Visual Analysis of the Air Pollution Problem in Hong Kong Huamin - - PowerPoint PPT Presentation

Visual Analysis of the Air Pollution Problem in Hong Kong

Huamin Qu, Wing-Yi Chan, Anbang Xu, Kai-Lun Chung, Kai-Hon Lau, and Ping Guo The Hong Kong University of Science and Technology Beijing Normal University

IEEE Visualization 2007 October 30, 2007, Sacramento, CA, USA

Outline

• Introduction
• Background and Motivation
• Uniqueness of Weather Data
• Challenges
• Related Work
• Visualization Modules
• Experimental Results
• Conclusions and Future Work

Background: Hong Kong

Background: Air Pollution Problem in Hong Kong

• Hong Kong air quality is decreasing tremendously
• Air pollution problem becomes one of the biggest social issues
• Causes are still unknown
• Hypotheses are proposed without any formal proof

Motivation

• Institute for the Environment of the Hong Kong University of Science

and Technology (HKUST)

• Maintain a comprehensive environmental database on Hong Kong

and surrounding regions

• Found correlations with classical analysis methods
• Could not obtain convincing results for high-level correlations with

mathematical techniques

• Demanded visualization techniques for analysis

Uniqueness of Weather Data

• Time-series
• Recorded hourly
• Span >10 years
• Geographic information
• Inherit from monitoring station
• Multi-dimensional
• Have typically >10 attributes
• Important vector field
• Formed by wind speed and

wind direction

• 1. Precipitation
• 2. Wind Direction
• 3. Air Temperature
• 4. Wind Speed
• 5. Dew Point
• 6. Relative Humidity
• 7. Sea Level Pressure
• 8. Respirable Suspended Particulates (RSP)
• 9. Nitrogen Dioxide (NO2)
• 10. Sulphur Dioxide (SO2)
• 11. Ozone (O3)
• 12. Carbon monoxide (CO)
• 13. Solar Radiation
• 14. Air Pollution Index (API)
• 15. Contributed Pollutant to API

Weather data attributes

Challenges

• People are familiar with the existing tools to represent wind profile
• Polar coordinates and orientated arrows are commonly used
• This constraints the design of visualization tool
• Data are of large size and high dimensionality
• Data are both multivariate and time-series
• It should provide an intuitive way to compare the data across time

and stations

Outline

• Introduction
• Related Work
• System Overview
• Experimental Results
• Conclusions and Future Work

Related Work

• Weather data visualization is rarely considered as a standalone

problem

• Uniqueness of weather data is sometimes overlooked
• Vector values
• Time-series nature

Related Work (cont.)

• Scientific visualization approaches [Treinish’00]
• Show the weather condition rather than the underlying data for

analysis propose

• Nonphotorealistic brushes and natural textures [Healey‘04, Tang’06]
• Generate effective results but is scalable to only 3 attributes
• General multivariate applications [Luo‘03, Guo’06, Wilinson‘06]
• Do not addressed the important wind factor
• Cannot be directly used for air quality analysis

[Treinish’00] [Healey’04] [Guo’06]

Outline

• Introduction
• Related Work
• Visualization Modules
• Experimental Results
• Conclusions and Future Work

Visualization Modules

• Polar system with embedded circular pixel bars
• Tailored parallel coordinates with S-shape axis
• Weighted complete graph

Visualization Module 1: Polar System

• Has been heavily applied by

domain scientists

• Allows query by wind speed

and direction

• Uses area-preserving mapping
• n distances
• Points are not over-

compressed in the center

Not Preserved Area Preserved

Polar System with Embedded Circular Pixel Bar

• Users select a sector to plot the inside-sector data of certain wind

direction and speed

• A complement plot of outside-sector data is blended underneath to

compare against overall distribution

Advantages of Embedded Circular Plots

• The corresponding wind direction and speed is obvious for rapid

comparisons between sectors

• Overall patterns are preserved in circular plots
• Users may perform accurate numerical analysis with the supplement

traditional layout

Visualization Module 2: Parallel Coordinates with S-Shape Axis

• Vertical axis is not good at encoding directions
• S-shape axis is introduced
• Similar to polar system
• Stands out among all axes to attract attention

Traditional layout (not intuitive) Circular layout (many overlapping)

S-style layout

Tailored Parallel Coordinates

• Scatter-plots are provided for detailed analysis

Visualization Module 3: Weighted Complete Graph

• Each node represents one data dimension
• Distance between nodes encodes their correlations
• Calculated with standard correlation coefficient
• Rendered by the LinLog energy model with the Barnes-Hut

algorithm A B C

correlated not really correlated

Weighted Complete Graph: Encoding Scheme

• Size of a node encodes accumulated correlations between this node

and all other nodes

• Bigger nodes may have strong relationship with other nodes
• Edges are included if the correlation is

above a threshold to reinforce the encoding of correlation

• Width:

Absolute value of correlation coefficient

• Color:

Positive / negative correlations

Weighted Complete Graph for Ordering Parallel Axes

• Helps explore overall relationship among dimensions
• Guides ordering of axes for parallel coordinates

(random) (guided)

Outline

• Introduction
• Related Work
• Visualization Modules
• Experimental Results
• Correlation Detection
• Similarities and Differences
• Time-Series Trend
• Discussion
• Conclusions and Future Work

Experiment 1: Correlation Detection

• Sector of high RSP (Respirable Suspended Particulates) is selected
• RSP is correlated with SO2 and O3 but not solar radiation
• Known:

API (Air Pollution Index) is correlated with O3 (red pixels) but not SO2

• Unknown:

A blue cluster appears behind a green one for SO2 and O3 plots

solar radiation sulphur dioxide SO2

• zone O3

Color = API (Air Pollution Index)

Experiment 2: Similarities and Differences

• External pollution from the factories on the Pearl River Delta

(northwest of Hong Kong) is generally believed to be the main source

• Local pollution is often ignored
• Power plants
• Vehicles and vessels

HK

(northwest)

Pearl River Delta

Comparing 9 Districts in Hong Kong

• High SO2 for most stations:
• Strong wind
• Northwest wind
• External Sources
• High SO2 for Kwai Chung:
• All wind speed
• Southwest wind
• Internal sources likely due

to cargo ships at Kwai Tsing Container Terminals

9 stations of 3 years data Color = amount of SO2

Detailed Comparisons with Sectors

• Sectors with high SO2 are selected for further studies
• Kwai Chung data generally shows a higher API value than Tung

Chung data

• But SO2 should not contribute much to API
• Local pollution is dominating in Kwai Chung

Experiment 3: Time-Series Trend

• Weighted complete graphs for 3-year Tung Chung data are generated
• More correlated attributes are placed closer in the parallel

coordinates display

Time-Series Trend with Parallel Coordinates

• Polar system is applied for filtering
• A time axis is added, with colors also denoting the time
• 2004 and 2005 plots are more similar
• In 2006 plot, temperature varies dramatically

Discussion: Using 3 Modules As A Whole

• Use weighted complete graph

for spotting correlated attributes and deciding the order of axes in parallel coordinates

• Observe general correlation

with parallel coordinates

• Study specific relationships

among 3 attributes for a subset

• f data using polar system with

embedded pixel bars

Discussion: Feedback from Domain Scientists

• Very positive feedback received from the domain scientists
• Polar system with embedded pixel bar
• Offers easy navigation to explore the data interactively
• Tailored parallel coordinates
• Show general relationships in a qualitative way
• Use S-shape axis to encode directions intuitively
• Supply scatter-plots for accurate analysis
• Weighted complete graph
• Provides correlation overview that is useful for initiating analysis
• Aids axis order selection in parallel coordinates for clearer results

Outline

• Introduction
• Related Work
• Visualization Modules
• Experimental Results
• Conclusions and Future Work

Conclusions

• Comprehensive System
• The first attempt designed for air quality analysis
• Novel Techniques
• Polar system with circular pixel bars: scalar + vector
• Enhanced parallel coordinates: vector + time axes
• Weighted complete graph: parallel axis order selection
• Significant Application
• Analyzed the air pollution problem of Hong Kong
• Revealed known findings effectively
• Detected unknown patterns by domain scientists

Future Work

• Continue as a long-term project with the Institute for the Environment
• f HKUST
• Make the visualization system available to the public on the Web
• Incorporate new datasets for further exploration
• Add animations and other visual aids for more revealing results

Acknowledgments

• Dr. Zibin Yuan at the Institute for the Environment of

the Hong Kong University of Science and Technology

Thank You!

Weighted complete graph Enhanced parallel coordinates with S-shape vector axis Polar system with embedded circular pixel bars