The Community and Environment Spatial Analysis Center

1
The Community and Environment Spatial Analysis
Center
Washington Chapter of the Urban and Regional
Systems Association MEMBERS WORKSHOP 2001
SESSION 4
Geographic Information From Images Eugene
Martin eugene_at_commenspace.org
2
Motivating questions
What is the process required to make reliable
information with images?
What are the possibilities and limitations?
How can we structure our information needs to
take advantage of the opportunities available
with imagery?
3
Food for thought
What processes and analysis produced the data
that you rely on?
What data do you need and how will you get it?
Project management
Hiring and cooperating with consultants or
contractors
Responsibility as a data author
4
Measuring landscapes with imagery

• Inexpensive, considering the alternatives

• Inherently pixel based

5
The Fuzzy Pixel
Indeterminate boundaries
Rotated/stretched during geometric processing
Can be contaminated by bright neighbors and/or
shadows
Think of the pixel as a estimate and as
sample of the landscape that has biases
6
The Fuzzy Pixel
7
Sensor Design Constraints

• Pixel size depends on the design parameters of
  the sensor

• Should be considered a strength AND a weakness

8
Bands and wavelengths

• The information content of a pixel is measured
  reflectance (light).

• Multiple samples of the electro-magnetic spectrum
  are the bands of an image

9
DN Digital Number
10
What part of the landscape do you want to
measure?

• Definitions

• What is the size of a phenomenon relative to
  the pixel

• Translate measurement needs into pixel and
  wavelength characteristics

• Measure the image to develop information about
  the landscape

• Interpretation and analysis a balance what is
  measured and the information available in the raw
  image

11
The pixel can only belong to one class

• Single element separated from all others

• Multiple elements compete to be a pixel

• Supervised classification

Identify and select representative pixels for
specific classes in the image
Assign each pixel to a class based on similarity

• Un-Supervised classification

Clump similar pixels together
Identify the clumps as known landscape types
12
Getting inside the pixel
Sub-pixel analysis
All pixels are mixed
Fixed number of mixable elements
Mathematical analysis of wavelength values assess
percent composition of elements
Layer cake set of results for further analysis
13
Example 1 Washington Gap Analysis
14
Example 2 Mapping Forest Cover
15
Example 3 Landscape Layer-cake

• Major Hurdles
• Data of appropriate scale, extent and content not
  available
• Measurement of features linked to physical
  processes
• Sub-pixel information required

• Data Sources
• LANDSAT Image from July, 1999 (30 Meter pixels)

• Solutions
• Develop new application of Spectral Mixing
  Analysis (SMA)
• Quantify sub-pixel composition of impervious
  surfaces, green vegetation, forest canopy, dry
  grass and bare soil.
• Verification and validity testing
• Overlay basin delineations to quantify landscape
  processes

16
Layer 1 Impervious Surface
17
Layer 2 Open Vegetation
18
Layer 3 Canopy Vegetation
19
Layer 4 Dry Grass/Leaves
20
Verifying and Testing
21
More Verifying and Testing
22
New Trends in Imagery
On demand high resolution panchromatic (Space
Imaging Ikonos)
On demand medium to high resolution visible and
IR
Hyperspectral airborne sensors lots of bands and
high resolution
FAST medium resolution global imaging
Consistency with historic images for time
sequence analysis (ASTER)
23
Take home messages..
Information from images is created by PEOPLE
People make decisions about landscapes and
analysis
The pixel is a slippery thing to work with
Is it eye-candy or information?
The most important information in an image is in
the bands
Pixel size has advantages and limitations
Verification and testing let you know the
information is reliable
A little knowledge about the analysis behind the
information helps determine suitability for use
Ask tough questions about image analysis methods
24
206 749 0112www.commenspace.orgeugene_at_commensp
ace.org
Questions?