Peer-Reviewed Articles
1229 A Comparison of Standard and Hybrid
Classifier Methods for Mapping Mortality in Areas Affected by “Sudden
Oak Death”
Maggi Kelly, David Shaari, Qinghua Guo, and Desheng Liu
Abstract Download
Full Article
The sudden oak death (SOD) epidemic in California has resulted in
hundreds of thousands of dead trees in the complex
of oak (Quercus) and tanoak (Lithocarpus) woodland that
exist in patches along the California coast. Monitoring SOD
occurrence and spread is an on-going necessity in the state.
Remote sensing methods have proved to be successful in mapping and
monitoring forest health and distribution when a
sufficiently small ground resolution is used. Supervised, unsupervised,
and "hybrid" classification methods were evaluated
for their accuracy in discriminating dead and dying tree
crowns from bare areas and the surrounding forest mosaic
utilizing 1-m ADAR imagery covering both tanoak/redwood
forest and mixed hardwood stands. In both study areas the
hybrid classifier significantly outperformed the other methods, producing
low omission and commission errors among
information classes. The hybrid method was then further refined by
varying three parameters of the algorithm (iteration
number, homogeneity threshold, and number of classes) and
accuracy was assessed. The results demonstrate that while the
hybrid method outperformed the other classifiers, the
parameters that yielded highest accuracy for the algorithm
differed between the two study areas. The use of a randomly
selected subsample of training pixels was compared to the use
of polygonal training areas, and we found that polygonal
training data provided better classification accuracies in both
cases.
1241 Knowledge-Based Approaches to
Accurate Mapping of Mangroves from Satellite Data
Jay Gao, Huifen Chen, Ying Zhang, and Yong Zha
Abstract Download
Full Article
Mangroves are difficult to map accurately from satellite data
by means of parametric classification because of their spectral
similarity to other coastal vegetation despite their habitat
being inside coastal waters. This study aims to improve the
mapping accuracy through incorporation of such spatial
knowledge about mangroves in the Waitemata Harbor of
Auckland, New Zealand, from SPOT data. The spatial knowledge was combined
with spectral knowledge in the mapping.
Supervised classification was found to map stunted and lush
mangroves at an accuracy of, respectively, 46.7 percent and
68.3 percent. These accuracy levels rose, respectively, to
83.3 percent and 96.7 percent after the spatial knowledge was
sequentially incorporated into the mapping. A similar accuracy level
was achieved from knowledge-based spatial reasoning. If integrated
simultaneously with spectral knowledge,
spatial knowledge did not improve the accuracy noticeably
because of difficulty in gaining quality spectral knowledge. It
is concluded that knowledge-based, post-classification processing considerably
improves the accuracy of mapping mangroves over parametric classification.
1249 Exurban Change Detection in Fire-Prone
Areas with Nighttime Satellite Imagery
Thomas J. Cova, Paul C. Sutton, and David M. Theobald
Abstract Download
Full Article
Fire-prone landscapes are increasingly being settled. Monitoring this
development is an emerging need, and a low-cost
method would benefit emergency managers. Existing change-detection methods
can be expensive and time consuming
when applied to low-density urban change in large, vegetated
areas. Nighttime satellite imagery is explored as means for
addressing this problem, and a case study is presented for
Colorado. The results indicate that from 1992-2000, Grand
County had the greatest absolute increase in ambient sprawl
into fire-prone areas (215 km2), but Teller County had the
greatest percentage increase (7.3 percent). In 2000, La Plata
County had the most ambient development in fire-prone areas
(909 km2), but Jefferson County had the greatest percentage
(42 percent). The paper concludes with a discussion of the
prospects and problems of the approach.
1259 Integrating JERS-1 Imaging Radar
and Elevation Models for Mapping Tropical Vegetation Communities
in Far North Queensland, Australia
Catherine Ticehurst, Alex Held, and Stuart Phinn
Abstract Download
Full Article
The "Wet Tropics World Heritage Area" in Far North Queensland,
Australia consists predominantly of tropical rainforest
and wet sclerophyll forest in areas of variable relief. Previous
maps of vegetation communities in the area were produced by
a labor-intensive combination of field survey and air-photo interpretation.
Thus, the aim of this work was to develop a new
vegetation mapping method based on imaging radar that incorporates topographical
corrections, which could be repeated frequently, and which would reduce
the need for detailed field assessments and associated costs. The method
employed a topographic correction and mapping procedure
that was developed to enable vegetation structural classes to
be mapped from satellite imaging radar. Eight JERS-1 scenes
covering the Wet Tropics area for 1996 were acquired from
NASDA under the auspices of the "Global Rainforest Mapping
Project." JERS scenes were geometrically corrected for topographic
distortion using an 80 m DEM and a combination of
polynomial warping and radar viewing geometry modeling.
An image mosaic was created to cover the Wet Tropics region,
and a new technique for image smoothing was applied to the
JERS texture bands and DEM before a Maximum Likelihood
classification was applied to identify major land-cover and
vegetation communities. Despite these efforts, dominant vegetation community
classes could only be classified to low levels of accuracy (57.5 percent)
which were partly explained by
the significantly larger pixel size of the DEM in comparison to
the JERS image (12.5 m). In addition, the spatial and floristic
detail contained in the classes of the original validation maps
were much finer than the JERS classification product was able
to distinguish. In comparison to field and aerial photo-based
approaches for mapping the vegetation of the Wet Tropics, appropriately
corrected SAR data provides a more regional
scale, all-weather mapping technique for broader vegetation
classes. Further work is required to establish an appropriate
combination of imaging radar with elevation data and other
environmental surrogates to accurately map vegetation communities across
the entire Wet Tropics.
1267 Filtering Airborne Laser Scanner
Data: A Wavelet-Based Clustering Method
T.Thuy Vu and Mitsuharu Tokunaga
Abstract Download
Full Article
Filtering the airborne laser scanner data is challenging due to
the complex distribution of objects on Earth's surface and it
is still in development stage. This problem has been investigated so
far with varieties of algorithms, but they suffer from
different magnitudes of drawbacks. This study proposed a new and improved
hybrid method based on multi-resolution
analysis. Wavelet was adopted in this multi-resolution clustering approach.
It enabled the classification of objects based
on their size and the efficiency to filter out unwanted information at
a specific resolution, and the proposed algorithm
is named the ALSwave (Airborne Laser Scanner Wavelet)
method. ALSwave has been tested on two data sets acquired
over the urban areas of Tokyo, Japan and Stuttgart, Germany.
The results showed a well-filtered, bare earth surface coupled
with acceptable computational time. The accuracy assessment was carried
out by comparison between the filtered bare
earth surface by ALSwave and the manually filtered surface.
The Root Mean Square Error (RMSE) follows a linear relationship with
respect to terrain slope. This wavelet-based approach has opened a
new way to filter the raw laser data that
subsequently generates fast and more accurate digital terrain
models.
1275 Evaluation of Impervious Surface
Estimates in a Rapidly Urbanizing Watershed
Mark Dougherty, Randel L. Dymond, Scott J. Goetz, Claire A. Jantz,
and Normand Goulet
Abstract Download
Full Article
Accurate measurement of impervious surface (IS) cover is an
essential indicator of downstream water quality and a critical
input variable for many water quality and quantity models.
This study compares IS estimates from a recently developed
satellite imagery/land cover approach with a more traditional
aerial photography/land use approach. Both approaches are
evaluated against a high-quality validation set consisting of
planimetric data merged with manually-delineated areas of
soil disturbance. The study area is the rapidly urbanizing
127 km2 Cub Run watershed in northern Virginia, located on
the fringe of the Washington, D.C. metropolitan region. Results show
that photo-interpreted IS estimates of land class
are higher than satellite-derived IS estimates by 100 percent
or more, even in land uses conservatively assigned high IS
values. Satellite-derived IS estimates by land class correlate
well with planimetric reference data (r = 0.95) and with published ranges
for similar sites in the region. Basin-wide mean
IS values, difference grids, and regression and density plots
validate the use of satellite-derived/land cover-based IS estimates over
photo-interpreted/land use-based estimates. Results of this site-specific
study support the use of automated,
satellite-derived IS estimates for planning and management
within rapidly urbanizing watersheds where a GIS system is in
place, but where time-sensitive, high quality planimetric data
is unavailable.
1285 Snail Density Prediction for Schistosomiasis
Control Using Ikonos and ASTER Images
Bing Xu, Peng Gong, Greg Biging, Song Liang, Edmond Seto, and
Robert Spear
Abstract Download
Full Article
Schistosomiasis is a water-borne parasitic disease endemic in
tropical and subtropical areas. Its transmission depends upon
the presence of snails, which serve as intermediate hosts for
the parasite. Some efforts have been made to classify snail
habitats with remotely sensed data, but not to estimate snail
abundance that is an important parameter in schistosomiasis
transmission modeling. In this research, snail density was
predicted by integrating the field survey and satellite images
of different spatial resolution. A mountainous environment
near Xichang city, in southwest Sichuan province, China, was
chosen as the test site. Land-cover and land-use information
extracted from 4 m resolution Ikonos data and elevation data
derived from ASTER (Advanced Space-borne Thermal Emission and Reflection
Radiometer) data were used as reference
for scaling up to greater spatial extents. Therefore, we estimated
land-cover and land-use fraction data at the 30 m resolution level
based on classification results from the Ikonos
data. Snail abundance for each 30 m resolution grid was then
predicted by regressing field survey data with land-cover and
land-use fractions. Subsequently, a snail density map was
generated using the territory of each of the over 200 residential groups
as a mapping unit. An R2 of 0.87 was obtained between the average
snail density predicted and that surveyed
for 19 groups. With such a model, we were able to extrapolate
scattered snail abundance surveyed at a limited number of
sites to the entire area. Spatial autocorrelation of snail distribution
was considered as one of the possible factors in predicting snail
density and tested for further model calibration.
1295 A Modeling Approach for Estimating
Watershed Impervious Surface Area from National Land Cover Data 92
David B. Jennings, S. Taylor Jarnagin, and Donald W. Ebert
Abstract Download
Full Article
We used National Land Cover Data 92 (NLCD 92), vector impervious surface
data, and raster GIS overlay methods to derive impervious surface coefficients
per NLCD 92 class in portions of the Mid-Atlantic physiographic region.
Sample areas
for the study were thirty-six subwatersheds ranging in size
from 2 km2 to 150 km2. A three-category rural-to-urban gradient design
was utilized due to the changing sub-pixel relationship of impervious
surface areas within developed/non-
developed areas. A gradient rule based on the NLCD 92
DEVELOPED% defined the sample areas as "rural" (<18 percent
'developed'), "intermediate" (18 percent-40 percent 'developed')
and "dense suburban" (40.01 percent-80 percent
'developed'). The gradient scheme produced three separate
sets of coefficients per NLCD 92 Level 1 and Level 2 class. Results show
distinct per-class coefficient groupings across the
rural-to-urban gradient with coefficients directly related to
the increasing level of development in a subwatershed. We
also developed a linear equation between the NLCD 92 DEVELOPED% and truth
percent impervious area. Results show a
relative accuracy of approximately 80 percent and a mean
absolute TIA% estimate error of approximately 2.0 percent
+/- 1.0 percent for both the Level 1 coefficients and the Level
2 coefficients. Results derived from the linear regression
model show a relative accuracy of 70 percent with a mean
absolute TIA% estimate error of approximately 2.0 percent +/-
1.0 percent. This suggests that a linear model can be used as
a rapid assessment tool to approximate TIA% from NLCD 92
data. Results are based on a spatial aggregation of pixels to
the subwatershed or "whole-area" scale and are most applicable
to "pour-point" models utilizing a single percent impervious
surface area parameter. The models reported here
have been tested only in the Mid-Atlantic region (USEPAFederal Region
3).
| Top | Home |