by 2.3 to 3 times, and inertial
and tested in CANSARP Scientific. While this algorithm accounts for inertial
buoys, and it still underestimates the total drift length by as much as the existing
estimation methods. Possible reasons for this are that the climatological currents are a
poor representation of the realistic current motion on the tested dates, the CMC GEM
winds predictions are inaccurate for the tested dates, or that the validation of the wind
(modeled after the Norwegian seach program "LEEWAY") was investigated. Because
91
this method has a stochastic component, it was thought that the search area produced by it
would be more accurate than those from the MiniMax Methods. This was not the case as
the predicted drift trajectory calculation is the same in MiniMax as Monte Carlo, and just
the search area determination varies producing the same issue of length underestimation.
The most potential for future development was found when search simulations
were run using model output data. In this thesis, CNOOFS and Mercator model output
were applied in CANSARP Scientific. The direct application of this data to search
surface, the CNOOFS data produced drift predictions of about 3 times shorter than the
SLDMB trajectories, while the Mercator predictions were on an average of2.7 times
shorter. Because of the nature of the data (based on actual environmental inputs rather
First a study was carried out that tested whether higher resolution in the CNOOFS
model output data would improve the search prediction. This experiment was run since
the CNOOFS model output files are depth-integrated and do not presently account for
true surface velocity; a feature that was suspected to affect search prediction trajectories.
and plotting them against the actual CNOOFS data. This study was inconclusive, but the
extrapolation ofthe CNOOFS velocity profiles to the surface layer indicate that even
with a higher vertical resolution model, the velocities would not impact the drift
92
A second study explored the possibility of increasing the search area by a factor
determined based on geography. Ideally this would be an excellent solution, but factors
determined from this study indicated that the search radius would be unrealistic to cover
in an average SAR mission. Further research into this idea may be useful for tweaking
A third experiment in an attempt to find a lost mooring provided the most insight
regarding drift patterns in the North Atlantic Ocean. As a result of an inexplicable drift
direction of the stranded mooring, altimetry (both sea surface velocity and height) data
was obtained in efforts to find some subscale process that may have caused the unusual
drift behavior. Although the altimetry displayed some general patterns of the current
flow, the daily-averaged data was not refined enough to observe any anomalous
processes. It is thought that the baroclinic structures that lead to velocities unresolved in
the ocean forecast model may be one cause of this behavior, but no proof of this is found.
The real-life mooring experiment proved that the model resolution is not
sufficient to represent all small scale behaviors in the ocean, and that even a transition
from 1/4° to 1/12° model output data will probably not resolve many of these. In lieu of
this, developing a search procedure that accounts for the errors characteristic of the model
output would provide accurate enough search regions to find the search object.
Resultantly a procedure was developed to calculate the model error of a dataset based on
a known drifter dataset. This procedure can be applied to any model output data of
93
current velocities before they are used in CANSARP, with the intention of incorporating
Overall, this thesis has provided insight into the shortcomings of the current
Canadian Search and Rescue drift calculation theories, and has brought a new search
concept based on model error determination into the forefront. With future work planned
to implement an improved search approach based on error analysis of model currents into
CANSARP Scientific, and to test it with a large drifter data set, it can be said that
improvements are certainly in the works for the search theories that originated in the
94
References
Allen, A.A., Howlett, E. (2008). SAROPS EDS Environmental Data Server. Retrieved
December 2, 2008, from www.marcoos.us/downloads/talks-
presentations/MARCOOS_PI_ 2/Sarops%200cean%20Sciences.ppt
Allen, A.A., Plourde, J.V. (1999). Review ofLeeway: Field Experiments and
Implementation, CG-D-08-99. Contract report prepared for the U.S. Department
ofTransportation and U.S. Coast Guard.
Breivik, 0., & Allen, A. (2008). An Operational Search and Rescue Model for the
Norwegian Sea and the North Sea. Journal ofMarine Systems, 69, 99-113.
doi: 10.10 16/j .jmarsys.2007 .02.0 10
Brown, K.Q. (1979). Voroni Diagrams from Convex Hulls. Information Processing
Letters, 9:5 6.
Canadian Coast Guard College, (2005). Total Water Current. Unpublished Lecture.
Chelton, D.B., DeSzoeke, R.A., Schlax, M.G., Naggar, K.E., & Siwertz, N. (1997).
Geographical Variability of the First Baroclinic Rossby Radius ofDeformation.
Journal ofPhysical Oceanography, 28, 433-460. doi: 10.1175/1520-
0485(1998)028<0433 :GVOTFB>2.0.C0;2
Cressie, N. (1990). The origins ofkriging. Mathematical Geology, 22(3) 239-252. doi:
10.1007/BF00889887
Department of Fisheries and Oceans Canada (2007). Fisheries and Oceans Canada
Integrated Science Data Management (ISDM). Retrieved August 2008 from
http://www .meds-sdmm.dfo-mpo.gc.ca/meds/Home_ e.htm
El-Sabh, M.I. (1976). Surface Circulation Patterns in the Gulf of St. Lawrence. Journal of
Fisheries Resources Board of Canada, 33:124-138.
95
Environment Canada (2007). The CMC Numerical Output Database Meteorological data
in GRIB format. Retrieved September 2008 from
http://www.weatheroffice.gc.ca/grib/index_e.html
Fox-Kemper, B. (2002). The Ekman Current, SAIL Weather and Sea Column, pp. 40.
Retrieved from November 2008 http://fox-kemper.com/files/pdfs/Sail10_02.pdf
Frost, J.R. (Soza & Co.) & Stone, L.D. (Metron Inc.). (2001). Review ofSearch Theory:
Advances and Applications to Search and Rescue Decision Support. Final Report.
Report Number CG-D-15-01. Contract report prepared for the United States
Department of Transportation.
Kom, G.A., Kom, T.M. (2000). Mathematical Handbook for Scientists and Engineers.
Dover Publications.
Koopman, B.O. (1980). Search and Screening: General Principles with Historical
Applications, Elmsford, NY: Pergaman Press.
Kundu, P.K. (1990). Fluid Mechanics. San Diego, California: Academic Press.
Lenn, Y.D. & Chereskin, T.K. (In Press). Observations ofEkman Currents in the
Southern Ocean. Journal ofPhysical Oceanography . doi:
10.1175/2008JP03943.1
Madsen, O.S. (1977). A Realistic Model of the Wind-Induced Ekman Boundary Layer.
Journal ofPhysical Oceanography, 7: 248-225.
Murphy, D.L. & Hanson, W.E. (1989). Drifting Buoy Measurements in the Labrador
Current. Proceedings of the Conference on Marine Data Systems (MDS 89),
Marine Technology Society, Washington, D.C. pp. 219-224.
Northrop Grumman Space & Mission Systems Corporation. (2008). USCG SAROPS
Version 1.1.0.1 For ArcGIS Version 9. 2 Windows XP, Windows Server 2003,
Under Contract Number HSCG23-07-JTED150. Contract report prepared for the
United States Coast Guard.
96
O'Donnell, J.D. et al. (2005). Integration of Coastal Ocean Dynamics Application Radar
(CODAR) and Short-Term Predictive Systems (STPS) Surface Current Estimates
into the Search and Rescue Optimal Planning System (SAROPS). Final Report.
Report Number CG-D-01-2006 Under Contract Number DTCG39-00-D-R000081
HSCG32-04-J-100052. Contract report prepared for the United States Department
ofHomeland Security.
Picaut, J., Hayes, S.P., & McPhadden, M.J. (1989). Use of the Geostrophic
Approximation to Estimate Time-Varying Zonal Currents at the Equator. Journal
of Geophysical Research, 94: 3228-3236.
Pollard, R.T., & Millard, R.C. (1970). Comparison Between Observed and Simulated
Wind-Generated Inertial Oscillations. Deep-Sea Research, 17(4), 813-821. doi:
10.1016/0011-7471(70)90043-4
Saucier, F.J., Roy, F., Gilbert, D., Pellerin, P., & Ritchie, H. (2003). Modeling the
formation and circulation process of water masses and sea ice in the Gulf of St.
Lawrence, Canada. Journal of Geophysical Research, 108, 813-821.doi:
10.1 029/2000JC000686
Seaconsult Marine Research Ltd. (1993). CANSARP V3.2 Technical Manual. Prepared
for the Canadian Coast Guard Search and Rescue.
Schaefer, J.T. (1973). On the Solution ofthe Generalized Ekman Equation. Monthly
Weather Review, 101(6), 535-537. doi: 10.1175/1520-
0493(1973)101 <0535:0TSOTG>2.3.C0;2
Soza and Company. ( 1996). The Theory ofSearch: A Simplified Explanation, Contract
Delivery Number 96-F-HNG040 Under Contract Number DTCG23-95-D-
HMS026. Contract report prepared for the Office of Search and Rescue US Coast
Guard. Retrieved September 2008 from
http://www.navcen.uscg.gov/MARCOMMS/geninfo/Theory_of_Search.pdf
Tang, C.L., et. al. (2008). BIO Ocean Forecasting Models and Forecasting Systems for
Eastern Canadian Waters. Canadian Technical Report ofHydrography and Ocean
Sciences. In press.
97
Yao, T., C. L. Tang, and I. K. Peterson (2000), Modeling the seasonal variation of sea ice
in the Labrador Sea with a coupled multicategory ice model and the Princeton
ocean model, J. Geophys. Res., 105(Cl), 1153- 1165.
98
Appendix A Running Cansarp Scientific
simulate the output of the CANSARP program of the Canadian Coast Guard Search and
Rescue. Unlike the CANSARP program, CANSARP Scientific does not run from a
Graphical User Interface (GUI), and settings have to be manually typed into functions,
Within the 'cansarp_sim' folder for CANSARP Scientific, there is a folder called
'settings_ scripts' . This folder contains three .m files that contain constant values and
four .m files that can be edited according to a desired simulation. To set-up a simulation
A-1
CANSARP- DRIFT- CONFIDENCE- FACTOR: Values of 0.125 or 0.3. It is assumed
to be 0.125, unless there is very little known about the drift in question. Default: 0.3.
SAVE_PLOT_OUTPUT: Set to 'no ', 'iterative' . or ' last' such that no plots, each
timestep plot, or just the final output plots are saved. Default: ' no'
MINIMAX_PLOT_TYPE: Set to 'all', ' last' , ' half, or 'quarter' indicating the
approximate number of iterations to plot. Default: 'all' .
A-2
CANSARP_CURRENT_TYPE: IfCANSARP_USE_TWC is marked as true, option
can be used to determine which current type to use. Options are 'constant', 'noofs' or
'mercator'. IfCANSARP_USE_TWC is false, this variable is not used.
Possible values:
1: Regional GRIB CMC file ('CMC GEM REGIONAL 15-km');
2: Global GRIB CMC file ('CMC GEM GLOBAL 0.9 deg');
3: CANSARP standard 2-degree CMC wind ('standard CANSARPV3.2 CMC 2 deg
wind');
41: CNOOFS winds ('NOOFS ')
42: CNOOFS winds with CSAR-wind ('NOOFS-wind + CSAR-wind')
5: CSAR-wind map and CMC REG GEM when available, old CMC wind otherwise
('CSAR-wind + CMC GEM REG');
6: Scatterometer winds, both ERS2 and QuickSCAT, and CMC REG GEM when
available, old CMC wind otherwise ('scatterometers wind+ CMC GEM REG');
A-3
61: QuickSCAT and CMC REG GEM when available, old CMC wind otherwise
('QuickSCAT wind+ CMC GEM REG');
62: ERS2 and CMC REG GEM when available, old CMC wind otherwise ('ERS2
wind+ CMC GEM REG');
99: Constant wind vector manually specified ('constant wind')
In path_ settings.m, the path representing the data that is to be used for winds,
currents, and field data must be specified to correspond to the selections made above
A-4
~---------------------------------------------------------------------------- ------
Running a Simulation
Once all of the settings are prepared, a simulation can be made in CANSARP
Scientific. To run such a simulation, the following variables must be declared in a script
drift start time: The starting date and time of the drift, input as:
datenum(yyyy,mm,dd,HH,MM,SS)
time interval: The total length of the drift in hours. If this is not input, the entire
data from the field experiment is applied, if available.
drift_start_position: The LKP for the drift given as [latitude longitude] in degrees.
Once these variables are declared, the following can be typed into the command
line in MATLAB (for simplicity, a calling function was composed for this project):
The results will yield a matrix of positions corresponding to the calculated drift
expressed in the wind_ drift variable and the drifter variable will contain a structure
including information about the drifter in the simulation including its name, LKP
position, LKP time, the number of errors associated with the simulation, any previous
errors, and the total drift time of the simulation. A MATLAB figure is also produced if
A-5
Appendix B Changes Made to the Monte Carlo Method
in CANSARP Scientific
While no conceptual changes have been made to the Monte Carlo Method, some
• Vectorization of the number of particles run was applied and tested. In the
original version of CANSARP Scientific, a loop exists such that for each particle
that is seeded, all computations must be done on each particle. It was attempted
to vectorize this process such that all particles undergo computations in a matrix,
but time savings were minimal and the time to reprogram outweighed the benefit
of vectorization.
• The scripts that were used to process the Monte Carlo Method using Mercator
currents were edited such that only current files for new dates are loaded, rather
than repeatedly loading the same files. This change resulted in notable time
B- 1
Time Cornpanson Between Old and Edrted Monte Carlo Script lor Dnftet 17316 for Varying Times with Mercator Curr11nts for 100 Particles
~ 2500 .--:.::--:Ti:::-rm-e-,-fo-r-::0-ld_V_e-re-,-on-,
"~ ...,._Time for New Version
1/l. 2!XXl - - · Overall Improvement
§
a:
E
~
"
-"
~
1-
...................... , ........................... ... .
~"
~ 500
tii
~
..
-' 00 10 15 20 25
U!ngth of Monte Carlo Cansarp Srmulation (Hours)
-;;;- • 10' Time Companson Between Old and Edrted Monte Carlo Scnpt for Dnfter 17316 with Mercator Currents lor 24 Hours and Varying Numbers of Partrcles
2
-g - + -lime for Old Version
~ - + -Time for New Version
~ . - - · Overall Improvement
15
a:
Figure B-1 Time improvements made in Mercator Monte Carlo script by number of hours of
simulation run and number ofparticles used.
B-2
Appendix C General Changes Made to CANSARP
Scientific
was in the way that currents were extracted from the CNOOFS files. The current
extraction method written for CANSARP Scientific was geared towards the Mercator
currents which are structured on a regular grid as seen in Figure C-1 . Applying this
method to the CNOOFS currents of irregular grid resulted in values for currents that were
sometimes in quite inaccurate locations, and thus produced incorrect drift simulations.
In the function called get_sea_cnoofs.m, a subgrid ofthe current data file was
being extracted according to the coordinates a square box, which of course does not apply
to the CNOOFS data. To resolve this issue, it has been programmed such that relative
coordinates to the known position are determined, and the surrounding indices are then
extracted.
C-1
A more general change implemented was the ability to select the level of the
model output data file to use in running a CANSARP Scientific simulation. Previously,
the only data that could be used was the surface level data (level 1), but now the number
of the level can be selected by the user. It is important to note though, that this is not the
depth of the current and to determine the depths that correspond to the levels, the data
C-2