1. Supervised Classification of Landsat TM Imagery of the Yolo Bypass
Diana Craft
American River College
Geography 342, Fall 2010

2. Project Goals
Perform supervised classification on Landsat TM imagery using Spatial Analyst in ArcGIS
Obtain images showing seasonal variations in the Yolo Bypass
Show that selected fields in the Yolo Bypass have different uses throughout the year, supporting agriculture, wildlife, and flood control

3. The Yolo Bypass 59,000 acre engineered flood bypass
Used for agriculture, wildlife habitat, flood protection
Diversion channel for Sacramento River; part of the Sacramento River Flood Control Project
Fed by Cache and Putah Creeks, Colusa Basin Drain, and Willow Slough
Crossed by Interstates 5 and 80 and railroads

4. Step 1: Acquire Images

5. Image Source USGS’s Global Visualization Viewer (GLOVIS)
Several years’ worth of monthly satellite images
Download issues: CTRL for Pop-ups; opening .tar.gz files (use Winzip, not 7-zip); some images have to be ordered (delay); cloudiness may limit choices

6. GLOVIS Images: Easy to view month-by-month; seasonal trends emerge
Screenshots from GLOVIS lined up to compare 2009 & 2010 month-to-month

7. Step 2: Preprocess Images

8. Composite Bands
Problem: Unzipped Landsat image is set of 6 single-band images that each display as B/W
Use ArcGIS Toolbox > Data Management > Raster > Raster Processing > Composite Bands
This creates new single multiple-band raster dataset from multiple single-band raster datasets

9. Clip & Symbolize
Problem: Landsat TM image is huge; want to zero in on area of interest
Use ArcGIS Toolbox > Data Management > Raster > Raster Processing > Clip
This clips the image to extent of given coordinates
Next, go to Symbolize in Layer Properties to assign image bands to color channels: R=4, G=5, B=2

10. Step 3: Select Training Sites

11. Attribute Table & Land Cover Types

12. Select Training Sites

13. Step 4: Perform Classification

14. Classification Use Spatial Analyst in ArcGIS Run Create Signatures
Run Dendrogram routine
Run Maximum Likelihood Classification

15. Evaluate Results

16. Step 5: Refine and Repeat

17. Refine & Re-evalulate

18. Make Changes to Attribute Table

19. Repeat Classification & Evaluate

20. Conclusion and Learning Outcomes

21. What went wrong: Supervised Classification
Using multiple images was too ambitious for the allotted time; zeroed in on one image only
Preprocessing took too much time (batch processing might have helped)
Image interpretation was difficult without actual field data (or experience)
Initial bad result: Flawed training site selection, or problem with Max Likelihood Classification?

22. Lessons Learned: Project Design
Diagram the whole process first
Do small sample tests (rather than run multiple images down a dead end)
Keep a log to document steps including those that don’t work (for future reference)
Allow enough time to work it out when things go awry
Refine and repeat as needed