2nd Annual CNY-ASPRS Remote Sensing Symposium
“Developing Partnerships in Remote Sensing and GIS”
A Joint GIS/SIG and CNY-ASPRS Symposium
Abstracts Using a Geospatial Library to Organize Your Imagery and
GIS Data
Joan Zelinski, PAR Government Systems Corporation, Rome, NY, 315-268-1608,
joan_zelinski@partech.com
An open architectural concept is presented, based on the FGDC metadata standard
and the Geographic Information Access Specification (GIAS), which allows users
and external organizations easy access to the digital geospatial data they
need. The distributed geospatial library provides for storage, retrieval,
and display of remotely sensed data, GIS data, and other data.
Land Cover Classification using Support Vector Machines: Effect of
Kernel Functions
Pakorn Watanachaturaporn, Manoj K. Arora, and Pramod K. Varshney, Department
of Electrical Engineering and Computer Science, Syracuse University, Syracuse,
NY 13244
Land cover is an important variable for a number diverse applications such
as forestry, hydrology, agriculture, environment, geology and ecology. Many
natural resource management, planning and monitoring programs depend on accurate
land cover information. Classification is the fundamental operation to retrieve
land cover information from remote sensing data. The limitations of conventional
statistical classifiers such as the most widely used maximum likelihood classifier
are well known. A range of potentially new classifiers such as neural networks,
rule based decision tree and fuzzy c-means, have been developed. However,
these classifiers along with the statistical classifiers have marked limitations
in extracting land cover information from hyperspectral remote sensing data.
Recently, Support Vector Machines (SVM), having its roots in machine learning
theory, has been proposed as an alternative to produce land cover classification.
Originally applied in optical character and handwritten digit recognition,
and face identification problems, SVM utilizes optimization tools that seek
to identify a linear optimal separating hyperplane to discriminate any two
classes of interest. When the classes are linearly separable, the linear
SVM performs adequately. Often, the classes are not separable linearly. In
such cases, the SVM maps the dataset to a higher dimensional feature space
and creates a linear optimal hyperplane using a kernel function. A number
of kernel functions such as polynomial, sigmoid and radial basis functions
can be used. The selection of appropriate kernel function and its parameters,
and the optimization technique may be critical to the implementation of SVM.
In this paper, we introduce SVM classifier for land cover classification
from both multi and hyperspectral remote sensing datasets. We also investigate
the effect of different kernel functions and their parameters on the accuracy
of classifications from Landsat TM and AVIRIS datasets. The experimental
results from both the datasets show a significant variation in classification
accuracy of the order of 25 to 35% with the use of different kernel functions
and their parameters. The results clearly demonstrate that the accuracy of
SVM classifier is highly sensitive to a particular type of kernel function
and its parameters.
Differences in Accuracy Assessments Resulting from Variability in Ground
Reference Protocols
Jason Cole, William Stiteler, Russel Aicher, and Paul Hopkins, SUNY ESF
Ground reference is an important aspect of remotely-sensed land cover classification,
but is often located based on practical constraints, rather than through
a theoretically superior and rigorous procedure. An appropriate sampling
protocol is expensive to implement, but ensures lack of bias and increases
the confidence users can place in results or accuracy assessment. Using
conveniently-located reference (whether it is located with ground visits
or using photo-interpretation) tends to skew ground reference towards homogeneous,
easily identified areas of land cover. Homogeneous reference used for training
data will tend to produce more homogeneous classifications. Homogeneous
reference used for accuracy assessment will tend to produce statistics that
are based on areas that are easiest to classify. This poster will present
results of land cover classifications that show that two seemingly valid
sets of ground reference, one derived from aerial photos in easy-to-distinguish
areas, the other from highly detailed ground reference plots, can return
significantly different classification results.
Implementing Data sharing between Utilities and Municipalities The
NSTAR Cape Cod, Massachusetts Model
Ralf Platte, Business Development, James W. Sewall Company, 154 Mountain
Rd., Salisbury NH 03268, Telephone: 603-648-2737, Fax: 603-648-6524, rplatte@jws.com
Presentation will focus on: Understanding the needs of both parties regarding
GIS data, looking at obstacle and how they were overcome, and developing
standards for future data sharing opportunities. NSTAR Gas and Electric has
contracted with the James W. Sewall Co. in Old Town , Maine to fly , provide
planimetric mapping, and address data for 39 town on Cape Cod in Massachusetts.
1,100 square miles of data will be collected at a resolution and accuracy
(1” = 100’ map accuracy w/ 4” pixel resolution) suitable for most municipal
GIS applications. We will discuss the logistics of the project and the process
used by NSTAR / Sewall to make this data available to the local communities.
Also covered will be the many obstacles a project like this presents and our
approach at overcoming them. From the beginning data sharing was one of the
main components of this project , details of which will provide a model for
future efforts similar in nature.
Overlay of imagery with ground-water model visualization using USGS-modelviewer
and VRML
Paul Misut,USGS,2045 Rt.112Coram, NY 11727, pemisut@usgs.gov
USGS-modelviewer is a computer program that displays the results of transient-state
three-dimensional ground-water models in local-model coordinates. (http://water.usgs.gov/nrp/
gwsoftware/modelviewer/ModelViewer.html) To enhance the evaluation of remotely-sensed
data, it would be useful to integrate remote-sensing-imagery visualization
with the capabilities of USGS-modelviewer. The following techniques are demonstrated:
export of a modelviewer object to a VRML file, and text-editing of the VRML
file to include a satellite image overlay . A hypothetical example on Long
Island, New York is shown (LANDSAT path 013, row 032 of April 14, 2001) and
the result is discussed in the context of public-water-supply management.
Creation and Demonstration of an Aerial Flight Visualization Over the
Torne Valley in the Town of Ramapo
Wayne Richter, Division of Fish, Wildlife and Marine Resources, New York
State Department of Environmental Conservation, 625 Broadway, Albany, New
York 12233-4754, wrichter@dec.state.ny.us, 518-402-8958
I demonstrate a one minute simulated flight over a three dimensional perspective
rthorectified photographic mosaic draped on a digital elevation model. In
addition to the photographic image, the flight shows vector information relevant
to the Department of Environmental Conservation's evaluation of the impacts
of a proposed project on the state endangered timber rattlesnake. Vector
information includes snake telemetry tracks, snake basking areas and the
project outline depicted in their landscape position. Use of three dimensional
perspective draping greatly facilitates understanding the landscape and potential
impacts in the high relief project area. I describe production, using moderately
priced, commercially available software and hardware, of a site visualization
movie that can be displayed at public meetings without specialized equipment.
This movie derives from aerial photographs taken with New York State's photogrammetric
mapping camera, scanned in house, and orthorectified with ERDAS Imagine OrthoBASE
software. ERDAS Imagine was used to create a composite from the rectified
photographs. ERDAS VirtualGIS was used to drape the mosaic over a 10 meter
digital elevation model, add ESRI format vector data, and record a fly-through
movie.
Land Use and Land Cover Summary Report And Recommendations
Land Use and Land Cover (LULC) Work Group, NYS GIS Standards and Data Coordination
Work Group, Ed Freeborn, Visual Geographics
A Land Use/Land Cover (LULC) Study Group was convened in the summer of 2002,
as a subgroup of the NYS GIS Standards and Data Coordination (S/DC) Work
Group. The purpose of this subgroup was to explore potential for coordination
of Land Use and Land Cover (LULC) activities within New York State. This
report represents a summary of the findings of that group; the primary recommendation
is that a Land Use and Land Cover Work Group be established under the auspices
of the NYS GIS Coordinating Body to receive direction from and advise the
Coordinating Body on matters concerning LULC in NYS and the region. Establishing
a LULC Work Group can assist in promoting and coordinating LULC mapping,
management and modeling efforts within the state. This can result in better
LULC information for national, state, private and research agencies across
New York State at greatly reduced costs in efforts, data and program resources.
Through surveys and phone outreach activities the LULC Study Group identified
nearly 60 projects that generate, utilize or will soon require LULC information
for the NY state region. Additional projects are being added to the list
weekly. Information content of these varying projects is broad, and spans
many topics such as: · Homeland security and emergency management
· Public health studies · Natural resource management · Urban
planning · Water quality protection Accurate, current, high resolution
LULC was identified as a priority by this Study Group. Detailed needs of
NY's LULC community, however, are not fully understood. Defining, coordinating
and helping to integrate these needs are valuable services that a LULC Work
Group could provide. The LULC Study Group recommends establishment of a
LULC Work Group, potentially modeled after the Digital Orthoimagery Work
Group, with the following goals and tasks: · Coordinate data development
with national programs and other states · Foster partnerships among
LULC developers and users · Ensure that data are made available in
readily usable forms · Facilitate collection and exchange of information
regarding LULC-related projects, methodologies, data sources and events · Promote
technical interchange and development · Identify and synthesize LULC-related
needs and requirements
Mapping Macrophyte Vegetation in Onondaga Lake Using Remotely Sensed
Imagery of Differing Spatial and Spectral Resolutions
Trevis Gigliotti1, Dr. Paul Hopkins1, Joseph Mastriano2, Dr. Elizabeth Moran3,
and Lindi Quackenbush1 1. State University of New York College of Environmental
Science and Forestry 2. Onondaga County Department of Water Environment
Protection 3. EcoLogic, LLC
The pollution level of Onondaga Lake is the subject of management efforts
and media attention at the local, state and national levels. On August 1,
1998, the Onondaga County Department of Water Environment Protection (OCDWEP)
began a 15-year Ambient Monitoring Program (AMP) involving Onondaga Lake.
The AMP is intended to evaluate the effectiveness of improvements to the
metropolitan sewage treatment plant and includes a requirement to monitor
the characteristics of macrophyte vegetation in the Lake. Assessing macrophyte
vegetation indicates reactions to nutrient loading. The macrophyte assessment
in June 2000 interpreted digitized aerial photography and proved to be expensive
and time consuming. The goal of this project was to develop cost-effective,
semi-automated strategies to map the abundance of macrophyte vegetation in
an urban lake environment. The study evaluated the effectiveness of IKONOS
and ASTER satellite imagery as an alternative to aerial photography for mapping
macrophyte vegetation. Initial image processing focused on isolating and
extracting the lake’s littoral zone using band ratios and digital bathymetry
data. The second phase of processing applied an ISODATA clustering algorithm
to a variety of image and derived image layers including Normalized Difference
Vegetation Index, chromaticity and texture. The derived image data was incorporated
to enhance subtle differences in water clarity, color and surface continuity
created by the presence of macrophytes. Finally, the macrophyte vegetation
maps were analyzed and standard error matrices and percent cover calculated.
Macrophyte maps produced by “heads-up” digitizing scanned aerial photographs
provided the baseline for evaluation. Initial accuracy measures show potential
for the use of satellite imaging in the OCDWEP’s macrophyte community assessment.
Future work will evaluate the utility of differing classification techniques
and the need for normalizing the image data for lake bottom reflectance.
Mapping and Monitoring of Submerged Aquatic Vegetation in the Hudson
river, New York
Eugenia Barnaba, Cornell Institute for Resource Information Systems, S. Findlay,
Institute of Ecosystems Studies, S. Hoskins, Cornell Institute for Resource
Information Systems, and C. Nieder, NYSDEC/National Estuarine Research Reserve
Resource managers, non-governmental organizations, educators and the general
public are being introduced to the value and use of remote sensing and geographic
information systems in the Hudson River, New York by way of a long term project
to map and monitor submerged aquatic vegetation (SAV) in the river. Long
recognized as an important component of a wide variety of aquatic ecosystems,
SAV contributes to primary productivity and as habitat for fish. Reliable
information on the abundance, distribution and ecological functions of SAV
is essential for understanding and managing what is considered to be an important
resource. The first of its kind in the Hudson, this project brings together
a group of collaborators with expertise from Cornell University, the Hudson
River National Estuarine Research Reserve/NYS Department of Environmental
Conservation, and the Institute of Ecosystem Studies. In a 120-mile stretch
of the river, true color aerial photographs were acquired at the 1:14,400
scale and at low tide. Focusing primarily on Vallisneria (water celery)
and Trapa natans (water chestnut ), plant beds were mapped to a base map
overlay. Ground-truth was accomplished with actual sampling of SAV beds
for quality assurance and to describe abundance, biomass, and species composition.
The Hudson River shoreline was mapped as a separate overlay. A geographic
information system (GIS) has been created that now includes polygon and shoreline
data. The project team has conducted workshops for resource managers, educators
and river users on the value, location and size of beds, hands-on instruction
in remote sensing and GIS for potential application in their respective areas
of responsibility. The team is presently engaged in a multi-year assessment
of ecological function of SAV beds, findings of which will be incorporated
into the database and distributed via CDROM and/or website.
Effective Partnerships in Land Cover: The Lake George Watershed Land
Cover Map
Emily Constantine, IAGT
As part of a joint effort, a 13-class land cover map of the Lake George Watershed
was developed by the Institute for the Application of Geospatial Technology
(IAGT), the Lake George Association (LGA), and the Town of Bolton, New York.
The land cover map was required as input data for a non-point source watershed
pollution model by the NY Department of State Division of Coastal Resources,
who also took part in portions of the land cover map development process.
The 13 unique land cover classes were defined by the LGA, and the map was
derived by the IAGT from Landsat 7 satellite imagery. Supervised classification
techniques were used to create 12 of the 13 classes. Existing GIS wetland
data sets were used to create the 13th class. The IAGT developed an accuracy
assessment plan, which was carried out by the LGA. The IAGT then analyzed
the accuracy data collected by the LGA team. The data used to assemble the
error metrics were created using both on-screen and field-based accuracy
assessment methods. While accuracies within the individual classes were lower
than NLCD averages, the error metrics were consistent with the uncertainty
found in the map source and reference data sets, and overall error percentages
were significantly improved when similar classes were combined. The map was
successful in identifying impervious surfaces, which was a high-priority
class with respect to the non-point source watershed pollution model. The
land cover map and accuracy assessment were successfully completed on a limited
budget, demonstrating how effective partnerships can overcome resource limitations
and can promote the use of remote sensing at the state and local government
level.
Origin and Status of the NY Statewide Digital Orthoimagery Program
Tim Ruhren, NYS OCSCIC
NYSDOP imagery collection started in April of 2000 and the first statewide
cycle is scheduled to be completed with image capture in April 2003. This
presentation will briefly cover the origins of the NYSDOP before discussing
the status of the program. The imagery collected in each annual lot will
be described as well as the availability of the imagery.
Quality Assurance for the Statewide Digital Orthoimagery Program
Ron Frederiks, NYS DOT
The specifications for the Statewide Digital Orthoimagery Program (DOP) are
the most stringent of any such program in the country. Quality assurance to
enforce these specifications requires a very rigorous, standards-based approach
that considers not only inspection of the visual quality of the submitted
orthoimagery, but also testing of the horizontal and vertical accuracy of
the orthoimagery. The process will be explained and illustrated with actual
examples of QA results and findings. This presentation will be of interest
to any user of the DOP data.
A Customer-Oriented View of the NYSDOP
Tim Ruhren, NYS OCSCIC
As more of the NYSDOP imagery becomes available, more issues arise that lead
to changes in how the data is presented. Security concerns are given more
weight than in the beginning of the program. Increased use of new software
has led to the production of new files. End-users have found new ways to
use the orthoimagery. This presentation will focus on how the imagery is
made available and how this has evolved to increase the usefulness of the
imagery. A sample of uses will be briefly described