Mapping Annual Urban Change
Using Time Series Landsat and NLCD
Heng Wan, Yang Shao, James B. Campbell, and Xinwei Deng
Abstract
Annual urban change information is important for an im-
proved understanding of urban dynamics and continuous
modeling of urban ecosystem processes. This study examined
Landsat-derived Normalized Difference Vegetation Index
(
NDVI
) time series for characterizing annual urban change. To
reduce impacts from cloud contamination and missing data,
United States Geological Survey (
USGS
) Landsat Analysis
Ready Data were processed to derive annual
NDVI
layers using
a maximum value composite algorithm. National Land Cover
Database land cover products from 2001 and 2011 were used
as references for generating a decadal urban change mask.
Within the decadal urban change mask and using annual
NDVI
as input, we examined three time-series change detec-
tion methods to pinpoint specific year of urban change: (a)
minimum-value method, (b) break-point detection, and (c)
simple-threshold identification. For accuracy assessment,
we divided change pixels into urbanization and urban-inten-
sification pixel groups, defined by initial land cover types.
We used Google Earth’s High-Resolution Imagery Archive
as primary reference data for detailed accuracy assessment.
Overall, the urbanization pixel group has good change detec-
tion accuracies of above 82% for all three change detection
algorithms. The break-point detection method resulted in the
highest overall accuracy of 88%. Overall accuracies for urban
intensification pixel group were in the range of 35%–76%,
depending on choice of change detection algorithm, length
of input time-series, and further division of pixel subgroups.
Introduction
Land use and land cover change (
LULC
as a main driver of global environment
consequences of
LULCC
span a wide ra
domains such as local-regional climate, air and water qual-
ity, hydrological cycle, biogeochemical fluxes, biodiversity,
and food production (Messina and Walsh 2001; DeFries
et
al.
2004; Foley
et al.
2005; Yuan 2008; Roy and Srivastava
2012; Staudt
et al.
2013; Deng
et al.
2014; Maimaitiyiming
et
al.
2014; Muñoz-Rojas
et al.
2015). Although urban area only
covers a small percentage of the earth’s land surface, urban-
ization is probably the most intensive type of
LULCC
to alter
local and regional environments. Conversion from forest and
other natural landscapes to urban are typically irreversible
and can lead to many ecological problems including deterio-
ration of water and air quality (Kalnay and Cai 2003; Yuan
2008), biodiversity loss (McKinney 2002), introduction and
spread of invasive species (Alston and Richardson 2006), and
habitat fragmentation (Radeloff
et al.
2005).
Consequences of urban growth have traditionally been
treated as local issues, but recent studies suggest that impacts
are far-reaching, with regional and global implications (Seto
et
al.
2012). Currently, over half of the world’s population lives in
urban environments, which is expected to grow at an unprec-
edented rate in coming decades, especially in developing coun-
tries (Cohen 2003; Seto
et al.
2009). Monitoring urban dynam-
ics is of critical importance for ecosystem service assessment.
Increasing awareness of the impact of urbanization on a
global scale has motivated numerous urban mapping efforts
using satellite remote sensing. Early satellite-based urban
mapping applied medium resolution data from Landsat Mul-
tispectral Scanner System (
MSS
), (79 m), but its coarse spatial
resolution presented a significant challenge for detailed urban
studies (Welch 1982; Barnsley
et al.
2003). With the continuous
improvement of sensor technology and image processing ca-
pabilities, remote sensing data with higher spatial and spectral
resolution, such as those from Landsat (Thematic Mapper (
TM
)/
Enhanced Thematic Mapper (
ETM
)/ Operational Land Imager
(
OLI
)) and
SPOT
, are now routinely used for urban mapping. The
Landsat series of satellite imagery are among the most widely
used because of their rich archive and open access (Woodcock
et al.
2008). Examples of applications include general land
cover mapping (Fung 1992; Zha
et al.
2003; Lo 2004; Lu
et al.
2011; Zhu
et al.
2012; Chen
et al.
2015), study of urban dynam-
ics (Masek
et al.
2000; Yuan
et al.
2005; Taubenböck
et al.
2012; Zhang and Weng 2016; Wu and Chin 2016), urban spatial
structure analysis (Herold
et al.
2002; Seto and Fragkias 2005;
Wang
et al.
2014), and quantification of urban thermal charac-
teristics (Weng
et al.
2004; Weng 2009; Xian and Crane 2006).
Most previous studies on urban mapping follow a “snap-
–10 year mapping intervals. For example,
- and national-scale urban mapping has
ucted every five years (e.g., 2001, 2006,
and 2011), as part of the National Land Cover Database (
NLCD
)
development (Homer
et al.
2007; Homer
et al.
2015). The
arbitrary mapping interval, however, is limiting for certain
applications requiring land cover data at a higher temporal
frequency (e.g., annual). For example, spatially distributed
landscape process models and system dynamic models typi-
cally favor high temporal frequency land cover map products
to support continuous modeling of ecosystem processes and
functions (Lunetta
et al.
2006; Winz
et al.
2009). Forecasting
future land cover distributions can also benefit greatly from
understanding historical and ongoing urban changes, ob-
served with high temporal frequency (Pontius
et al.
2008).
Annual urban map products can be derived using a
number of image classification and change detection tech-
niques. One approach is to conduct image classification
for each mapping year. Intrinsic image classification errors
and their accumulations over all mapping years make this
Heng Wan, Yang Shao, and James B. Campbell are with
Virginia Tech, College of Natural Resources and Environment,
Geography Department, 115 Major Williams Hall, Blacksburg,
VA 24061 (
.
Xinwei Deng is with Virginia Tech, Department of Statistics,
211 Hutcheson Hall, Blacksburg, VA 24061.
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 10, October 2019, pp. 715–724.
0099-1112/19/715–724
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.10.715
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
October 2019
715
