Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.	EPA/600/A-04/084
9.3 HIGH-RESOLUTION DATASET OF URBAN CANOPY PARAMETERS FOR HOUSTON, TEXAS
*Steven J. Burian1, Stephen W. Stetson2, WooSuk Han1, Jason Ching3, and Daewon Byun4
department of Civil & Environmental Engineering, University of Utah, Salt Lake City, Utah
2Global Environmental Management, Inc., Camden, Maine
3ASMD, ARL, NOAA, on assignment to U.S. Environmental Protection Agency
institute for Multi-Dimensional Air Quality Studies (IMAQS), University of Houston, Houston, Texas
1.	INTRODUCTION
Mesoscale meteorological, urban dispersion and air
quality simulation models applied at various horizontal
scales require different levels of fidelity for specifying
the characteristics of the underlying surfaces. As the
modeling scales approach the neighborhood level
(~1 km horizontal grid spacing), the representation of
urban structures and surface cover properties requires
much greater detail. To provide the most accurate
surface characterization possible for an air quality
modeling study of Houston, Texas, airborne LIDAR
(Light Detection and Ranging) data were obtained at 1-
m horizontal grid cell spacing for Harris County, Texas,
an area of approximately 5800 km2. The gridded
dataset of full-feature elevation data was processed
using GIS analysis techniques to determine more than
20 urban canopy parameters (UCPs) including building
height statistics and histograms, height-to-width ratio,
plan area density function, frontal area density function,
roughness length, displacement height, mean
orientation of streets, and sky view factor. In an effort to
improve the efficiency and accuracy of the roughness
length derivation, an alternative gridded dataset of
roughness length was produced using satellite data
collected by Synthetic Aperture Radar (SAR)
instrumentation. The comparison of the SAR and
morphometric (LIDAR) roughness lengths suggested an
integration of the satellite and airborne LIDAR datasets
may provide an efficient means to derive a more
accurate roughness length gridded dataset. In this
paper, we describe the high-resolution Houston UCP
dataset, report on the variability of the UCPs across the
Houston urban terrain, and present the comparison of
the morphometric and SAR roughness lengths.
2.	BACKGROUND
Urban canopy parameterizations have been used to
represent urban effects in numerical models of
mesoscale meteorology, the surface energy budget, and
pollutant dispersion.	The urban canopy
parameterization accounts for the drag exerted by urban
roughness elements, enhanced production of turbulent
kinetic energy, and alteration of the surface energy
budget (Brown 2000). Accurate representation of urban
effects in numerical simulations using urban canopy
parameterizations requires the determination of surface
* Corresponding author address'. Steve Burian, 122 S.
Central Campus Dr., Suite 104, Salt Lake City, UT
84112; E-mail: burian@eng.utah.edu
cover and geometric parameters describing the urban
terrain (e.g., building height and geometry
characteristics).
A handful of researchers over the years have
pioneered the work on obtaining surface cover and
morphological parameters for cities (e.g., Ellefsen 1990;
Grimmond and Souch 1994; Voogt and Oke 1997;
Cionco and Ellefsen 1998; Grimmond and Oke 1999).
These studies have provided much useful information
on building and vegetation parameters, focusing mostly
on residential areas and, for a few cities, industrial and
commercial areas as well. Past work involved detailed
in situ studies, using visual surveys in an area
encompassing a few city blocks and extrapolating the
results to the entire city. With the recent availability of
digital 3D building and vegetation datasets and high-
resolution imagery, calculation of morphological and
surface cover parameters has become automated using
image processing and geographical information system
(GIS) software allowing larger areas to be analyzed
much more efficiently (e.g., Ratti and Richens 1999;
Ratti et al. 2001; Burian et al. 2002; Long et al. 2003).
3. HOUSTON URBAN TERRAIN DATABASES
The CMAQ/MM5/DA-SM2-U modeling system was
chosen for this neighborhood scale air quality modeling
project. The modeling domain is centered on the
Houston metropolitan area in southeast Texas (see
Figure 1) and encompasses an 82,368-km2 area
covered by approximately two-thirds land surface and
one-third water surface (primarily the Gulf of Mexico).
The land use/land cover (LULC) for the modeling
domain is based on the GIRAS LULC dataset for the
conterminous U.S. at 1:250,000 available from the U. S.
Environmental Protection Agency (EPA) and U.S.
Geological Survey (USGS) (see Figure 2). The land
use and cover information represented in the GIRAS
LULC dataset dates to the late 1970s and early 1980s;
therefore, to better represent current conditions the
dataset was updated using high-resolution aerial
photographs dating to 2000. Overall the land surfaces
of the modeling domain are predominantly rural,
consisting of significant fractions of Cropland & Pasture
and Forest Land. The highest concentration of urban
land use is the Houston metropolitan area located at the
left center of Figure 2.
Most of the UCPs were derived by processing an
airborne LIDAR full-feature digital elevation model
(DEM) obtained from TerraPoint, LLC. LIDAR
technology produces x, y, z representation of
topography via airborne lasers. TerraPoint provided the
1

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
DEM as a distribution of data points according to an
evenly spaced grid at 1-m resolution (i.e., a raster
dataset). Horizontal accuracy of the DEM was 15 to 25
cm, while the vertical accuracy was approximately 10
cm. The DEM represented all terrain elements including
buildings and vegetation, but did not differentiate
between elements.
roughness was Synthetic Aperture Radar, or SAR, data
collected from the ASAR instrument aboard the
European Space Agency's satellite, ENVISAT. The
ASAR instrument is an all-weather, day-and-night, high-
resolution imaging instrument that provides radar
backscatter measurements indicative of terrain
structure, surface roughness, and dielectric constant.
The surface roughness measured by the ASAR
instrument is defined as the variation of surface height
within an imaged resolution cell (see Figure 3). A
surface appears rough to microwave illumination when
the height variations become larger than a fraction of
the radar wavelength. The fraction is qualitative, but
may be shown to decrease with incidence angle. The
SAR data was processed to remove anomalies due to
cross-scene illumination differences due to incidence
angle as well as to remove anomalies associated with
data spikes and drop-outvalues.
side-looking geometry
spacecraft
Figure 1. Houston metropolitan area vicinity map and
modeling domain. The inner grid of the modeling
domain is shown as a red box.
spacecraft
altitude
Dallas - Fort Worth
San Antonio
Houston

0	100 200 400 Kilometers
1	' i i I i i i I
slant range to
different scatterers on
the ground
ground range
0 25 50 100 Kilometers
Land Use
| Urban
Agriculture
I Bare Soil arid Rock
| Forested
pj Rangeland
Wetland
| Water
Gulf of Mexico
Figure 2. Land use and land cover of modeling domain.
The LIDAR DEM only covered 5800 kmz of the
modeling domain, the left central part of the domain
containing the Houston metropolitan area. At 1-m
resolution the uncompressed dataset in its various
forms was immense (accumulated it was hundreds of
gigabytes) and caused numerous data management
and processing issues working as described by Burian
et al. (2003), To define the UCPs outside of the DEM
coverage an extrapolation approach was devised based
on the correlation between the UCPs and land use type
within the 5800 km2 DEM coverage. The extrapolation
method and the accuracy of the results are described
later.
One other dataset used for the derivation of surface
transmitted and received signal
received echo
'transmitted
pulse
time delay
(slant range)
Figures. Illustration of SAR data collection concept.
For this project UCPs were defined for full canopy
morphology (buildings and vegetation), building-only
morphology, and vegetation-only morphology.
Therefore, the raster DEM had to be intersected with
another dataset that would differentiate between
buildings and vegetation. The dataset chosen to
differentiate buildings and vegetation was a digital
building footprint layer obtained from the City of Houston
(COH). The COH building footprint dataset is based on
1983 aerial photographs with small updates in the mid
1990s, not adequate for this project since the LIDAR
data represent November 2001 conditions. The COH
building dataset was therefore modified by overlaying it
onto a series of high-resolution aerial photos covering
the Houston metropolitan area collected in 2000.
Buildings shown in the aerial photo but not contained in
the COH dataset were added and buildings in the COH
dataset not shown in the aerial photos were removed.
This tedious process was performed using the ESRI
2

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
ArcGIS software package for a 1653 km2 area
containing most of the defined Houston metropolitan
area and large tracts of outlying rural, forested, and
agricultural regions. The original COH dataset
contained 523,920 building footprints, while the modified
dataset contains 664,861 building footprints. The
building footprints defined the DEM cells representing
buildings and all other DEM cells were assumed to be
vegetation or non-building structures (e.g., roadway
overpasses). Further details about the datasets used to
derive the UCPs are provided by Burian et al. (2003).
4. URBAN CANOPY PARAMETER DATASET
This project required the calculation of the following
canopy, building, vegetation, and other UCPs:
Canopy UCPs:
•	Mean canopy height
•	Canopy plan area density
•	Canopy top area density
•	Canopy frontal area density
•	Roughness length
•	Displacement height
•	Sky view factor
Building UCPs:
•	Mean building height
•	Standard deviation of building height
•	Building height histograms
•	Building wall-to-plan area ratio
•	Building height-to-width ratio
•	Building plan area density
•	Building rooftop area density
•	Building frontal area density
Vegetation UCPs:
•	Mean vegetation height
•	Vegetation plan area density
•	Vegetation top area density
•	Vegetation frontal area density
Other UCPs:
•	Mean orientation of streets
•	Plan area fraction surface covers
•	Percent directly connected impervious area
•	Building material fraction
Calculation procedures were based on the GIS
approach described by Burian et al. (2002). The mean
orientation of streets was determined for each grid cell
within the modeling domain, the canopy only UCPs
were determined for the 5800 km2 DEM coverage, the
building and vegetation specific parameters were
determined for the 1653 km area covered by building
footprints. Within the 5800 km2 DEM coverage and
1235 km2 of the 1653 km2 building footprint zone the
average UCP values for each USGS Level 2 land use
type were determined (the other 418 km2 of the 1653
km2 building footprint zone was used for extrapolation
validation as described below). The USGS land use in
these regions was updated based on the year 2000
aerial photographs. The mean UCP values per land use
type were then extrapolated to areas outside the data
coverage using an area-weighted average based on
underlying land use amounts within each grid cell.
The accuracy of the extrapolation process was
assessed for most of the UCPs for a 418 km2 area
within the 1653 km2 building footprint coverage. The
relative similarity of the 418 pairs of "calculated" and
"extrapolated" UCPs was measured using bias, root
mean square error (RMSE), and cumulative relative
error (CRE) statistics and visualized using scatter plots:
Bias = -Y(uCPi - UCp)	(1)
^ 7 = 1
RMSE = J (lJCI>! - UCPt )2	(2)
Y\ucp7 -ucp\
CRE = !00 •—		(3)
fucFi
Z=1
where UCP, is the extrapolated UCP for the ith grid cell,
UCPi is the calculated UCP for the ilh grid cell, and n is
the number of grid cells (418). Table 1 contains the
accuracy assessment statistics for the building height
UCPs. On average the extrapolation will produce higher
building height parameters, with the most significant
error associated with the standard deviation of building
height estimate. Table 2 contains a comparison of the
average UCPs per land use type calculated in the
derivation (1253 km2) area and the validation area (418
km2). The Residential and Industrial land use types
have the greatest differences.
Table 1. Building height UCP accuracy assessment
statistics.

Bias
RMSE
CRE
Mean Building Height
1.02
(+23%)
1.64
28%
Standard Deviation of
Building Height
1.47
(+75%)
1.84
83%
Footprint Area-Weighted
Mean Height
1.15
(+23%)
2.39
34%
Wall-to-Plan Area Ratio
0.009
(+9%)
0.05
31%
Height-to-Width Ratio
0.004
(+11%)
0.02
37%
Scatter plots of extrapolated versus calculated
UCPs were created using data for the 418 grid cells in
the defined validation area. These plots were meant to
identify limitations of the extrapolation process and
define future enhancements. Figure 4 contains the
scatter plot for mean building height. One readily
3

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
apparent observation is the "floor" limiting the lower end
of the extrapolation values. This observation was
expected because the use of the average value
prevents the accurate prediction of extremely low or
high values in the distribution of parameter values.
Possible remedies to this limitation are to incorporate (1)
other base data layers into the extrapolation process
that can be correlated to morphological parameters
(e.g., population) or (2) the variability of the parameter
into the extrapolation. Both of these remedies are being
evaluated in other work.
The summary statistics (bias, RMSE, and CRE)
were determined and scatter plots created for most of
UCPs included in the dataset. The complete results of
the assessment are too extensive to include in this
paper, but are summarized in more detail by Burian et
al. (2003). More will be described in the presentation.
Several noteworthy observations regarding the UCP
gridded dataset for Houston are:
•	All UCPs were over-estimated on average by the
extrapolation, except the UCPs that were
determined as a function of height.
•	Upper and lower limits to the estimated UCP
values caused by the extrapolation were noted
for many of the UCPs (as was described above
for the mean building height)
•	Morphometric roughness length extrapolation
was less accurate than the morphometric
displacement height extrapolation. This is initially
addressed below with an alternative method to
derive roughness length using satellite data and
continues to be addressed in on-going work.
•	The range of errors is significant and in some
cases the errors resulting from the extrapolation
can be quite large, but on average the
magnitudes of the errors are moderate (within the
30-50% range predominantly).
The final UCP dataset for the Houston modeling domain
contained more than 80 million parameter values
corresponding to grid cells with specified coordinates.
Overall, the UCPs contained in the gridded dataset were
concluded to be more accurate and representative of
surface properties than could have been obtained by
simply using literature values correlated to land use type
and have been shown to improve model results (e.g.,
Dupont et al. 2004).
15
¦c 10
5
0
0
5
10
15
Calculated
Figure 4. Scatter plot of extrapolated versus calculated
mean building height (m) for the 418 grid cells in the
validation area.
5. UCP VARIABILITY
The initial Houston UCP dataset extrapolation using
the mean UCP value correlated to land use provides
reasonable estimates, but improvements can be made.
In an effort to develop an improved method to perform
the extrapolation, the variability of the UCPs for the land
use types was quantified. The motivation for this effort
was to use the quantified UCP variability as a function of
land use type in an enhanced extrapolation that
incorporated the variability using a stochastic approach.
The first step of quantifying the variability is reported
here.
The same updated land use and building datasets
covering the 1653-km2 area that were used to derive the
gridded UCP coverage described above were also used
to quantify the parameter variability for selected land
use types (see Figures 5 and 6). The study area has a
large fraction of Residential land use (38%), with smaller
amounts of Commercial & Services, Industrial, Cropland
& Pasture, and Forest Land.
The UCP variability was determined for the
following seven land use types: Residential; Commercial
& Services; Industrial; Transportation, Communication,
Utility; Other Urban or Built-up; and Cropland & Pasture.
These land uses were selected because they had
sufficient coverage in the study area to provide
adequate samples to quantify variability. The land use
dataset of the study area is divided into 5779 polygons,
each representing one land use type. Of the 5779 total
polygons, 5447 represent the seven land use types
included in the study. Table 3 lists the summary
characteristics of the 5447 polygons.
Table 2. Comparison of mean building height per land
use type in the derivation and validation zones.	
USGS Level 2 Land
Use Name
Mean
Building
Height (m) -
Derivation
Zone
Mean
Building
Height (m) -
Validation
Zone
Residential
5.70
4.74
Commercial &
Services
6.05
5.85
Industrial
6.09
4.95
Transportation,
Communications,
Utility
4.81
4.17
Other Urban or Built-
Up Land
4.95
4.68
Cropland & Pasture
5.02
4.94
Deciduous Forest
Land
7.32
5.67
4

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
UCP Variability
Study Area
50 Kilometers
Figure 5. Location and extent of UCP variability study
area.
Figure 6. Land use of UCP variability study area.
Table 3. Summary characteristics of the land use
All seven land use types have a sufficient number
of samples to derive meaningful statistics and perform
scientific visualizations to quantify and express the
relative variability of the UCP values. The majority of
land use samples are Residential, which also have the
largest average size. The Other Urban or Built-up land
use has the smallest average polygon size. The
minimum land use sample size is slightly smaller than
0.5 hectares, while the largest size is 1268 hectares.
The 1268 hectare polygon is a relatively unique sample
because it represents the Bush Intercontinental Airport.
For each land use sample polygon, the set of
UCPs were calculated and the variability of the UCPs
was then quantified for each land use type by
calculating a series of statistics representing the central
tendency and spread of the distribution of values. The
following UCPs were included in this analysis:
•	Number of buildings per hectare
•	Mean building height
•	Standard deviation of building height
•	Plan-area-weighted mean building height
•	Plan area fraction of buildings
•	Building frontal area index
•	Building height-to-width ratio
•	Roughness length (simple & Raupach) - based
on buildings and canopy
•	Displacement height (simple & Raupach) - based
on buildings and canopy
•	Mean canopy height
•	Standard deviation of canopy height
•	Plan area fraction of canopy
Summary statistics for the UCP values for each
land use type were determined. Table 4 contains the
statistics for the Residential land use type. The
summary of the other UCPs and land use types are too
extensive to include in this paper; complete results are
available in Burian (2004). The summary statistics for
all land use types indicated three general observations:
•	The coefficient of variation (Cv) suggests the
Residential land use has the smallest UCP
variability of the seven land uses studied, while
the Transportation, Communication, Utility land
use has the largest.
•	UCP values equal to zero indicated no buildings



10 Kilometers
Land Use
Residential
m Commercial & Services
Industrial
Mixed Industrial & Commercial
Transportation, Communication, Utility
Mixed Urban Or Built-up
Other Urban Or Built-up
| Cropland & Pasture
|B| Forest Land
Rangeland
Wetland
m Non-vegetated Open Space
¦¦ Water
polygons included in the UCP variability analysis.
Land Use
Number
Avg.
Size
(ha)
Median
Size
(ha)
Residential
1450
44
29
Commercial & Services
762
15
8
Industrial
690
23
11
Transportation
290
21
12
Other Urban Or Built-up
818
11
7
Cropland & Pasture
762
39
17
Forest Land
675
33
16
5

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
(or canopy) were present in the sample polygon.
This was uncommon for the Residential,
Commercial & Services, and Industrial
categories, but was common for the Other Urban
or Built-up, Transportation, Cropland & Pasture,
and Forested land use types.
• Of the three primary urban land use types
(Residential, Commercial & Services, and
Industrial) the Commercial & Services had the
highest UCP variability.
Table 4. Summary UCP statistics for Residential land
UCP
Mean
Median
Standard
Deviation
Cv
No. of Buildings
per hectare
9.01
8.64
5.12
0.57
Mean bldg ht. (m)
5.45
5.42
1.13
0.21
St. dev. bldg. ht.
(m)
2.24
2.18
1.02
0.46
Plan-area-
weighted mean
bldg. ht. (m)
5.32
5.26
1.10
0.21
Bldg. Xp
0.17
0.17
0.10
0.59
Bldg. ht-to-width
0.05
0.05
0.02
0.40
Bldg. XF
0.07
0.07
0.04
0.52
Roughness length
(simple; bldgs) (m)
0.55
0.54
0.11
0.21
Displacement ht.
(simple; bldgs) (m)
2.72
2.71
0.56
0.21
Roughness length
(Raupach; bldgs)
(m)
0.32
0.31
0.18
0.56
Displacement ht.
(Raupach; bldgs)
(m)
2.01
2.01
0.69
0.35
Mean canopy ht.
(m)
6.38
6.20
1.76
0.28
St. dev. canopy ht.
(m)
4.46
4.43
1.19
0.27
Canopy Xp
0.61
0.62
0.11
0.19
Roughness length
(Simple; Canopy)
(m)
0.64
0.62
0.18
0.28
Displacement ht.
(Simple; Canopy)
(m)
3.19
3.10
0.88
0.28
The first observation would be expected in most
cities because residential development is guided by
fairly uniform building codes and builders will typically
follow a small number of plans when constructing new
houses in a subdivision. The uniformity of most
subdivisions significantly reduces the UCP variability
from one location in a subdivision to another, but also
limits the variability from one residential location to
another. The third observation also would be expected
in other cities because Commercial & Services is a
broader definition and will not contain as uniform
building types and site layouts as Residential and
Industrial land uses. Direct comparison of the
coefficients of variation of each UCP for all land uses
was conducted by extracting the numbers from the
tables and creating bar chart plots. Figure 7, for
example, shows the plot for the mean building height
UCP (all plots available in Burian (2004)). The plots
support the observations derived from the analysis of
the UCP summary statistics for each land use type.
Res. Comm. & Ind. Trans. Other Crop Forest
Serv.	Urban
Figure 7. Plot of coefficients of variation for the mean
building height UCP.
6. ROUGHNESS LENGTH DERIVATION
The representation of surface roughness is a critical
first step in many meteorological, wind engineering, and
atmospheric dispersion modeling activities. It provides
an estimate of the drag and turbulent mixing associated
with the underlying surface. The roughness length (z0)
is a key parameter in the logarithmic velocity profile
based on similarity theory and is commonly used in
many models to specify boundary conditions above
built-up areas. The roughness length is directly related
to the overall drag of the surface. Mathematically, it
represents the distance above the displacement height
plane at which the velocity goes to zero.
z0 is difficult to estimate with certainty by
experiment or theory. Grimmond and Oke (1999)
reviewed methods to calculate the z0 of urban areas
based on building and vegetation morphology. They
compared the predictions of the morphological methods
to those obtained from wind measurements in urban
areas and found significant differences. However,
collecting and analyzing wind measurements to
determine the roughness length for a large number of
model grid cells covering a heterogeneous urban area is
not practical. Even collecting measurements for
representative urban land use types is not feasible for
most modeling projects. Methods must be developed
that can efficiently and accurately produce gridded
coverage of roughness parameters.
For the Houston UCP database five morphological
estimates of z0 were used and three were compared to
investigate the relative differences between the
6

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
methods. Unfortunately, there is no "true" roughness
length dataset to make comparisons to determine which
method is the most accurate and under what
circumstances. Manual interpretation of the relative
differences can however provide useful observations
and potential recommendations for deriving more
accurate roughness parameter data layers.
The first morphological technique used to derive the
roughness length coverage for the 82,368-km2 inner
modeling domain is a simple approach that defines the
roughness length to be one-tenth of the mean height of
roughness elements (buildings and vegetation). The
second technique is based on morphometric equations
introduced by Raupach (1994):
i^ = l-
1 exp[ (c 2i,) ']
icdl2Aff5
(4)
and
£^ =
ZH
where
f z\ f
1 - =
V
exp
-H
/ U
-k— + Wk
it.
\
(5)
/
K
U
= min
(6'.v + CR^f}"
U
(6)
and Zd is the displacement height, zH is the mean
canopy height. If is the frontal area index, y/k is the
roughness sublayer influence function, U and u< are the
large-scale wind speed and the friction velocity,
respectively, cs and cR are drag coefficients for the
substrate surface at height zh in the absence of
roughness elements and of an isolated roughness
element mounted on the surface, respectively, and Cdi is
a free parameter. Raupach (1994) suggested i//* =
0.193, (U'/U)max = 0.3, Cs = 0.003, Cr = 0.3, and Cdi =
7.5. Using these values, a von Karman constant (k) of
0.4, and the values computed for the mean building
height and the frontal area index for a north wind
direction the roughness length was determined.
The third approach computes the surface
roughness using data collected from satellite using
Synthetic Aperture Radar (SAR) instrumentation
(Stetson 2004). The SAR sensors provided backscatter
measurements indicative of surface roughness at 150 m
resolution. Following the data calibration, a surface
roughness coefficient for each ground sample, or pixel,
was generated based on the range of SAR values within
each separate LULC class. The roughness coefficient
for each pixel was then applied to the standard
roughness-length value associated with each LULC
class using a look-up-table derived from peer-reviewed
literature sources. For each 150 m pixel, the roughness
coefficient was generated as the quotient of the
individual pixel's SAR value divided by the mean SAR
value for its respective LULC class. Within each LULC
class the roughness coefficient was used to derive the
roughness length:
z- = RG * z
(7)
where RC is the roughness coefficient and z0« is a
standard roughness length for the given land use or
cover class obtained from the literature. The final step
was to resample the 150 m resolution roughness to 1-
km to correspond to the model domain. It should be
noted that the derivation of the SAR roughness
coverage used a newer and improved LULC dataset
than that used to derive the other roughness length
values, although the differences in the dataset are not
significant in the metropolitan area.
The study area for comparison of the derived
roughness lengths was centered on Houston, Texas
(see Figure 8). The study area encompasses eight
counties of the Texas Gulf Coast covering an area of
approximately 23,100 km2. Note that Harris County
shown in the figure contains the Houston metropolitan
area.

, Chambers
Fori Brno
Eight County
Study Area
Figure 8. Location of roughness length calculation
comparison study.
The three roughness length coverages were first
compared by calculating summary statistics for the
entire study area (see Table 5), The DEM methods
(simple ratio and Raupach) have mean and maximum
values that are approximately two times the values for
the SAR method. The minimum's are all the same
(nearly zero) and are correctly assigned to water and
flat non-vegetated areas by the three approaches.
Interestingly, the measure of variability of values across
the study area (standard deviation) is identical for all
three estimation techniques. The maximum values are
important because for both the simple method and the
Raupach method the values correspond to the location
of the downtown core area, while the maximum values
for the satellite data do not.
Figure 9 shows histograms of the roughness length
values for the three estimation methods. In this form,
one notes the weight of the distribution of the values
based on the satellite estimation method is towards 0,0
- 0.5. In fact, 60% of the 1-km2 grid cells are estimated
from the satellite data to have a roughness length in this
7

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
range; while less than 20% of the grid cells fall within
this range for the simple method and the Raupach
method. More than 99% of the roughness lengths
estimated from the satellite data are less than 1.0, but
approximately 20% of the values estimated using the
DEM data (both methods) are greater than 1.0.



¦ Satellite
~ Simple


¦ Raupach
I

1
-
n

E
i

J


11


i
, r-n, ,
Roughness Length (m)
Figure 9. Comparison of roughness length histograms.
The second phase of the roughness length
comparison focused on smaller areas and specific
locations with selected land use samples. One location
of interest was the downtown core area of Houston.
The street network in the downtown area is angled off of
the north-south : east-west directions. The west end of
the downtown contains the dense high-rise buildings.
Equally as dense, yet shorter buildings occupy the east
and south ends. The major sports facilities are located
on the eastern end. The roughness lengths using the
three techniques for six 1-km resolution grid cells
covering the downtown area were computed. The
DEM-based roughness lengths (simple and Raupach)
are much higher in the tall building region of the
downtown than the satellite method. The satellite
roughness of 0.97 m for the grid cell containing the
tallest buildings is at the upper end of the range of
values for the satellite, suggesting that the satellite data
may be correctly identifying a location of a relative
maximum, although it does not identify it as the absolute
maximum in the dataset. The values predicted by the
DEM procedures (2.9 m) are more consistent with
values calculated using morphology and wind profile
measurements for the downtown core areas of other
cities (Burian et al. 2003).
Similar comparisons were made for several other
locations for multiple land use types (urban and non-
urban). A summary comparison of the roughness length
values estimated per land use and cover type was also
determined and is shown in Table 6. The synthesis of
results suggests that the gridded fields of roughness
lengths produced by the three techniques are
predominantly similar for Residential, Industrial,
Transportation, and Other Urban or Built-up land uses.
The most noteworthy difference corresponded to
Commercial land use types, while Cropland & Pasture
and Forested land uses also exhibited significant
differences.
Table 6. Comparison of roughness lengths per land use
7. SUMMARY
The project described in this paper involved the
processing of high-spatial resolution digital terrain
datasets using GIS and image processing software and
other computational tools. The objective was to derive
an accurate gridded set of urban canopy parameters for
use in the CMAQ/MM5/DA-SM2-U modeling system.
The first generation dataset has the following
characteristics:
•	16 UCPs required one value per grid cell; 82,368
grid cells [1,317,888 total values]
•	9 UCPs (Plan Area Densities, Top Area
Densities, and Frontal Area Densities) are given
as a function of height (one value per meter for a
range of 33 meters to 297 meters) for each grid
cell [~74,000,000 total values]
•	2 UCPs (Land Cover Fraction and Building
Material Fraction) have five values per grid cell
[823,680 total values]
Table 5. Summary statistics for entire study area
comparing roughness length calculation methods.	

Satellite
DEM-
DEM-

Simple
Raupach
Mean (m)
0.41
0.70
0.73
Standard
Deviation (m)
0.32
0.32
0.32
Minimum (m)
0.00
0.00
0.00
Maximum (m)
1.20
2.90
2.90
and cover type.
Land Use
Mean z0
(m)
Satellite
Mean z0
(m)
Simple
Mean z0
(m)
Raupach
Residential
0.72
0.72
0.80
Commercial &
Services
0.65
0.74
0.85
Industrial
0.47
0.62
0.65
Mixed Industrial &
Commercial
0.47
0.74
0.83
Transportation,
Communication,
Utility
0.41
0.69
0.74
Mixed Urban Or
Built-up
0.62
0.74
0.80
Other Urban Or
Built-up
0.67
0.69
0.76
Cropland &
Pasture
0.25
0.44
0.47
Forest Land
0.67
0.94
0.93
Rangeland
0.16
0.57
0.51
Wetland
0.25
0.42
0.44
Non-vegetated
Open Space
0.59
0.29
0.39
Water
—
—
—
8

-------
Fifth Symposium on the Urban Environment, American Meteorological Society,
Vancouver, BC, Canada 23-28 August 2004.
•	1 UCP (Building Height Histograms) has 62
values per grid cell (62 height increments)
[5,106,816 total values]
•	And the land use fraction has 29 values per grid
cell [2,388,672 total values]
The total number of UCPs in the first generation dataset
is more than 80 million, not including the multiple
roughness length values and other derivative products.
The UCP variability assessment indicated that the
Residential land use has the least amount of parameter
variability across the Houston metropolitan area, while
the Commercial & Services land use type had the
greatest. The comparison of the three roughness length
derivation approaches indicated that methods to
estimate roughness length have a significant degree of
variability and the accuracy of any method is
questionable because a gridded dataset of true values
does not exist for comparison. The new satellite
approach to estimating roughness lengths introduced by
Stetson (2004) was found to be a promising technique
because it produced comparable results to the methods
based on morphometric equations even using standard
values from the literature. Possible future
improvements to the satellite approach include
incorporating a calibration step and using city-specific
morphometric estimates of roughness lengths per land
use class in the extrapolation process. As future work
improves the accuracy of the UCP values, the Houston
database will be updated.
Disclaimer
The United States Environmental Protection Agency
through its Office of Research and Development
partially funded and collaborated in the research
described here. It has been subjected to Agency review
and approved for publication.
8. REFERENCES
Brown, M. J., 2000: Urban parameterizations for
mesoscale meteorological models. Mesoscale
Atmospheric Dispersion. Ed., Z. Boybeyi., pp. 193-
255
Burian, S. J., 2004: Urban canopy parameter
assessment for Houston, Texas. Research report
prepared for Daewon Byun, Institute for
Multidimensional Air Quality Studies (IMAQS),
University of Houston, Houston, Texas.
Burian, S. J., M. J. Brown, and S. P. Linger, 2002:
Morphological analysis using 3D building databases:
Los Angeles, California. LA-UR-02-0781. Los Alamos
National Laboratory, 66 pp.
Burian, S. J., W. Han, S. P. Velegubantla, S. R. K.
Maddula, 2003: Development of gridded fields of
urban canopy parameters for Models-3/CMAQ/MM5.
U.S. EPA Internal Report, RTP, North Carolina.
Cionco, R. M. and R. Ellefsen, 1998: High resolution
urban morphology data for urban wind flow modeling.
Atm. Env., 32(1), 7-17.
Dupont, S., J. Ching, and S. Burian, 2004: Introduction
of urban canopy parameterization into MM5 to
simulate urban meteorology at neighborhood scale.
Preprints, Symposium on Planning, Nowcasting, and
Forecasting in the Urban Zone, 84th AMS Annual
Meeting, Seattle, WA.
Ellefsen, R., 1990: Mapping and measuring buildings in
the canopy boundary layer in ten U.S. cities. Energy
and Buildings, 15-16, 1025-1049.
Grimmond, C. S. B. and C. Souch, 1994: Surface
description for urban climate studies: A GIS based
methodology. Geocarto Int., 9, 47-59.
Grimmond, C. S. B. and T. Oke, 1999: Aerodynamic
properties of urban areas derived from analysis of
surface form. J. Appl. Met., 38, 1262-1292.
Long, N., S. Kermadi, C. Kergomard, P. G. Mestayer,
and A. Trebouet, 2003: Urban cover modes and
thermodynamic parameters from urban database and
satellite data: A comparison for Marseille during
ESCOMPTE, Proceeding, Fifth Int. Conf. Urban
Climate, Lodz, Poland, September 2003.
Ratti, C. and P. Richens, 1999: Urban texture analysis
with image processing techniques, Proceedings of the
CAADFutures99 Conference, Atlanta, June 1999.
Ratti, C., S. Di Sabatino, R. E. Britter, M. J. Brown, F.
Caton, and S. Burian, 2001: Analysis of 3-D urban
databases with respect to pollution dispersion for a
number of European and American cities. Preprints,
Third Int. Conf. On Urban Air Quality, March 2001,
Loutraki, Greece.
Raupach, M. R., 1994: Simplified expressions for
vegetation roughness length and zero-plane
displacement height as functions of canopy height
and area index. Boundary-Layer Meteorology, 71:
211-216.
Stetson, S. W., 2004: Surface roughness and z0
parameter measured from satellite-based synthetic
aperture radar. Research Report. Global
Environmental Management, Inc.
Voogt, J. and T. Oke, 1997: Complete urban surface
temperatures. J. Appl. Met., 36, 1117-1132.
9

-------