Environmental Engineering Reference
In-Depth Information
The housing-unit counts from the 1990 and 2000 U.S.
Censuses (US Census Bureau, 1991, 2001a) were used to derive
HUD for each county subdivision within the South Atlantic divi-
sion in each of the two years separately. All county subdivisions
were separated into four categories of settlement densities includ-
ing urban (
shrubland, grassland, and agriculture categories (Table 19.2).
These documented values were used in estimating
ε
g
of BUILT-
UP and WETLAND with a few assumptions applied to each
category. For WETLAND,
ε
g
was calculated as the average of
ε
g
for DECIDUOUS FOREST,CONIFEROUS FOREST,MIXED FOREST,
SHRUBLAND,andGRASSLAND, assuming equal distribution of
wooded and herbaceous species given the absence of detailed
species composition information in wetland areas. For BUILT-
UP,
ε
g
was estimated as the area weighted average of tree, lawn,
impervious surface, and water.
ε
g
of impervious surface and
waterwereassumedtobezero.
ε
g
of tree was calculated as the
average of
ε
g
for DECIDUOUS FOREST and CONIFEROUS FOREST,
which is equal to 1.29 g C MJ
−
1
.
ε
g
of lawn was assigned as
the value for GRASSLAND. The area proportion of tree, lawn,
impervious surface and water for the BUILT-UP category was
estimated by examining percentage of canopy coverage and
impervious surface within areas identified as BUILT-UP in 2001.
Canopy coverage and impervious surface were extracted from
the NLCD 2001 Imperviousness and Tree Canopy data, which
provide the pixel-wise Landsat-based estimation of canopy and
impervious proportions at the 1-km resolution (Huang, Homer
and Yang, 2003; Yang
et al
., 2003). The average tree (or canopy)
cover was found to be 20.3%within the BUILT-UP area across the
South Atlantic division; the number was 21.5% for impervious
surface. The area percentage of lawn or water was calculated as
half of the rest area (i.e., 29.1% each). Therefore,
ε
g
of the BUILT-
UP type was estimated as the weighted average of tree (1.29 g
CMJ
−
1
; 20.3%), lawn (0.86 g C MJ
−
1
; 29.1%), and impervious
surface or water (0; 50.6%).
0
.
1 hectares per housing unit), suburban (0.1-0.69
hectares per housing unit), exurban (0.69-16.2 hectares per
housing unit), and rural (
>
16.2 hectares per housing unit) den-
sities following the definition of Theobald (2005). This yielded
maps of the four settlement densities for 1990 and 2000 with
Census county subdivisions as mapping units in size ranging
from 0.62 to 3069 km
2
(on average 214 km
2
or 21 437 hectares
per county subdivision). Both maps were exported to raster for-
mat with the spatial resolution of 100 m, and used in change
analysis that detects pixel-wise persistent or transitional status
concerning settlement densities in 1990 and 2000. To minimize
the sliver problem of GIS (Goodchild, 1978), the change data
were clumped by types of settlement-density conversion using
the four-neighbor rule and clumps smaller than 0.25 km
2
were
merged into nearby larger clumps.
There were 16 possibilities of settlement-density conversion
between 1990 and 2000, including four persistent types (i.e.,
Persistent Urban, Persistent Suburban, Persistent Exurban, and
Persistent Rural) and 12 transitional types. Some of the transition
was introduced by changes in boundaries of Census units between
the twoCensus dates (Goodchild, Anselin andDeichmann, 1993).
I attempted to minimize this type of transition, which obscures
the actual changes in HUD, by excluding county subdivisions
with area variation equal to or higher than 10%between 1990 and
2000. This resulted in a removal of 646 county subdivisions in the
2000 Census throughout the South Atlantic division, equivalent
to 19.4% of the total count of county subdivisions and 16.4% of
the total area of the South Atlantic division. Among the removed
units, 414 county subdivisions were located within Virginia and
West Virginia. These removed county subdivisions were masked
out for further analysis.
≤
19.3.3
Estimating APAR, GPP and
changes in GPP
To estimate GPP in 1992 and 2001, APAR was calculated for each
year separately as the product of PAR and fAPAR (Fig. 19.1).
PAR was the downward shortwave radiation, which came from
the monthly average climate forcing data based on the NOAH
land-surface model (Rodell
et al
., 2004), multiplied by a scalar
0.45. The 1-degree PAR data were re-sampled to 1-km resolu-
tion to be used jointly with vegetation maps and fAPAR data
prepared at this finer resolution for the estimation of GPP.
fAPAR was estimated based on the biweekly AVHRR NDVI and
maps of vegetation and land uses (compiled in the previous
step), using the look-up table approach (Knyazikhin
et al
., 1999).
The NDVI data were processed to minimize cloud cover, sen-
sor degradation, and atmospheric effects (USGS EROS Data
Center, 2006) before being applied to this study. Broadleaf for-
est, needleleaf forest, shrubland, and grassland/cereal crops in
Knyazikhin
et al
. (1999) approximated the DECIDUOUS FOREST,
CONIFEROUS FOREST,SHRUBLAND and GRASSLAND categories in
this study, whereas the NDVI-fAPAR look-up values for MIXED
FOREST was derived as the average between broadleaf forest and
needleleaf forest. AGRICULTURE was approximated as the average
of broadleaf crop and grassland/cereal crops; and WETLAND as
the average of broadleaf forest, needleleaf forest, shrubland, and
grassland/cereal crops. The NDVI-fAPAR look-up for the BUILT-
UP category was estimated as the area weighted average of MIXED
FOREST (20.3%), GRASSLAND (29.1%), and impervious surface
or water (50.6%), using the similar approach for the estimation
of
ε
g
for the BUILT-UP type.
19.3.2
Preparing vegetationmaps
and light-use efficiency parameters
The estimation of GPP based on LUE models requires data on
light-use efficiency (
ε
g
), solar radiation, vegetation indices, and
types of vegetation (Fig. 19.1). In this study, types of vegeta-
tion were inferred from the National Land Cover Dataset 1992
(Vogelmann
et al
., 2001) and National Land Cover Database
2001 (Homer
et al
., 2007), each providing the nationwide land-
cover/land-use classification close to Census 1990 and Census
2000. The National Land Cover Dataset/Database (NLCD) land-
cover/land-use classification was grouped into ten vegetation and
land-use categories (Table 19.2), each differing from the other
categories in terms of their ecological functions (Gower, Kucharik
and Norman, 1999). This resulted in a map of 10 vegetation and
land-use categories at the spatial resolution of 30 m across the
South Atlantic study area in 1992 and 2001, respectively. The
30-m vegetation and land-use data were aggregated to 1 km,
where the coarse pixel held proportions of individual vegetation
and land-use categories by area within that pixel.
To determine values of light-use efficiency for individual
vegetation and land-use categories, parameters documented in
previous research (Yang
et al
., 2007) were applied to the forest,
Search WWH ::
Custom Search