LIMNOLOGY

and
OCEANOGRAPHY: METHODS Limnol. Oceanogr.: Methods 14, 2016, 586–599
C 2016 Association for the Sciences of Limnology and Oceanography
V
doi: 10.1002/lom3.10114

The particle tracking and analysis toolbox (PaTATO) for Matlab

Erick Fredj,*1 Daniel F. Carlson,†2 Yael Amitai,3 Avi Gozolchiani,4 Hezi Gildor3
1
Department of Computer Science, Jerusalem Institute of Technology, Jerusalem, Israel
2
Department of Earth, Ocean, and Atmospheric Science, Florida State University, Tallahassee, Florida
3
The Institute of Earth Sciences, The Hebrew University, Jerusalem, Israel
4
Department of Environmental Physics, Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sede
Boqer Campus, Israel

Abstract
Lagrangian particle tracking and Lagrangian coherent structures (LCS) analysis tools aid in the studies of
fluid flow and are especially helpful in understanding the role of transport in marine ecosystems. However,
most existing particle tracking and analysis tools operate in conjunction with a specific model and/or require
execution in multiple programming languages. The Particle Tracking and Analysis TOolbox (PaTATO) for
Matlab aims to increase the availability of particle tracking and analysis techniques to a wider audience.
PaTATO is compatible with many different types of velocity data and can compute forward and backward tra-
jectories in two and/or three dimensions. PaTATO can compute standard LCS metrics, like Lyapunov expo-
nents and relative dispersion, as well as the maximal extent of a trajectory (MET), a relatively new metric.
Most importantly, PaTATO is computationally efficient and easy-to-use. We describe PaTATO, and present
examples using the time-periodic double gyre, high frequency radar surface current observations, the
Massachusetts Institute of Technology general circulation model (MITgcm), altimeter-derived geostrophic
velocities (AVISO), and Nucleous for European Modelling of the Ocean (NEMO).

Lagrangian particle tracking tools are commonly used in the Regional Ocean Modeling System (ROMS), or approach
oceanography and limnology to study transport and mixing the subject from an applied mathematics perspective (i.e.,
of tracers by observed or modeled fluid flows. Example appli- Onu et al. 2015). Furthermore, most popular particle track-
cations include ocean mixing (Poje and Haller 1999; Wiggins ing tools employ specific advection and turbulence parame-
2005; Kirwan 2006; Koshel and Prants 2006; Mancho et al. terization schemes, and/or require additional knowledge of
2006; Lehahn et al. 2007; Carlson et al. 2010a; Aharon et al. programs like Fortran. External software for analysis and vis-
2012), larval dispersal (Paris et al. 2005; Lett et al. 2007; ualization is often necessary, requiring the user to switch
Schlag et al. 2008), hydrodynamic connectivity (Berline between several different programming languages to progress
et al. 2014; Rossi et al. 2014), dissolved marine pollutants from a velocity field to a Lagrangian metric, like a plot of rel-
(Perianez 2004; Lekien et al. 2005; Havens et al. 2009), ative dispersion. As a result, the non-specialist may find
marine debris (Yoon et al. 2010; Kako et al. 2010, 2011; implementation of existing Lagrangian particle tracking and
Kataoka et al. 2013; Mansui et al. 2015), oil spills (Beegle- analysis software intimidating.
Krause 2001; Abascal et al. 2009; Havens et al. 2009; Mezić We developed the Particle Tracking and Analysis TOolbox
et al. 2010; Mariano et al. 2011), search and rescue (Breivik (PaTATO) in Matlab, an easy-to-use scripting language, to
and Allen 2008), and more. The diversity of applications is make Lagrangian particle tracking and analysis methods
reflected in the number of Lagrangian particle tracking tools widely accessible to a diverse audience without the need for
available (see Table 1). Most existing particle tracking tools extensive computer science experience. While PaTATO is
were developed around a specific ocean model, for instance intended to for use by novices, treating it as a “black box”
can lead to erroneous results, especially when computing
Additional Supporting Information may be found in the online version of Lagrangian coherent structures. For an in-depth review of
this article. particle tracking theory and coastal ocean applications see
† Lynch et al. (2015). PaTATO can compute forward and back-
Present address: Arctic Research Centre, Aarhus University, Aarhus,
Denmark ward trajectories in two or three dimensions from any veloc-
*Correspondence: [email protected] ity field on a regular grid. To date, PaTATO has been used to

586
Fredj et al. PaTATO for Matlab

Table 1. Summary of selected Lagrangian particle tracking tools.

Name Language Model dependent Purpose Reference

LTRANS Fortran 90 Yes Larvae Schlag et al. (2008)
ICHTHYOP Java Yes General Lett et al. (2008)
Ariane Fortran 90/95 Yes General Blanke and Raynaud (1997)
Connectivity-Modeling-System Fortran 90 and C No Larvae and General Paris et al. (2013)
GNOME C11 No Oil Beegle-Krause (2001)
HFR Progs Matlab No General Kaplan and Largier (2006)
LCS Tool Matlab No LCS Onu et al. (2015)

putation module quickly and efficiently computes many

(105 2106 ) particle trajectories on an ordinary laptop. The
visualization and analysis module contains a number of
tools commonly used in dynamical systems theory as well as
marine ecology. The theory and numerics upon which
PaTATO is based are well-established and are referenced in
the text. Detailed information for each script in PaTATO is
available in the help section of each .m file and in the
PaTATO user guide. PaTATO dependencies are also described
in the user guide.
Compatibility with multiple data sources, deterministic
and stochastic particle tracking options, embedded visualiza-
tion and analysis tools, and options for customization set
PaTATO apart from other particle tracking software. The
novelty of PaTATO, therefore, lies in its simplicity, versatil-
ity, customizability, and ease of use for both the experienced
ocean modeler as well as the novice. The remainder of this
paper describes each module of PaTATO, provides a sum-
mary of Lagrangian coherent structures analysis methods,
and presents example applications.

PaTATO pre-processing module

Fig. 1. The double gyre velocity field as defined by Eqs. 12–14 for (a) The toolbox can be used with any velocity data set on a
t 5 0 and (b) t 5 7. Every 5th velocity vector is plotted for clarity.
regular cartesian or spherical grid. When the gridded velocity
field changes in time, it is assumed to be arranged in snap-
compute trajectories from the time-dependent double-gyre shots, where the entire field is updated between subsequent
toy model (Carlson et al. 2010a), HF radar observations snapshots. Note that snapshots are not assumed to be regu-
(Carlson et al. 2010a,b), altimeter-derived geostrophic veloc- larly time-sampled. The pre-processing module formats the
ities (Mundel et al. 2014), and velocity fields from both the input data to the standard PaTATO format. In geophysical
Massachusetts Institute of Technology general circulation applications, the preprocessing module also defines the geo-
model (MITgcm; Fine et al. 2013) and the ROMS (Carlson graphical area, including coastlines, bathymetry, and points
et al. 2016). PaTATO goes beyond mere trajectory computa- of interest (HF radar antennae installations, major cities,
tion and offers a variety of analysis and visualization tools, etc.), for visualization.
from Lagrangian coherent structures to connectivity The user must define the shape and size of the particle
matrices. patch, particle density, distribution, times of initial release
Three modules make up PaTATO: pre-processing, trajec- (and re-release, if desired), time span of the integration,
tory computation, and visualization/analysis. During pre- external time step, ODE solver, and integration options
processing, the user selects the velocity field and specifies (absolute and relative tolerances, and maximum internal
the integration options, including ordinary differential equa- time step). The aforementioned parameters strongly influ-
tion (ODE) solver and optional turbulence parameterization, ence the outcome of the particle tracking results and, there-
and specifies particle initial conditions. The trajectory com- fore, must be defined with care, especially when performing

587
Fredj et al. PaTATO for Matlab

Fig. 3. Top: Custom, user-defined initial positions created by defining a

polygon in the active plot window. The double gyre velocity field at
time t 5 0 is shown for reference. Bottom: Selected particle trajectories
color-coded with time since release (the color bar shows time and
Fig. 2. (a) The absolute dispersion (AD), (b) relative dispersion (RD), applies to the bottom panel).
and (c) finite time Lyapunov exponent (FTLE) computed using particles
seeded in a time-periodic double gyre toy model. All units are velocities at two adjacent grid points. For the zonal (x-direc-
nondimensional. tion) velocity u, in a 3D field, one obtains
 
Lagrangian coherent structures analyses. Furthermore, sensi- x2xi21
ui;j;k 5ui21;j;k 1 ðui;j;k 2ui21;j;k Þ (1)
tivity tests may also be necessary to quantify the influence Dx
of variations in particle tracking parameters on computed
where Dx is the grid spacing. The interpolation coefficients,
trajectories. The initial positions of the particles in 2D can
i.e., the counterparts of x2x
Dx of Eq. 1 in either 2D or 3D, are
i21
be specified either by the user by specifying a shape on a
calculated and arranged in a data structure before the trajec-
map or the coordinates of the corners. In 3D the user must
tories are calculated for each given snapshot of the velocity
specify the coordinates and depth of the initial positions.
field. The time stepping method assumes that the tracer
Matlab?s ode45 (Runge/Kutta 4/5 order solver) is used by
velocity is constant during a short time interval, chosen by
default, but any standard Matlab ODE solver can be used.
the adaptive time stepping of the ODE solver. The tracer
Custom solvers, for instance the Euler-Mayurama solver
velocities are then updated successively to each new particle
(Higham 2001) used by Carlson et al. (2010a), can also be
location and time. The time step of the ODE solver is much
used. The time direction (forward or backward), total time
shorter than the time between subsequent snapshots, thus
span, time step, and magnitude of stochastic term (if used)
the velocity of each particle is calculated twice during each
are defined in a “config” file in the pre-processing module.
ODE solver step based on the coefficients of the two nearest
snapshots, and a weighted average between these two veloc-
Trajectory computation module ities is then carried over. Given an updated location after
The velocities and particle initial positions defined in the each ODE solver step, Matlab’s griddedInterpolant class (see
PaTATO configuration file are used to compute trajectories. Matlab help) finds the relevant cell using a rounded division
Velocities defined on both Arakawa A and C grids (Arakawa of its new position by the grid spacing, when the grid is uni-
and Lamb 1977) can be used in PaTATO. For the A grid, the form (typical for the horizontal location), or by a binary
zonal and meridional velocity components are defined at the search, when the grid spacing is non uniform (typical for the
center of each grid point (i, j, k) but the velocity at a grid vertical location). 3D particle tracking could be optimized
point can be defined by interpolating between discretized further by re-gridding the vertical velocity positions to a

588
Fredj et al. PaTATO for Matlab

Fig. 4. Top row: The AD (a) and RD (b) of deterministic trajectories Fig. 5. Trajectories of 30 particles released in the Gulf Stream at depths
computed from surface currents in the northern Gulf of Eilat during 1–2 ranging from 10 m to 110 m and tracked for 7 yr. Particles were
January 2006. Bottom row: The AD (c) and RD (d) of stochastic trajec- advected northward in the Gulf Stream and turned east in the North
tories computed using the same surface current data with Atlantic Gyre. The particle positions are color-coded to represent their
K51029 m2 s21 . depth in meters. After 7 yr, some particles reached depths exceeding
200 m.
uniform grid. Particles advected to locations with no velocity
data (onto land or at open boundaries) remain fixed at their vðx; tÞ5vM ðx; tÞ1v0 (3)
last valid location.
where vM ðx; tÞ and v0 correspond to the spatially and tempo-
rally varying measured or modeled velocity field and the
Stochastic Lagrangian models
Deterministic particle tracking assumes that the velocities velocity due to unresolved turbulent diffusion, respectively.
PaTATO includes an optional Zeroth order (random walk)
correctly represent the velocities of the fluid at the spatial
stochastic Lagrangian model (SLM) to simulate the transport
scales relevant to the particular Lagrangian study. In other
of particles under the influence of both advection, via the
words
input velocity field, and diffusion, through the addition of a
dx
vðx; tÞ5 (2) fluctuating velocity term (LaCasce 2008). Two variants of the
dt
Zeroth order SLM can be constructed. The first uses the sta-
However, the input data used for particle tracking, regardless tistics of the flow field to determine the maximum magni-
of the source, will have limited spatial resolution as well as tude of the stochastic term
measurement and/or computation error. For example, Ocean pffiffiffiffiffiffi
Global Circulation Model (OGCM)s do not resolve turbulent dXðtÞ5UðXðtÞ; tÞdt1 2rdw (4)
motion at scales smaller than the grid scale of the model,
where X, U, t, r, and w represent the position, time and
and the background diffusivity used to achieve numerical
space-dependent velocity, velocity variance of the flow field,
stability is often different than the observed diffusivity. One
and a white-noise random number with zero mean, respec-
method to account for unresolved, sub-grid processes and/or
tively (LaCasce 2008). Alternatively, the magnitude of the
measurement errors is to decompose the velocity into mean
stochastic term can be related to the eddy diffusivity K
and fluctuating components but such a decomposition
pffiffiffiffiffiffiffiffiffiffiffi
depends on the temporal or spatial scale over which the dXðtÞ5UðXðtÞ; tÞ; dt1 2Kdt dw (5)
averaging is performed. Selection of an appropriate length or
time scale is complicated when there is no well-defined spec- Use of Eq. 5 requires adequate resolution of the large-scale
tral gap between either low and high frequency or large and time and space dependent velocity UðXðtÞ; tÞ and the probabil-
small scale motions (Falco et al. 2000; LaCasce 2008; Carlson ity distribution function of the eddy velocity should resemble
et al. 2010a). Alternatively, Eq. 2 can be modified to include a Gaussian (Carlson et al. 2010a). A modified Euler-Mayurama
a stochastic term solver (Higham 2001) computes stochastic particle trajectories.

589
Fredj et al. PaTATO for Matlab

Fig. 6. The absolute dispersion (AD), the maximal extent of a trajectory (MET), the maximal extent in X and the maximal extent in Y computed
using particles seeded in an open sea region in the Eastern Mediterranean.

Analysis and visualization of predators using LCS to find prey (Kai et al. 2008; d’Ovidio
PaTATO plots two or three dimensional virtual particle et al. 2013). LCS analysis has also aided in understanding
trajectories and also computes and plots the Lagrangian met- the effects of ocean currents on phytoplankton distributions
rics described below. PaTATO can compute and plot single (Perez-Munuzuri and Huhn 2010) and harmful algal blooms
particle Lagrangian statistics, such as the absolute dispersion (Olascoaga et al. 2014). The resolution, and subsequent
(AD), integral time and length scales, and diffusivity. Multi- detection, of LCS depends on the total integration time, ini-
ple particle metrics, like the relative dispersion (RD), Lyapu- tial particle density, and the temporal and spatial resolutions
nov exponents, and the extreme (maximal/minimal) extent of the underlying velocity field (see Discussion in Carlson
of a trajectory and some of its variants (Mundel et al. 2014), et al. (2010a)). Therefore, these parameters must be specified
can be computed to identify Lagrangian coherent struc- carefully in the pre-processing module. Furthermore, the
tures (LCS). The analysis tools are versatile and can be sensitivity of the results to different parameter settings
applied to other data in addition to virtual particle trajecto- should be investigated. Given the wide-spread use of LCS in
ries (for instance, surface drifters). PaTATO contains more marine science, we now summarize several of the key met-
analysis tools that are described in the user guide. Here we rics used to identify them in the fluid flows.
briefly summarize the main analysis tools used in the
examples. Maximal extent of a trajectory (MET)
Lagrangian coherent structures analysis The MET uses extremal values of an observable function
Particle and propagule transport is a central feature of U for each particle along a time segment of a trajectory. The
fluid flows and even simple two-dimensional velocity fields MET is a recent addition to the Lagrangian diagnostic family
can exhibit chaotic behavior (Aref 1990). Understanding (Mundel et al. 2014) that enables the detection of LCS and
mixing properties of the flow and identifying mixing barriers their dynamics in two (and potentially three) dimensional
is of great interest for geophysicists and marine ecologists unsteady flows in both bounded and open domains. The
alike. Lagrangian coherent structures (LCS) analysis uses the MET is straightforward to compute, unlike other Lagrangian
advection of particles or tagged water masses to identify diagnostics, and provides new insights regarding the mixing
regions with similar transport properties (Shadden 2005; properties on both short and long time scales and on both
Samelson 2013; Haller 2015). LCS analysis has been used in spatial plots and distribution diagrams. PaTATO computes
the study of marine trophic dynamics, with some evidence the MET, following Mundel et al. (2014),

590
Fredj et al. PaTATO for Matlab

Fig. 7. Upper panel: Surface trajectories of particles released in six

coastal regions in the Adriatic Sea and tracked for 62 d. Trajectories are
color-coded with respect to source region. Lower panel: Particle posi-
tions prior to entering an example region in the Adriatic Sea. Positions
are color coded according to days since to 1 May 2000.

(
maxt2½t1 ;t1 1s xðt; t0 Þ  r
Mr6 ðs; x0 ; t1 Þ 5 (6)
mint2½t1 ;t1 1s xðt; t0 Þ  r
Fig. 8. Upper panel: Connection percentages, or the percentage of
particles released in a source region that travel to a given destination
which represents the maximal/minimal extents visited by region, for six coastal regions in the Adriatic Sea. Lower panel: Average
the particle during the extremal window ½t1 ; t1 1s in given transit times in days between six coastal regions in the Adriatic Sea.
shift
direction r. PaTATO also calculates Mr ðs; x0 ; t1 Þ or the Transit times were only computed if at least 10 particles connected a
difference between the maximal and minimal extents, given region pair.
called the maximal shift, and the mean of the maximal
and minimal extents Mrmean ðs; x0 ; t1 Þ hereafter referred to as Relative dispersion (RD)
the mean extrema of extents (MEEX; see Mundel et al. The RD describes how particles with some known initial
(2014)). separation disperse over time (Haller and Yuan 2000;

591
Fredj et al. PaTATO for Matlab

Table 2. Connection percentage (%), upper and lower bounds (UB, LB), and average and standard deviations of transit times
(TAVG, TSTD) for a source region (SRC) and destination regions (DST).

SRC 5 1 SRC 5 2

DST % LB UB TAVG STD DST % LB UB TAVG STD

2 84.00 79.43 87.71 13.40 5.19 1 - - - - -
3 36.67 31.41 42.26 21.59 3.22 3 35.67 30.46 41.24 17.93 6.53
4 - - - - - 4 - - - - -
5 - - - - - 5 - - - - -
6 - - - - - 6 - - - - -

SRC 5 3 SRC 5 4

DST % LB UB TAVG STD DST % LB UB TAVG STD

1 8.00 5.43 11.63 55.67 0.92 1 - - - - -
2 29.00 24.16 34.37 58.15 3.01 2 - - - - -
4 - - - - - 3 - - - - -
5 - - - - - 5 0.33 0.06 1.86 42.00 0.00
6 - - - - - 6 - - - - -

SRC 5 5 SRC 5 6

DST % LB UB TAVG STD DST % LB UB TAVG STD

1 - - - - - 1 - - - - -
2 - - - - - 2 - - - - -
3 - - - - - 3 - - - - -
4 - - - - - 4 - - - - -
6 2.67 1.36 5.17 19.75 3.24 5 - - - - -

LaCasce 2008). The RD is a commonly used and simply cal- approximation to k with positive values indicating hyper-
culated multiple particle metric that requires knowledge bolic trajectories, or exponential separation of nearby par-
only of initial and final particle positions. Ridges or lines of ticles. Ridges, or large FTLE values, can indicate barriers to
high RD denote regions with large sensitivity to the initial transport (Boffetta et al. 2001).
position of each particle and can indicate barriers to mixing
Absolute dispersion (AD)
(Boffetta et al. 2001; LaCasce 2008; Gildor et al. 2009;
The AD is a single particle metric that computes the total
Carlson et al. 2010a). PaTATO computes the RD from the
displacement from a particle’s initial position zð0Þ5ðxð0Þ; yð0ÞÞ
four nearest neighbors initially located at zk ð0Þ5ðxk ; yk Þ using
at later times zðtÞ5ðxðtÞ; yðtÞÞ (Haller and Yuan 2000; LaCasce
their locations at some later time 2008). Thus,
1X 4
RD2k 5 ðzj ðtÞ2zk ðtÞÞ2 (7) AD2 ðtÞ5ðzðtÞ2zð0ÞÞ2 (10)
4 j51
For a group of particles, the AD provides indications of their
Lyapunov Exponents
drift and spread (LaCasce 2008). When a limited number of
The maximum Lyapunov exponent (MLE; k) characterizes
trajectories are available, as is common in surface drifter
the distance between initially proximal particles by describ-
studies, the AD is a commonly used metric.
ing their exponential separation
 
1 dðtÞ
k5 lim lim ln (8) Example applications
t!1 dð0Þ!0 t dð0Þ
Five different velocity fields are used to demonstrate the
where utility of PaTATO. The example applications presented here
dðtÞ5jjz1 ðtÞ2z2 ðtÞjj (9) have been selected from in-depth studies of LCS and Lagran-
is the distance between two particles at time t (Boffetta et al. gian transport (Carlson et al. 2010a,b, 2016; Gildor et al.
2001). The finite time Lyapunov exponent (FTLE) is a finite 2009, 2010; Mundel et al. 2014). For the sake of clarity,

592
Fredj et al. PaTATO for Matlab

Table 3. Trajectories computation speed comparison between HFR Progs and PaTATO toolboxes as a function of the released par-
ticles amount in the double gyre experiment.

# of particles HFR Progs computation time [s] PaTATO computation time [s] Speed up
100 0.412% 0.9 6 11% 0.5
400 1.2 6 0.1% 0.8 6 5% 1.5
900 2.7 6 0.2% 1.1 6 12% 2.5
1600 5.8 6 17% 1.8 6 14% 3.2
2500 10.4 6 0.1% 2.6 6 2% 4.0
4900 16.3 6 15% 3.8 6 12% 4.3
10,000 36.6 6 15% 7.3 6 9% 5.0
40,000 134.9 6 2% 34.1 6 9% 4.0
90,000 325.8 6 0.4% 79.3 6 2% 4.1
160,000 627.0 6 0.2% 139.4 6 2% 4.5
250,000 1029.7 6 0.1% 208.6 6 1% 4.9

sensitivity tests carried out in the original studies are not 2012) have used the time-dependent double gyre as a rela-
reproduced here. However, we stress that particle tracking tively simple demonstrative velocity field. The following def-
parameters like total integration time, time step, particle initions are taken from Shadden et al. (2005). The flow field
density, and ODE solver, are application-dependent and the is defined by the stream function
settings used in these example applications cannot necessar-
wðx; y; tÞ5Asin ðpf ðx; tÞÞsin py (11)
ily be applied to other velocity fields. In other words,
PaTATO should not be used as a “black box” trajectory gen- where
erator and, instead, the user should carefully define particle
tracking parameters and perform necessary sensitivity tests f ðx; tÞ5aðtÞx2 1bðtÞx (12)
(see Lynch et al. (2015)). aðtÞ5sin ðxtÞ (13)
First we consider the time-dependent double gyre. The bðtÞ5122sin ðxtÞ (14)
time-dependent double gyre has been used extensively to
study chaotic advection and LCS (Shadden 2005; Carlson over the domain ½0; 23½0; 1. The double gyre is stationary
et al. 2010a; Lipinski and Mohseni 2010; Onu et al. 2015) for  5 0 and time-periodic for  > 0. The velocities are
and is used here to demonstrate features of the analysis and defined by the stream function
visualization module. In the second example we consider a @w
u52 52pAsin ðpf ðx; tÞÞcos ðpyÞ (15)
geophysical application, computing two-dimensional trajec- @y
tories using surface currents observed by a HF radar network @w df
v5 5pAcos ðpf ðx; tÞÞsin ðpyÞ (16)
in the northern Gulf of Eilat/Aqaba (Red Sea) with a SLM @x dx
sub-model to demonstrate PaTATO’s ability to parameterize
sub-grid-scale lateral mixing. The third example demon- Time, velocity, and computed quantities are nondimensional
strates two-dimensional MET analysis using altimeter-derived in the double gyre example. The following double gyre dem-
geostrophic currents, obtained from the AVISO project, in onstration presents a simplified application of the PaTATO
the Eastern Mediterranean Sea. Three-dimensional particle toolbox as the velocities are computed in the demo script.
tracking is demonstrated using MITgcm simulations of the Here, we demonstrate defining integration options and ini-
global ocean. Finally, Lagrangian transport and transit times tial positions, plotting trajectories, and visualizing LCS using
between coastal regions in the Adriatic Sea is presented using single and multiple particle metrics. The reader can replicate
the results by running demo_2D_double_gyre.m.
surface velocities from the Nucleous for European Modelling
First, the demo plots the Eulerian velocity field using the fol-
of the Ocean (NEMO). The scripts for each example can be
lowing parameters in Eqs. 11–-14: x5 2p T ; A 5 0.1; 50:25;
found in the demo directory of PaTATO.
T 5 10. x is the frequency of oscillation, T is the tidal period,
Double gyre and A is the amplitude of the velocity. In this example, the time
In this example the flow field is two-dimensional, area span considered is [0, 17] and dt 5 0.1. The velocity fields at t 5 0
preserving, and the velocity field is specified in space and and t 5 7 are shown in Fig. 1. We seed the entire domain with
time as it is derived from a known stream function. Advan- 500 3 250 particles and define the integration options and ODE
ces in LCS (Poje and Haller 1999; Shadden et al. 2005; solver. In this example, we use absolute and relative tolerances
Carlson et al. 2010a; Aharon et al. 2012; Schindler et al. of 1023 , a maximum time step of 1/2, and Matlab’s ode45

593
Fredj et al. PaTATO for Matlab

(Runge-Kutta 4/5 order scheme). The trajectories are then com- time step and saves the animation as an .avi file. Next, particle
puted using the uniform initial positions and the velocities over tracking options are specified. We seed the entire HF radar cover-
the time interval [0, 17] using Eqs. 15, 16. Resolution and identi- age area in the northern Gulf with a uniform spacing of dx5dy5
1 o
fication of LCS in time-varying double gyre depend strongly on 360 (approximately 250 m). The time span is the entire 48 h
the total integration time and particle density. Carlson et al. period and Matlab’s ode45 solver is used (all options are speci-
(2010a) showed how the width of the mixing barrier varied fied in the configuration file config_GulfofEilat_HFR.m). The
with particle density. The reader can easily vary the total inte- trajectories are computed and stored in a structure and saved as
gration time and particle density to see first-hand how these a .mat file. The visualization and analysis functions animate
parameters affect the resulting LCS. selected particle trajectories and compute and display the AD
An animation of particle trajectories reveals mixing between and RD after 5 h (Fig. 4).
the two gyres due to chaotic advection. LCS are visualized by The AD and RD of the deterministic trajectories are consistent
computing the AD, RD, and FTLE (Fig. 2). The AD shows that with observed particle trajectories in the animation and show
particles that were seeded on the outer edges of both gyres expe- maximal values around the center of the west coast that extend
rienced the largest displacements. The RD shows how initially across the gulf, dividing the region into northern and southern
adjacent particles spread over time with large values of RD corre- sections (Fig. 4a,b). In addition to large particle spreading in the
sponding to large dispersion indicative of hyperbolic points in central gulf and along the western shore the RD reveals fila-
the flow. Continuous features of high RD suggest the presence ments of high RD suggestive of rapid particle spread as well as
of a mixing barrier and the FTLE results are similar to the RD, as quiescent regions (Fig. 4b). Thin filaments with RD of  0.5–
shown in Carlson et al. (2010a). 0.6 km can be seen in the southern half of the northern Gulf
Next, we demonstrate the ability of PaTATO to construct enclosing an area in the southeastern part of the domain with
a user-specified polygon and seed it with particles. The user very little particle spreading (Fig. 4b).demo_2D_Eilat_Stochas-
creates the polygon by calling the generateTracerLocation.m b).demo_2D_Eilat_Stochastic.m computes stochastic trajectories
function in an active figure window. The spacing between according to Eq. 5, with K51029 m2 s21 using the same input
particles is defined in the function call and the boundaries data and initial positions over the same time period. The config-
of the polygon are defined by clicking on the active plot. uration file config_GulfofEilat_HFR_SLM.m specifies K and
When finished, the user is prompted to continue, or erase changes the ODE solver to a constant time step Euler-Mayurama
the current polygon and start again. Next, the initial posi- method. For comparison with the deterministic trajectories we
tions are plotted in the polygon and the user is prompted to show an animation of the particle trajectories and compute the
start over or continue. Finally, the user is prompted to save AD and RD using particle positions after 5 h. The AD and RD of
the initial positions in a separate file, or to continue without the stochastic trajectories are quite similar to the deterministic
saving. Here, we construct a diamond shaped polygon (Fig. trajectories in this example Fig. 4). The added stochastic
3), seed it with evenly spaced particles, and track these par- motions increase the number of particles that encounter a
ticles using the same time-periodic velocity field, time inter- boundary, thereby decreasing the number of particles available
val, and integration options as used earlier in this example. to compute AD and RD.
A scatter plot of selected particles shows their trajectories Raw HF radar surface currents should be interpolated
color-coded with time (Fig. 3). before use with PaTATO as particles that encounter spatial
gaps in the raw velocity field are removed from the integra-
HF radar surface currents
tion. In addition to OMA, other methods to fill gaps in HF
A 42 MHz Codar SeaSonde HF radar system observed sur-
radar data exist, including the recent method of Fredj et al.
face currents in the northern Gulf and produced total veloc-
(in press) and others (Kim et al. 2007; Yaremchuk and
ity vector maps at approximately 300 m resolution every 30
Sentchev 2011).
min (Gildor et al. 2009; Carlson et al. 2010b). Spatial gaps in
the HF radar data were interpolated using the open- MITgcm
boundary modal analysis (OMA) technique (Lekien et al. We demonstrate three-dimensional particle tracking using
2004; Kaplan and Lekien 2007; Lekien and Gildor 2009; a simulation of the global ocean from the MITgcm tutorial
Carlson et al. 2010b). For background on the dynamics of in the script demo_3D_GlobalOcean_MITgcm. The MITgcm
the Gulf of Eilat see Berman et al. (2000), Manasrah et al. solves an approximated form of the thermodynamic and
(2004), Monismith and Genin (2004), Ashkenazy and Gildor momentum equations known as the hydrostatic primitive
(2009), Gildor et al. (2009), Carlson et al. (2010a, 2012), equations (HPEs) on a regular spherical coordinate grid
Biton et al. (2008), and Biton and Gildor (2011a,b). (Marshall et al. 1997a,b), configured here with realistic geog-
We use PaTATO to compute both deterministic and stochas- raphy and bathymetry on a 48 3 48 spherical polar grid. Fif-
tic trajectories using OMA-interpolated HF radar observations of teen vertical levels are used, ranging in thickness from 50 m
surface currents during 1–2 January 2006. demo_2D_Eilat.m at the surface to 690 m at depth, giving a maximum model
loads the velocity data and animates the surface currents at each depth of 5200 m. The model is forced with climatological

594
Fredj et al. PaTATO for Matlab

wind stress data from Trenberth et al. (1990) and National In this demo we seed 14,000 particles in the region 25.58–
Centers for Environmental Prediction (NCEP) surface flux 348E/328–34.58N of the Eastern Mediterranean (Fig. 6) and
data from Kalnay et al. (1996). Levitus climatology (Levitus compute their trajectories during September 2010. The parti-
and Boyer 1994) is used to initialize the stratification in the cle trajectories are then used to calculate AD, MET, and
model. These fluxes are combined with the NCEP climato- MEEX. In this example we also introduce the option to calcu-
logical estimates of surface heat flux, resulting in a mixed late backward-trajectories by setting a flag in the configuration
boundary condition of the style described in Haney (1971). file. The trajectories are then calculated on an inverted veloc-
The horizontal grid of the global ocean tutorial places the ity field from the last to the first dates given by the user.
international reference meridians 08 and 3608 on the western
Lagrangian transport and transit times in the Adriatic Sea
and eastern boundaries of the model, respectively. Latitude is
Following Carlson et al. (2016), surface transport between
defined from 808S to 808N. The longitude and latitude matrices
six coastal regions in the Adriatic Sea is investigated using
must be transferred to a grid system on a Mercator projection,
NEMO surface velocities from May-June 2000 (62 d). Surface
which requires that the international reference meridian be
velocity data were obtained from the NEMO-based Mediter-
placed at the center of the zonal grid. The function fit2merca-
ranean Sea Physics Reanalysis provided by the Copernicus
tor.m performs this operation in this demonstration.
Marine Environment Monitoring Service (www.marine.
MITgcm diagnoses velocity fields on Arakawa C-grid, by
copernicus.eu). The velocity data were computed daily, with
default. However, we choose to use velocity field that is specified 1o
a horizontal resolution 16 and 72 vertical levels. The tempo-
on Arakawa A-grid in the 3D demo of trajectories computation
ral and horizontal resolutions of the model are marginal for
in the north Atlantic Ocean (Fig. 5). Therefore, a procedure of
a complex marginal sea like the Adriatic and are simply used
averaging two face velocities in each direction to the velocity in
to demonstrate PaTATO’s ecological analysis tools.
the cell center, is performed in the pre-processing part of
First, 300 virtual particles were released in six coastal regions
demo_3D_GlobalOcean_MITgcm.m. Additional option is to that enclose existing or planned marine protected areas and
provide three grid systems corresponds to three velocity fields. A tracked for 62 d (Fig. 7, upper panel). PaTATO was then used to
MITgcm user can also activate the diagnostics package of the compute the connection percentage (percentage of particles
model and output all velocity components in the cell center that left a given source region and reached a given destination
(i.e., in Arakawa A grid) directly. region) and average transit times between regions. Note that we
After running the MITgcm Global Ocean tutorial for 1000 present the connection percentage, but this example could be
yr to obtain a 3D global circulation pattern, we seeded 180 easily extended to compute connectivity estimates. PaTATO dis-
particles in the western side of the north Atlantic (708–758W/ plays simplified graphical results (Fig. 8) and outputs a detailed
308–358N) in various depths ranging from 10 m to 170 m LaTex table, complete with confidence intervals, that can be
and tracked them for 7 yr. Figure 5 shows that the particles inserted into a manuscript (Table 2). Next, backwards trajecto-
were advected north by the Gulf Stream and then eastward ries are computed to find the origins of particles located in
in the North Atlantic Gyre (Schmitz and McCartney 1993). Region 3 on 1 July 2000 (Fig. 7, lower panel).
Some particles reached depths > 250 m, following the well-
known Atlantic meridional overturning circulation, in which Assessment
water advected north by the Gulf Stream cools, become
Lagrangian particle tracking programs for oceanographic and
denser and sinks (Kuhlbrodt et al. 2007).
limnological studies have been in use for decades. The PaTATO
AVISO geostrophic velocities toolbox for Matlab differs from existing products in its compati-
Our next example uses surface geostrophic velocities cal- bility with multiple velocity fields, from HF radar observations
culated from sea surface height observations compiled by to three-dimensional GCMs. Additionally, the PaTATO analysis
the Archiving Validation and Interpretation of Satellite Data module computes standard LCS metrics, as well as the MET and
in Oceanography (AVISO). AVISO processes, validates, and MEEX. Currently, PaTATO is the only particle tracking toolbox
merges all available altimetry data from the Jason-1, TOPEX/ that can compute the MET-family of LCS metrics. Furthermore,
Poseidon, European Remote Sensing Satellite-2 (ERS-2), Envi- the PaTATO toolbox provides the only generic environment to
sat and Geosat satellites. We focus on LCS in the Eastern evaluate the connection percentage and average transit times
Mediterranean given the abundance of features in this between regions for ecological purposes.
region (Matteoda and Glenn 1996; d’Ovidio et al. 2004; Most notably, the efficiency of PaTATO’s trajectory com-
Pascual et al. 2007; Amitai et al. 2010; Efrati et al. 2013). In putation algorithm significantly reduces computation time
the Mediterranean, the AVISO velocity field is provided when compared to other particle tracking software. Enhance-
o o
weekly with a spatial resolution of 18 318 . LCS in the Eastern ment with respect to other toolboxes that use the same
Mediterranean are difficult to detect and define, yet they approach of piecewise interpolation in space and time is
exert a significant influence on circulation in the region reached by trading time for memory. To this end, we calcu-
(d’Ovidio et al. 2004; Amitai et al. 2010; Efrati et al. 2013). late spatial interpolation coefficients for all grid cells ahead

595
Fredj et al. PaTATO for Matlab

of time stepping, and save them in a lookup table. Thus, ini- Berline, L., A. Rommou, A. Doglioli, A. Molcard, and A.
tiation of more particles on more grid cells in larger domains Petrenko. 2014. A connectivity-based eco-regionalization
yields the largest speedups. To demonstrate PaTATO’s com- method of the Mediterranean Sea. PLoS One 9. doi:
putational efficiency, we generated trajectories using the 10.1371/journal.pone.0111978
PaTATO and HFR (Kaplan and Largier 2006) toolboxes on Berman, T., N. Paldor, and S. Brenner. 2000. Simulation of
the (2 dimensional) double gyre setup (see above). Each run wind-driven circulation in the Gulf of Elat (Aqaba). J.
was repeated 10 times, to show a statistically significant Mar. Syst. 26: 349–365.
tendency. A comparison of the mean run speeds with differ- Biton, E., J. Silverman, and H. Gildor. 2008. Observations
ent number of particles was performed on Fujitsu computer and modeling of a pulsating density current. Geophys.
using Linux 64 bits with intel core i7-4770 CPU @ 3.40GHz Res. Lett. 35: L14603. doi:10.1029/2008GL034123
x8 and 7.7GB memory. The results are presented in Table 3, Biton, E., and H. Gildor. 2011a. The coupling between
where the speed-up reaches 5 for the case of 13104 tracers. exchange flux through a strait and dynamics in a small
convectively driven marginal sea: The Gulf of Aqaba (Gulf
of Eilat). J. Geophys. Res. 116: C06017. doi:10.1029/
Summary 2011JC006944
We described the structure and operation of PaTATO and Biton, E., and H. Gildor. 2011b. The general circulation of
demonstrated its versatility and computational efficiency the Gulf of Aqaba (Gulf of Eilat) revisited: The interplay
through five different example applications. We also sum- between the exchange flow through the Straits of Tiran
marized the theory behind the Lagrangian analysis metrics. and surface fluxes. J. Geophys. Res. 116, C08020, doi:
In addition to the five examples shown in this paper, 10.1029/2010JC006860
PaTATO is fully customizable and, therefore, applicable to a Blanke, B., and S. Raynaud. 1997. Kinematics of the Pacific equa-
range of scientific problems, from physical oceanography to torial undercurrent: A Eulerian and Lagrangian approach
marine ecology, and beyond. PaTATO puts Lagrangian parti- from GCM results. J. Phys. Oceanogr. 27: 1038–1053. doi:
cle tracking and Lagrangian analysis techniques, including 10.1175/1520-0485(1997)027<1038:KOTPEU>2.0.CO;2
LCS analysis, into the hands of a large user group. Finally, Boffetta, G., G. Lacorata, G. Redaelli, and A. Vulpiani. 2001.
Detecting barriers to transport: A review of different tech-
we stress again the importance of proper parameter selection
niques. Physica D 159: 58–70. doi:10.1016/S0167-
and the importance of sensitivity tests.
2789(01)00330-X
References Breivik, Ø., and A. A. Allen. 2008. An operational search and
rescue model for the Norwegian Sea and the North Sea. J.
Abascal, A. J., S. Castanedo, R. Medina, I. J. Losada, and E.
Mar. Syst. 69: 99–113. doi:10.1016/j.jmarsys.2007.02.010
Alvarez-Fanjul. 2009. Application of HF radar currents to
Carlson, D. F., E. Fredj, H. Gildor, and V. Rom-Kedar. 2010a.
oil spill modelling. Mar. Pollut. Bull. 58: 238–248. doi: Deducing an upper bound to the horizontal eddy diffusiv-
10.1016/j.marpolbul.2008.09.020 ity using a stochastic Lagrangian model. Environ. Fluid
Aharon, R., V. Rom-Kedar, and H. Gildor. 2012. When com- Mech. 10: 499–520. doi:10.1007/s10652-010-9181-0
plexity leads to simplicity: Ocean surface mixing simpli- Carlson, D. F., P. A. Muscarella, H. Gildor Jr., B. L. Lipphardt,
fied by vertical convection. Phys. Fluids 24: 056603. doi: and E. Fredj. 2010b. How useful are progressive vector
10.1063/1.4719147 diagrams for studying coastal ocean transport? Limnol.
Amitai, Y., Y. Lehahn, A. Lazar, and E. Heifetz. 2010. Surface Oceanogr.: Methods 8: 98–106. doi:10.4319/lom.2010.8.98
circulation of the eastern Mediterranean Levantine basin: Carlson, D. F., E. Fredj, H. Gildor, E. Biton, J. V. Steinbuck,
Insights from analyzing 14 years of satellite altimetry S. G. Monismith, and A. Genin. 2012. Observations of
data. J. Geophys. Res. 115: C10058. doi:10.1029/ tidal currents in the northern Gulf of Eilat/Aqaba (Red
2010JC006147 Sea). J. Mar. Syst. 102–104: 14–28. doi:10.1016/
Arakawa, A., and V. R. Lamb. 1977. Methods of computa- j.jmarsys.2012.04.008
tional physics, V. 17. Academic Press. Carlson, D., and others. 2016. Observed and modeled surface
Aref, H. 1990. Chaotic advection of fluid particles. Philos. Lagrangian transport between coastal regions in the
Trans. R. Soc. Lond. 333: 273–288. doi:10.1098/ Adriatic Sea with implications for marine protected areas.
rsta.1990.0161 Cont. Shelf Res. 118: 23–48. doi:10.1016/j.csr.2016.02.012
Ashkenazy, Y., and H. Gildor. 2009. Long-range temporal ndez, E. Herna
d’Ovidio, F., V. Ferna ndez-Garcıa, and C. Lo pez.
correlations of ocean surface currents. J. Geophys. Res. 2004. Mixing structures in the Mediterranean Sea from
114: 1–10. doi:10.1029/2008JC005235 finite-size Lyapunov exponents. Geophys. Res. Lett. 31:
Beegle-Krause, J. 2001. General NOAA oil modeling environ- L17203. doi:10.1029/2004GL020328
ment (GNOME): A new spill trajectory model. Interna- d’Ovidio, F., S. D. Monte, A. D. Penna, C. Cotte, and C. Guinet.
tional Oil Spill Conference Proceedings 2001, p. 865–871. 2013. Ecological implications of eddy retention in the open

596
Fredj et al. PaTATO for Matlab

ocean: A Lagrangian approach. J. Phys. A: Math. Theor. 46: Kalnay, E., and others. 1996. The NCEP/NCAR 40-year rean-
254023 pp 21. alysis project. Bull. Am. Meteorol. Soc. 77: 437–471. doi:
Efrati, S., and others. 2013. Intrusion of coastal waters into 10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2
the pelagic eastern Mediterranean: In situ and satellite- Kaplan, D. M., and J. Largier. 2006. HF radar-derived origin
based characterization. Biogeosciences 10: 3349–3357. and destination of surface waters off Bodega Bay, Califor-
doi:10.5194/bg-10-3349-2013 nia. Deep-Sea Res. II 53: 2906–2930. doi:10.1016/
Falco, P., A. Griffa, P.-M. Poulain, and E. Zambianchi. 2000. j.dsr2.2006.07.012
Transport properties in the Adriatic Sea as deduced from drifter Kaplan, D. M., and F. Lekien. 2007. Spatial interpolation and
data. J. Phys. Oceanogr. 30: 2055–2071. http://dx.doi.org/10. filtering of surface current data based on open-boundary
1175/1520-0485(2000)030<2055:TPITAS>2.0.CO;2 modal analysis. J. Geophys. Res. 112: C12007. doi:
Fine, M., H. Gildor, and A. Genin. 2013. A coral reef refuge 10.1029/2006JC003984
in the Red Sea. Glob. Change. Biol. 19: 3640–3647. doi: Kataoka, T., H. Hinata, and Y. Nihei. 2013. Numerical esti-
10.1111/gcb.12356 mation of inflow flux of floating natural macro-debris
Fredj, E., H. Roarty, J. Kohut, M. Smith, and S. Glenn. 2016. into Tokyo Bay. Estuar. Coast Shelf Sci. 134: 69–79. doi:
Gap filling of the coastal ocean surface currents from HFR 10.1016/j.ecss.2013.09.005
data: Application to the mid Atlantic bight HFR network. Kim, S., E. Terrill, and B. Cornuelle. 2007. Objectively map-
J. Atmos. Oceanic Technol. 33: 1097–1111. doi:10.1175/ ping HF radar-derived surface current data using measured
JTECH-D-15-0056.1 and idealized data covariance matrices. J. Geophys. Res.
Gildor, H., E. Fredj, J. V. Steinbuck, and S. Monismith. 2009. 112: C06021. doi:10.1029/2006JC003756
Evidence for submesoscale barriers to horizontal mixing Kirwan, A. 2006. Dynamics of critical trajectories. Prog. Oce-
in the ocean from current measurements and aerial pho- anogr. 70: 448–465. doi:10.1016/j.pocean.2005.07.002
tographs. J. Phys. Oceanogr. 39: 1975–1983. doi:10.1175/ Koshel, K. V., and S. V. Prants. 2006. Chaotic advection in
2009JPO4116.1 the ocean. Phys. Usp. 49: 1151.
Gildor, H., E. Fredj, and A. Kostinski. 2010. The Gulf of Eilat/ Kuhlbrodt, T., A. Griesel, M. Montoya, A. Levermann, M.
Aqaba: A natural driven cavity? Geophys. Astrophys. Fluid Hofmann, and S. Rahmstorf. 2007. On the driving proc-
Dynam. 104: 301–308. doi:10.1080/03091921003712842 esses of the Atlantic meridional overturning circulation.
Haller, G. 2015. Lagrangian coherent structures. Annu. Rev. Rev. Geophys. 45: RG2001. doi:10.1029/2004RG000166
Fluid Mech. 47: 137–162. LaCasce, J. H. 2008. Statistics from Lagrangian observations.
Haller, G., and G. Yuan. 2000. Lagrangian coherent struc- Progr. Oceanogr. 77: 1–29. doi:10.1016/j.pocean.2008.02.002
tures and mixing in two-dimensional turbulence. Physica Lehahn, Y., F. d’Ovidio, M. Levy, and E. Heifetz. 2007. Stir-
D 147: 352–370. doi:10.1016/S0167-2789(00)00142-1 ring of the northeast atlantic spring bloom: A lagrangian
Haney, R. L. 1971. Surface thermal boundary condition for analysis based on multisatellite data. J. Geophys. Res.
ocean circulation models. J. Phys. Oceanogr. 1: 241–248. Oceans 112: C08005. doi:10.1029/2006JC003927
doi:10.1175/1520-0485(1971)001<0241:STBCFO>2.0.CO;2 Lekien, F., C. Coulliette, R. Bank, and J. Marsden. 2004.
Havens, H., M. E. Luther, and S. D. Meyers. 2009. A coastal Open-boundary modal analysis: Interpolation, extrapola-
prediction system as an event response tool: Particle tion, and filtering. J. Geophys. Res. 109: C12004. doi:
tracking simulation of an anhydrous ammonia spill in 10.1029/2004JC002323
Tampa Bay. Mar. Pollut. Bull. 58: 1202–1209. doi: Lekien, F., C. Coulliette, A. J. Mariano, E. H. Ryan, L. K.
10.1016/j.marpolbul.2009.03.012 Shay, G. Haller, and J. Marsden. 2005. Pollution release
Higham, D. J. 2001. An algorithmic introduction to numeri- tied to invariant manifolds: A case study for the coast of
cal simulation of stochastic differential equations. SIAM Florida. Physica D 210: 1–20. doi:10.1016/
Rev. 43: 525–546. doi:10.1137/S0036144500378302 j.physd.2005.06.023
Kai, E., and others. 2008. Top marine predators track Lagran- Lekien, F., and H. Gildor. 2009. Computation and approxima-
gian coherent structures. Proc. Natl. Acad. Sci. U. S. A. tion of the length scales of harmonic modes with applica-
106: 8245–8250. www.pnas.org/cgi/doi/10.1073/pnas. tion to the mapping of surface currents in the Gulf of Eilat.
0811034106 J. Geophys. Res. 114: C06024. doi:10.1029/2008JC004742
Kako, S., A. Isobe, S. Seino, and A. Kojima. 2010. Inverse esti- Lett, C., P. Penven, P. Ayo  n, and P. Freon. 2007. Enrich-
mation of drifting-object outflows using actual observa- ment, concentration and retention processes in relation
tion data. J. Oceanogr. 66: 291–297. to anchovy Engraulis ringens eggs and larvae distributions
Kako, S., A. Isobe, S. Magome, H. Hinata, S. Seino, and A. in the northern Humboldt upwelling ecosystem. J. Mar.
Kojima. 2011. Establishment of numerical beach-litter Syst. 64: 189–200. doi:10.1016/j.jmarsys.2006.03.012
hindcast/forecast models: An application to Goto Islands, Lett, C., and others. 2008. A Lagrangian tool for modelling
Japan. Mar. Pollut. Bull. 62: 293–302. doi:10.1016/ ichthyoplankton dynamics. Environ. Model. Softw.
j.marpolbul.2010.10.011 1210–1214. doi:10.1016/j.envsoft.2008.02.005

597
Fredj et al. PaTATO for Matlab

Levitus, S., and T. P. Boyer. 1994. World ocean atlas 1994. Paris, C. B., R. K. Cowen, R. Claro, and K. C. Lindeman.
volume 4. temperature. Tech. Rep. NESDIS–4, National 2005. Larval transport pathways from cuban snapper (Lut-
Environmental Satellite, Data, and Information Service, janidae) spawning aggregations based on biophysical
Washington, DC (United States). modeling. Mar. Ecol.-Prog. Ser. 296: 93–106.
Lipinski, D., and K. Mohseni. 2010. A ridge tracking algo- Paris, C., J. Helgers, E. van Sebille, and A. Srinivasan. 2013.
rithm and error estimate for efficient computation of Connectivity modeling system: A probabilistic modeling
Lagrangian coherent structures. Chaos 20: 017504. doi: tool for the multi-scale tracking of biotic and abiotic vari-
10.1063/1.3270049 ability in the ocean. Environ. Model. Softw. 42: 47–52.
Lynch, D., D. Greenberg, A. Bilgili Jr, D. J. Mcgillicuddy, J. doi:10.1016/j.envsoft.2012.12.006
Manning, and A. Aretxabaleta. 2015. Particles in the coastal Pascual, A., M. Pujol, G. Larnicol, P. Y. Le Traon, and M.
ocean: Theory and applications. Cambridge University Press. Rio. 2007. Mesoscale mapping capabilities of multisatellite
Manasrah, R., M. Badran, H. U. Lass, and W. Fennel. 2004. altimeter missions: First results with real data in the Medi-
Circulation and winter deep-water formation in the terranean Sea. J. Mar. Syst. 65: 190–211. doi:10.1016/
northern Red Sea. Oceanologia 46: 5–23. j.jmarsys.2004.12.004
Mancho, A. M., D. Small, and S. Wiggins. 2006. A tutorial on Perez-Munuzuri, V., and F. Huhn. 2010. The role of meso-
dynamical systems concepts applied to lagrangian transport scale eddies time and length scales on phytoplankton pro-
in oceanic flows defined as finite time data sets: Theoretical duction. Nonlinear Process. Geophys. 17: 177–186. doi:
and computational issues. Phys. Rep. 437: 55–124. doi: 10.5194/npg-17-177-2010
10.1016/j.physrep.2006.09.005 Perianez, R. 2004. A particle-tracking model for simulating
Mansui, J., A. Molcard, and Y. Ourmieres. 2015. Modelling the pollutant dispersion in the Strait of Gibraltar. Mar. Pollut.
transport and accumulation of floating marine debris in the Bull. 49: 613–623. doi:10.1016/j.marpolbul.2004.04.003
Mediterranean basin. Mar. Pollut. Bull. 91: 249–257. doi:
Poje, A., and G. Haller. 1999. Geometry of cross-stream mix-
10.1016/j.marpolbul.2014.11.037
ing in a double-gyre ocean model. J. Phys. Oceanogr. 29:
Mariano, A. J., V. H. Kourafalou, A. Srinivasan, H. Kang, G. R.
1649–1665. http://journals.ametsoc.org/doi/pdf/10.1175/
Halliwell, E. H. Ryan, and M. Roffer. 2011. On the model-
1520-0485(1999)029%3C1649%3AGOCSMI%3E2.0.CO%3B2
ing of the 2010 Gulf of Mexico oil spill. Dyn. Atmos.  pez, and E. Hernandez-Garcıa.
Rossi, V., E. Ser-Giacomi, C. Lo
Oceans 52: 322–340. doi:10.1016/j.dynatmoce.2011.06.001
2014. Hydrodynamic provinces and oceanic connectivity
Marshall, J., A. Adcroft, C. Hill, L. Perelman, and C. Heisey.
from a transport network help designing marine reserves.
1997a. A finite-volume, incompressible Navier Stokes
Geophys. Res. Lett 41: 2883. doi:10.1002/2014GL059540
model for studies of the ocean on parallel computers. J.
Samelson, R. 2013. Lagrangian motion, coherent structures,
Geophys. Res. 102: 5753–5766. doi:10.1029/96JC02775
and lines of persistent material strain. Ann. Rev. Mar. Sci.
Marshall, J., C. Hill, L. Perelman, and A. Adcroft. 1997b.
5: 137–163. doi:10.1146/annurev-marine-120710-100819
Hydrostatic, quasi-hydrostatic, and nonhydrostatic ocean
Schindler, B., R. Peikert, R. Fuchs, and H. Theisel. 2012.
modeling. J. Geophys. Res. 102: 5733–5752. doi:10.1029/
Ridge concepts for the visualization of Lagrangian coher-
96JC02776
ent structures. In Topological methods in data analysis
Matteoda, A., and S. Glenn. 1996. Observations of recurrent
and visualization II, mathematics and visualization.
mesoscale eddies in the Eastern Mediterranean. J. Geo-
phys. Res. 101: 20687–20709. doi:10.1029/96JC01111 Springer-Verlag. 221–235. doi:10.1007/978-3-642-23175-
Mezić, I., S. Loire, V. A. Fonoberov, and P. Hogan. 2010. A 9_15
new mixing diagnostic and Gulf oil spill movement. Sci- Schlag, Z., W. North, and K. Smith. 2008. Larval TRANSport
ence 330: 486–489. doi:10.1126/science.1194607 Lagrangian model (LTRANS) user’s guide. Tech. rep., Uni-
Monismith, S. G., and A. Genin. 2004. Tides and sea level in the versity of Maryland Center for Environmental Science,
Gulf of Aqaba (Eilat). J. Geophys. Res. Oceans 109: C04015. Horn Point Laboratory. Cambridge, MD. (United States).
doi:10.1029/2003JC002069 Schmitz, W. J., and M. S. McCartney. 1993. On the north
Mundel, R., E. Fredj, H. Gildor, and V. Rom-Kedar. 2014. Atlantic circulation. Rev. Geophys. 31: 29–49.
New Lagrangian diagnostics for characterizing fluid flow Shadden, S. C. 2005. Lagrangian coherent structures, p. 59–
mixing. Phys. Fluids 26: 126602. doi:10.1063/1.4903239 89. In R. Grigoriev [ed.], Transport and mixing in laminar
Olascoaga, M. J., F. J. Beron-Vera, L. E. Brand, and H. Koçak. flows: From microfluidics to ocean currents. Wiley-VCH
2014. Tracing the early development of harmful algal Verlag GmbH and Co. KGaA.
blooms on the west Florida shelf with the aid of Lagran- Shadden, S., F. Lekien, and J. Marsden. 2005. Definition and
gian coherent structures. J. Geophys. Res. 2008; 113: properties of Lagrangian coherent structures from finite-time
c12014. doi:10.1029/2007JC004533 Lyapunov exponents in two-dimensional aperiodic flows.
Onu, K., F. Huhn, and G. Haller. 2015. LCS Tool: A compu- Physica D 212: 271–304. doi:10.1016/j.physd.2005.10.007
tational platform for Lagrangian coherent structures. J. Trenberth, K. E., W. G. Large, and J. G. Olson. 1990. The
Comp. Sci. 7: 26–36. doi:10.1016/j.jocs.2014.12.002 mean annual cycle in global ocean wind stress. J. Phys.

598
Fredj et al. PaTATO for Matlab

Oceanogr. 20: 1742–1760. doi:10.1175/1520-0485(1990) Acknowledgments

020<1742:TMACIG>2.0.CO;2 HG and EF were supported by a grant from the Israeli Ministry of Sci-
Wiggins, S. 2005. The dynamical systems approach to lagran- ence and Technology and the Taiwan Ministry of Science and Technol-
gian transport in oceanic flows. Annu. Rev. Fluid Mech. ogy. DFC was supported in part by a grant from BP/The Gulf of Mexico
Research Initiative.
37: 295–328. doi:10.1146/annurev.fluid.37.061903.175815
Yaremchuk, M., and A. Sentchev. 2011. A combined EOF/
Submitted 11 January 2016
variational approach for mapping radar-derived sea sur-
Revised 15 April 2016
face currents. Cont. Shelf Res. 31: 758–768. doi:10.1016/
Accepted 28 April 2016
j.csr.2011.01.009
Yoon, J.-H., S. Kawano, and S. Igawa. 2010. Modeling of
marine litter drift and beaching in the Japan Sea. Mar. Pol-
lut. Bull. 60: 448–463. doi:10.1016/j.marpolbul.2009.09.033 Associate editor: John Crimaldi

599