Ground truthing of aerial and space images

1
Ground truthing of aerial and space images

• B. J. Turner and T.A. Groen

2
What is ground truthing?

• Comparing results of classification of images
  into land cover categories with reality (truth)
• To assess accuracy of classification
• To quantify accuracy
• Quantification
• By an Error Matrix

3
Principles

• Choose a number of sample points on image
• How?
• How many?
• Locate same points on ground (or on ground
  truth)
• How?
• What is ground truth?
• Field visit
• Higher resolution imagery

4
Choosing the sample

• Desirable
• Random or stratified random
• As many as possible (at least 30)
• About equal numbers in each class
• Practical
• Without bias
• Representative
• Access problems
• Cost

5
What is ground truth?

• Field verification difficulties
• Locating same point Use GPS if possible
• Defining same class one based on aerial view,
  other on ground view
• Use existing data points, e.g., from NFI plots,
  if available
• Using higher resolution imagery
• E.g., if evaluating satellite-derived imagery,
  may use aerial photos
• Assumes classes can be accurately assessed on
  higher res imagery

6
The Error Matrix

• Reference (ground truth)

• Image
• classes

7
Types of Classification Accuracy

• Accuracy Types
• Overall accuracy
• User accuracy
• Producer accuracy
• Error Types
• Errors of commission
• Errors of omission

8
Combining classes to improve accuracy Revised
error matrix

• Reference (ground truth)

• Image
• classes

9
Combining classes to improve accuracy
Legend
Forest type A
Degraded forest type A
Forest type B
Degraded forest type B
Other
10
Combining classes to improve accuracy
Legend
Forest
Degraded forest
Other