Estimating Error in an Analysis of Forest Fragmentation Change Using North American Landscape Charac

1
Estimating Error in an Analysis of
ForestFragmentation Change Using NorthAmerican
Landscape Characterization(NALC) Data

• Daniel G. Brown, Jiunn-Der Duh, and Scott A.
  Drzyzga
• ??? ???

2
Objectives

• 1. calculate the measurement error in estimating
  change in four different metrics of forest amount
  and fragmentation as a result of differences in
  the characteristics of MSS scenes.
• test the effects of landscape generalization
  through postprocessing, to remove small-scale
  variability, and scale on the error in
  fragmentation metrics.
• develop a statistical model to evaluate potential
  causes of error and to predict the amount of
  error in a change analysis.

3
Study Areas
4
Methods

• (1) classify each of the twelve images,
• (2) subdivide the overlap study areas into
  landscape partitions,
• (3) calculate several forest fragmentation
  metrics for each of the landscape partitions in
  each image,
• (4) evaluatethe differences between the landscape
  metrics for images of the same epoch.

5
Dates
6
Methods

• Image Classification
• Doing unsupervised classification to assign
  pixels to one of four classes forest, nonforest,
  water, and other.

7

• Classification Accuracies

8
Methods

• Landscape Partitioning
• subdividing the regional landscape into
  equally sized and shaped landscape partitions.

9
Methods

• Metric Calculation
• Four metrics of forest fragmentation were
  selected for the analysis.

10

• Landscape Patterns Metrics Used

11
Methods

• Characterizing Measurement Error
• measure of error is the difference between
  metric values calculated for the same landscape
  partition within the same epoch.

12

• Compare RMSE RE

13
Results

• Effect of Landscape Size
• It illustrates the effects of increasing the
  landscape partition size on the error
• calculated for each metric in each study area in
  each.

14
(No Transcript)
15

• Effects of Sieving and Filtering
• It illustrates the effects that sieving at
  three levels of polygon size and filtering at two
  window sizes had on the error in a change
  analysis.

16
(No Transcript)
17

• A Model of Error in Metric Differences

18
(No Transcript)
19
(No Transcript)
20
(No Transcript)