classification, separating developed land into four classes
(21-open space, 22-low-intensity, 23- medium-intensity, and
24-high-intensity) (Table 1). According to Homer
et al
. (2007),
these four developed classes are derived by establishing
thresholds within
NLCD
’s imperviousness data product. The
NLCD
developed classes include roads where impervious pix-
els intersect ancillary vector data from the U.S. Department
of Transportation’s Bureau of Transportation Statistics (
BTS
)
roads (US Department of Transportation Bureau of Transpor-
tation, 2016). The
NLCD
Level I developed class aggregates the
Level II classes 21, 22, 23, and 24 (Table 1).
Formal accuracy estimates and a qualitative assessment of
NLCD
against satellite imagery and aerial photography suggest
that
NLCD
does an effective job mapping developed lands,
with Level I user’s accuracy ranging from 73 to 74 percent
and producer’s accuracy ranging from 72 to 74 percent (2001
and 2006, respectively) (Wickham
et al
., 2013). A characteris-
tic of the
NLCD
dataset is that rural roads are not differentiated
from developed areas that are more traditionally considered
part of the urban footprint. This addition of rural roads to
core developed areas results in estimates of developed land
far higher than the actual area of urban land. Additionally,
commission errors of the developed class occur because the
road network includes primary and secondary roads that are
delineated three or more 30-meter pixels wide, and tertiary
roads including gravel tracks and logging routes. Most roads
are not wider than 90 meters, which means that
NLCD
often
over-represents actual road area. Road density is inconsis-
tent, with large differences between the western and eastern
United States stemming from the use of automated processes
in the west to reduce the mapping of dirt tracks through rural
areas. Finally,
NLCD
treats the road network as a static input
across the 2001, 2006, and 2011 map dates, unlike the rest of
the developed footprint that changes dynamically over time.
While including rural roads as part of the developed foot-
print may be desirable in some applications focused on exam-
ining all impervious lands (Bierwagen
et al
., 2010; Booth
et
al
., 2002; Powell
et al
., 2008; Weng, 2012; Yang
et al
., 2003),
end users often prefer a finer level of detail and autonomy to
distinguish between urban development, rural development
and roads (Brown
et al
., 2005; Endreny and Thomas, 2009;
Hilferink and Rietveld, 1999; Leyk
et al
., 2014; Theobald,
2014). In select cases, the aforementioned problems with
how roads are represented in
NLCD
have required researchers
working on precise land use assessments and reliable empiri-
cal models to edit out roads to create quality developed maps
as an input (Schwarz
et al
. 20001; Sleeter
et al
., 2017; US
Environmental Protection Agency, 2017). In many of these as-
sessments and modeling efforts, the proportion of the devel-
oped footprint characterized by higher-intensity urban use is
of particular interest. To more accurately represent this urban
component of developed lands and better understand how
urban land changes over time, errors attributable to including
rural roads in the
NLCD
developed class must be resolved.
Other notable, large-scale attempts have been made to
improve
NLCD
developed classes by adding more developed
classes and resolving classification problems (Claggett
et al
.,
2013; Falcone, 2015; Theobald, 2014). For example,
USGS
NAWQA
Wall-to-Wall Anthropogenic Land Use Trends (
NWALT
)
is derived from a multi-tiered process incorporating ancillary
data to combine land use products and to modify
NLCD
by
thinning rural roads (Falcone, 2015). These modified products
fix some errors associated with the inclusion of roads in the
NLCD
products (Endreny and Thomas, 2009) but neither pro-
vide a comprehensive, consistent product representative of
urban land cover with corresponding accuracy estimates, nor
describe a benchmark method that can be replicated.
The objective of our study is to generate more reliable maps
of urban lands in the United States by removing rural roads
from the
NLCD
developed class for each map date while retain-
ing developed lands represented by urban infrastructure and
urban roads. The road removal process relies on merging
NLCD
developed maps with readily available geospatial data, apply-
ing multiple filtering processes, and performing automated and
manual editing to correct the urban footprint. We create urban
maps for
CONUS
for four dates (1992, 2001, 2006, and 2011).
Finally, an accuracy assessment is performed on the newly cre-
ated urban maps and the
NLCD
maps for the years 2001 and 2006
to evaluate how well the different products capture urban lands.
The new urban maps provide accurate spatially explicit data on
urban land and urban change critical to research on the causes
and consequences of urban expansion across the country.
Materials and Methods
Removing Rural Roads from NLCD
In general, the removal of rural roads from the
NLCD
devel-
oped classes was achieved by applying several neighborhood
filtering operations and various manual and automated editing
procedures to resolve omission and commission errors (Figure
1). We did not attempt to remove roads from urban areas be-
cause roads within high-density development are considered
part of the urban class. The first step in the process was to
aggregate the 2001
NLCD
(Homer
et al
., 2007) Level II devel-
oped classes into just one class (Level I developed) to provide
a base from which to identify urban areas (Level I and Level
II are based on Anderson (1976)). Unlike the 1992
NLCD
map
(Vogelmann
et al
., 2001), the 2001
NLCD
map is the first year
that roads were included in the developed classes. The Level
I developed class provided an ideal starting point for delineat-
ing urban areas because the four Level II developed classes
coalesce into larger clumps or contiguous groups of pixels
when viewed as one thematic layer (Figure 2a). Many of these
clumps were clearly urban areas where roads, residential,
commercial, and industrial areas had merged. The
NLCD
Level
I developed class was then intersected with a circa-2000
BTS
roads dataset (US Department of Transportation Bureau
of Transportation, 2016) to identify which
NLCD
developed
pixels corresponded to roads defined by
BTS
, and to include
Table 1. USGS National Land Cover Database level II developed classes.
NLCD Class NLCD Class Definition
21
Developed, Open Space - areas with a mixture of some constructed materials, but mostly vegetation in the form of lawn grasses.
Impervious surfaces account for less than 20% of total cover. These areas most commonly include large-lot single-family housing
units, parks, golf courses, and vegetation planted in developed settings for recreation, erosion control, or aesthetic purposes.
22
Developed, Low Intensity - areas with a mixture of constructed materials and vegetation. Impervious surfaces account for 20%
to 49% percent of total cover. These areas most commonly include single-family housing units.
23
Developed, Medium Intensity - areas with a mixture of constructed materials and vegetation. Impervious surfaces account for
50% to 79% of the total cover. These areas most commonly include single-family housing units.
24
Developed High Intensity - highly developed areas where people reside or work in high numbers. Examples include apartment
complexes, row houses and commercial/industrial. Impervious surfaces account for 80% to 100% of the total cover.
102
February 2018
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
