Visualizing Gridded Datasets with Large Number of Missing Values

Post on 04-Jan-2016

29 views 3 download

description

Visualizing Gridded Datasets with Large Number of Missing Values. Suzana Djurcilov and Alex Pang University of California, Santa Cruz. OVERVIEW. Motivation NEXRAD Background Visualization Options Conclusions and Suggestions Future Directions. Motivation. - PowerPoint PPT Presentation

Transcript of Visualizing Gridded Datasets with Large Number of Missing Values

UCSC

Visualizing Gridded Datasets with Large Number of Missing

Values

Suzana Djurcilov and Alex Pang

University of California, Santa Cruz

UCSC

OVERVIEW• Motivation

• NEXRAD

• Background

• Visualization Options

• Conclusions and Suggestions

• Future Directions

3UCSC

Motivation

• Known visualization tools (e.g. VTK) often assume full grid

• Filling grids with arbitrary values causes incorrect visualizations

4UCSC

Background

• NEXRAD (WSR-88D) is a 3D radar

• Output is a conical grid with usually no more than 4% filled

• Standard viz methods are 2D

5UCSC

NEXRAD

6UCSC

Incorrect contours when using arbitrary values

-99.99 99.995 5

31 13

Threshold = 2.0

7UCSC

What can be done ?

P o in tC lo ud

D e lau n ayT ria n gu la tion

S u rfa ceR e con s truc tion

P o lygo n ize

S ca tte red In te rp o la teto fu ll g rid

V o lu m eR e n de ring

M o d if iedG ra d ie n t

M o d if iedS u rfa ce

S m o o th ed

Iso surfa ce C u tt ingP la n es

G ridd ed

V isu a liza tionO p tio ns

8UCSC

Point Cloud

• Draw a point or sphere at point location

• Advantage: quick and simple

• Disadvantage: cluttering, poor depth perception

9UCSC

Point Cloud

10UCSC

Interpolation

• Very useful for evenly distributed data

• Many choices: Shepard’s, Multiquadrics, Krigging etc.

• Need to be careful to preserve desired properties in the data

11UCSC

Interpolation methods

Method Troubles Good forShepard’s Many artifacts Simple tasks

Multi-quadrics Out-of-range values Small datasets

Thin-platesplines

Expensive Low-variabilitydatasets

Krigging User-specifiedvarigram

High-variabilitydatasets

12UCSC

Interpolation - Distribution types

Clustered Uniform

13UCSC

Interpolation - artifacts

Stack-of-pancakes artifact from Shepard’s

14UCSC

Delaunay

• Take a subset around a certain treshold

• Connect the points using Delaunay triangulation

• Advantage: widely available

• Disadvantage: connected regions, convex shapes

15UCSC

Delaunay

16UCSC

Surface reconstruction

• Hoppe et al. 1992 - treat the subset as unorganized points

• Recreate the surface using tangent-planes incident to the mesh points

• Advantage: plausible surface from a subset

• Disadvantage: choppy edges

17UCSC

Surface reconstruction

18UCSC

Modified Normals

• Take an average of neighboring normals

• Use only available data

10111111

ijkkijjkiijkkijjki

ijk

VVVVVVV

19UCSC

Modified Normals

before after

20UCSC

Modified Isosurface

• Take an average of neighboring gradients• Move surface vertices in direction of the gradient• Takes out very sharp features

21UCSC

Modified Isosurface

before after

22UCSC

Smoothed Isosurface

• Taubin 1995 - Gaussian smoothing of vertex points

• Alternative inward and outward steps

• Advantage: takes out sharp edges

• Disadvantage: possibility of excessive smoothing

23UCSC

Smoothed Isosurface

24UCSC

Conclusions

• Sparse gridded datasets can be handled as gridded or scattered

• Standard methods need adjustments for missing values

• We present two options for improving isosurfaces

25UCSC

Suggestions

• For very sparse data use scattered methods

• Interpolation best for uniform distribution

• Clustered data better treated raw

• With high-frequency data post-process isosurfaces with smoothing

26UCSC

Future Work

• Expand into other physical sciences

• Experiment with vector algorithms

• Apply a variety of gradient filters

27UCSC

Acknowledgements

• Wendell Nuss, NPS, Monterey

• ONR grant N00014-96-0949, NSF grant IRI-9423881, DARPA grant N66001-97-8900, NASA grant ncc2-5281

• Santa Cruz Laboratory for Visualization and Graphics (SLVG)

UCSC

http://www.cse.ucsc.edu/research/slvg/nexrad.html

Point Cloud Delaunay

Surface Reconstruction Smoothed Isosurface

29UCSC

Volume Visualization

Default transfer function Transfer function notincluding missing values