Peer-Reviewed Articles
129 Incorporating Remote Sensing Information in Modeling House Values: A Regression Tree Approach
Danlin Yu and Changshan Wu
Abstract Download
Full Article
This paper explores the possibility of incorporating remote
sensing information in modeling house values in the City
of Milwaukee, Wisconsin, U.S.A. In particular, a Landsat
ETM+ image was utilized to derive environmental characteristics,
including the fractions of vegetation, impervious
surface, and soil, with a linear spectral mixture analysis
approach. These environmental characteristics, together
with house structural attributes, were integrated to house
value models. Two modeling techniques, a global OLS
regression and a regression tree approach, were employed
to build the relationship between house values and house
structural and environmental characteristics. Analysis of
results indicates that environmental characteristics generated
from remote sensing technologies have strong influences
on house values, and the addition of them improves
house value modeling performance significantly. Moreover,
the regression tree model proves as a better alternative to
the OLS regression models in terms of predicting accuracy.
In particular, based on the testing dataset, the mean average
error (MAE) and relative error (RE) dropped from 0.202 and
0.434 for the OLS model to 0.134 and 0.280 for the regression
tree model, while the correlation coefficient between the
predicted and observed values increased from 0.903 to
0.960. Further, as a nonparametric and local model, the
regression tree method alleviates the problems with the
OLS techniques and provides a means in delineating urban
housing submarkets.
139 An Integrated Approach to Wildland Fire Mapping of California, USA Using NOAA/AVHRR Data
Peng Gong, Ruiliang Pu, Zhanqing Li, James Scarborough, Nicolas Clinton, and Lisa M. Levien
Abstract Download
Full Article
To map wildland fires for emission estimation in California,
this paper presents an integrated approach to wildfire
mapping using daily data of the Advanced Very High
Resolution Radiometer (AVHRR) on board a National Oceanic
and Atmospheric Administration’s (NOAA) satellite. The
approach consists of two parts: active fire detection and
burnt area mapping. In active fire detection, we combined
the strengths of a fixed multi-channel threshold algorithm
and an adaptive-threshold contextual algorithm and modified
the fire detection algorithm developed by the Canada
Center for Remote Sensing (CCRS) for fire detection in boreal
forest ecosystems. We added a contextual test, which
considers the radiometric difference between a fire pixel and
its surrounding pixels, and a sun glint elimination test to
the CCRS algorithm. This can effectively remove false alarms
caused by highly reflective clouds and surfaces and by warm
backgrounds. In burnt area mapping, we adopted and
modified the Hotspot and NDVI Differencing Synergy (HANDS)
algorithm, which combines the strengths of hotspot detection
and multi-temporal NDVI differencing. We modified the
HANDS procedure in three ways: normalizing post-fire NDVI
to pre-fire NDVI by multiplying an NDVI ratio coefficient,
calculating mean and standard deviation of NDVI decrease of
land-cover types separately, and adding a new iteration
procedure for confirming potential burnt pixels. When the
integrated method was applied to the mapping of wildland
fires in California during the 1999 fire season, it produced
comparable results. Most of the wildfires mapped were
found to be correct, especially for those in forested ecosystems.
Validation was based both on limited ground truth
from the California Department of Forestry and Fire Protection
and on interpreted burnt areas from Landsat 7 TM
scenes.
151Control Patches for Automatic Single Photo Orientation
Jen-Jer Jaw and Yi-Shen Wu
Abstract Download
Full Article
Traditional aerial triangulation has long relied on control
points for orientating the photo models into a ground-based
coordinate system, reducing the distortion effect when tying
photos by imperfect photo measurements, and calibrating
camera parameters. Field surveys to provide an adequate
number of control points and manual measurements of the
control points afterwards in the photos incur considerable
cost both in labor and expense. The past decade has seen
the development of digital photogrammetry as a result of
integrating into the photogrammetric discipline both the
advantages of image processing techniques and the rapid
computational efficiency of the computer. Such a development
has made possible database-supplied control entities,
as well as the automation of matching control data between
object space and image space. Inspired by this capability,
we employ a control patches database where the control
points are found on older imagery and matched in the new
photo. The successful implementation of the proposal lies
in an effective control patches database, robust matching
methodology, and a reliable orientation approach. Single
photo resection is applied whenever no less than three
matched control patches are available. The experiments
under this project suggest the potential efficiency of automatic
control point measurements from the control patches
database and a reliable photo orientation solution.
159 Comparison of Automated Watershed Delineations: Effects on Land Cover Areas, Percentages, and Relationships to Nutrient Discharge
Matthew E. Baker, Donald E. Weller, and Thomas E. Jordan
Abstract Download
Full Article
We compared manual delineations with those derived from
ten automated delineations of 420 watersheds in four
physiographic provinces of the Chesapeake Basin. Automated
methods included commercial DEM-based routines
and different parameterizations of four enhanced methods:
stream burning, normalized excavation, surface reconditioning,
and normalized reconditioning. Un-enhanced methods
resulted in individual watershed boundaries with some gross
discrepancies in watershed size relative to manual delineations
(error rate of 0.22 > 25 percent difference compared
to manual) and significantly different watershed size
distributions (Mann-Whitney U p = 0.012). Integrating
mapped streams through enhanced methods substantially
improved correspondence with manual watersheds (error
rates of only 0.08–0.02 > 25 percent difference). Analysis of
cropland area among methods showed a significant difference
between manual estimates and un-enhanced estimates
(p = 0.049) that was corrected using enhanced algorithms.
Subsequent analysis of percent cropland revealed that
measurements of land cover proportions were not always
affected by delineation errors. However, differences were
large enough to influence regressions with stream nitrate-N
at the 90 percent confidence level within one physiographic
province. Enhanced delineations produced statistical relationships
between percent cropland and nitrate-N concentrations
consistent with manual delineations. The results
provide support for enhanced automated watershed delineation
within the Chesapeake Basin and suggest that normalized
excavation can be an effective augmentation of
existing stream burning and reconditioning procedures.
169 Mapping Prairie Pothole Communities with Multitemporal Ikonos Satellite Imagery
Rick Lawrence, Rebecca Hurst, T. Weaver, and Richard Aspinall
Abstract Download
Full Article
We evaluated the ability of Ikonos imagery from August
and October 2000 to classify prairie pothole community
types of the Missouri Coteau of North Dakota. Classification
tree analyses were conducted to create land-cover maps at
three levels of detail. The analyses successfully distinguished
broad cover types (potholes including emergent vegetation
versus upland vegetation) at 92 percent overall accuracy.
Overall accuracy dropped to 80 percent when upland
vegetation was segregated into woody and grassy communities
and to 71 percent when we attempted to classify at the
species or near-species levels. The use of two image dates
was of importance in the classifications; the failure to
acquire early season imagery, therefore, might have impaired
our results.
175 Assessing Accuracy of Net Change Derived from Land Cover Maps
Stephen V. Stehman and James D. Wickham
Abstract Download
Full Article
Net change derived from land-cover maps provides important
information for environmental monitoring and modeling.
To better target the objectives of net change accuracy,
we require modifications of the sampling design and
analysis protocols typically implemented for assessments
focusing on single date or gross change maps. Mean
absolute deviation estimated for user-defined reporting
domains is suggested to characterize net change accuracy.
Stratified sampling is often desirable to improve precision
for high priority estimates (e.g., high net change domains),
but decisions regarding the number and identity of strata
must be made recognizing the precision trade-offs among
the multiple estimates of interest in a net change assessment.
The accuracy assessment strategy and a protocol for
evaluating sampling design options are demonstrated using
a population of map and reference net change derived from
existing land-cover maps and representing change from 1990
to 2000.
187 Population Density and Image Texture: A Comparison Study
XiaoHang Liu, Keith Clarke, and Martin Herold
Abstract Download
Full Article
The correlation between census population density and
Ikonos image texture was explored. The spatial unit for the
analysis was census blocks with homogenous land-use. Ikonos
image texture was described using three methods: the graylevel
co-occurrence matrix (GLCM), semi-variance, and spatial
metrics. Linear regression was conducted to explore the
correlation between image texture and population density. It
was found that although correlation exists, its degree varies
depending on the method used to describe image texture. The
highest correlation is given by the spatial metrics method.
This result suggests that the correlation between texture and
population density is not strong enough to predict or forecast
residential population. However, image texture does provide a
base to refine census-reported population distribution using
remote sensing. High-resolution satellite images therefore
have the potential to support “smart interpolation” programs
to estimate human population distribution in areas where
detailed information is not available.