An aerial photographic analysis of landscape change in riparian areas in Kakadu National Park, Austr

1
An aerial photographic analysis of landscape
change in riparian areas in Kakadu National
Park, Australia

• An investigation of landscape scale changes in
  the World Heritage Kakadu National Park, Australia

Aaron Petty, Daniel Banfai, Caroline Lehmann,
Lynda Prior, David Bowman, Michael Douglas
2
Top End Geography
3
Vegetation of Kakadu

• Savanna (80)

4
Vegetation of Kakadu

• Rainforest

5
Vegetation of Kakadu
Floodplain sedgeland
Melaleuca forest
6
Climate

• Mean Annual Rainfall 1100 1500 mm pa

Data derived from McDonald and McAlpine (1991).
Floods and Droughts The Northern Climate in
Monsoonal Australia Landscape ecology and Man
in the Northern Lowlands. Haynes, et al. eds.
pp. 19-29.
7
Fire in Kakadu

• Rapid fuel load
• build-up

8
Fire characteristics

• Low relief ? intense fires can spread 10s kms

• Low intensity

9
1990s intensified effort to reintroduce
traditional burning- usually by means of burning
earlier
10
History of buffalo in tidal interface region

• Spread from Cobourg Peninsula and were well
  established along Alligator Rivers Region by
  1850.
• After 1950 massive expansion of population.
• Likely reached carrying capacity by 1970.
• 1972 1975 Campaign to eradicate buffalo in
  Woolwonga Reserve.
• 1984 1988 BTEC campaign.

11
Gindjala 1979
Gindjala 2003
12
Aerial photos used in project

• Four time layers

• 1950 150 000 B/W
• 1964 116 000 B/W
• 1984 125 000 Color
• 2004 125 000 Color

13
Why use aerial photos?

• Very cost effective for high resolution.
• Centimeter pixel resolution for large scale
  photos (gt 110 000).
• Photos in this project scanned at 1 pixel .5 m.
• Provide extensive time depth
• Earliest photos of Kakadu are from 1936.
• Earliest comprehensive photos are from 1950.

14
Overlay 10m x 10m grid
15
Grid is stationary in space
16
Why use manual classification?

• Time equal to auto-classification methods such as
  E-Cognition.
• Avoids many of the problems with aerial photos
• Differences in scale, color, and time of year
  between time layers.
• Differences in contrast, developing and time of
  day between frames.
• Distortion from film, lenses, etc.
• The brain is, and will likely always be, the best
  pattern recognition system available.

17
Classification by dots

• Grassland lt ¼ of square
• Open Forest ? ¼ and lt ½ of square
• Closed Forest ? ½ of square

Will attempt classification based on vegetation
types from field data
18
Shade presents a difficulty with classification
Use of stereo-pairs should help clarify shade, as
well as resolve shrub-layer.
19
1964 Cover
20
1984 Cover
21
Coincidence Table

•   

22
Spatial Modeling
23
Statistical Modeling of Dots

• Some independent variables
• Behavior of neighboring dots.
• History of dot.
• Fire history (satellite derived from 1980).
• Buffalo density (estimated).
• Rainfall and flow rates (guaging station data).
• Invariant ecological characteristics slope,
  aspect, distance from water.

24
Conclusion

• Consider Aerial Photos
• Readily available.
• Fantastic historical resource.
• Less expensive for high resolution images.
• Vegetation classification is well established.
• Stereo pairs allow three dimensional resolution.