PROSPECTIVE CHANGE DETECTION 2000 2004 IN KENTUCKY IMPERVIOUSNESS LEXINGTON, KY AND CANOPY CLOSURE P

1
PROSPECTIVE CHANGE DETECTION (2000 2004) IN
KENTUCKY IMPERVIOUSNESS (LEXINGTON, KY) AND
CANOPY CLOSURE (PULASKI COUNTY, KY)

• Demetrio P. Zourarakis(1)
• Michael Palmer(2)
• Andrew Brenner(3)
• Susan C. Lambert(4)
• (1) Ph.D., Remote Sensing/GIS Analyst
• Commonwealth Office of Technology (COT)
• Kentucky Division of Geographic Information
• (2) GIS/Remote Sensing Analyst, SANBORN
• (3) Ph.D., General Manager, SANBORN
• (4) Geographer, GISP, Principal Investigator
• KLS, KLC, KWMIP COT

Biloxi, MS 16-18 May 2005
2
Kentucky Landscape Snapshot Project(NASA-funded
1.8 M 2002-2005)
Prospective Change Detection

• Problems
• No comprehensive picture of the forest, urban
  and rural landscape
• No baseline from which to measure KY changing
  landscape
• No geographic information input for future land
  planning
• Difficulty in measuring rates of landscape change
• Few tools available for decision makers to use
  these data
• Little use of remote sensing technologies within
  Kentucky governments
• Objectives
• Establish a snapshot of the forest, urban and
  rural landscape
• Establish an accurate landscape baseline
• Provide input data for federal, state and local
  land planning
• Establish an operational change detection program
• Create tools and training for KY personnel to use
  the resource
• Promote use of remote sensing imagery and methods
  within Kentucky governments

3
KLS Prospective Change Detection

• Impervious Classification
• Lexington, Kentucky

4
Prospective Change Detection
2001 Landsat

• The impervious prospective change detection was
  completed with a 2001 Landsat image and a 2004
  SPOT image
• The two images were rectified to the 1995 USGS
  DOQQ mosaic using Autowarp to assure that the two
  images are lined up correctly

1995 USGS DOQQ Mosaic
2004 SPOT
5
Prospective Change Detection
Downtown Lexington
Lexington Airport

• Impervious training data was derived from Space
  Imagings IKONOS satellite using eCognition to
  create segments to create the binary
  classification
• The training images are 16-bit 4-band
  multispectral images. The downtown image is 4km
  x 4km and the airport image is 1km x 1km
• The training data must be edited to be correct
  for each date to avoid errors in classification

6
Prospective Change Detection
2001 Landsat Impervious Estimate

• Full scene canopy estimates were created using
  USGS provided CART software as well as
  Rulequests Cubist classifier
• The 2001 classification has an average error of
  10.9 and the 2004 classification has an average
  estimated error of 12.5
• The method used to create these classifications
  are consistent with the methods used in created
  the USGS NLCD 2001 classifications

2004 SPOT Impervious Estimate
7
Prospective Change Detection

• Assuming that there is little or no loss of
  impervious areas, the change classification was
  set to values of 0-100 to coincide with the
  values of change from the 2001 and 2004
  classifications
• The change classification was processed using
  eCognition to create segments and select areas of
  real change

8
Prospective Change Detection

• Based on the change image, segments were selected
  that correspond to real changes, since much of
  the change indicated on the change image is
  erroneous change due to sensor, season, and
  atmospheric differences

Change with eCognition Polygons
2001 Landsat with eCognition Polygons
Real Change Polygons
2004 Spot with Real Change Polygons
9
Prospective Change Detection
Landsat
SPOT
Change
10
Prospective Change Detection

• Full 2004 SPOT image with change mask

11
Prospective Change Detection

• Masked areas of change around the Lexington area

12
Prospective Change Detection

• Selected segments are then summarized by the
  means of the change values inside their
  boundaries
• Red colors indicate low positive areas of change
  and yellow to green colors indicate increasing
  values of positive impervious change

13
KLS Prospective Change Detection

• Canopy Classification
• Pulaski County, Kentucky

14
Prospective Change Detection
2001 Landsat Mosaic

• The canopy prospective change detection was
  tested on a 2001 Landsat mosaic and a 2004 SPOT
  image.
• Each of the images were rectified to a 1995 USGS
  DOQQ mosaic using Autowarp.
• This was done to ensure that each of the images
  were rectified to a consistent base map

2004 SPOT
1995 USGS DOQQ Mosaic
15
Prospective Change Detection

• Change detection training imagery used was from
  National Agriculture Imagery Program (NAIP),
  which are an 8-bit, true color aerial images
• Canopy classification was created using
  eCognition segments and unsupervised
  classifications created from Erdas Imagine
• Two NAIP scenes of 3km x 3km were classified to
  be used by both dates of imagery
• To ensure accurate training data for the early
  date, the hi-res classification must be edited to
  correct any areas that may have changed

2004 NAIP Imagery
2004 NAIP Imagery With Canopy Classification
16
Prospective Change Detection
2001 Landsat 30m Canopy Estimate Classification

• Full scene canopy estimates were created using
  USGS provided CART software as well as
  Rulequests Cubist classifier
• The classifications both have an average error of
  12.5
• The method used to create these classifications
  are consistent with the methods used in created
  the USGS NLCD 2001 classifications

2004 SPOT 30m Canopy Estimate Classification
17
Prospective Change Detection
Change Image

• The 2001 classification was then subtracted from
  the 2004 classification to find the percent
  change
• The next step was to input the change
  classification into eCognition and create
  segments

Change Image with Segments
18
Prospective Change Detection

• Based on the change image, segments were selected
  that correspond to real changes, since much of
  the change indicated on the change image is
  erroneous change due to sensor, season, and
  atmospheric differences

Change with eCognition Polygons
2004 Spot with eCognition Polygons
Real Change Polygons
2001 Landsat with Real Change Polygons
19
Prospective Change Detection

• More areas of change

Landsat
SPOT
Change
20
Prospective Change Detection

• 2004 SPOT with change mask

21
Prospective Change Detection

• Masked areas of change around the Williamsburg,
  KY area

22
Prospective Change Detection

• Selected segments are then summarized by the
  means of the change values inside their
  boundaries
• Green color indicates positive canopy change
  (re-growth) and orange and red colors indicate
  negative canopy change (cuts, disease, etc.)

23
Summary and Conclusions
Prospective Change Detection

• KLS deliverable met
• Methodology on the right track
• Land development patterns spur local
  governments
• interest in change detection tools for better
  governance
• Recent drastic changes in logging patterns in
  Kentucky
• justifies intensification of change detection
  work
• On-line change detection masks will be served
  out by the
• Kentucky Landscape Census (KLC) portal

24

• Questions or comments?
• Surf over to http//kls.ky.gov
• or
• Call/email us at
• Susan C. Lambert, P.I.
• susan.lambert_at_ky.gov
• 502-573-0342
• Demetrio P. Zourarakis, Technical Lead
• demetrio.zourarakis_at_ky.gov
• 502-573-1450 ext. 224