Title: A Fuzzy Accuracy Assessment of the MidAtlantic Gap Analysis Land Cover Map Using Airborne Videograph
1A Fuzzy Accuracy Assessment of the Mid-Atlantic
Gap Analysis Land Cover Map Using Airborne
Videography
Confusion, Fuzziness, and Random Thoughts
- D. Ann Rasberry and Alexa J. McKerrow
- 3 August 2002
2Middle-Atlantic GAP
3Example from Assateague/Chincoteague
- Division Vegetated
- Order Tree Dominated
- Class Closed Tree Canopy
- Subclass Evergreen
- Group Temperate or Subpolar Needle-leaved
- Subgroup Natural
- Formation Rounded Crowns
- Alliance Pinus taeda Forest
- Association Pinus taeda/Symplocos tinctoria
Myrica cerifera Vaccinum elliotti
4(No Transcript)
5(No Transcript)
6Video Flight Lines 1996-1997
7Ancillary Data
NLCD93 NWI Elevation, Slope, Aspect STATSGO State
Natural Heritage
Program
Inventories DOQQs Data collected coincident with
video interpretation key development
8Maryland, Delaware, New Jersey GAP Land Cover
Map 1993
9Error Assessment
- Conventional Confusion Matrix
- Conditional Probability Accuracies
- Fuzzy Set Assessment
10Video Classes
Map Class
11 12Conditional Probability
- Weighting the accuracies based on the proportion
of samples in a class and the proportion of the
area represented by a class. - We used this to account for non-proportional
sampling. - We attempted to get a minimum number of points
per class and this adjusts for variation in the
area represented by the assessment points.
13Conditional Probability
nii
(N) (A)
where ni sample size in class i ai area
represented in class i nii samples correctly
classified in class i N total sample in the
study area A total area in the study area
14(No Transcript)
15(No Transcript)
16Fuzzy Set Assessment
- Expands analysis of error, or uncertainty, in the
map - Provides information about the nature, frequency,
magnitude, and source of error in the thematic
map - Improves the understanding of the relationships
between classes in the map - Gopal and Woodcock, 1994
17Linguistic Scale
- Level 5 Absolutely correct
- The mapped class exactly matches that which is on
the ground, or that which has been assessed from
the video - Level 4 Incorrect, but very similar
- The mapped class does not match that which has
been assessed, but the two are ecologically very
similar, sharing the same dominant species, the
same hydrological regime, comparable substrates,
and like vegetation structure
18Linguistic Scale
- Level 3 Incorrect, but somewhat similar
- The mapped class does not match that which was
assessed, but the two are somewhat similar, in
that they share the same hydrological regime and
like vegetation structure - Level 2 Incorrect, but understandable
- The mapped class does not match that which has
been assessed, but the two share a similar
vegetation structure, creating a plausible
ecological explanation for the confusion - Level 1 Absolutely incorrect
- The mapped class does not match that which has
been assessed, and the confusion is ecologically
inexplicable
19(No Transcript)
20(No Transcript)
21(No Transcript)
22(No Transcript)
23Potential Sources of Error
- Time differential between imagery, video, and
field work - Positional errors
- Interpretation errors on videography
- A single interpreter throughout the process
- Error propagation from ancillary datasets
- Assessing a polygon using a site
- May not be a homogenous polygon
24Conclusions
- Using video works well for accuracy assessment
- Need to develop a good interpretation key with
ground photos - Make the key permanent
- Invest the time in training the interpreter
color, shape, size, texture, and juxtaposition - Not all projects will be able to fly such
extensive video - Improvements in technology should greatly enhance
this method and its utility to land cover mapping - We need to do some additional analysis to look at
class confusion - Dont think this is necessarily a video issue
- Learned ways to make better use of ancillary data
25(No Transcript)