Urban Impervious Surface Estimation from
Remote Sensing and Social Data
Yan Yu, Jun Li, Changyu Zhu, and Antonio Plaza
Abstract
We propose an inspiring approach for accurate impervious
surface estimation based on the integration of remote sens-
ing and social data. The proposed approach exploits the
strengths of two kind of heterogeneous features, i.e., physi-
cal features and social features, where the former ones are
derived by a morphological attribute profiles-guided spectral
mixture analysis model using remote sensing imagery, and
the latter ones are obtained from the normalized kernel den-
sity of point of interest and vector road datasets. These two
features are then integrated using a multivariable linear re-
gression model to estimate impervious surfaces. The proposed
method has been tested in the main urban area of Guang-
zhou, China, in pixel level and parcel level, respectively. The
obtained results, with the overall
RMSE
of 10.98% and 10.90%
for pixel level and parcel level, respectively, demonstrate
the good performance of integrating remote sensing imagery
and social data for mapping of urban impervious surface.
Introduction
Impervious surfaces (
IS
), defined as the surfaces that water
cannot infiltrate, are usually made up of anthropogenic mate-
rials, e.g., rooftops, parking lots, streets, and outdoor facilities
(Slonecker
et al
., 2001). These increasing
IS
have been identi-
fied as a significant indicator of ecological condition and
urbanization (Brabec
et al
., 2002; Hasse and Lathrop, 2003),
as they have direct impacts on water quality of neighboring
water bodies, hydrologic cycles, natural temperature cycles,
and land surface temperatures (Schueler, 1994; Adams
et al
.,
1993; Geiger
et al
., 2009; Yuan and Bauer, 2007). The demand
for current and accurate
IS
maps has greatly increased. To
date, different approaches have been applied to characterize
and quantify
IS
using ground-measured data and remote sens-
ing data (Weng, 2012).
Remote sensing data have been used for the estimation of
IS
since the 1970s. Many techniques have been proposed to ex-
ploit the spatial, spectral, texture and context information of
remote sensing data (Weng, 2012). For instance, Møller-Jensen
(1990) applied a linear segmentation model which incorpo-
rates texture and context information of Landsat-TM imagery
to cover urban areas. Deng
et al
. (2012) used a linear spectral
un-mixing model to extract
IS
information from Landsat im-
agery. Research on imperviousness estimation from multi-
sensor and multi-source data has also attracted interest. Yang
et al
. (2003) quantified urban
IS
by using Landsat-7
ETM+
and
high-resolution imagery. Liu
et al
. (2013) integrated night-time
light luminosity, land surface temperature and multi-spectral
reflectance data to enhance
IS
while suppressing other un-
wanted land covers. Huang
et al
. (2017) studied the subtle
urban changes using multi-view satellite imagery. Although
remote sensing data brings desirable properties (large cover-
age, information of spectral reflectance, etc.), impervious
surface estimation is still a difficult task due to the complex-
ity of urban/suburban land cover, as well as the limitations of
spectral and spatial resolution of remote sensing imagery (Lu
and Weng, 2006). In this sense, medium-resolution satellite
imagery can help to map urbanization areas at a large spatial
scale, but it may lead to the underestimation of
IS
because of
the heterogeneity of urban landscapes (e.g., soil, grass and
water body). High-resolution remote sensing imagery (Ikonos,
GF
-2, aerial photography, etc.), provides an alternative. Nu-
merous pixel-based methods and object-based methods have
been explored to map the
IS
with high-resolution imagery.
Sawaya
et al
. (2003) utilized several pixel-based methods to
map the imperviousness of Eagan City. Lu and Weng (2006)
estimated urban impervious surface using decision tree classi-
fier and linear spectral mixture analysis model. Li
et al
. (2011)
explored the object-based method to map urban impervious
surface using very high resolution imagery. Zhang and Huang
(2018) monitored the change of impervious surface using the
multi-feature objected-based approach. In general, the accura-
cy of impervious fractions generated by traditional estimation
methods, such as those presented in Bauer
et al
., 2004; Lu
et
al
., 2011; Wu and Murray, 2003, is mostly dependent on the
quality of multi/hyper-spectral imagery. These signature-based
methods however have limitations in low-albedo impervious-
ness like old town areas, shadow, and urban greening areas.
The fast development of social technologies with global
navigation satellite system results in a wide availability of
social data, such as taxi
GPS
data, smart card data, social
media data, and volunteered web maps. These heterogeneous
datasets, with the attributes of geographic and human activ-
ity, provide unprecedented possibilities for the improvement
of urban study (Hu
et al
., 2016; Liu
et al
., 2015; Lu and Liu,
2012). Point of interest (
POI
) data, composed of geographic lo-
cation and their particular place-based information are widely
available on the Web like
Google Place
1
,
Facebook Place
2
,
Gaode Place
3
(in China), which use their own taxonomy of
categories or tags (Rodrigues
et al
., 2012). Different from the
check-in data gathered from social media platforms,
POI
data
are usually associated with their certain and detailed informa-
tion as names, addresses, coordinates (latitudes and longi-
tudes), categories, etc., which can reflect the land use type of
a certain place. They therefore provide a new direction for
Yan Yu, Jun Li, and Changyu Zhu are with the Guangdong
Provincial Key Laboratory of Urbanization and Geo-
simulation, School of Geography and Planning, Sun Yat-sen
University, Guangzhou, 510275, China (corresponding author:
Jun Li;
)
Antonio Plaza is with the Hyperspectral Computing
Laboratory, Department of Technology of Computers and
Communications, Escuela Politécnica de Cáceres, University
of Extremadura, Cáceres, Spain.
Photogrammetric Engineering & Remote Sensing
Vol. 84, No. 12, December 2018, pp. 771–780.
0099-1112/18/771–780
© 2018 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.84.12.771
1.
/
2.
/
3.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2018
771
