Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.

Digital Object Identifier 10.1109/ACCESS.2017.DOI

Visualizing Three-Dimensional Ocean

Eddies in Web Browsers
LI LIU1 , DEBORAH SILVER1 , AND KAREN BEMIS.2
1
Department of Electrical and Computer Engineering, Rutgers University, New Brunswick, NJ 08901 USA
2
Department of Marine and Coastal Sciences, Rutgers University, New Brunswick, NJ 08901 USA
Corresponding author: Li Liu (e-mail: l.liu6819@gmail.com).
This work has been funded by US National Science Foundation (NSF) Grants OCE-1049088.

ABSTRACT Despite the importance of mesoscale eddies, and other three-dimensional features discernable
in numerical simulations of ocean circulations, to the understanding of oceanic heat transport and climate
change, there are few dedicated three-dimensional visualization tools for scientists to use. This paper
presents a framework for the visualization of three-dimensional (3D) ocean eddies in web browsers. The
framework includes a feature-based eddy computation algorithm implemented by a standalone C++ program
and a Web Graphics Library (WebGL)-based 3D eddy viewer. Combining eddy extraction, tracking, query,
and interactive visualizations in one workflow, the proposed framework is designed to facilitate the visual
exploration and analysis of ocean eddies and thus help domain experts get more insight into the process
of ocean circulation. In the framework, the feature-based algorithm significantly reduces the storage
requirement of data and the rendering cost of visualization tasks by exploiting the coherence between
consecutive data frames and focusing on just those regions of interest. The 3D eddy viewer presents
eddy visualizations within a standard web browser, which enables the easy sharing of the tool and the
results between scientific collaborators at different locations on diverse platforms without the need to install
specialized software. In this paper, the general considerations in the design process as well as the technical
facts of the framework, the application of the framework to a high-resolution regional ocean simulation, the
general usage scenarios of the 3D eddy viewer, and the user evaluation are discussed.

INDEX TERMS Ocean Eddies, Geospatial Visualization, Feature Tracking, 3D, WebGL

I. INTRODUCTION eddies interact with ocean stratification? How do such in-

ESOSCALE ocean eddies are typically understood
M as vortices that are 50 - 200 km in diameter, may
live for several months, and can propagate across an ocean
basin [23]. They play an important role in transporting heat,
salt, and nutrients in the ocean, both horizontally and ver-
tically. In the past few decades, the computational power
to support three-dimensional computer modeling of ocean
circulation has greatly increased, enabling oceanographers
to study ocean eddies in three dimensions at unprecedented
precision and scope. However, publications in oceanography
typically still show only simple two-dimensional plots of
eddy variables on the surface and in vertical sections [1] [3]
[10]. One major reason is the limitations of the dedicated
visualization tools that are available to the oceanographic
community. While 2D eddy representations efficiently and
simply illustrate the dominantly horizontal dynamics in the
ocean, they do not address all questions that are important
to the oceanographic communities. For example: How do
teractions affect the observed degree of taper? How does
an eddy evolve from its birth to disappearance? What are
the differences between cyclonic and anti-cyclonic eddies in
distribution and motion? These questions, about the structure
of eddies, their internal variation in physical properties (e.g.,
temperature and nutrients), and their motions over time and
space, are fundamentally three-dimensional.
Driven by the above needs and the interests of scientists,
three-dimensional visualization technologies have been de-
veloped and used in the analysis of ocean eddies. Previous
studies used visualization to provide a census of the eddies
in a given dataset, visualize the spatial distribution of eddies
[12] [23] [25], and show the migration paths of eddies [12]
[26]. Three-dimensional ray casting techniques and special-
ized colormaps were used to visualize the eddy structure
and depth [15]. Our previous work in [7] combines the
hand-drawn figurative depiction of mesoscale ocean eddies
with time-dependent (time-varying) feature-based visualiza-

VOLUME 4, 2016 1

tion techniques to automatically produce illustrative visual-
izations from a high-resolution regional ocean model. This
combination allows the entire dataset to be summarized into a
single concise view but still enables structural understanding
of ocean eddies and their dynamics.
Existing data visualization software such as ParaView,
VisIt, and EnSight can often support the visualization of
time-varying volume data (e.g., ocean data). However, these
tools are usually proprietary, require licenses and installation
on a specified operating system and software environment,
and have a steep learning curve [24]. Another issue with the
existing visualization applications is that they often display
the sequence of data frames (i.e., timesteps) as animations
without considering the temporal coherence between data
frames. Last but not least, these visualization tools are not
designed specifically for exploring ocean eddies and thus
fail to satisfy many common requirements of scientists. De-
spite the above-mentioned successful eddy visualizations that
have been developed, a dedicated eddy visualization tool
that provides a one-stop solution from data processing to
visualization without the need to install specialized software
is still desperately needed by domain researchers.
The major challenges of developing such a tool include: a.
incorporating different data formats and coordinate systems
into one workflow; b. efficiently presenting large amounts
of data without the loss of the details; and c. assuring easy
distribution and installation of the tool given the range of
users and potential platforms. To address these challenges,
this study extends the illustrative eddy visualizations in our
previous work [7] to an integrated framework for the visu-
alization and analysis of three-dimensional ocean eddies in
web browsers. Combining eddy extraction, tracking, query,
and interactive visualizations in one workflow, the proposed
framework is designed to facilitate the visual exploration and
analysis of ocean eddies using the efficiency of feature-based
algorithms, a customized visualization design, the ubiquity of
web browsers, and the capabilities of Web Graphics Library
(WebGL) as well as GPUs and thus to help domain experts
get some insight into the process of ocean circulation.
The main contributions of this work include: (1) Investi-
gating ocean eddies and their motions as independent enti-
ties, which highlights their internal properties and individual
movements. (2) Representing the process of eddy migration
as a series of temporal-spatial patterns called eddy paths and
efficiently querying eddy paths from tracking eddies over
their life cycles. (3) Proposing web-compatible eddy meta-
data to reduce the storage requirement of data and the ren-
dering cost of visualization tasks. (4) Developing a WebGL-
based viewer to present the three-dimensional visualizations
of eddies within any standard web browsers. The viewer has
the benefit of not requiring a client-side installation, offering
centralized maintenance, and ensuring a common environ-
ment for all users, which promotes collaborative research and
customized extensions.
The remainder of the paper is organized as follows: Sec-
tion II reviews the related work and Section III introduces
2 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
the proposed framework for eddy visualization. We elaborate
our eddy feature computation method in Section IV and
describe the ocean data and the implementation performance
in Section V. The 3D eddy viewer is presented in Section VI.
In Section VII, we demonstrate the usage scenarios of the
eddy viewer. A user evaluation is provided in Section VIII.
Finally, conclusions and future work are given in Section IX.
II. RELATED WORKS
Previous research related to the methodologies adopted in
this paper includes time-varying volume visualization, ocean
eddy visualization, and data visualization software.
Visualizations of time-varying volume data usually fall
into two major categories: time-independent algorithms and
time-dependent algorithms. Time-independent algorithms
process multiple timesteps independently and display the se-
quence of timesteps as an animation. Related methods gener-
ally include reducing the storage size of data to make it more
manageable (e.g., compression, temporal subsampling, and
contouring), preselecting transfer functions for direct volume
rendering, and interactive hardware-accelerated volume ren-
dering [9]. Most existing data visualization software adopt
time-independent algorithms as default ways of temporal
navigation. In contrast, time-dependent algorithms introduce
a dependency between neighboring timesteps. Two common
methods are particle tracing and feature-based method (also
known as feature extraction and tracking) [24]. Particle trac-
ing is a fundamental method for the visualization of flow
field data and is usually computational complex and inten-
sive. The feature-based method (e.g., [19] [20]) exploits the
data coherence between consecutive timesteps and focuses
on just those regions of interest. The resulting significant
reduction in the storage requirement of data and rendering
cost of visualization tasks makes the feature-based method
highly suitable for WebGL-based visualizations. Thus, this
approach is followed in this work.
Understanding the motion of ocean eddies and visualizing
the eddies along with their motions in three-dimensions is
of increasing interest to the oceanographic, environmental,
and meteorological communities. Zhu and Moorhead [26]
tracked eddy instances over time by calculating their corre-
lation probabilities between successive timesteps and placed
the reconstructed three-dimensional structures of eddies in
an ocean bathymetry of the Pacific Ocean off the coast of
Japan. Studies have also visualized the structure of ocean
eddies. Grant et al. [5] visualized the vertical structure and
the life cycle of large anti-cyclonic eddies associated with
the loop current by showing the raising and sinking of the
thermocline associated with eddy genesis and decay. Petersen
et al. [12] tracked eddies from a high-resolution simulation
of the global ocean and presented the shape of eddies as iso-
surfaces of the Okubo-Weiss parameter. Three-dimensional
ray casting techniques and specialized colormaps were used
to visualize the eddy structure and depth [15]. Our previous
work in [7] combines the hand-drawn figurative depiction
of mesoscale ocean eddies with feature-based visualization
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
techniques to automatically produce three-dimensional illus-
trative visualizations from a high-resolution regional ocean
model. This study extends some of the techniques in [7] to
accommodate presenting visualizations in web browsers.
Data visualization software such as ParaView, VisIt, and
EnSight can extend their applications to time-varying volume
data. These programs display the sequence of timesteps as
animation and support temporal navigation in the form of
linear time slider and VCR-like controls (i.e., play, stop, step
forward, and step backward) [24]. Three different animation
modes are usually provided: a sequence of defined or se-
lected frames (e.g., the keyframe animation in EnSight), an
animation of a specific duration (e.g., the sequence mode
in ParaView), and a sequence of all timesteps. Multiple
operations (e.g., filtering, cutting, changing color schemes)
can be applied to individual timesteps. However, these vi-
sualization tools do not consider the temporal correlations
between consecutive data frames and are not specifically
designed for exploring ocean eddies. They usually fail to
satisfy some very basic tasks in studying ocean eddies such
as observing the interactions between different eddies and
the variations of multiple physical properties along with
the motion of eddies. Additionally, existing visualization
software all require installation and have specifications on
the software environment such as the versions of OpenGL.
An installation-free and dedicated eddy visualization tool is
needed by the domain experts, which motivates this study.
III. DESIGN AND APPROACH
The overarching goal of this study is to present a frame-
work for the exploratory visualization of ocean eddies in
web browsers. An integrated framework that combines eddy
extraction, tracking, and visualization in one workflow can
provide researchers with a one-stop solution to explore eddies
in 3D from ocean data so that they can focus on more
advanced questions in research.
A. VISUALIZATION DESIGN
The potential audience for the eddy visualizations include
scientists and students in oceanography, meteorology, and
environmental sciences. As part of our design process, we
interviewed four domain experts (two from environmental
sciences, two from oceanography) and five graduate students
(two from environmental sciences, three from oceanography)
to collect their requirements. We found that all of the po-
tential users expect an intuitive and interactive method to
facilitate the exploratory visualization of ocean eddies. They
want to observe the existing ocean eddies in a given day
and further to see a big picture of eddies along with their
motions in a geographic context within a period of time.
The scientists particularly want to track ocean eddies over
their life cycles to understand the dynamics of eddies in time
and space, while the students were especially interested in
exploring the multiple geometric (e.g., volume) and physical
properties (e.g., temperature) of individual eddy instances.
Based on the requirement analysis and the characteristics of
VOLUME 4, 2016
time-dependent feature-based algorithm, four visualizations
were designed to help users explore eddies from an ocean
data in three-dimensions. These visualizations are:
Daily eddy instance view which places all eddy instances
of a day in the bounding box of the data frame to give users
some basic idea of the relative positions, sizes, shapes, and
distributions of eddies before locating an individual eddy
from the visualization for more details.
Individual eddy path view which illustrates the motion of
an individual eddy in its life cycle as a series of consecutive
timesteps in space (i.e., an eddy path) with a directional line
above the eddy path to convey the direction of eddy motion
and to provide a sense of continuity. Users can adjust the dis-
tance between the consecutive timesteps, isolate a timestep
from the eddy path, and compare different timesteps.
Eddy skeleton overview where eddy instances in multiple
days or timesteps of multiple eddy paths are represented as
simple glyphs (cylinders) and are placed in the context of a
geographic map (e.g., ocean basin and coastline). This visu-
alization permits observation of a large portion of the dataset
and intends to provide an efficient and succinct overview of
the vertical extent and spatial (horizontal) distribution of the
visualized eddies with eddy vertical transfer highlighted.
3D eddy path overview which presents an “overview +
detailed” view where the loaded eddy paths are presented
in the context of a geographic map (the same as that of
eddy skeleton overview) and allows users to isolate each
individual eddy path from the visualization for more details.
In the visualization, timesteps of eddy paths are presented
as isosurfaces of eddy features to highlight the interactions
between eddy paths, add a sense of continuity, and bring a
more real and detailed overview of eddy motions to users.
Some of these visualizations have been published in our
previous work [7]. Table 1 lists the accompanying example
figures of these visualizations in this paper.
TABLE 1: Example Figures of The Visualizations
Visualization Name Example Figures
Daily Eddy Instance View Figure 8
Individual Eddy Path View Figure 9
Eddy Skeleton Overview Figure 10
3D Eddy Path Overview Figure 11
B. APPROACH OVERVIEW
An overview of the proposed eddy visualization framework
is illustrated in Figure 1. The framework is divided into
two major parts: eddy computation and eddy visualization.
In the eddy computation part, time-dependent feature-based
algorithms are chosen to generate web-compatible metadata
for efficient loading and rendering of a large amount of data
in web browsers and to enable a structural understanding
of ocean eddies and their motions. The workflow of the
computation starts from reading and preprocessing ocean
3
 Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

Ocean Data Frames User

Presentation
1 2 3 ...... N
Interactions
Interactions

WebGL API
General & I/O Control
Daily Eddy Instance View

Individual Eddy Path View

Identification & Extraction
Eddy Polygon
Metadata Eddy Skeleton Overview Meshes Pixels
Tracking Over Time
3D Eddy Path Overview
Query & Isolation

GPU
Polygon Mesh Creation
Eddy Metadata Creation

HTML CSS JS
Three.js
Feature-based Computation 3D Eddy Viewer Interactive Visualizations in a Web Browser

FIGURE 1: An overview of the proposed eddy visualization framework. The framework includes a feature-based eddy
computation algorithm and a 3D eddy viewer: the computation algorithm generates web-compatible eddy metadata for
visualization; the viewer reads and converts the eddy metadata to a series of polygon meshes and renders these polygon
meshes as visualizations in a web browser using the WebGL API and the capabilities of GPUs. Users’ interactions with the
graphic user interface and the visualizations are supported in the 3D eddy viewer.

data (a series of data frames), identifying and extracting eddy
features (i.e., regions of interest) from every data frame,
tracking the extracted eddy features over time, and isolating
eddy paths from the tracking history and the metadata of
extraction. Important physical and geometric properties are
computed in the process. In the visualization part, the four
designed visualizations are implemented by combining the
WebGL techniques with standard visualization techniques
such as surface rendering (i.e., rendering the isovalue surface
of a volumetric object) so that users can explore the metadata
of eddy computation at different levels from a single eddy
instance to multiple eddy paths in a web browser. Presenting
visualizations inside a standard web browser waives the
requirement of a client-side installation, ensures a common
environment for all users by centralized maintenance, and
facilitates the easy sharing of the tools and the interactive
parametrization of visualizations between research collabo-
rators at different locations on diverse platforms.
IV. FEATURE-BASED EDDY COMPUTATION
The eddy computation algorithm described in this section
is based on the time-dependent feature-based visualization
techniques. These techniques can significantly reduce the
amount of data that needs to be visualized by focusing on just
those regions of interest. The process consists of identifying
interesting regions or coherent structures of a dataset (i.e.,
eddies in this work), automatically extracting these features
from each timestep using a variant of segmentation, and
matching extracted features from a timestep to the next using
volume overlap, pattern matching or feature-vector matching.
A. EDDY IDENTIFICATION AND EXTRACTION
In identifying ocean eddies, we choose one of the ocean
community’s canonical methods: the Okubo-Weiss (OW) pa-
4 VOLUME 4, 2016
rameter [11] [22]. Generally, the OW parameter describes the
relative dominance of deformation with respect to rotation
of the flow. It divides the ocean into regions dominated by
vorticity, regions dominated by strain or deformation, and a
background where neither effect is dominant. Regions domi-
nated by vorticity are potentially eddies. Mathematically, the
OW is defined as:
OW = Sn2 + Ss2 − ω 2 (1)
where Sn and Ss are the normal and shear components of
strain respectively, and w is the relative vorticity of the flow.
They are different combinations of partial differentiations of
the horizontal velocities (u and v) in the x and y directions.
∂u ∂v
Sn = − (2)
∂x ∂y
∂v ∂u
Ss = + (3)
∂x ∂y
∂v ∂u
ω= − (4)
∂x ∂y
Eddies can be identified where the calculated OW values
exceed a given threshold. Given a positive threshold value,
the OW parameter allows partitioning of the ocean data into
strain dominated and vorticity dominated domains.
To extract vorticity dominated features, a region-growing
algorithm [16] [20] with a threshold of 0.2σOW is applied
to the OW parameter field in every data frame. Note that
the threshold value changes in each data frame as the OW
parameter depends on time. Connected components of points
with OW values below the threshold are extracted from
the data as vorticity-dominated features which are potential
eddies. Attributes of each extracted eddy feature, such as
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

Sea Surface

10 m
II I
III IV
Depth Volume

Y The chosen plane Vorticity

Orientation
Radius
FIGURE 2: A method to filter out non-eddy high-vorticity
features and to determine the spin orientations of eddies. The
directions of velocity vectors are checked at every data point
on the plane 10 meters below the ocean surface. The direction
angles are discretized into four axis quadrants.

volume, mass, centroid, and bounding box are also computed

during the process of extraction.
Though the OW parameter can identify eddies both at and
below the surface of the ocean, it also identifies high-vorticity
features that should not be considered as eddies, particularly
meanders in the Gulf Stream and other strong currents [23].
To filter out non-eddy high-vorticity features, we check the
directions of velocity vectors on a plane at about 10 meters
below the sea surface (Figure 2) for each extracted feature.
The direction is estimated as the angle between each point’s
velocity vector and the positive X axis (a vector pointing to
the east direction). As shown in Figure 2, these angles are
discretized into four quadrants: 0 to 90 degrees (Quadrant I),
90 to 180 degrees (Quadrant II), 180 to 270 degrees (Quad-
rant III), and 270 to 360 degrees (Quadrant IV). As a circular
flow, an eddy should contain velocity vectors in all four angle
domains. If the velocity vectors are consistent in the clock-
wise direction, the feature is labeled an anti-cyclonic eddy. If
the vectors are consistent in the counter-clockwise direction,
the feature is labeled a cyclonic eddy. Features with incon-
sistent vectors are discarded as non-eddies. Here, cyclonic
and anti-cyclonic are vorticity orientations. In the northern
hemisphere, a cyclonic eddy spins counter-clockwise and an
anti-cyclonic eddy spins clockwise.
After extracting eddy features from the raw data, three
geometric properties (i.e., depth, radius, and volume) are
computed for each eddy feature (Figure 3) based on the
metadata of eddy features (e.g., centroids, bounding box).
We also pull temperature and salinity from the raw data
frames and compute the horizontal velocity magnitude using
velocity vectors u and v:
V elocityM agnitude = u2 + v 2
To reduce the rendering cost for an efficient visualization,
the Marching Cubes algorithm [8] is applied to compute an
isovalue surface (i.e., isosurface, a 3D analog of an isovalue
VOLUME 4, 2016
FIGURE 3: Four geometric properties are computed for each
eddy feature in the process of eddy computation: depth,
volume, radius, and vorticity orientation.
line) of each eddy feature. Using isosurfaces, a volume
dataset can be rendered by a simple polygonal model (i.e.,
surface rendering). The metadata of each data frame is stored
in two text files: one saves data points of eddy features
(point metadata); the other saves isosurfaces of eddy features
(surface metadata). The computed geometric and physical
properties are written in both files.
B. EDDY TRACKING
Since an eddy is considered as a temporal phenomenon,
the complete description must include all of its instances
along the time axis and their associated properties [26]. Once
eddies have been identified and extracted from the raw data,
they are correlated over time with an overlap-based feature
tracking algorithm [20] to determine which eddy instances
in consecutive timesteps belong to the same eddy. We refer
to this spatial-temporal process of an eddy migration over its
life cycle as an eddy path. In an eddy path, the eddy may go
through five states from its birth to death: birth, continuation,
merge, split, and death [10] as defined below:
Birth - An eddy in the current timestep does not corre-
spond to any eddies in the previous timestep.
Continuation - An eddy from the previous timestep corre-
(5) sponds to just one eddy in the current timestep (no interaction
with other eddies).
Merge - Two eddies from the previous timestep corre-
spond to one eddy in the current timestep.
5

Split - An eddy from the previous timestep corresponds to
two eddies in the current timestep.
Death - An eddy from previous timestep does not corre-
spond to any eddies in the current timestep.
The result of eddy tracking is a text-based correspondence
list (tracking history) which records the evolution of eddies
in each timestep (day).
C. EDDY PATH ISOLATION
Tracking is only the first step in detecting eddy paths. With
numerous eddy features spanning hundreds of timesteps, a
method is needed to isolate each eddy path from the extrac-
tion and tracking metadata. An isolated eddy path contains a
sequence of eddy features (Figure 4). Our method applies a
three-step eddy path isolation algorithm as described below.
Step 1 - Query the correspondence list to create a list of
eddy sequences. The algorithm searches “death” events in
every timestep from the correspondence list of eddy features.
Whenever a “death” event is found, the algorithm will iter-
atively trace this eddy back to the previous timesteps until
it finds the “birth” event of this eddy. A list is created to
record each eddy path as a sequence of triplet: timestep,
index of the eddy feature in this timestep, and the state in
the evolution process. In the computation, we set a minimum
lifetime cutoff on the duration of eddy paths, usually to 30
days (but this can be changed by users).
Step 2 - Extracts every timestep of each eddy path from
the metadata of eddy extraction and save the eddy paths as
independent files based on the created list.
Step 3 - Compute the duration (the number of days) and
migration distance of each eddy path. The migration distance
of an eddy path is computed by accumulating the Euclidian
distance of center points (Cx, Cy, 0) on the ocean surface
between each pair of timesteps:
n p
X thus more web-compatible than the widely used OBJ file
MD = (Cxi+1 − Cxi )2 + (Cyi+1 − Cyi )2 (6)
k=1
where MD indicates the migration distance and n is the
number of timesteps (days) in the eddy path.
D. PROGRAM DESIGN AND ARCHITECTURE
A standalone C/C++ program with the support of VTK
libraries (https://www.vtk.org/) was developed to implement
the feature-based eddy computation algorithm introduced
previously. The program consists of four modules to handle
the general control and the three major steps of feature-based
eddy computation. Figure 5 shows the architecture of the pro-
gram, the pipeline of implementation, and the file format of
the metadata for web-based visualization. As shown in Figure
5a, the naming protocol for the source code files follows
meaningful and descriptive schemes. To run the program,
users need to specify some configuration information such
as the locations of the input and output files, the number of
timesteps, variable names, threshold, and the geographical
scope of the data.
6 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
Sequence 1:
…
Sequence 2:
…
Sequence 3:
…
…
FIGURE 4: Sequences of timesteps from three example eddy
paths are juxtaposed in three rows. An eddy may go through
five states in its life cycle: birth, continuation, merge, split, and
death.
The output files of the program are listed in Table 2 with
brief descriptions. The isosurfaces of eddy features in a day
or in an eddy path, which are saved as *.day files and *.path
files respectively, are the metadata for visualization. In a
feature isosurface file (Figure 5b), each feature entry starts
with a special line “155 155 155” and is followed by the
geometric properties and the ranges of the variables. The
feature is then written as a sequence of vertices and triangles:
a “vertex line” contains the coordinates of this vertex and the
interpolated values of non-primary variables (temperature,
salinity, and velocity magnitude in this implementation) at
this vertex; a “triangle line” saves the vertex indices of a
triangle. This file format is believed to be more efficient and
format which uses the first character of each line to specify
the type of command (i.e., “v” for vertices, “f” for faces of a
triangle).
V. DATA AND COMPUTATION PERFORMANCE
A. DATA
The ocean data used for this study is from a multi-decadal
regional ocean model simulation [6]. The simulation was per-
formed using the Regional Ocean Modeling System (ROMS)
[18], which solves the incompressible, hydrostatic, primitive
equations with a nonlinear free surface. It employs orthogo-
nal curvilinear coordinates in the horizontal and a generalized
topography following coordinate in the vertical [6]. The
simulation domain covers the major path of the Gulf Stream
through the northwest Atlantic (Figure 6).
In this study, the implementation covers the most recent
three years (2005 - 2007) in the simulation with 1095 daily
NetCDF [13] files (approximately 1.67 GB per file). The
vertical coordinate and stretching functions follow the ROMS
conventions. For details about this particular simulation,
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

Eddy Eddy Eddy Path ### FILE HEADER ###

General Control Extraction Tracking Isolation 155 155 155 //the start of a feature
# Geometric properties
# Ranges of variables
UOCD Number of nodes
x0 y0 z0 var1 var2 var3 ……
Configuration *.sequence file x1 y1 z1 var1 var2 var3 ……
Node Cell Extract *.uocd files Track *.trackTable file Isolate x2 y2 z2 var1 var2 var3 ……
..
.
Number of triangles
vertex ID vertex ID vertex ID
vertex ID vertex ID vertex ID
DAY
PATH .
..
Input Main Extract Track Isolate 0 //the end of this feature
155 155 155
*.day files *.path files ..
Ocean Data .

(a) (b)

FIGURE 5: (a) The architecture and pipeline of the eddy computation program. The program includes a module for general
control and three modules to implement the feature-based computation. (b) The file format of the feature isosurface file which is
used for surface rendering in the visualization. File extensions and file headers are used to differentiate eddy features in a data
frame (*.day files) and timesteps of an eddy path (*.path files).

TABLE 2: Output Files and File Descriptions

55
File Extension Description

Saving data points of the extracted eddy features in 50

*.uocd every data frame, these files are used for tracking
eddy features by their volume overlap. 45

Saving isosurfaces of eddy features in every data

Latitude

40
*.day frame, these files are used for eddy path isolation
and surface rendering in eddy visualization.
35

This file saves the evolution history (correspondence

*.trackTable 30
information) of each eddy feature frame by frame.
25
This file saves each eddy path as a sequence of
*.sequence triples: time step, feature ID, and evolution state 20

(i.e., birth, continuation, merge, split, and death).

15

Saving isosurfaces of eddy features in every eddy 260 270 280 290 300 310 320 330
*.path path, these files are used for surface rendering in Longitude
eddy visualization.

FIGURE 6: The ROMS data which was used to illustrate our

please see [6] [18]. The data contains 56 variables and uses
a geographic coordinate system with the spatial coverage
from 258.34◦ to 315.19◦ in longitude, 8.78◦ N to 53.83◦ N
in latitude, and 0 meter to 5500 meters in altitude. Each data
frame has 760 * 360 grid points with a horizontal spacing of
7 km and 40 vertical terrain-following levels stretched toward
the surface so as to resolve the surface boundary layers.
B. COMPUTATION PERFORMANCE
Our implementation of the eddy computation program was
performed on a Dell XPS desktop with an Intel Core i7-4770
(3.40 GHz * 8) CPU, 16 GB memory, and a Nvidia GTX 760
graphics card (2 GB video memory), running under Ubuntu
16.04 LTS.
The 1095 ROMS data frames (1.78 TB in total) were
downloaded from the data server and were stored in a local
folder. Some important parameters of the program running
are set as follows: a) depths of eddy features > 100 meters; b)
volume of eddy features > 50 data points; c) duration of eddy
VOLUME 4, 2016
framework covers the major path of the Gulf Stream through
the Northwest Atlantic Ocean. The red box in the map is an
approximate bound of the eddy computation.
paths > 20 days. The performance is evaluated by the elapsed
times and the output files of the computation. In the eddy
computation program, three timer functions were set to count
the elapsed times of eddy extraction (including data reading
and preprocessing), eddy tracking, and eddy path isolation
(including metadata creation). The generated files were saved
in a new folder. Table 3 lists the elapsed times and the file
sizes of the output (the tracking table and eddy path list are
negligible in file size and so are not listed in the table).
VI. 3D EDDY VIEWER
We developed EddyViz, a browser-based viewer to imple-
ment the designed eddy visualizations by surface rendering
the isosurfaces of eddy features within web browsers. The
viewer supports interactive surface rendering of 3D objects
(i.e., eddy features in this study) and overlaying 3D geo-
7

TABLE 3: Elapsed Times and Output Files
Elapsed Times File Sizes of the Output Files
Extraction Tracking Isolation *.day *.uocd
1095 files 1095 files
2h 46m 2h 33m 5h 41m
10.52 GB 8.67 GB
graphic maps through the WebGL API. WebGL has been
adopted by most modern browsers, which allows users to
create beautiful and high-performance three-dimensional ap-
plications, using the capabilities of the GPU in computers.
The viewer is written in JavaScript language with the support
of Three.js library [2], a JavaScript 3D Library for WebGL.
As a browser-running and open source product, EddyViz is
easy to get running and gives users a unique opportunity to
monitor what happening in the code and permits the adaption
of the code into their particular interests whenever needed.
As the user interface of EddyViz shown in Figure 7, the
four eddy visualizations are implemented in four independent
webpages. Each webpage contains a viewport for visualiza-
tion and a control panel to manipulate file I/O, variables
changes, and fire functions on the fly. In the viewer, users can
zoom in/out, rotate, and pan a visualization by mouse or key-
board. The scale of vertical exaggeration in the visualizations
is set to 1000. By default, when openning up eddy skeleton
overview and 3D eddy path overview, a 3D geographic map
will be automatically loaded as the background. The map
is generated by combining ocean basin bathymetry with
land surface topography for the east coast of the US and
creating a terrain elevation surface using Delaunay Triangu-
lation [14]. As currently implemented, the terrain elevation
surface creation uses a high-resolution elevation dataset from
National Oceanic and Atmospheric Administration (NOAA,
http://www.noaa.gov/). Users can also create their own maps
and replace the default one. In EddyViz, the isosurfaces of
eddy feature can be either rendered as a solid color or be
FIGURE 7: The user interface of the 3D eddy viewer “Ed-
dyViz". EddyViz implements the designed interactive visual-
izations of eddies in four independent webpages.
8 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
color-coded by any of the three physical properties chosen
in the eddy computation: temperature, salinity, and velocity
magnitude, with different color maps. Minor coding updates
*.path could provide a different selection of physical properties.
68 files
VII. USAGE SCENARIOS
225.67 MB
In this section, we illustrate the functionality and utility of the
developed 3D eddy viewer EddyViz by using three typical
usage scenarios with the key tasks that the viewer as well as
the included visualizations are designed to support. The data
for exploration is the isosurfaces of eddy features (surface
metadata) and covers three years (2005 - 2007) in the ROMS
data. More features and functionalities of EddyViz can be
found in the accompanying video to this paper.
A. EXPLORING EDDY INSTANCES IN A DAY
In the first scenario, student A was doing a daily analysis of
eddies in the Spring of 2005. She first opened the webpage
“Daily Eddy Instances View” from the user interface (as
shown in Figure 7) of EddyViz, then clicked the “File Open”
button in the control panel. In the opened file selection dialog,
she selected a *.day file stored in the local hard drive. Once
the data file was loaded, the basic information of this data
frame (i.e., file name, date, and the number of eddies) was
displayed in the control panel. As shown in Figure 8a, all
of the eddy instances in the day Aprill 30, 2005 (i.e., eddy
features in the corresponding data frame) were listed on the
right-hand side of the webpage and spun clockwise (anti-
cyclonic eddies) or counter-clockwise (cyclonic eddies).
These eddy instances were also visualized as a daily eddy
instance view in the center of the webpage. The visualization
gave the student an overall daily distribution and orientation
of cyclonic (green) and anti-cyclonic (purple) eddies. The
student found that most eddies have a tapering shape and the
two types of eddies were almost equal numbered in this day.
She also noticed that the anti-cyclonic eddies were larger in
size and were positioned to the north of cyclonic eddies. To
verify her judgement, the student clicked the button “Export
Data Information” in the control panel to request the basic
information and statistics of the eddies in this day which was
displayed in a separated webpage.
After having some overall understanding of the eddy
instances in this day, student A double clicked an eddy
feature (the 24th on the list) in the visualization to get
more details (note: an individual eddy feature can also be
chosen from the list on the right-hand side). The webpage
then switched to a "single eddy mode" where the selected
eddy was visualized independently with its basic information
(i.e., location, spin orientation, radius, depth, and volume)
displayed in the control panel. To observe different physical
properties of the chosen eddy, she color-coded the eddy by
temperature, salinity, and velocity magnitude successively
with different colormaps. Figure 8b illustrates the variation
of temperature on the surface of the eddy. As can be found
from the figure, the temperature of this eddy first reduces
as the eddy penetrates into the ocean then stays constant
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
(a)
(b)
FIGURE 8: Exploring eddy instances in a day. (a) The daily
eddy instance view presents all of the eddy instances in this day
(data frame) in the bounding box of the data frame. Two colors
are used to differentiate cyclonic (green) and anti-cyclonic
(purple) eddies. An eddy (the 24th in this data frame) is
selected from the visualization and presented independently
in (b). The eddy is color-coded by the temperature variation
on its surface. A color bar on the top of the webpage shows
the range of the temperature variations.
after the depth went beyond some level which is defined as
the isothermal curve. Finally, the student switched back to
the initial mode of the visualization and continued to pick
another eddy feature for analysis.
B. EXPLORING TIMESTEPS OF AN EDDY PATH
In the second scenario, scientist B wanted to study the general
process of eddy migration by focusing on the motion of an
individual eddy over its life cycle. In the webpage “Individual
Eddy Path View” of EddyViz, he selected a *.path file from
the data folder and loaded the file into the webpage. The
information in the control panel indicated that this eddy had
a counter-clockwise orientation and lasted 44 days from May
9 to June 22 in the year 2005. All timesteps that this eddy
went through in its life cycle were listed on the right-hand
side of the webpage and were presented as an individual eddy
path view without any distracting context. To avoid occlusion
VOLUME 4, 2016
(a)
(b)
FIGURE 9: Exploring timesteps of an eddy path. (a) The
individual eddy path view presents all of the timesteps of an
eddy path as isosurfaces which are color-coded by veloc-
ity magnitude. The arrow above highlights the continuation
and direction of the eddy motion. (b) The first and the last
timesteps of the eddy path are color-coded by temperature,
salinity, and velocity magnitude and are represented as wire-
frames in a coordinated view for comparison. The colorbars
and the red brackets below indicate the variation ranges of
the three physical properties on these two timesteps. On the
bottom left corner of the webpage, a small inset presents a
miniature of the eddy path with the two chosen timesteps
highlighted in solid red color.
by successive timesteps, the scientist adjusted the distance
between each two neighboring timesteps to a factor of ten
relative to the actual distance. He then color-coded the visu-
alized features by different physical properties and rotated the
visualization to observe the eddy path from different angles.
The individual eddy path view in Figure 9a illustrates the
changes of velocity magnitude on the surface of this eddy
along with its motion. From the visualization, he noticed that
the eddy seemed to go through an intensification process:
the features became more colorful and bigger in size and the
spaces between the consecutive timesteps were increasing as
the eddy evolved.
To summarize the structural and temporal changes of this
eddy in its life cycle, the scientist compared the first and the
9
 Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

FIGURE 10: This eddy skeleton overview summarizes all of the eddy instances that had depths penetrating into the ocean at least
1727 meters in the year 2007. In the visualization, each timestep of an eddy path is represented as a simple glyph (skeleton)
and is placed in a context of the ocean basin and coastline. It is clear that the cyclonic eddies (yellow) which mostly appeared in
the south were outnumbered by the anti-cyclonic eddies (purple) which gathered in the north.

last timestep of the eddy path in a coordinated view where the
two chosen timesteps were both color-coded by temperature,
salinity, and velocity magnitude and were represented as
wireframes to highlight the topological structures (Figure
9b). As shown in Figure 9b, the three representations of the
first timestep are listed on the first row while those of the last
timestep are listed on the second row. The color bars on the
top of the visualization show the ranges of variations in these
physical properties; the two red brackets below the color bars
indicate the ranges for the two selected timesteps. From the
visualization and the geometric information in the control
panel, the scientist found that the eddy became bigger, higher,
hotter, saltier, and faster at the end of its life cycle compared
with its initial stage.
C. SUMMARIZING EDDIES IN A PERIOD OF TIME
In the third scenario, scientist C was doing a longitudinal
study of eddy evolution in the North Atlantic. Considering
the large portion of data that would be involved, she believed
that studying eddies day by day or focusing on individual
eddies was no longer effective. Instead, she would like to see
a big picture of eddies in a long time period. In EddyViz,
she opened up the webpage “Eddy Skeleton Overview”, then
selected and loaded all of the *.day files of the year 2007
through the control panel in the webpage. All of the eddy
instances saved in the loaded files were visualized as an
eddy skeleton overview. From the visualization, the scientist
noticed that the cyclonic eddies (yellow) were more likely to
appear in the south while the anti-cyclonic eddies (purple)
were mostly gathered in the north. Both types of eddies were
10
almost equal numbered. Also, most eddies extended less than
2000 meters below the sea surface (the deepest point of the
ocean basin is 5500 meters). Using a scrollbar in the control
panel, she slowly filtered the eddy cylinders (skeletons) in
the visualization by the depths (penetration into the ocean) of
eddies from 300 meters to 2000 meters to oberve the changes
in the distributions and numbers of the visualized eddies. She
found that the cyclonic eddies were gradually outnumbered
by the anti-cyclonic eddies when the depth threshold went
beyond 900 meters. In Figure 10, the eddy skeleton overview
presents all of the eddy instances that have depths penetrated
into the ocean at least 1727 meters in the year 2007.
Although the eddy skeleton overview presents an efficient
and simplified overview of eddies, the scientist felt some
important information such as the size, shape, and interaction
of eddies were missing from the visualization. She decided to
use another visualization, which is 3D eddy path overview,
to summarize the motion of eddies in the year 2007. All
of the eddy paths in 2007 were visualized as a 3D eddy
path overview as shown in Figure 11. From the figure, she
found that the cyclonic eddies (green) were more likely to
move westward while anti-cyclonic eddies (purple) tended to
move to different directions. She then isolated a cyclonic and
an anti-cyclonic eddy paths from the overview visualization
successively and presented them in a separate view (as shown
on the top of Figure 11) to get more details of these two
eddies. Some basic information (i.e., date, duration, and
distance) of the selected eddy paths were displayed in the
control panel. Finally, she requested a statistic summary of
all of the eddy paths in the year 2007.
VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
1
2007/06/12 – 2007/08/25
FIGURE 11: This 3D eddy path overview presents a bunch of selected eddy paths in the context of ocean basin and coastline.
A red ball is added above the last time step of each visualized eddy path to convey the moving direction of the eddy. The
visualization shows that cyclonic eddies (green) tend to move westward while anti-cyclonic eddies (purple) move to different
directions. A cyclonic and an anti-cyclonic eddy paths are isolated by double-clicking them in the visualization and are displayed
separately as shown on the top of the figure.
VIII. USER EVALUATION
The effectiveness of the visualizations and the utility of the
viewer EddyViz was evaluated by a survey study to determine
how well the method proposed in this paper satisfies the
requirements of the potential users.
A. THE PROCEDURE
The study involved 22 graduate students and postdoctoral
fellows (12 female, 10 male) from the Department of Marine
and Coast Science and the Department of Environmental
Sciences at Rutgers University. They all had research or study
experience with ocean eddies before. After a brief intro-
duction to the project and a demonstration on the EddyViz
(approximately 10 minutes in total), the study participants
were asked to complete three basic tasks (similar to the
user scenarios in the previous section) to make sure they
used the viewer sufficiently. Next, the participants used the
viewer to explore the data based on their individual interests.
Finally, they were invited to fill out a survey assessing the
visualizations and their user experiences. The survey con-
tains five statements about the visualization design and five
statements about the utility of EddyViz (see Table 4). Each
statement in the survey uses a five-point Likert scale (1 =
very poor, 2 = poor, 3 = fair, 4 = good, 5 = excellent). We
strongly encouraged the participants to add comments to each
VOLUME 4, 2016
2
2007/07/26 – 2007/12/25
statement and to suggest new functionalities for the viewer.
B. THE RESULTS
Figure 12 summarizes all of the participants’ ratings on the
ten survey statements (blue-colored bars for the statements of
visualization and green-colored bars for the statements of the
utility of EddyViz). Statistically, the visualization aspects re-
ceived ratings of over 4 (good) on four statements and over 3
(fair) on the remaining item. The free-form written responses
and the oral comments also tended towards positive attitudes.
(a) Most participants (20 out of 22) thought the visualizations
were well-designed and effective in general, which received
an average score of 4.45 with a standard deviation σ of ±0.67
in the statement S1. The students were amazed to see and
to interact with the eddy visualizations. Many students com-
mented on the visualizations as “innovative” and “provide a
close view to ocean eddies and their movements”.
(b) The statements S2, S3, and S5 received ratings over 4
(good). For S2, 15 participants agreed that the daily eddy
instance view can facilitate the exploration of eddy instances
in a day and the statement received an average score of 4 (σ
= ±0.81). For S3, 19 participants believed that the individual
eddy path view can help them get a better understanding of
the motion of an eddy (the average score is 4.26, with σ =
11

TABLE 4: User Study Survey Statements.
Visualization Aspects
The visualizations provided by the viewer are well-
S1
designed and effective in general.
The Daily Eddy Instance View (1st visualization) can
S2
facilitate the exploration of eddy instances in a day.
The Individual Eddy Path View (2nd visualization) helps
S3
me get a better understanding of the motion of an eddy.
The Eddy Skeleton Overview (3rd visualization) provides
S4
a succinct summarization of eddy distribution.
The 3D Eddy Path Overview (4th visualization) gives me
S5
a big picture of eddies and their motions in a region.
Usability and User Experience Aspects
S6 The viewer has a rational design.
S7 I have fun using the viewer.
The viewer could be useful and helpful for my research
S8
and study in future.
S9 The eddy data can be easily opened and handled.
S10 The navigation is smooth and comprehensible.
±0.73). One research fellow told us “showing the trajectories
of an eddy in 3D is more visual and intuitive than the
tranditional 2D plot with arrows”. Another student called this
visualization as “an effective and succinct way to observe the
motion of an eddy”. For S5, almost all of the participants (21
out of 22) accepted that the 3D eddy path overview gave them
a big picture of eddy migrations in a region, which resulted
in an average score of 4.21 (σ = ±0.54). One comment on
this statement reads “it summarizes different life histories of
eddies and how these eddies are affected by interactions with
each other and the ocean basin”.
(c) In contrast, the statement S4 received a fair rating (the
average score = 3.37, with σ = ±0.83). 14 participants held a
positive attitude toward the statement and one student partic-
ularly liked the eddy skeleton overview as “it provides a sim-
ple overview of the distributions of cyclonic and anticyclonic
eddies and the thresholds can be used to adjust the eddies in
the ocean bottom topography.” However, some participants
raised criticisms such as “the density of the plotted items on
the map” and “the relative simplicity of the functionality”.
The utility and user experience aspects of EddyViz ex-
posed more limitations and received lower ratings on aver-
age. The critical comments (both written and verbal) from
the students usually focused on basic operations that would
increase the “user-friendliness” of the viewer such as a more
advanced and reliable interaction in locating and filtering
12
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
objects from visualizations and making the navigation more
intuitive and controlled.
(d) The ratings on the statements S6 and S7 are high. 16
out of 22 participants agreed that the design of EddyViz was
rational (S6’s average score is 4, with σ = ±0.69) and 18 par-
ticipants had fun using the viewer (S7’s average score is 4.14,
with σ = ±0.71). Participants were generally impressed with
a dedicated and installation-free tool for eddy visualization.
Frequently repeated descriptive terms for the viewer in the
written or oral comments include: visual, interactive, clear,
innovative, and cool.
(e) Surprisingly, the usefulness and helpfulness of EddyViz
(S8) were not much favored by the participants (average
score is 3.55, with σ = ±1.06). Five participants commented
on how the tool could benefit their research in future. One of
them stated, “It is nice to have a noncommercial and web-
based tool for this special domain. This tool enables more
interactive and knowledge gain compared to traditional data
slice browsing.” However, some participants concerned about
the separation of data processing from the viewer and the
generality of the viewer. As one student would have liked
to “get the visualizations directly from the original ocean
simulation.” Another student wanted to “use the tool for
visualizing eddies from other ocean data”.
(f) The participants rated the 9th statement from 1 to 5, which
resulted in a low average score (2.91) and a high standard
deviation (±1.23). This might be because of the abstractness
(if not confusing) of the file names, the constraints of web
browsers, and some limitations of the data reading function-
ality in EddyViz under current implementation. We found
that the browser experienced a very long responding time
when a participant tried to load all of *.day files to the viewer.
A research fellow commented that “It is hard to find a data
file by the date of eddy occurrence. I can not tell which files
belong to 2005 or 2006.” Four other participants also wished
“the ability to load eddy data from remote hard drives”.
(g) The participants had few to no problems with the smooth-
ness of navigation in the viewer and thought the viewer was
self-explanatory, which results in an average score of 3.41 (σ
= ±0.96) for S10.
The overall ratings and the free-form written responses in
the surveys and the participants’ verbal comments during or
after the study confirm the overall success of the designed
visualizations and support the utility of EddyViz in general.
Although the absence of eddy computation in the study and
the reliance on expressed opinions limit the interpretation of
results, the study suggests that the potential users liked an
installation-free and dedicated 3D eddy viewer for exploring
eddies from ocean data and favored the novelty and effec-
tiveness of our method in general. The study gave us much
more qualitative information regarding how researchers can
benefit from the visualizations to gain a better understanding
of the ocean eddies in different perspectives (e.g., distri-
bution, shape, physical properties, dynamics). Many of the
VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
S1 S2 S3 S4
12 12
11
8 8 8 8
7 7
4 4
3 3
2 2
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
4.45 ± 0.67 4 ± 0.82 4.26 ± 0.73 3.37 ± 0.83
FIGURE 12: Results of survey study. Six out of the ten survey statements receive an average rating over 4 (good) and the
remaining four statements were scored over 3 (fair).
comments from the participants have been being used or
considered for the improvement of our method.
IX. CONCLUSION AND FUTURE DEVELOPMENT
The main motivation of this paper is to provide a one-stop
solution from processing the ocean data to visual exploration
of 3D ocean eddies and to make the ocean data as well as
the visualization tool more accessible and interpretable to a
wider audience. For this reason, the focus has been on devel-
oping a browser-based visualization tool that does not require
any special software to be installed on the client side and
thus enables sharing the tools and the visualization results
easily with remote collaborators working in different plat-
forms or software environments. The proposed visualization
framework in this paper contains a time-dependent feature-
based algorithm implemented by a standalone C++ program
and a browser-based 3D eddy viewer named EddyViz. The
visualization design was based on specific requirements from
potential users and also considered the constraints imposed
by web browsers. The feature-based algorithm generates the
isovalue surfaces of eddies to reduce the rendering cost of
visualization and also supports a structural understanding
of ocean eddies and their motions. EddyViz presents 3D
eddy visualizations within a standard web browser using the
capabilities of WebGL as well as GPUs. In this paper, the
application of the framework to a high-resolution regional
ocean simulation of the northwest Atlantic Ocean and the
general usage scenarios of EddyViz are discussed. A user
evaluation was conducted and the results confirm the success
of the visualization design and the potential of the proposed
framework in helping researchers understand ocean eddies.
Although the proposed visualization framework provides
a considerable set of useful features and fulfills the initial
design goal, we have become aware of some of the limitations
in implementing the framework. For the future developments,
four major aspects can be explored to advantage. First of
all, for the feature-based eddy computation program, we
plan to convert the C++ code to Matlab code and han-
dle the input parameters to the program through a Matlab
graphic user interface (GUI). Matlab is a well-documented
standard code scripting language (popular in both academia
VOLUME 4, 2016
S5 S6 S7 S8 S9 S10
13
12
11
10
9
7
5 6 6 5 5
5 5 5
4 4 4
1 2 2
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
4.21 ± 0.54 4 ± 0.69 4.14 ± 0.71 3.55 ± 1.06 2.91 ± 1.23 3.41 ± 0.96
and industry) and is fully readable even by novice users.
This is also a preferred coding method by most scientists.
Secondly, for EddyViz, we decide to make the visualizations
which are currently implemented in independent webpages
more coordinated so that users can switch between different
visualizations. Thirdly, we have only applied the framework
to Gulf Stream eddies from the particular ROMS model in
this work. In consideration of the universality, we plan to
apply the framework to other ocean eddies (e.g., mesoscale
eddies in the Northwestern Pacific Ocean). We also anticipate
improvements to the data reading and preprocessing parts
in expanding the applications. Finally, we intend to process
data from the simulation environment or other powerful
centralized infrastructure and visualize the results in web
browsers. This client-server architecture will leave the heavy
computation tasks in the server side so that the client side
can execute as few operations as possible to obtain a directly
renderable representation.
ACKNOWLEDGMENT
We wish to thank the Department of Environmental Sciences
and the Department of Marine and Coastal Sciences at Rut-
gers University for providing the ocean simulation data. We
particularly thank Dr. Enrique Curchitser and Dr. Kang for
their valuable suggestions as well as many useful feedbacks
along the way. We thank all students and researchers who
participated in the user studies. We also thank the anonymous
reviewers for the useful suggestions.
REFERENCES
[1] Amores, A., Melnichenko, O. and Maximenko, N., 2017. Coherent
mesoscale eddies in the North Atlantic subtropical gyre: 3-D structure and
transport with application to the salinity maximum. Journal of Geophysical
Research: Oceans, 122(1), pp.23-41.
[2] Cabello, R., 2010. Three. js. URL: https://github. com/mrdoob/three.js.
[3] Castelao, R.M., 2014. Mesoscale eddies in the South Atlantic Bight and the
Gulf Stream recirculation region: vertical structure. Journal of Geophysical
Research: Oceans, 119(3), pp.2048-2065.
[4] Chaigneau, A., Le Texier, M., Eldin, G., Grados, C. and Pizarro, O., 2011.
Vertical structure of mesoscale eddies in the eastern South Pacific Ocean:
A composite analysis from altimetry and Argo profiling floats. Journal of
Geophysical Research: Oceans, 116(C11).
[5] Grant, J., Erlebacher, G. and O’Brien, J., 2002, October. Case study:
visualizing ocean flow vertical motions using Lagrangian-Eulerian time
13
 Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

surfaces. In Proceedings of the conference on Visualization’02 (pp. 529- data (e.g., *.day files, *.path files, etc.) for eddy visual
532). IEEE Computer Society. exploration and data analysis. More information about
[6] Kang, D. and Curchitser, E.N., 2013. Gulf Stream eddy characteristics in
a high-resolution ocean model. Journal of Geophysical Research: Oceans, the program, the access to the source code, and the in-
118(9), pp.4474-4487. troduction on how to run the code can be found at:
[7] Liu, L., Silver, D., Bemis, K., Kang, D. and Curchitser, E., 2017, June. Il- http://vizlab.rutgers.edu/eddyprogram The website also con-
lustrative Visualization of Mesoscale Ocean Eddies. In Computer Graphics
Forum (Vol. 36, No. 3, pp. 447-458). tains some example data for testing the program. Please note,
[8] Lorensen, W.E. and Cline, H.E., 1987, August. Marching cubes: A high the ROMS data is not made available to the public at the
resolution 3D surface construction algorithm. In ACM siggraph computer moment. For the access to the ROMS data, please contact:
graphics (Vol. 21, No. 4, pp. 163-169). ACM.
[9] Ma, K.L., 2003. Visualizing time-varying volume data. Computing in Dr. Dujuan Kang: dujuan@marine.rutgers.edu Dr. Enrique
Science and Engineering, 5(2), pp.34-42. Curchitser: enrique@marine.rutgers.edu
[10] Matsuoka, D., Araki, F., Inoue, Y. and Sasaki, H., 2016. A new approach
to ocean eddy detection, tracking, and event visualizationâĂŞapplication
to the northwest pacific ocean. Procedia Computer Science, 80, pp.1601- EDDYVIZ
1611. EddyViz reads the *.day and *.path files that are gener-
[11] Okubo, A., 1970, June. Horizontal dispersion of floatable particles in the
ated from the eddy computation program and presents these
vicinity of velocity singularities such as convergences. In Deep sea re-
search and oceanographic abstracts (Vol. 17, No. 3, pp. 445-454). Elsevier. files as visualizations in web browsers. A link to EddyViz:
[12] Petersen, M.R., Williams, S.J., Maltrud, M.E., Hecht, M.W. and Hamann, http://vizlab.rutgers.edu/eddyviz.html
B., 2013. A three-dimensional eddy census of a high-resolution global
ocean simulation. Journal of Geophysical Research: Oceans, 118(4),
All of the functionalities in EddyViz have been tested
pp.1759-1774. in Firefox and Google Chrome. The basic operations can
[13] Rew, R. and Davis, G., 1990. NetCDF: an interface for scientific data be found from the accompanying video of this manuscript.
access. IEEE computer graphics and applications, 10(4), pp.76-82.
Please note, the number of opened tabs or webpages in a
[14] Ruppert, J., 1995. A Delaunay refinement algorithm for quality 2-
dimensional mesh generation. Journal of algorithms, 18(3), pp.548-585. browser will affect the speed of data loading and visualization
[15] Samsel, F., Petersen, M., Abram, G., Turton, T.L., Rogers, D. and Ahrens, rendering.
J., 2015, November. Visualization of ocean currents and eddies in a high-
resolution global ocean-climate model. In Proceedings of the International
Conference on High Performance Computing, Networking, Storage and
Analysis (Vol. 2, No. 3).
[16] Samtaney, R., Silver, D., Zabusky, N. and Cao, J., 1994. Visualizing
features and tracking their evolution. Computer, 27(7), pp.20-27.
[17] Schroeder, W.J., Lorensen, B. and Martin, K., 2004. The visualization
toolkit: an object-oriented approach to 3D graphics. Kitware.
[18] Shchepetkin, A.F. and McWilliams, J.C., 2003. A method for computing
horizontal pressureâĂŘgradient force in an oceanic model with a non-
aligned vertical coordinate. Journal of Geophysical Research: Oceans,
108(C3).
[19] Shen, H.W., 2006. Visualization of large scale time-varying scientific data.
In Journal of Physics: Conference Series (Vol. 46, No. 1, p. 535). IOP
Publishing.
[20] Silver, D. and Wang, X., 1997. Tracking and visualizing turbulent 3d
features. IEEE Transactions on Visualization and Computer Graphics,
3(2), pp.129-141.
[21] Sood, V., John, B., Balasubramanian, R. and Tandon, A., 2005, September.
Segmentation and tracking of mesoscale eddies in numeric ocean models.
In Image Processing, 2005. ICIP 2005. IEEE International Conference on
(Vol. 3, pp. III-469). IEEE.
[22] Weiss, J., 1991. The dynamics of enstrophy transfer in two-dimensional
hydrodynamics. Physica D: Nonlinear Phenomena, 48(2-3), pp.273-294.
[23] Williams, S., Hecht, M., Petersen, M., Strelitz, R., Maltrud, M., Ahrens, J.,
Hlawitschka, M. and Hamann, B., 2011, June. Visualization and analysis
of eddies in a global ocean simulation. In Computer Graphics Forum (Vol.
30, No. 3, pp. 991-1000). Blackwell Publishing Ltd.
[24] Wolter, M., 2010. Navigation in time-varying scientific data (Doctoral
dissertation, RWTH Aachen University).
[25] Woodring, J., Petersen, M., SchmeiΚer, A., Patchett, J., Ahrens, J. and
Hagen, H., 2016. In situ eddy analysis in a high-resolution ocean climate
model. IEEE transactions on visualization and computer graphics, 22(1),
pp.857-866.
[26] Zhu, Z. and Moorhead, R.J., 1995, September. Extracting and visualizing
ocean eddies in time-varying flow fields. In Proceedings of the 7th Inter-
national Conference on Flow Visualization (pp. 206-211). Seattle, WA.
.

APPENDIX
EDDY COMPUTATION PROGRAM
This program reads time varying "Regional Ocean Mod-
eling System" (ROMS) data and creates a set of meta-
14 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS

LI LIU received his Ph.D. degree in Computer

Engineering from Rutgers, The State University
of New Jersey, his M.S. degree in Signal and
Information Processing from Beijing Normal Uni-
versity (in Beijing, China), and B.S. degree in
Electronic and Communications Engineering from
Donghua University (in Shanghai, China). His
research focuses on scientific visualization, com-
putational fluid dynamics (CFD) visualization,
tracking motions in scientific simulations, volume
graphics, machine learning, image processing and understanding, informa-
tion visualization, and UI/UX design.

DEBORAH SILVER is a professor in the De-

partment of Electrical and Computer Engineering
at Rutgers, The State University of New Jersey
and the Executive Director of the Professional
Science Master’s Program. She received a B.S.
from Columbia University School of Engineering
and an M.S. and Ph.D. from Princeton University
in Computer Science. Her area of research is in
Scientific Visualization. She has been a PI in the
Vizlab at the CAIP Center, Rutgers University
since joining the faculty. She has taught courses in Computer Graphics,
Visualization, Data Structures, Software Engineering and Robotics. She is
involved in different visualization projects including oceanic visualization,
medical visualization, CFD visualization, and volume graphics. She has been
co-chair of the papers session and program co-chair of the yearly IEEE
Visualization conference, she has been Vice Chair of operations for the
IEEE Technical Committee on Computer Graphics (1993-2000), and she has
been on the editorial committee of the IEEE Transaction on Visualization
and Computer Graphics (1995-2000). She was the Associate Dean for
Continuing and Professional Education for the School of Engineering from
2008-2010.

KAREN BEMIS received her B.S. from Rice

University (in Houston, TX) in geophysics in 1988
and went on to the MIT/WHOI Joint Program
in Oceanography to study heat flux. There, she
learned to love hydrothermal systems and fluid
dynamics, while working with Dick Von Herzen
and Debbie Smith. She left the Joint Program
in 1991 with a master’s degree in oceanography.
She received her Ph.D. in Geological Sciences at
Rutgers in 1995 on the morphology of volcanos,
especially small explosive volcanoes called cinder cones. After a short
internship at the NJ Geological Survey studying groundwater flow in the
Passaic River System, she returned to hydrothermal systems in 1996 to
work with Peter Rona as a postdoc in the Institute of Marine and Coastal
Sciences at Rutgers. Through Peter, she was introduced to Deborah Silver
and scientific visualization. Currently, she’s a Assistant Research Professor
in the Department of Marine and Coastal Sciences. She continues to study
cinder cones in Guatemala and sonar imaging of hydrothermal systems while
exploring other uses of scientific visualization in geologic problems and the
psychology of understanding and using visualizations.

VOLUME 4, 2016 15