This is a copy of the entrant's web pages. The original web site may be found at http://www.yeraze.com/neworleans.

IEEE Visualization 2005 Contest Entry


PDF: viz05contest.pdf (254Kb) (Link fixed, 4:45 Aug3)
Video: composite.avi (19.7Mb)
Images: Image Gallery
Instructions: Instructions for running the application

Visualization of the New Orleans Windfield Dataset

Authors:

Contest Entry Visualization System

The visualization system was designed by myself entirely using open source tools.. Developed for Windows Platforms using Bloodshed Software's Dev-C++ IDE and OpenGL, it uses a combination of novel and well established algorithms to enable a fast and accurate representation of the data on nearly any machine. It is capable of colormapped surface renderings, pathline generation, and isosurface extration of full resolution data at interactive framerates.

Task 1: Interactive Exploration

The application was run on my personal computer, an Athlon 3Ghz system with 1 Gig of RAM and an NVidia 5900 Graphics card, running Windows XP. The application loads in approximately 20 seconds to the following screen.

The starting window size is 1024x768 with full 24bit color (+8 bit transparency, which is unused). The window is resizable however, to any screen size. Visualization is done on the entire dataset, with no decimation or subsampling. On my system, performance remains around 23frames per second at almost all times, dropping only when using highly complex isosurfaces. Mostly for information to the judges, the framerate is shown in the lower left at all times. Three major visualization modes are supported:
Boundary Surface Visualization with Colormapped data Pathline Visualization of the Wind Flow from any user specified area Isosurface visualization of multiple fields at user specified values

Task 2: Static Presentation

Static presentation of this application consisted of simply using screencaptures of the application. By pressing the spacebar, the user can hide the User Interface windows showing just the visualization. Gimp for Windows (an open-source image editing suite similar to Photoshop) was used to both capture the images and edit them before saving. By raising my screen resolution to 1600x1200 and maximizing the application window, I was able to grab high-resolution images better suited to editing.

Task 3: Data Specific Tasks

#1: What is the structure of the wind field?

Dropping a large generator of 50 pathlines in the position shown, it's obviously that the structure of the windfield is largely in the one indicated direction. Wind seems to large continue in a straight line until obstructed by a building. There are however, a few exceptions of isolated vortices forming at high altitudes without any obstructions.
The windflow across the city Closer view of the windflow,
showing a free-standing vortex
Closer view of the windflow,
showing interactions with the geometry

#2: From a source at various locations in New Orleans (such as the Riverwalk mall), what would the time-evolution of the pollutant be?

By placing pathline sources at various points, you can directly see how a pollutant would flow through the city. Various locations were used, and some interesting discoveries were made.
  1. For the most far-reaching ground-based release, release in a wind shielded area allows the pollutant to rise to a higher altitude and spread further before being caught by high winds.
  2. In most areas, the pollutant does not spread very far along the ground due to obstructions.
  3. A release at the corner of a building near a highspeed wind flow results in alot of initial dispersal, followed usually by very long travel times. The vortices at the corners of a building spread the pollutant out, allowing for larger dispersal. Also, the corners of buildings are opportunities for the pollutant to gain altitude as it follows the windflows in upward spiraling vortices.
Release between 2 small buildings creating a wind-shielded Alley.
Here you can see how the pollutant rises over the buildings to be caught in the high speed winds and transported long distances.
Release at the corner of a building.
Here you can see the large initial dispersion of the pollutant due to the vortex at the corner of the building. The pollutant is further dispersed by the sharp corner of the obstructing building, and travels far, even exiting the city.
A close-up of a pollutant released near the edge of a building
Here you can see how the pollutant uses the wind vortices and currents near building corners to gain altitude and travel further.

#3: Are their any vortices or large wind-shielded areas in the data?

Vortices abound in this dataset along the edges of buildings. These seem to be formed as the wind whips around the corners of the building, leaving low pressure areas that form vortices. However, there are a few other vortices separate from the boundary geometry, such as the ones shown in the pictures below. These seem to have no cause, but are significantly larger and more powerful than the ones near the buildings.

Wind shielded areas can also be considered areas of low wind speed or momentum. These can simply be found using an isosurface of a low momentum magnitude, as shown below. The wind shielded areas aren't found until you look near the area of the winds exiting the city. This is possibly due to a simulation error, or possibly because the portion of the city near the wind "entry" point is the first contact point of the wind. It may require alot of obstructions before the wind is sufficiently slown down enough for any visible difference.
Release in a courtyard
Here you can see several large vortices that would quickly oversaturate an area with a pollutant.
Isosurface of Momentum Magnitude
Here you can see the wind shielding effect in the alleys between several buildings, as well as on a rooftop of a short building near a taller one.
Release in a wind shielded alcove
Here you can see a wind shielded alcove created by several buildings nearby, and the erratic windflow created in the center

Additional Comments

Unique Capabilties of this visualization system:
  1. Texture Mapped Colormap - By using a 1-D texture for the colormap instead of the usual per-vertex color approach, the colors are correctly interpolated between data points. This is covered in more depth in the accompanying PDF paper. Example shown below.

    The top of the superdome, colored by Momentum Magnitude. At the joining point of the Wireframe & Solid fill portions you can see where the grid cells show both red and blue areas. Most visualizations would linearly interpolate between the two colors, but my system correctly inserts the green values that exist between them.
  2. 2nd Order Runge Kutta Pathlines - All pathline computations are done with 2nd order Runge Kutta interpolations to improve accuracy.
  3. Binary Space Blocks - the volumetric data was subdivided into Axis Aligned subblocks to aid in localization. This reduced the complexity of the pathline calculations significantly as localizing a point to it's containing tetrahedra went from a brute force comparison of all 20 million tetrahedra, down to a comparison of a maximum of 100,000 tetrahedra and 256 square bounding blocks.
  4. Cached Localization Information - Since pathlines move very slowly through the dataset, I was able to exploit basic caching algorithms to further improve the pathline performance. Every time a point is localized to a tetrahedra, the located Block & Tetrahedra are remembered and used as first guesses on the next search. This results in significant time savings.
  5. Generic Support - Due to the system design, this application should work on any 4-surface FieldFlow data. With very minor modifications, it could work with any number of surfaces.