Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
Combining Combining Automated and Manual Automated and Manual
Techniques for Accurate Techniques for Accurate NWINWI MappingMapping
Pamela SwintPamela Swint [email protected]@vt.edu
Kevin McGuckinKevin McGuckin [email protected]@vt.edu
Conservation Management InstituteConservation Management InstituteBlacksburg, VABlacksburg, VA
Problem OverviewProblem Overview
• Wetland managers experience:– An increased demand for high resolution
wetland data– A need for a faster turn-around time
• We explore the efficiency of combining:1) Computer-assisted delineations2) Manual QA/QC
Photo-interpretation/AttributionImage PUBHx
PUBHx
PFO1A
PSS1B
PUBHh
PFO1E
Classified MapTraditional Approach
Image
Photo-interpretation/Attribution
Data Editing/Photo-interpretation/
Attribution
Image
Feature Analyst
PUBHx
PUBHx
PFO1A
PSS1B
PUBHh
PFO1E
Classified MapTraditional Approach
Hybrid Computer-Manual Approach
PUBHx
PUBHx
PFO1A
PSS1B
PUBHh
PFO1E
Classified Map
= AutomatedProcess
DeleteClutter
Hybrid Computer-Manual Approach
= AutomatedProcess
PUBHx
PUBHx
PFO1A
PSS1B
PUBHh
PFO1E
Classified Map
ClutterDeletion
Image Feature AnalystData Editing/
Photo-interpretation/Attribution
Feature Analyst InterfaceFeature Analyst Interface
Feature Analyst InterfaceFeature Analyst Interface
Feature Analyst InterfaceFeature Analyst Interface
Feature Analyst InterfaceFeature Analyst Interface
Example Search Patterns
Bullseye 3 (3x9) Bullseye 4 (3x13)Bullseye 1 (3x9)
Foveal Max. (3x33)
Bullseye 2 (3x17)
Circle Max. (3x97)
Manhattan Min. (3x5) Manhattan Max. (3x85)
Square Min. (3x9) Square Max. (3x81)
How does FA compare How does FA compare to hand delineations?to hand delineations?
• Pond drawn at a zoom scale of 1:2,500
Yellow = FARed = Hand Delineation
1:1,000
Case Study: Shirley PlantationCase Study: Shirley Plantation
Step 1: Create Training SamplesStep 1: Create Training Samples
• Train FA on dark water features
Step 2: Run Classification using Step 2: Run Classification using Default SettingsDefault Settings
Step 3: Create Additional Training Step 3: Create Additional Training SetSet
• To capture the remaining water features
Step 4: Classification Result using Step 4: Classification Result using Default SettingsDefault Settings
Step 5: Merge ClassificationsStep 5: Merge Classifications
Case Study: Northern AlaskaCase Study: Northern Alaska• ~ 530 square miles
Alaska Case StudyAlaska Case Study• 1 Training Set; 30,000 features identified• ~ 3 hours total (2.5 hours for trial and
error, ½ hour to train and classify data)
Final ProductFinal Product
Hybrid Computer-Manual Approach
= AutomatedProcess
PUBHx
PUBHx
PFO1A
PSS1B
PUBHh
PFO1E
Classified Map
ClutterDeletion
Image Feature AnalystData Editing/
Photo-interpretation/Attribution
Examples of ClutterExamples of Clutter
Raw FA ResultsRaw FA Results-- clipped by soils, slopeclipped by soils, slope
~ 4500 polygons 1230 polygons
Final Wetland MapFinal Wetland Map-- with manual editingwith manual editing
Hybrid Computer-Manual Approach
= AutomatedProcess
PUBHx
PUBHx
PFO1A
PSS1B
PUBHh
PFO1E
Classified Map
ClutterDeletion
Image Feature AnalystData Editing/
Photo-interpretation/Attribution
Use of Ancillary GIS DataUse of Ancillary GIS Data• Datasets involved
– SSURGO Hydric soils– NHD– USGS Topographic maps– USGS 10m NED
Comparison of Final Product Comparison of Final Product vsvsOriginal NWIOriginal NWI
Yellow = FARed = Original NWI
Yellow = FARed = Original NWI
Comparison of Final Product Comparison of Final Product vsvsOriginal NWIOriginal NWI
Minimum Mapping Unit Minimum Mapping Unit ComparisonComparison
• PUBH only; upstate New York area
LLWW AttributionLLWW Attribution
• Landscape Position, Landform, WaterflowPath, Waterbody Type (LLWW)
• LLWW attribution uses both manual and automated methods
• LLWW Modifiers are added during wetland attribution/photo-interpretation process
LLWW ModelLLWW Model• Model Inputs
– Cowardin wetlands, modifiers, flow assignments, and buffered stream network
LLWW ModelLLWW Model• Step 1
– Assign landscape position and Waterbodytype based on Cowardin classification
Lakes Rivers Ponds Expand Codes Marine Estuarine
LLWW ModelLLWW Model• Step 2
– Spatial and photo-interpreted attributes
Stream Value
Lakes Rivers
LLWW ModelLLWW Model• Step 3
– Additional Attributions
Modifiers Landform Pond Modifiers
Clean-up
How long does mapping wetlands How long does mapping wetlands this way take?this way take?
Upstate NY
Per USGS 1:24k Quad
Upstate NY
How long does mapping wetlands How long does mapping wetlands this way take?this way take?
Upstate NY
Per USGS 1:24k Quad
Upstate NY
Low Contra
st, H
igh Complex
ity
High Contrast, Little Complexity
How long does mapping wetlands How long does mapping wetlands this way take?this way take?
Per USGS 1:24k Quad
Inland VA
Upstate NY
How long does mapping wetlands How long does mapping wetlands this way take?this way take?
Per USGS 1:24k Quad
Inland VA
Upstate NY
Task BreakTask Break--downdown
ConclusionConclusion
• Combination of manual and automated techniques provide efficiency
• Processing time varies with the amount of wetland complexity involved
ConclusionsConclusions
• High contrast features One-pass Feature Analyst classification
• Low contrast features Multi-pass FA classification and/or manual delineation
• CMI is looking to asses FA in other regions (southwest, mountainous terrain...etc)
AcknowledgementsAcknowledgements
U.S. Fish and WildlifeRalph Tiner, John Swords, Herb Bergquist
Virginia TechBeth Pokorski, Staci Hudi, Tim Brown, Jason
Herman, Arvind Bhuta, Hanna Mabe
Contact InfoContact Info
Pamela Swint ([email protected])Kevin McGuckin ([email protected])
Scott Klopfer ([email protected])
Conservation Management Institute1900 Kraft Drive, Suite 250
Blacksburg, VA 24061540-231-7348