Professional Documents
Culture Documents
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
VOLUME 4, 2016 1
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
tion techniques to automatically produce illustrative visual- the proposed framework for eddy visualization. We elaborate
izations from a high-resolution regional ocean model. This our eddy feature computation method in Section IV and
combination allows the entire dataset to be summarized into a describe the ocean data and the implementation performance
single concise view but still enables structural understanding in Section V. The 3D eddy viewer is presented in Section VI.
of ocean eddies and their dynamics. In Section VII, we demonstrate the usage scenarios of the
Existing data visualization software such as ParaView, eddy viewer. A user evaluation is provided in Section VIII.
VisIt, and EnSight can often support the visualization of Finally, conclusions and future work are given in Section IX.
time-varying volume data (e.g., ocean data). However, these
tools are usually proprietary, require licenses and installation II. RELATED WORKS
on a specified operating system and software environment, Previous research related to the methodologies adopted in
and have a steep learning curve [24]. Another issue with the this paper includes time-varying volume visualization, ocean
existing visualization applications is that they often display eddy visualization, and data visualization software.
the sequence of data frames (i.e., timesteps) as animations Visualizations of time-varying volume data usually fall
without considering the temporal coherence between data into two major categories: time-independent algorithms and
frames. Last but not least, these visualization tools are not time-dependent algorithms. Time-independent algorithms
designed specifically for exploring ocean eddies and thus process multiple timesteps independently and display the se-
fail to satisfy many common requirements of scientists. De- quence of timesteps as an animation. Related methods gener-
spite the above-mentioned successful eddy visualizations that ally include reducing the storage size of data to make it more
have been developed, a dedicated eddy visualization tool manageable (e.g., compression, temporal subsampling, and
that provides a one-stop solution from data processing to contouring), preselecting transfer functions for direct volume
visualization without the need to install specialized software rendering, and interactive hardware-accelerated volume ren-
is still desperately needed by domain researchers. dering [9]. Most existing data visualization software adopt
The major challenges of developing such a tool include: a. time-independent algorithms as default ways of temporal
incorporating different data formats and coordinate systems navigation. In contrast, time-dependent algorithms introduce
into one workflow; b. efficiently presenting large amounts a dependency between neighboring timesteps. Two common
of data without the loss of the details; and c. assuring easy methods are particle tracing and feature-based method (also
distribution and installation of the tool given the range of known as feature extraction and tracking) [24]. Particle trac-
users and potential platforms. To address these challenges, ing is a fundamental method for the visualization of flow
this study extends the illustrative eddy visualizations in our field data and is usually computational complex and inten-
previous work [7] to an integrated framework for the visu- sive. The feature-based method (e.g., [19] [20]) exploits the
alization and analysis of three-dimensional ocean eddies in data coherence between consecutive timesteps and focuses
web browsers. Combining eddy extraction, tracking, query, on just those regions of interest. The resulting significant
and interactive visualizations in one workflow, the proposed reduction in the storage requirement of data and rendering
framework is designed to facilitate the visual exploration and cost of visualization tasks makes the feature-based method
analysis of ocean eddies using the efficiency of feature-based highly suitable for WebGL-based visualizations. Thus, this
algorithms, a customized visualization design, the ubiquity of approach is followed in this work.
web browsers, and the capabilities of Web Graphics Library Understanding the motion of ocean eddies and visualizing
(WebGL) as well as GPUs and thus to help domain experts the eddies along with their motions in three-dimensions is
get some insight into the process of ocean circulation. of increasing interest to the oceanographic, environmental,
The main contributions of this work include: (1) Investi- and meteorological communities. Zhu and Moorhead [26]
gating ocean eddies and their motions as independent enti- tracked eddy instances over time by calculating their corre-
ties, which highlights their internal properties and individual lation probabilities between successive timesteps and placed
movements. (2) Representing the process of eddy migration the reconstructed three-dimensional structures of eddies in
as a series of temporal-spatial patterns called eddy paths and an ocean bathymetry of the Pacific Ocean off the coast of
efficiently querying eddy paths from tracking eddies over Japan. Studies have also visualized the structure of ocean
their life cycles. (3) Proposing web-compatible eddy meta- eddies. Grant et al. [5] visualized the vertical structure and
data to reduce the storage requirement of data and the ren- the life cycle of large anti-cyclonic eddies associated with
dering cost of visualization tasks. (4) Developing a WebGL- the loop current by showing the raising and sinking of the
based viewer to present the three-dimensional visualizations thermocline associated with eddy genesis and decay. Petersen
of eddies within any standard web browsers. The viewer has et al. [12] tracked eddies from a high-resolution simulation
the benefit of not requiring a client-side installation, offering of the global ocean and presented the shape of eddies as iso-
centralized maintenance, and ensuring a common environ- surfaces of the Okubo-Weiss parameter. Three-dimensional
ment for all users, which promotes collaborative research and ray casting techniques and specialized colormaps were used
customized extensions. to visualize the eddy structure and depth [15]. Our previous
The remainder of the paper is organized as follows: Sec- work in [7] combines the hand-drawn figurative depiction
tion II reviews the related work and Section III introduces of mesoscale ocean eddies with feature-based visualization
2 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
techniques to automatically produce three-dimensional illus- time-dependent feature-based algorithm, four visualizations
trative visualizations from a high-resolution regional ocean were designed to help users explore eddies from an ocean
model. This study extends some of the techniques in [7] to data in three-dimensions. These visualizations are:
accommodate presenting visualizations in web browsers. Daily eddy instance view which places all eddy instances
Data visualization software such as ParaView, VisIt, and of a day in the bounding box of the data frame to give users
EnSight can extend their applications to time-varying volume some basic idea of the relative positions, sizes, shapes, and
data. These programs display the sequence of timesteps as distributions of eddies before locating an individual eddy
animation and support temporal navigation in the form of from the visualization for more details.
linear time slider and VCR-like controls (i.e., play, stop, step Individual eddy path view which illustrates the motion of
forward, and step backward) [24]. Three different animation an individual eddy in its life cycle as a series of consecutive
modes are usually provided: a sequence of defined or se- timesteps in space (i.e., an eddy path) with a directional line
lected frames (e.g., the keyframe animation in EnSight), an above the eddy path to convey the direction of eddy motion
animation of a specific duration (e.g., the sequence mode and to provide a sense of continuity. Users can adjust the dis-
in ParaView), and a sequence of all timesteps. Multiple tance between the consecutive timesteps, isolate a timestep
operations (e.g., filtering, cutting, changing color schemes) from the eddy path, and compare different timesteps.
can be applied to individual timesteps. However, these vi- Eddy skeleton overview where eddy instances in multiple
sualization tools do not consider the temporal correlations days or timesteps of multiple eddy paths are represented as
between consecutive data frames and are not specifically simple glyphs (cylinders) and are placed in the context of a
designed for exploring ocean eddies. They usually fail to geographic map (e.g., ocean basin and coastline). This visu-
satisfy some very basic tasks in studying ocean eddies such alization permits observation of a large portion of the dataset
as observing the interactions between different eddies and and intends to provide an efficient and succinct overview of
the variations of multiple physical properties along with the vertical extent and spatial (horizontal) distribution of the
the motion of eddies. Additionally, existing visualization visualized eddies with eddy vertical transfer highlighted.
software all require installation and have specifications on 3D eddy path overview which presents an “overview +
the software environment such as the versions of OpenGL. detailed” view where the loaded eddy paths are presented
An installation-free and dedicated eddy visualization tool is in the context of a geographic map (the same as that of
needed by the domain experts, which motivates this study. eddy skeleton overview) and allows users to isolate each
individual eddy path from the visualization for more details.
III. DESIGN AND APPROACH In the visualization, timesteps of eddy paths are presented
The overarching goal of this study is to present a frame- as isosurfaces of eddy features to highlight the interactions
work for the exploratory visualization of ocean eddies in between eddy paths, add a sense of continuity, and bring a
web browsers. An integrated framework that combines eddy more real and detailed overview of eddy motions to users.
extraction, tracking, and visualization in one workflow can Some of these visualizations have been published in our
provide researchers with a one-stop solution to explore eddies previous work [7]. Table 1 lists the accompanying example
in 3D from ocean data so that they can focus on more figures of these visualizations in this paper.
advanced questions in research.
TABLE 1: Example Figures of The Visualizations
A. VISUALIZATION DESIGN Visualization Name Example Figures
The potential audience for the eddy visualizations include
scientists and students in oceanography, meteorology, and Daily Eddy Instance View Figure 8
environmental sciences. As part of our design process, we Individual Eddy Path View Figure 9
interviewed four domain experts (two from environmental Eddy Skeleton Overview Figure 10
sciences, two from oceanography) and five graduate students
(two from environmental sciences, three from oceanography) 3D Eddy Path Overview Figure 11
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 B. APPROACH OVERVIEW
want to observe the existing ocean eddies in a given day An overview of the proposed eddy visualization framework
and further to see a big picture of eddies along with their is illustrated in Figure 1. The framework is divided into
motions in a geographic context within a period of time. two major parts: eddy computation and eddy visualization.
The scientists particularly want to track ocean eddies over In the eddy computation part, time-dependent feature-based
their life cycles to understand the dynamics of eddies in time algorithms are chosen to generate web-compatible metadata
and space, while the students were especially interested in for efficient loading and rendering of a large amount of data
exploring the multiple geometric (e.g., volume) and physical in web browsers and to enable a structural understanding
properties (e.g., temperature) of individual eddy instances. of ocean eddies and their motions. The workflow of the
Based on the requirement analysis and the characteristics of computation starts from reading and preprocessing ocean
VOLUME 4, 2016 3
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
WebGL API
General & I/O Control
Daily Eddy Instance View
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 rameter [11] [22]. Generally, the OW parameter describes the
features (i.e., regions of interest) from every data frame, relative dominance of deformation with respect to rotation
tracking the extracted eddy features over time, and isolating of the flow. It divides the ocean into regions dominated by
eddy paths from the tracking history and the metadata of vorticity, regions dominated by strain or deformation, and a
extraction. Important physical and geometric properties are background where neither effect is dominant. Regions domi-
computed in the process. In the visualization part, the four nated by vorticity are potentially eddies. Mathematically, the
designed visualizations are implemented by combining the OW is defined as:
WebGL techniques with standard visualization techniques
such as surface rendering (i.e., rendering the isovalue surface OW = Sn2 + Ss2 − ω 2 (1)
of a volumetric object) so that users can explore the metadata
where Sn and Ss are the normal and shear components of
of eddy computation at different levels from a single eddy
strain respectively, and w is the relative vorticity of the flow.
instance to multiple eddy paths in a web browser. Presenting
They are different combinations of partial differentiations of
visualizations inside a standard web browser waives the
the horizontal velocities (u and v) in the x and y directions.
requirement of a client-side installation, ensures a common
environment for all users by centralized maintenance, and ∂u ∂v
facilitates the easy sharing of the tools and the interactive Sn = − (2)
∂x ∂y
parametrization of visualizations between research collabo-
rators at different locations on diverse platforms. ∂v ∂u
Ss = + (3)
∂x ∂y
IV. FEATURE-BASED EDDY COMPUTATION
The eddy computation algorithm described in this section ∂v ∂u
ω= − (4)
is based on the time-dependent feature-based visualization ∂x ∂y
techniques. These techniques can significantly reduce the Eddies can be identified where the calculated OW values
amount of data that needs to be visualized by focusing on just exceed a given threshold. Given a positive threshold value,
those regions of interest. The process consists of identifying the OW parameter allows partitioning of the ocean data into
interesting regions or coherent structures of a dataset (i.e., strain dominated and vorticity dominated domains.
eddies in this work), automatically extracting these features To extract vorticity dominated features, a region-growing
from each timestep using a variant of segmentation, and algorithm [16] [20] with a threshold of 0.2σOW is applied
matching extracted features from a timestep to the next using to the OW parameter field in every data frame. Note that
volume overlap, pattern matching or feature-vector matching. the threshold value changes in each data frame as the OW
parameter depends on time. Connected components of points
A. EDDY IDENTIFICATION AND EXTRACTION with OW values below the threshold are extracted from
In identifying ocean eddies, we choose one of the ocean the data as vorticity-dominated features which are potential
community’s canonical methods: the Okubo-Weiss (OW) pa- eddies. Attributes of each extracted eddy feature, such as
4 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
Sea Surface
10 m
II I
III IV
Depth Volume
…
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. FIGURE 4: Sequences of timesteps from three example eddy
Whenever a “death” event is found, the algorithm will iter- paths are juxtaposed in three rows. An eddy may go through
five states in its life cycle: birth, continuation, merge, split, and
atively trace this eddy back to the previous timesteps until
death.
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 output files of the program are listed in Table 2 with
the evolution process. In the computation, we set a minimum brief descriptions. The isosurfaces of eddy features in a day
lifetime cutoff on the duration of eddy paths, usually to 30 or in an eddy path, which are saved as *.day files and *.path
days (but this can be changed by users). files respectively, are the metadata for visualization. In a
Step 2 - Extracts every timestep of each eddy path from feature isosurface file (Figure 5b), each feature entry starts
the metadata of eddy extraction and save the eddy paths as with a special line “155 155 155” and is followed by the
independent files based on the created list. geometric properties and the ranges of the variables. The
Step 3 - Compute the duration (the number of days) and feature is then written as a sequence of vertices and triangles:
migration distance of each eddy path. The migration distance a “vertex line” contains the coordinates of this vertex and the
of an eddy path is computed by accumulating the Euclidian interpolated values of non-primary variables (temperature,
distance of center points (Cx, Cy, 0) on the ocean surface salinity, and velocity magnitude in this implementation) at
between each pair of timesteps: this vertex; a “triangle line” saves the vertex indices of a
triangle. This file format is believed to be more efficient and
n p
X thus more web-compatible than the widely used OBJ file
MD = (Cxi+1 − Cxi )2 + (Cyi+1 − Cyi )2 (6) format which uses the first character of each line to specify
k=1 the type of command (i.e., “v” for vertices, “f” for faces of a
where MD indicates the migration distance and n is the triangle).
number of timesteps (days) in the eddy path.
V. DATA AND COMPUTATION PERFORMANCE
D. PROGRAM DESIGN AND ARCHITECTURE A. DATA
A standalone C/C++ program with the support of VTK The ocean data used for this study is from a multi-decadal
libraries (https://www.vtk.org/) was developed to implement regional ocean model simulation [6]. The simulation was per-
the feature-based eddy computation algorithm introduced formed using the Regional Ocean Modeling System (ROMS)
previously. The program consists of four modules to handle [18], which solves the incompressible, hydrostatic, primitive
the general control and the three major steps of feature-based equations with a nonlinear free surface. It employs orthogo-
eddy computation. Figure 5 shows the architecture of the pro- nal curvilinear coordinates in the horizontal and a generalized
gram, the pipeline of implementation, and the file format of topography following coordinate in the vertical [6]. The
the metadata for web-based visualization. As shown in Figure simulation domain covers the major path of the Gulf Stream
5a, the naming protocol for the source code files follows through the northwest Atlantic (Figure 6).
meaningful and descriptive schemes. To run the program, In this study, the implementation covers the most recent
users need to specify some configuration information such three years (2005 - 2007) in the simulation with 1095 daily
as the locations of the input and output files, the number of NetCDF [13] files (approximately 1.67 GB per file). The
timesteps, variable names, threshold, and the geographical vertical coordinate and stretching functions follow the ROMS
scope of the data. conventions. For details about this particular simulation,
6 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
(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).
40
*.day frame, these files are used for eddy path isolation
and surface rendering in eddy visualization.
35
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.
(a) (a)
(b) (b)
FIGURE 8: Exploring eddy instances in a day. (a) The daily FIGURE 9: Exploring timesteps of an eddy path. (a) The
eddy instance view presents all of the eddy instances in this day individual eddy path view presents all of the timesteps of an
(data frame) in the bounding box of the data frame. Two colors eddy path as isosurfaces which are color-coded by veloc-
are used to differentiate cyclonic (green) and anti-cyclonic ity magnitude. The arrow above highlights the continuation
(purple) eddies. An eddy (the 24th in this data frame) is and direction of the eddy motion. (b) The first and the last
selected from the visualization and presented independently timesteps of the eddy path are color-coded by temperature,
in (b). The eddy is color-coded by the temperature variation salinity, and velocity magnitude and are represented as wire-
on its surface. A color bar on the top of the webpage shows frames in a coordinated view for comparison. The colorbars
the range of the temperature variations. 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
after the depth went beyond some level which is defined as miniature of the eddy path with the two chosen timesteps
highlighted in solid red color.
the isothermal curve. Finally, the student switched back to
the initial mode of the visualization and continued to pick
another eddy feature for analysis. by successive timesteps, the scientist adjusted the distance
between each two neighboring timesteps to a factor of ten
B. EXPLORING TIMESTEPS OF AN EDDY PATH relative to the actual distance. He then color-coded the visu-
In the second scenario, scientist B wanted to study the general alized features by different physical properties and rotated the
process of eddy migration by focusing on the motion of an visualization to observe the eddy path from different angles.
individual eddy over its life cycle. In the webpage “Individual The individual eddy path view in Figure 9a illustrates the
Eddy Path View” of EddyViz, he selected a *.path file from changes of velocity magnitude on the surface of this eddy
the data folder and loaded the file into the webpage. The along with its motion. From the visualization, he noticed that
information in the control panel indicated that this eddy had the eddy seemed to go through an intensification process:
a counter-clockwise orientation and lasted 44 days from May the features became more colorful and bigger in size and the
9 to June 22 in the year 2005. All timesteps that this eddy spaces between the consecutive timesteps were increasing as
went through in its life cycle were listed on the right-hand the eddy evolved.
side of the webpage and were presented as an individual eddy To summarize the structural and temporal changes of this
path view without any distracting context. To avoid occlusion eddy in its life cycle, the scientist compared the first and the
VOLUME 4, 2016 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 almost equal numbered. Also, most eddies extended less than
two chosen timesteps were both color-coded by temperature, 2000 meters below the sea surface (the deepest point of the
salinity, and velocity magnitude and were represented as ocean basin is 5500 meters). Using a scrollbar in the control
wireframes to highlight the topological structures (Figure panel, she slowly filtered the eddy cylinders (skeletons) in
9b). As shown in Figure 9b, the three representations of the the visualization by the depths (penetration into the ocean) of
first timestep are listed on the first row while those of the last eddies from 300 meters to 2000 meters to oberve the changes
timestep are listed on the second row. The color bars on the in the distributions and numbers of the visualized eddies. She
top of the visualization show the ranges of variations in these found that the cyclonic eddies were gradually outnumbered
physical properties; the two red brackets below the color bars by the anti-cyclonic eddies when the depth threshold went
indicate the ranges for the two selected timesteps. From the beyond 900 meters. In Figure 10, the eddy skeleton overview
visualization and the geometric information in the control presents all of the eddy instances that have depths penetrated
panel, the scientist found that the eddy became bigger, higher, into the ocean at least 1727 meters in the year 2007.
hotter, saltier, and faster at the end of its life cycle compared
with its initial stage. 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
C. SUMMARIZING EDDIES IN A PERIOD OF TIME of eddies were missing from the visualization. She decided to
In the third scenario, scientist C was doing a longitudinal use another visualization, which is 3D eddy path overview,
study of eddy evolution in the North Atlantic. Considering to summarize the motion of eddies in the year 2007. All
the large portion of data that would be involved, she believed of the eddy paths in 2007 were visualized as a 3D eddy
that studying eddies day by day or focusing on individual path overview as shown in Figure 11. From the figure, she
eddies was no longer effective. Instead, she would like to see found that the cyclonic eddies (green) were more likely to
a big picture of eddies in a long time period. In EddyViz, move westward while anti-cyclonic eddies (purple) tended to
she opened up the webpage “Eddy Skeleton Overview”, then move to different directions. She then isolated a cyclonic and
selected and loaded all of the *.day files of the year 2007 an anti-cyclonic eddy paths from the overview visualization
through the control panel in the webpage. All of the eddy successively and presented them in a separate view (as shown
instances saved in the loaded files were visualized as an on the top of Figure 11) to get more details of these two
eddy skeleton overview. From the visualization, the scientist eddies. Some basic information (i.e., date, duration, and
noticed that the cyclonic eddies (yellow) were more likely to distance) of the selected eddy paths were displayed in the
appear in the south while the anti-cyclonic eddies (purple) control panel. Finally, she requested a statistic summary of
were mostly gathered in the north. Both types of eddies were all of the eddy paths in the year 2007.
10 VOLUME 4, 2016
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
1
2
2007/06/12 – 2007/08/25
2007/07/26 – 2007/12/25
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 statement and to suggest new functionalities for the viewer.
The effectiveness of the visualizations and the utility of the
viewer EddyViz was evaluated by a survey study to determine B. THE RESULTS
how well the method proposed in this paper satisfies the Figure 12 summarizes all of the participants’ ratings on the
requirements of the potential users. ten survey statements (blue-colored bars for the statements of
visualization and green-colored bars for the statements of the
A. THE PROCEDURE utility of EddyViz). Statistically, the visualization aspects re-
The study involved 22 graduate students and postdoctoral ceived ratings of over 4 (good) on four statements and over 3
fellows (12 female, 10 male) from the Department of Marine (fair) on the remaining item. The free-form written responses
and Coast Science and the Department of Environmental and the oral comments also tended towards positive attitudes.
Sciences at Rutgers University. They all had research or study
(a) Most participants (20 out of 22) thought the visualizations
experience with ocean eddies before. After a brief intro-
were well-designed and effective in general, which received
duction to the project and a demonstration on the EddyViz
an average score of 4.45 with a standard deviation σ of ±0.67
(approximately 10 minutes in total), the study participants
in the statement S1. The students were amazed to see and
were asked to complete three basic tasks (similar to the
to interact with the eddy visualizations. Many students com-
user scenarios in the previous section) to make sure they
mented on the visualizations as “innovative” and “provide a
used the viewer sufficiently. Next, the participants used the
close view to ocean eddies and their movements”.
viewer to explore the data based on their individual interests.
Finally, they were invited to fill out a survey assessing the (b) The statements S2, S3, and S5 received ratings over 4
visualizations and their user experiences. The survey con- (good). For S2, 15 participants agreed that the daily eddy
tains five statements about the visualization design and five instance view can facilitate the exploration of eddy instances
statements about the utility of EddyViz (see Table 4). Each in a day and the statement received an average score of 4 (σ
statement in the survey uses a five-point Likert scale (1 = = ±0.81). For S3, 19 participants believed that the individual
very poor, 2 = poor, 3 = fair, 4 = good, 5 = excellent). We eddy path view can help them get a better understanding of
strongly encouraged the participants to add comments to each the motion of an eddy (the average score is 4.26, with σ =
VOLUME 4, 2016 11
Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS
S1 S2 S3 S4 S5 S6 S7 S8 S9 S10
13
12 12 12
11 11
10
9
8 8 8 8
7 7 7
5 6 6 5 5
5 5 5
4 4 4 4 4
3 3
2 2 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 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 4.21 ± 0.54 4 ± 0.69 4.14 ± 0.71 3.55 ± 1.06 2.91 ± 1.23 3.41 ± 0.96
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 and industry) and is fully readable even by novice users.
considered for the improvement of our method. This is also a preferred coding method by most scientists.
Secondly, for EddyViz, we decide to make the visualizations
IX. CONCLUSION AND FUTURE DEVELOPMENT which are currently implemented in independent webpages
The main motivation of this paper is to provide a one-stop more coordinated so that users can switch between different
solution from processing the ocean data to visual exploration visualizations. Thirdly, we have only applied the framework
of 3D ocean eddies and to make the ocean data as well as to Gulf Stream eddies from the particular ROMS model in
the visualization tool more accessible and interpretable to a this work. In consideration of the universality, we plan to
wider audience. For this reason, the focus has been on devel- apply the framework to other ocean eddies (e.g., mesoscale
oping a browser-based visualization tool that does not require eddies in the Northwestern Pacific Ocean). We also anticipate
any special software to be installed on the client side and improvements to the data reading and preprocessing parts
thus enables sharing the tools and the visualization results in expanding the applications. Finally, we intend to process
easily with remote collaborators working in different plat- data from the simulation environment or other powerful
forms or software environments. The proposed visualization centralized infrastructure and visualize the results in web
framework in this paper contains a time-dependent feature- browsers. This client-server architecture will leave the heavy
based algorithm implemented by a standalone C++ program computation tasks in the server side so that the client side
and a browser-based 3D eddy viewer named EddyViz. The can execute as few operations as possible to obtain a directly
visualization design was based on specific requirements from renderable representation.
potential users and also considered the constraints imposed
by web browsers. The feature-based algorithm generates the ACKNOWLEDGMENT
isovalue surfaces of eddies to reduce the rendering cost of We wish to thank the Department of Environmental Sciences
visualization and also supports a structural understanding and the Department of Marine and Coastal Sciences at Rut-
of ocean eddies and their motions. EddyViz presents 3D gers University for providing the ocean simulation data. We
eddy visualizations within a standard web browser using the particularly thank Dr. Enrique Curchitser and Dr. Kang for
capabilities of WebGL as well as GPUs. In this paper, the their valuable suggestions as well as many useful feedbacks
application of the framework to a high-resolution regional along the way. We thank all students and researchers who
ocean simulation of the northwest Atlantic Ocean and the participated in the user studies. We also thank the anonymous
general usage scenarios of EddyViz are discussed. A user reviewers for the useful suggestions.
evaluation was conducted and the results confirm the success
of the visualization design and the potential of the proposed REFERENCES
framework in helping researchers understand ocean eddies. [1] Amores, A., Melnichenko, O. and Maximenko, N., 2017. Coherent
Although the proposed visualization framework provides mesoscale eddies in the North Atlantic subtropical gyre: 3-D structure and
transport with application to the salinity maximum. Journal of Geophysical
a considerable set of useful features and fulfills the initial Research: Oceans, 122(1), pp.23-41.
design goal, we have become aware of some of the limitations [2] Cabello, R., 2010. Three. js. URL: https://github. com/mrdoob/three.js.
in implementing the framework. For the future developments, [3] Castelao, R.M., 2014. Mesoscale eddies in the South Atlantic Bight and the
Gulf Stream recirculation region: vertical structure. Journal of Geophysical
four major aspects can be explored to advantage. First of Research: Oceans, 119(3), pp.2048-2065.
all, for the feature-based eddy computation program, we [4] Chaigneau, A., Le Texier, M., Eldin, G., Grados, C. and Pizarro, O., 2011.
plan to convert the C++ code to Matlab code and han- Vertical structure of mesoscale eddies in the eastern South Pacific Ocean:
dle the input parameters to the program through a Matlab A composite analysis from altimetry and Argo profiling floats. Journal of
Geophysical Research: Oceans, 116(C11).
graphic user interface (GUI). Matlab is a well-documented [5] Grant, J., Erlebacher, G. and O’Brien, J., 2002, October. Case study:
standard code scripting language (popular in both academia visualizing ocean flow vertical motions using Lagrangian-Eulerian time
VOLUME 4, 2016 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
VOLUME 4, 2016 15