GEOMORPHIC CHANGE DETECTION April 30 th to May 1 st , 2014 Sponsored by: I. THRESHOLDING ALTERNATIVES IN GCD
GEOMORPHIC CHANGE DETECTION
April 30th to May 1st , 2014
Sponsored by:
I. THRESHOLDING ALTERNATIVES IN GCD
WORKSHOP PLAN – DAY 1… (AFTERNOON)
E. Essential Best Practices to Support GCD
F. Traditional Approaches to Change Detection
G. Change Detection in Raster Calculator vs. GCD
H. Getting to know GCD Software
I. Thresholding Alternatives
J. Recap & Preview of Day 2
1:00 to 1:45
A Plan.. Not a Contract
1:45 to 2:15
2:15 to 3:00
3:15 to 4:00
4:00 to 4:45
4:45 to 5:00
3:00 to 3:15 BREAK
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 6
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting
GCD 6 THRESHOLDING
• Simple defined minLoD
• Propagated Errors
• Probabilistic Confidence Interval
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 5
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting
APPLICATION OF A MINLoD
• You take original DoD, and remove all changes <= minLoD
• For example +/- 20 cm
VARYING minLoD THRESHOLDS
22zzz
newold DEMDEM
© Wheaton 2008
newoldDEMDEMf z,zz
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 5
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting
MINLoD USING ERROR PROPAGATION
• Distinguish those changes that are real from noise
• Use standard Error Propagation
• DEM Errors can vary temporally and spatially
z z DEM old
2
z DEM new
2
Elevation (Time 1)
Elevation (Time 2)
See •Brasington et al (2000): ESPL
•Lane et al (2003): ESPL
•Brasington et al (2003): Geomorphology
z 10 2 20
2 22.36
22.36 cm ≈ 8.8 in
e.g.
z DEM old
10cm
z DEM new
20cm
ERROR PROPAGATION GETS APPLIED SAME WAY AS MINLoD
• Does not matter whether the minLoD is specified, or calculated from error propagation
• Just on a cell-by-cell basis!
• In background a perror grid is produced
WHAT ARE TYPICAL ERRORS?
• LiDaR : +/- 12 to 25 cm• Aerial Photogrammetry : +/- 10 to 15 cm
• Total Station Surveys : +/- 2 to 10 cm
• GPS: : +/- 3 to 12 cm• Terrestrial Laser Scanning: +/-
0.5 to 4 cm
Ground-Based Surveys
Remotely Sensed or Aerial Surveys
SO WHAT WOULD PROPAGATED ERRORS BE?
• LiDaR : +/- 12 to 25 cm (17 to 36 cm
minLoD)• Aerial Photogrammetry : +/- 10 to 15
cm(14 to 22 cm minLoD)
• Total Station Surveys : +/- 2 to 10 cm (3 to 14 cm minLoD)
• GPS: : +/- 3 to 12 cm (4 to 17 cm minLoD)
• Terrestrial Laser Scanning: +/-0.5 to 4 cm (0.7 to 6 cm minLoD)
Ground-Based Surveys
Remotely Sensed or Aerial Surveys
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 5
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting
HOW COULD I REPRESENT AS PROBABILITY?
• Using inferential statistics, we’ll calculate a t-score
• σDoD is the characteristic uncertainty
– In this case σDoD =
minLoD
• Just the ratio of actual change to minLoD change
• Assuming two-tailed test, t is significant at:
– 68% confidence limit when t= 1
– 95% confidence limit when t=1.96
t zDEM new
zDEM old
DoD
PROBABILITY THAT CHANGE IS REAL
Even when minLoD is spatially constant, probability varies in space… why?
© Wheaton (2008)
SENSITVITY OF THRESHOLD?
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 5
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting
WHERE DOES THIS FIT?
• Choosing the Threshold Method
GCD 6 THRESHOLDING - TUTORIAL
• Simple defined minLoD
• Propagated Errors
• Probabilistic Confidence Interval
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 5
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting
CONCEPT: EXCLUSION OR SUBTRACTION?
• LoD raster must be applied independently to cut raster and fill raster to preserve signage.
• Is it better to exclude points whose LoD > DoD or subtract LoD from DoD?
Slide from Greg Pasternack(2012) – USBoR Lecture
COMPARISON OF EXCLUSION VS. SUBTRACTION
• Blue is deposition; grey is within 1’; others are erosion
• Subtraction is always a more conservative estimate…
Slide from Greg Pasternack(2012) – USBoR Lecture
Exclusion raster
Subtraction raster
Net Change33,456 yds3
Net Change17,555 yds3
Raw DoD
WEIGHTING BELOW THE THESHOLD
• Instead of subtraction or exclusion, values below the threshold can be weighted based on their probabilities…
• For example, if the probability that change is real is 80%, for a vertical change of + 10 cm, the value used could be 8 cm.
See Lane SN, Westaway RM and Hicks DM (2003). DOI: 10.1002/esp.483.
DETAIL PLAN – H.
H. THRESHOLDING ALTERNATIVES FOR GCD
1. Three Types in GCD 5
2. Recall minLoD
3. Error Propagation
4. Probabilistic Thresholding
5. Tutorial
6. Thresholding by Subtraction vs. Exclusion vs. Weighting