Historical classification of land cover at the Cedarburg Bog Jason Schroeder CES major – UWM 1 credit BIO SCI 699 with Dr. Erica Young
May 14, 2015
Historical classification of land cover at the Cedarburg Bog
Jason Schroeder
CES major – UWM
1 credit BIO SCI 699 with Dr. Erica Young
Introduction
Part of a study of long term ecological changes in Cedarburg Bog
Long term studies will measure land cover changes within Cedarburg Bog (in response to climate and land use changes)
Project looks at land cover classification during time points 1941 & 2000
Introduction continued
Land cover classification can reveal wildlife and plant corridors Provide avenues for animals and invasive species
to and from the Bog
Comparison of two time points can provide insight into ecological changes Anthropogenic alteration of landscape
Farm abandonment Human settlement
Project Objectives
Use historical aerial photographs to map land use changes in Cedarburg Bog and surroundings
Use GIS to classify land cover categories:1. Forest2. Agriculture3. Water4. Wetland
Identify major areas of land use changes and corridors surrounding Cedarburg Bog
Methods
Georeferenced 1941 aerial photos in ‘ERDAS Imagine’ software package
Classified 1941 land cover
Compare land use changes between 1941 - 2000
Georeferencing
Apply a coordinate system to raw photos Distorts raw image to fit projected coordinate
system Corrected for curvature of earth
Coordinate system allows for exact location of objects Allows for ‘mosaicking’ of aerial photos Classification, area determination, queries
Georeferencing
1941 – raw unreferenced photo 2000 – georeferenced photo
Ground Control Points
Create points in raw photo that mirror georeferenced photo (next slide)
20 – 40 points per photo
1941 – raw unreferenced photo
Ground Control Points
2000 – georeferenced photo
Points in this photo mirror points in raw photo
Raw photo now has coordinate system imbedded
1941 Raw Image No spatial
information
1941 GEOTIFF Coordinates imbedded
Mosaic process Merge georeferenced photos to make a
seamless map
Detail of Mosaic process: merging overlapping images
Overlap from two photos
Software generates a “cutline”
Looks for pixels from each image that are similar to create seamless merge
Seamless merge – finished mosaic
Final mosaic – 15 photos merged together
Automatic classification limitations with grayscale images
Grayscale colors limited (256 shades of gray) Land cover patterns with similar pixel shades
classified in same group Water & forest Agriculture & wetland
Water overrepresented Wetland underrepresented
‘Supervised’ classification Blue: water
Green: forestTan: agriculture
By comparing this classification to the next slide, it is apparent that water is overrepresented - dark forest pixels are mistaken for water pixels.
Going forward
Manual classification of 1941 image More time effective More accurate
Provide a base map from which to compare changes
Compare to DNR WISCLAND land cover map (1992) or National Land Cover Dataset NLCD (2001)
Animated GIF from 1941 - 2000
1941
2000Areas of major change:
Conclusions
Automatic classification of grayscale images difficult and results in inaccurate results
Manual classification may take less time, provide more accuracy
1941 images can be compared to published WISCLAND or NLCD maps
Comparison of two time points shows that agriculture has decreased, forest and human settlement increased
Future results will be statistical
Acknowledgements
Dr. Jason E. Mills and Dr. Erica B. Young
Department of Biological Sciences
University of Wisconsin-Milwaukee
References WISCLAND, http://dnr.wi.gov/maps/gis/datalandcover.html NLCD, http://www.mrlc.gov/mrlc2k_nlcd.asp