ar-
hs.
Ry,
1a
of
ges
ize
ous
[2]
ues
rval
Was
rate
are based on data acquired during 1994 and 1995. The
values of the OLS data for this data set, designated as
OLS-pct, were computed as the ratio of the number of
scenes in which light was observed at a given pixel to the
number of cloud free scenes for that pixel (Elvidge et al.,
1997). :
The second DMSP-OLS data product was based on
digitally calibrated (to units of Watts/cm?/sr/um) OLS data
acquired over a portion of North America from 16-23
March 1996.. The OLS calibrated product, designated
"OLS-cal' and OLS-pct data were remapped from the
Interrupted Goode Homolosine projection to the LAEA
projection of the AVHRR data.
Landsat MSS LULC Classified Data
For purposes of validation of the above parameters, it is
necessary to have an objective characterization of
urbanized area in order to facilitate spatial correlation.
Landsat MSS images, prepared for a North American
Landscape Characterization project (USGS, 1997) were
obtained for the DFW region. The MSS data were also
remapped, from a UTM projection to the LAEA
projection of the AVHRR data. Additionally, the MSS
data were resampled to a resolution of 50 m? to precisely
fall within the larger 1 km? grid cell resolution of the
AVHRR and DMSP-OLS data (i.e., 400 MSS 50 m! grid
cells per 1 km" AVHRR or DMSP-OLS grid cell). While
the MSS scene included most of the Dallas-Ft. Worth
metropolitan region, only two of the climate stations were
located within the borders of the MSS scene.
The classification scheme of Anderson et al. (1976) was
applied to a Landsat MSS image of the Dallas-Ft. Worth
region for a MSS scene acquired on 8 October 1992. The
resultant classes derived included agricultural, urban,
water, forested, rangeland, and bare soil, as well as an
unclassified class.
The procedure for image classification began by first
inputting four channels of the resampled Landsat MSS
data (0.5-0.6; 0.6-0.7; 0.7-0.8; and 0.8-1.1 um) into an
unsupervised classification procedure. Twenty-four
spectrally separate clusters were subsequently output with
accompanying covariance matrices. Heterogeneous
clusters with large off-diagonal covariance values were
merged into the unclassified class. The other clusters
were merged into the six Anderson Level I classes using
ground truthing guidance from the USGS 200 meter
Anderson Level II data set (USGS, 1990).
Using the Anderson classified Landsat MSS data, subpixel
percentages of each class were aggregated to the same 1
km? grid used in deriving AVHRR and DMSP-OLS
parameters. The aggregation for each 1 km? grid cell was
determined as:
Class percentage — 100 (Number of MSS 50 m! class
pixels / 400) .
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
Classes were defined as predominant within a 1 km? grid
cell when they were at least 5094 present in the grid cell.
RESULTS
A 110 km transect that intersected the DFW region was
examined in greater detail (Figure 1). The transect is 3 km in
width and lies west to east through the DFW region. The
transect passes through Fort Worth approximately between
the 30 to 50 km interval on the transect, and Dallas between
the 60 to 100 km interval. The dips in the NDVI data at 72
km and again between 101 and 103 km along the transect are
associated with a river and reservoir, respectively. The dip in
the OLS data at the 80 to 89 km interval on the transect is
associated with a city park. Generally the NDVI values are
lower within the portion of the transect (20 to 90 km) that is
indicated as urban within the OLS classified MSS data. The
urban and rural relationships between the NDVI and DMSP
data suggest that the urban-rural differences in DMSP data
may provide a valuable tool for the assessment of the urban
heat-island effect.
It is noteworthy that, compared to the NDVI values in
Figure 1, the OLS values seem to better describe
neighborhood-scale details that are verifiable with the high
resolution classified Landsat MSS data. In an analysis of
the transect (Figure 1) data, the OLS-cal values were
associated with 10% more (28%) of the variation in the
percentage urban values (derived from the classified MSS
data) than the NDVI values (18%). Thus, at the 1 km?
resolution at which AVHRR and DMSP-OLS data can
readily be obtained, the results suggest that the OLS data
better characterize the location of urban related features
in well-lit cities and metropolitan areas.
The ratio of Tsfc to NDVI (derived from the AVHRR
data), displayed as a function of NDVI are displayed in
Figure 2. NDVI values associated with climatological
stations in urban environments typically are lower when
compared to the values associated with rural stations. Tsfc
values usually display the opposite trend such that the
values associated with stations in urban environments are
greater compared to those of rural stations. Stations that
have large values of the ratio of Tsfc to NDVI (large
values of Tsfc and small values of NDVI) would be
expected to be located in urban environments.
The AVHRR-derived Tsfc and NDVI data available for
all seven stations were used to assess the association of
the stations with the predominant Landsat-MSS derived
classes within the study area. The ratio of Tsfc to NDVI
for the seven climatological stations was compared to the
values obtained for those grid cells associated with
precominantly urban, agricultural and forested cells within
the MSS scene. The ratio of Tsfc to NDVI for station 285
is similar to that associated with the grid cells classified as
overwhelmingly urban (> 80% urban), while stations 280
and 282 display Tsfc/NDVI values similar to the
predominantly forested/agricultural classes (> 50%
forested or agricultural). Stations 279, 281, 283, and 284
411
