Improving the Classification of Plant Functional Types Using Evidential Reasoning on RemotelySensed

1
Improving the Classification of Plant Functional
Types Using Evidential Reasoning on
Remotely-Sensed Imagery of Lower Michigan

• Greg Lowman
• Grand Valley State University
• GPY 495 Senior Thesis

2
Purpose of Study

• Improve the classification accuracy of existing
  remotely sensed plant functional types using
  multisource evidential reasoning for lower
  Michigan
• Eliminate gaps and missing data, creating a more
  accurate map of plant functional types that can
  be used in subsequent research involving climate
  change

3
Plant Functional Types (PFT)

• Groups of plant species that share similar
  functioning at the organismic level, similar
  responses to environmental factors, and/or
  similar effects on ecosystems
• Eleven different plant function types

4
MODIS

• MODIS stands for Moderate Resolution Imaging
  Spectroradiometer
• Gives almost complete two day coverage of land,
  ocean, and atmosphere
• The only remotely sensed satellite data set
  available for plant functional types
• Significant gaps and missing data in plant
  functional type data

5
Dempster-Shafer theory of evidence

• Narrows down a hypothesis set as more evidence is
  found
• A way to use mathematical probability to judge
  what the outcome may be
• Given two or more belief functions, Dempsters
  combination rule computes a new belief function
  that represents the impact of the combined
  evidence
• In this study multiple lines of data will be
  combined creating a new data set

6
Methods

• Creating an appropriate climate map for Lower
  Michigan using ArcGIS 9.3.

7
Methods

• Run nine different programs in Matlab that
  convert distances to probability for the nine
  sources.
• Each source image is 1200X1200 pixels which are
  subset into sixteen 300X300 pixels using ERDAS
  Imagine.

8
Methods

• In order to get results, the Dempster-Shafer
  program is run.
• The program combines all 9 sources together.
• The results are put into text files.
• Using ERDAS Imagine the 16 subset text files are
  converted into a tiff file and are mosaic
  together creating the final plant functional type
  map.

9
Results
10
Results
11
Limitations

• Mean value for each climate type was only a
  sample
• Urban and barren classifications
• Resolution of images used

12
Future Research

• After improvements on mean vector for the climate
  types, check other states and compare with MODIS
  data

13
Questions?