1 Accuracy Assessment Methods Accuracy Assessment Methods and Results for Two Satellite and Results for Two Satellite Imagery Derived Landcover and Imagery Derived Landcover and Landuse Datasets Landuse Datasets Raymond Crew - GIS Researcher and MS Raymond Crew - GIS Researcher and MS Candidate Candidate Rick L. Day - Associate Professor Soil Rick L. Day - Associate Professor Soil Science Science Hanxing Pu - Penn State University Hanxing Pu - Penn State University Image based on photo by Peggy Greb, USDA ARS (Richard Lowrance in foreground)
33
Embed
1 Accuracy Assessment Methods and Results for Two Satellite Imagery Derived Landcover and Landuse Datasets Raymond Crew - GIS Researcher and MS Candidate.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Accuracy Assessment Methods and Accuracy Assessment Methods and Results for Two Satellite Imagery Results for Two Satellite Imagery Derived Landcover and Landuse Derived Landcover and Landuse
DatasetsDatasets
Raymond Crew - GIS Researcher and MS CandidateRaymond Crew - GIS Researcher and MS CandidateRick L. Day - Associate Professor Soil ScienceRick L. Day - Associate Professor Soil Science
Hanxing Pu - Penn State UniversityHanxing Pu - Penn State University
Image based on photo by Peggy Greb, USDA ARS (Richard Lowrance in foreground)
2
• Part of a study to inventory the riparian land cover conditions for the Chesapeake Bay watershed
• Highlighting accuracy assessment of a GIS based riparian land cover inventory
Introduction
3
• Largest estuary in the US
• Watershed covers 178,000 km2 and contains 464,000 km of stream banks and shorelines
• New York, Pennsylvania, Maryland, Delaware, Washington DC, West Virginia and Virginia
• 15 million residents growing to 18 million by 2020
The Chesapeake Bay Watershed
4
• Collaborative group coordinating efforts to protect the ecological health of the Chesapeake Bay. Formed in 1983.
– Chesapeake Bay Commission– U.S. Environmental Protection Agency– Maryland– Virginia– Pennsylvania– District of Columbia
Chesapeake Bay Program Office
5
• The Executive Council:– Governors of Maryland, Pennsylvania, and Virginia– Mayor of DC– Administrator of the EPA– Chair of the Chesapeake Bay Commission
• Issued a directive in 1994 asking the Chesapeake Bay Program Office to increase the focus on riparian stewardship, especially riparian forest buffers
Chesapeake Executive Council Directive on Riparian Buffers
6
• Riparian buffers are area of vegetated land adjacent to streams, rivers, marshes or shorelines forming a transition between the land and water environments. (Vellidis and Lowrance, 2004) (Lowrance, Altier et. Al, 1995)
• Both forests and wetlands are effective buffers (EPA, 2005)
What are riparian buffers
photo David J. Welsch USDA
7
• Reducing and Filtering Runoff• settles out sediments, nutrients and pesticides• increase infiltration rates• root systems stabilizing stream-banks and reduce erosion
• Nutrient Uptake• increases absorption of fertilizers and other pollutants by storing
bio-mass• de-nitrification via soil bacteria
• Canopy and Shade• moderate stream temperatures • leaf canopy filtering dust from wind erosion
• Habitat and Habitat Corridors
Why buffers are important
8
Study Objectives
1. Develop an improved automated inventory method
2. Conduct an inventory for two time periods
3. Validate method to determine reliability
4. Study differences in buffer conditions by stream order
5. Study environmental pressures associated with land-use surrounding the riparian zones
9
Automated Inventory
• Determine and collect the GIS datasets required for inventory acquire datasets
• land-cover• streams and water body polygons• watershed divides
• Calculate the inventory using an automated method• 100 and 300 foot buffers• stream miles – single sided and dual sided buffers• riparian miles
10
Streams and Water Body Data
• Hybrid dataset produced by the USGS and Chesapeake Bay Program Office
• 1:24K Resolution National Hydrography Dataset (Higher Resolution NHD)
• 1:24K Reach files produced by the CBPO
11
Land-cover Data
• Original plan:
• Have the Mid-Atlantic Regional Earth Science Applications Center (RESAC) produce a year 1990 and 2000 dataset using the same methods and type of imagery
• Unfortunately only a year 2000 dataset was produced
12
Land-cover Near Dulles Airport
13
2000 Land-cover Data
• Based on three year 2000 satellite images
• 30 meter resolution
• 21 land-cover and use classes - 4 pertinent to this study
• forest• urban forest• wetlands• water
14
Solution
• Use the best available land-cover data available for the year 2000 and the year 1992
• 2000• Use the RESAC landcover/landuse data
• 1992• Use the Multi-Resolution Land Characteristics
(MRLC) landcover data. Commonly called the NLCD
15
1992 MRLC Land-cover Data
• 18 land-cover classes in the Chesapeake Bay - 3 pertinent to this study
• forest• wetland• water
• based on primarily early 1990’s satellite images, however, some imagery as old as 1985 was used
• 30 meter resolution
16
Problems Using These Two Datasets
• Different Methods uses to derive land-cover categories
• example: large differences between classification of open water vs. wetlands
• No urban forest category in the NLCD as in the RESAC
• unsure if urban forest will be classified as forest or as something else
• NLCD based on a wider temporal range of images
17
The Sampling Process
1. Splits streams into 300’ (91.44 m) segments
2. Orients a transect perpendicular to each stream segment at the center of the segment
18
Algorithm Outputs
3. Establishes sampling locations every 50 feet (15.24 m) along the transect
4. Collects the land-cover informationfrom the satellite imagery at each sample location
5. Calculates buffer statistics for each watershed
• Forest• Forest or urban forest• Forest, urban forest, or wetlands • Forest, urban forest, wetlands or water
19
Riparian Buffer Algorithm
• Stream Miles• One or more sides buffered 100 feet or more• One or more sides buffered 300 feet or more• Both sides buffered 100 feet of more• Both sides buffered 300 feet or more
• Riparian Miles• Buffers of 100 feet or more• Buffers of 300 feet or more
20
Statistics by
State
21
Statistics by Watershed
22
Year 2000 Inventory
Percent of streams with at least 100 feet of a forested buffer
23
Validation Study
• Need to validate automated inventory method versus a traditional method to generate buffer inventories
• Must also test the reliability of land cover classifications in riparian zones
• Must validate and changes shown between the 1992 and 2000 inventories
24
Validation Study
• Generate a forest land cover inventory using air photo interpretation
• Compare each air photo inventory to the same year automated inventory
• Compare the air photo inventories to each other
25
Validation Study
• Randomly selected 50 locations throughout Chesapeake Bay Watershed
• Stratified to include all States and physiographic regions
• Used Hawth’s Analysis Tools 2.1 for ArcGIS
26
Acquire Air Photos
• Purchase overlapping stereo sets of two or three photos from the National Aerial Photography Program (NAPP)– 9” by 9” prints grayscale prints– 1:40,000 scale– Flown every 5 to 7 years– Only $5 per print
• Two or three photos cover ¼ of the extent a standard USGS topographic map
27
Air Photos
• Purchase two sets of photos, one centered near 2000 and the other near 1992
• Had to remove NY, no photos available later than 1998
28
Scanning Stereoscope
• Used a scanning & magnifying stereoscope
• All photos work by one individual, Hanxing Pu
• His work compared for quality to another tech, Chihiro Mather
29
• Coded every visible stream bank and waterbody outline into four classes
Red: Forest buffer exceeding 300’
Blue: Forest buffer between 100’ and 300’
Black: Forest buffer less than 100’
Yellow: No buffer visible
Photo Work
30
Digitizing
• Next digitize all the buffers see via air photo work
• Can only be accomplished for those steams seen on the air photos and those in the digital data
• Any stream missing from one of the two sets is not compared
31
Digitizing
• Removed the automated method results from the datasets to prevent any bias
32
Results
33
Thanks to:
• Dr. Anne Hairston-Strang – Maryland DNR Forest Service • J. Michael Foreman - Virginia Department of Forestry• Gene Odato - PA Bureau of Forestry • Sally Claggett - Chesapeake Bay Program Coordinator• Peter Claggett - Chesapeake Bay Program Office - USGS Land
Data Manager
• Dr. David Mortensen – Associate Professor of Weed Ecology • Dr. Doug Miller – Assistant Professor of Geography • Dr. Gary Petersen - Distinguished Prof of Soil Science