minnesota land cover mapping projectwgl.asprs.org/wp-content/uploads/2015/02/wgl_2015_knight.pdf ·...
TRANSCRIPT
Minnesota Land Cover Mapping Project
Joe Knight, Marv Bauer, Lian Rampi, Keith Pelletier Remote Sensing and Geospatial Analysis Lab
Department of Forest Resources University of Minnesota
Thanks to Marv, Keith, Lian, and Leif Olmanson for contributing slides.
Outline
- Brief history of MN land cover mapping
- Challenges funding mapping work
- Overview of lidar and OBIA benefits
- Methods for 2013-14 project
- Project status, early results
MN Land Cover Mapping
Statewide (RSGAL):
– 1990, 2000
Metro only (RSGAL)
– 1986, 1991, 1998, 2002, 2007, 2011
Minneapolis, St. Paul, Woodbury only (RSGAL)
– 2009
National Land Cover Data (MRLC)
– 2001, 2006, 2011
2000 Land Cover Level 3
0 % Impervious 100
Legend
92% accuracy
2011 Land Cover Level 2
Upland
County
0 % Impervious 100
Woodbury 2000 Land Cover Level 3
Woodbury 2011 Land Cover Level 2
Urban Tree Cover Woodbury Land Cover
Funding Challenges
- Federal agencies generally not interested
- State agencies interested, but lack funds and not coordinated
- LCCMR: political, relevance, uncertain future
Methods for 2013-14 Project
- Using statewide lidar, derived products, and ancillary data
- Object-based classification
- Ecoregion-based
Point Cloud for the Minnesota State Capitol area
Optical image for the Minnesota State Capitol area
Digital Surface Model for the Minnesota State Capitol area
Digital Elevation Model (DEM) for the Minnesota State Capitol area
Return
LIDAR-Derived Products DEM
Maximum
Vegetation Height
Mean
Topographic Position Index (TPI)
Compound Topographic Index (CTI) Slope
d = ( z - z(min)) / (z(max) - z(min) )
Buildings
What do you see?
Predominant Land-use?
Predominant Land-use?
Shape
Height
Tone/Color
Texture
Size
Association
Location
Pattern
Shadow
Knowledge-based Workflow
Segmentation
Integration
Classification
Image Objects
Morphology
Worldview-2 image, Richfield, MN
Selected class objects shown in blue
Objects Have Characteristics
• Color: What are the spectral values and how do they vary? • Size: How big or small is the object? • Shape: Roundness, Length/Width Ratio, etc. • Texture: Contrast, Homogeneity • Context: How does the object relate to its neighbors? Neighbors, Relative Location, Sub Objects, Super Objects • The big picture: How do the characteristics interrelate?
Landsat 8 Mosaic false color, Summer 13-14 Landsat 8 Mosaic false color, Fall 13-14
Optical Data
Optical Data
TCMA false color, Summer 13-14
TCMA false color, Summer Fall 13-14
Summer 13-14
Fall 13-14
Normalized Difference Vegetation Index (NDVI)
NIR - Red / NIR + Red
Optical Data
Lidar Data
Lidar derived layers :
• Digital Elevation Model (DEM) • Digital Surface Model (DSM) • Digital Terrain Model (DTM) • Normalized Digital Surface Model (nDSM) • Normalized Digital Terrain Model (nDTM) • Compound Topographic Index (CTI) • Slope • Building footprints
Other Data
• Major roads
• City streets
• Updated NWI
• Airport runway
• Railroads
Object Based Image Analysis
Design of a customized ruleset with the Cognition Network Language (CNL) within the Definiens eCognition
Hybrid approach:
• Segmentation
• Classify “easy” classes with ruleset
• Random Forest Classifier (Random Trees in eCognition)
Random Forest Classifier
N e
xam
ple
s
....…
....…
Take the majority
vote
M features
Figure from Oznur Tastan
Classes
(Very) Preliminary Results
Do you always want to use OBIA?
0 % Impervious 100 Landsat image
Summary
• The integration of OBIA and lidar is expanding the applicability and accuracy of image-based analysis.
• The technology and data are mature.
• We are optimistic that the statewide map will be of high quality.