ARTICLES
www.scichina.com www.springerlink.com 977
Chinese Science Bulletin 2006 Vol. 51 No. 8 977—986
DOI: 10.1007/s11434-006-0977-8
Remote sensing of spatial
patterns of urban renewal
using linear spectral mixture
analysis: A case of central
urban area of Shanghai
(1997―2000)
YUE Wenze1
, XU Jianhua2
, WU Jiawei2
& XU Lihua3
1. College of Southeast Land Management, Zhejiang University, Hangzhou
310029, China;
2. Department of Geography, East China Normal University, Shanghai
200062, China;
3. Zhejiang Forest University, Lin’an 311300, China
Correspondence should be addressed to Xu Jianhua (email: jhxu@
geo.ecnu.edu.cn)
Abstract It is very important to integrate remote
sensing with urban geography that the spectral mixture
analysis technique is applied to urban land cover
evolvement and its eco-environmental effect. Urban
land cover is mainly composed of complicated artificial
materials, which is the key factor to limit the development
of the spectral mixture analysis technique.
There are two main aspects in which the technique of
spectral mixture analysis is applied to urban geography:
one is to calculate vegetation fraction; the
other is to build quantitative model of the urban impervious
surface obtained from the combination between
high albedo fraction and low albedo fraction.
The technique of spectral mixture analysis is firstly
applied to study urban renewal pattern, scale and
mode which happened in Shanghai City from 1997 to
2000.
Keywords: linear spectral analysis, endmember, urban renewal,
Shanghai.
Spectral Mixture Analysis (SMA) was utilized for
calculating land cover fractions within a pixel and involves
modeling a mixed spectrum as a combination of
spectra for pure land cover types, called endmembers.
SMA, whose core is to mine higher scale level information
from relatively low scale data, is a hotspot in
the field of processing remote sensing data, which can
enrich information and improve the efficiency of remote
sensing data[1 ― 5]
. The linear spectral mixture
model bases on the assumption that the mixed spectrum
is a linear combination of endmembers. It also assumes
that the same type of land cover is of similar spectrum,
which can be added up together[6]
. There were abundant
studies about linear spectral mixture analysis (LSMA)
in which the most popular method is based on constrained
least-squares and minimum noise fraction
(MNF) transformation[1]
. For example, Small showed
his way to calculate the vegetation fraction by the
means of principle component (PC) transformation
based on linear mixture model in 2001[4]
. And then, in
2004, he adopted the MNF transformation to remove
the noise in image data and got the vegetation fraction
by the same linear mixture model[7]
. Besides, Tao et al.
studied the unmixed results after K-L, MNF transformations,
and conducted research on the accuracy of
SMA, which was of significance[1]
. Applications of
LSMA to urban land cover and ecological effects are a
hotspot in the field of international remote sensing
study. It is very important to combine remote sensing
with urban geography. However, urban land surface is
mainly composed of complicated artificial materials so
that the low accuracy of LSMA is the main limitation to
its development. Meanwhile, current key problem lies
in the selections of endmembers: the range, dimensionality,
and the criterion.
According to the Vegetation-Impervious Surface-Soil
(V-I-S) conceptual model promoted by Ridd
in 1995, urban land cover can be divided into four basic
elements: vegetation, high albedo, low albedo, and
soil[8]
. All pixels in one image can then be interpreted
as the combinations of these four elements, which set
up a theory foundation for application of pixel unmixture
in cities[4,7―9]
. As an individual endmember, vegetation
plays a distinct role in geography on the basis of
the model. Vegetation fraction/abundance was used in
many studies to reveal urban green space and its ecological
effects. Besides, water was masked from low
albedo by some scholars and quantitative model was
built combining with high albedo to get urban impervious
surface fraction. Similar to vegetation fraction,
urban impervious surface influences urban thermal environment
and atmospheric condition so it also has distinct
effect in geography. However, high albedo and
low albedo have not been seriously introduced. This
study, based on spectra data measured in field and samples
checkup on aerial photograph, finds that low albedo
endmember (except water) corresponds to dated
buildings in the old city zone while high albedo end-
ARTICLES
member corresponds to newly built construction. This
phenomenon suggests that changes of high albedo and
low albedo endmembers in different periods can represent
the spatio-temporal evolvement of urban landscape
patterns, especially urban renewal.
978 Chinese Science Bulletin Vol. 51 No. 8 April 2006
Shanghai (Fig. 1), with large amount of shanties
where per capita living space is less than 4 m2
in 1980’s,
once was the most difficult in project of urban renewal,
which is of great value for every municipality to improve
civil living conditions and life environment. After
two periods’ efforts, Shanghai becomes the successful
model in urban renewal project[10]
. The technique
of remote sensing is of significance for detecting
the spatial patterns of the old city area renewal. Firstly,
the spatial pattern of the urban renewal reveals the
changing characters of urban structure and the adjusting
direction of urban function, which can provide basic
ideas for urban plan in future. Secondly, with the increased
urban scale, applying remote sensing to studying
urban renewal can supply a low cost but high effective
means for programming, implementing and evaluating
the renewal engineering. Finally, the study promoted
the development of the integration between remote
sensing and urban geography. Based on the improved-endmember-selection
technology, this paper
tries to use LSMA to analyze the spatio-temporal pattern
and discuss the application of remote sensing in
urban geography from 1997 to 2000.
Fig. 1. The geographic map of Shanghai.
1 Method and technology
1.1 LSMA and its improvement
LSMA is a method physically based on image processing,
assuming that spectrum measured by a sensor is
a linear combination of the spectra of all components
within the pixel, and can be expressed as[5]
:
(1)
1
,
n
i k ik
k
R f R ER
=
= +∑ i
m
n
where (number of spectral bands);
(number of endmembers); is the spectral
reflectance of band which contains one or more
endmembers;
1,...,i =
1,...,k = iR
i
kf is the proportion of endmember k
within the pixel; is the known spectral reflectance
of endmember k within the pixel on band i; and
is the error for band i. Root mean square was used to
measure the accuracy of solution:
ikR
iER
2
1
,
m
i
i
RMS ER m
=
⎛ ⎞
= ⎜ ⎟
⎝ ⎠
∑ (2)
From eq. (2), we can see that the smaller the RMS is,
the smaller the error will be.
A constrained inverse least squares deconvolution
model was used in this research, assuming that the following
two conditions are satisfied simultaneously:
and 0 , (3)
1
1
n
k
k
f
=
=∑ 1kf≤ ≤
This will restrict each endmember fraction between
0 and 1; all the summation is equal to 1, which can
avoid any fraction larger than 1 or less than 0.
Many studies concluded that the MNF transformation
can reduce the correlation between bands and concentrate
most information on the first several components,
which can improve the accuracy of LSMA as a
result[11,12]
. According to the analysis above, selecting
suitable endmembers is the key factor determining the
overall accuracy during the unmixing process. An optimal
approach for selecting endmembers is to use
laboratory-based measurements of endmember’s spectra,
referred to as “reference endmember”. Although
substantial problems exist in correcting atmospheric
conditions in satellite sensor data, Gillespie et al., Wu
et al. and others, solved these problems by considering
the mixed spectra as a linear combination of endmembers
derived from an image (image endmember)[9,13]
.
Each image endmember is a combination of reference
endmembers including atmospheric scattering and ab-
sorption[14]
. As the spectral information of urban land
cover is so complicated, and atmosphere also influences
the spectra in addition, image endmember can
give a higher accuracy[4―6,9]
. Thus, in this study, image
ARTICLES
endmembers were chosen and derived from the
TM/ETM+ image.
In this paper, to improve the accuracy of LSMA,
pixel purity index is introduced to constrict the selection
range to very few pure pixels. Besides, we formed
a three-dimensional feature space, in which endmember
selection interacts with original image, which can also
improve the accuracy of results.
(i) Pixel purity index (PPI) The PPI is used to
find the most “spectrally pure” or extreme pixels in
multispectral and hyperspectral images[15]
. Separating
purer from more mixed pixels reduces the number of
pixels analyzed for endmember determination and
makes separation and identification of endmembers
easier. The PPI is computed by repeatedly projecting
n-dimensional scatter plots onto a random unit vector
as shown in Fig. 2. The extreme pixels in each projection
are recorded and the total number of times each
pixel marked as extreme is noted. In the process of iteration,
it asymptotically approaches a flat line (zero
slope), which means that all of the extreme pixels have
been found. A projection process is performed as fol-
lows:
(4)
1
pixel[ ] skewer[ ],
n
i
dp i i
=
= ⋅∑
Fig. 2. Schematic view of scatters, which were projected to random
skewers in 2D space.
A PPI image is created in which the DN of each
pixel corresponds to the number of times that pixel was
recorded as extreme. In our study, the number of pixels
is 1665292, and when the number of iteration is 10000,
the selected pure pixel is 71046 which accounts for
4.26% in all. The number of iteration increases, so does
the number of pure pixels selected. When the number
of iteration jumps to 15000, the number of pure pixels
becomes stable, with 87608 pixels selected, which accounts
for 5.26%. A mask gained from the PPI image
was used to filter “pure pixels” in which endmembers
were selected. Through PPI, the range of selection was
reduced by 95%.
(ii) Endmember selection in three-dimensional feature
space and unmixing For the components of
urban land cover, many studies considered that three or
four endmembers can be included, that is, high albedo
mainly composed of new building materials such as
glass, concrete and light encrust; low albedo including
water surface, asphalt, dated roof, etc.; green vegetation
such as lawn and woods; soil which in studies of belt
between town and country is considered to be the
fourth endmember[8]
. According to Small’s analysis
based on the global ETM+ data[5]
, generally, three
typical covers can well depict remote sensing data in
urban environment, they are, vegetation, high albedo,
low albedo. The paper uses three endmembers for that
inside Shanghai City. Fig. 3 depicts the feature space
formed by MNF component 1 and MNF component 2.
Endmembers “always” present themselves at the corner
of triangle. Theoretically, if all of the pixels are within
the triangles formed by endmembers, the mixture
model can be considered an ideal linear model[9]
. The
clear delineation of three MNF components (Fig. 4)
suggests that the reflectance spectra of the TM or ETM
+ image might be best represented by a three-endmember
linear mixing model.
Traditional method for endmembers selection only
considers two dimensions of MNF component 1 and
Fig. 3. Pixel scatters plots with a combination of the first two MNF
components.
www.scichina.com www.springerlink.com 979
ARTICLES
MNF component 2 (Fig. 3)[13,16]
. This paper picks out
pure pixels by PPI and forms three-dimensional feature
space, in which each dot represents a pixel (Fig. 4). The
feature space can interact with the original image and
the PPI result so that we can see the areal characters
and purity. In the selection, we change the angle and
direction of coordinate axis, observe pixels polyhedron
and its projections on three sides, and check its land
cover’s purity from PPI and high-resolution aerial photograph.
Through continuous filtration, we selected 100
dots for each endmember and the average reflectance of
endmembers selected was specified to each ETM+
bands range (Fig. 5).
urban landscape, we measured the spectrum reflectivity
on the spot of several typical urban building surfaces by
spectrometer. The number of measured points is 171
while the number of spectrum is 2305. Three types of
urban building surfaces can be classified: roads, rooftops
and soil which in summer is too little in central
city to consider. The spectrum reflectivity of glass or
metal overlaying on high buildings and large mansions
preponderates over the range of spectrometer. New
buildings are often covered by concrete whose spectrum
reflectivity is higher than 35% between 400 and
900nm, much higher than old cement surface, roof and
asphalt surface. For old city zone in Shanghai, the main
building surface cyan-tiled roof’s spectrum reflectivity
is obviously lower than old cement surface1)
. In addition,
asphalted rooftops and roads surface which has
low spectrum reflectivity were adopted in the past instead
of concrete, building decorative materials, glass,
metal and green vegetation now. So we conclude that
the evolvement of high albedo and low albedo can reflect
urban renewal and sprawl.
Furthermore, from Fig. 6 we can see that there lies
contrary trend between actual spectrum of old construction
surface and the results of low albedo in Fig. 5
derived from LSMA, for the unmixed image without
atmosphere correction, which was caused by the following
two reasons: the air path radiation (including
Rayleigh scatter and Mie scatter) caused by grains in
atmosphere will be cut down with the increase of
wavelength; generally, steam in atmosphere has higher
absorption to the radiation in infrared band than in visible
band. But all these would produce little influence
on the result using high or low albedo to represent new
or old building.
Fig. 4. Pixel scatter plot in 3D space of MNF1―3.
Fig. 5. The reflectance of three endmembers at six bands.
1.2 Checkup of the land surface characteristics of
high and low albedo endmembers
(i) Spectrum reflectivity checkup by spectrometer
There are still general rules for complicated urban
land cover. To reveal reflectance spectral features of
Fig. 6. The spectrum reflectivity on the spot of several typical urban
building surfaces.
1) Xu, L. H., Air path radiation and atmospheric quality of Shanghai, Thesis for doctor degree of East China Normal University, 2005.
980 Chinese Science Bulletin Vol. 51 No. 8 April 2006
ARTICLES
(ii) High-resolution measurement of high albedo
fraction To reveal the relationship between the
newly-built area and high albedo, we selected 75 samples
randomly in aerial colorful photograph. Screen
digitization was conducted to specify new buildings.
Comparing these samples with values of high albedo
endmember fraction from LSMA, we found a good
consistency between them (Fig. 7). But there are several
problems: temporal differences, new building
judgment, and digitization error, but the consistency is
so obvious that the mean error between both is only
0.044. So we conclude that high albedo endmember can
represent new buildings.
Fig. 7. Comparison of high albedo fraction between digitized and mod-
eled.
1.3 Validation
(i) Root mean square (RMS) During the process
of “unmixing”, errors from the data preparation, endmembers
selection and the linear mixing model by least
squares will reduce the accuracy of vegetation frac-
tion[7]
. The RMS for every image pixel was calculated
in order to assess the performance of this moedl. According
to formula (2), obviously, the lower the RMS is,
the better fit the model will be. Every pixel’s RMS of
2000 is shown as Fig. 8. The maximal value is 0.0867
and most are lower than 0.01 according to RMS frequency
histogram. 98.5% of RMS is smaller than 0.02,
which shows a better result than the former studies[4]
.
But the distribution of RMS is not symmetrical. As a
whole, the values located in urban area are higher than
in suburb, and in the old city area are also higher than
in newly-built city area. Comparatively, mixed pixels
are more inclined to have large RMS than pure pixels.
The more complicated the inside mixed pixel is, the
large the RMS will be, which suggests there may be
another spectral endmember to be further studied.
(ii) Accuracy assessment According to LSMA
theory above, the relationship between the three endmembers
is linear, and the total fraction should be equal
to 1 in a unit pixel. So, if we can determine the accurate
fraction of one or two endmembers, the accuracy of
results is relatively high on the whole. In the urban
central area, impervious surface can be considered as
summation of high albedo and low albedo except for
water[9]
. We digitized impervious surface fraction on
aerial photograph within random square samples
(masked off water) and got statistical results of LSMA.
As a result, we had a digitized impervious surface fraction
(actual fraction) and statistical impervious surface
fraction (modeled fraction). Comparison is made between
them for rigorous purpose.
In order to assess accuracy of impervious surfaces
estimation, a random sampling method was applied. We
randomly selected 127 rectangular windows with
150m×150m in color aerial photograph based on the
principle that each window should contain some urban
Fig. 8. The image and frequency plot of root mean square.
www.scichina.com www.springerlink.com 981
ARTICLES
buildings. We then digitalized the impervious surface
manually in the 127 windows, whose actual abundance
then can be obtained. These windows were then used to
extract the sum of high and low albedo above and got
the results of LSMA with the above models. Fig. 9
shows the comparison of the two results and the residual.
Except for some windows with the residual higher
than 0.4, most were within ± 0.2 and decreased as the
abundance of impervious surface increased. The accuracy
was evidently increased compared to the similar
researches[9]
.
Fig. 9. Accuracy assessment result of impervious surface estimation.
2 Results and analysis
2.1 The results of LSMA
According to discussion above, the LSMA results of
1997 and 2000 are shown as Figs. 10 and 11. Shanghai
City was in the stage of rapid expanding and changing
between 1997 and 2000. Many buildings were built
with the materials such as metal, concrete and glass
with high reflectivity. It is obvious that there is a great
difference between urban and suburb based on the fraction
image of high albedo. In the brim of urban, high
values appear less than the central area in high albedo
fraction imagery except for a few old towns. In addition,
the distribution of high albedo fraction inside urban
area focuses on newly built urban areas mainly at the
frontier of urban sprawling and residents in renewal
areas. However, low albedo, representing land covers
such as water, old buildings and heavy industry areas
seriously polluted for quite a few dusts, most shows
low values except the following areas, old urban area
between the People Square and the bund along with
both sides of Suzhou River, and Wusongkou area
giving priority to industry land, which have rather
higher values. The main characters of urban evolvement
is that with the old city area’s renewal and urban
sprawl, in the central city area many old buildings
showing low albedo are replaced by vegetation and
new buildings having high albedo. While in suburb,
farmland paddies were transformed to new buildings
(high albedo). With the importance of urban ecological
environment emerging, vegetation abundance inside
city is obviously increasing especially newly-built park
and large lawn. Vegetation fraction reached to 60%
above in New Jiangwan Town, Youthful Forest Park,
Lujiazui Park and Changfen Park in 2000.
2.2 The pattern and mode of urban renewal in center
urban area (1997―2000)
According to the analysis above, we drew the conclusion
that LSMA with improved endmember selection
had relatively high accuracy. Two ways of spectral
test and sample-in-field test proved that the abundance
of high and low albedo could reflect different characters
of the construction surface. Therefore, the changes
during different periods could be used to analyze the
Fig. 10. Three-component endmember fraction images of Shanghai City in 1997.
982 Chinese Science Bulletin Vol. 51 No. 8 April 2006
ARTICLES
Fig. 11. Three-component endmember fraction images of Shanghai City in 2000.
urban sprawl and the renewal of the old city area. This
paper selects two TM/ETM+ images of 1997 and 2000
for the urban renewal analysis. Since this period was
just when Shanghai underwent the first round urban
renewal, this paper focuses on the old city area which is
on the west of Huangpu River inside the inner ring road
and discusses the pattern and mode of the urban re-
newal.
If the low albedo fraction in one pixel increases significantly,
we can think that urban renewal is conducted
in the pixel because water varies little in temporal images
while old buildings then influence the fraction
obviously. Considering that one pixel is represented by
three components, we chose 15% fraction as a threshold.
Within one pixel, if the increase of low albedo
fraction goes over15%, we regard that urban renewal
happened similarly, the evolvement of vegetation and
high albedo is judged by the same principle. If the increase
of vegetation fraction goes over 15%, we regard
that the city renewal happened, that is, the green vegetation
replaced old buildings. If the increase of high
albedo fraction goes over 15%, we regard that the city
renewal happened; new buildings played a prevailing
role. If the increase of high albedo fraction and vegetation
fraction goes less than 15%, we regard that the city
renewal happened; new buildings and vegetation were
both conducted. Fig. 12 shows us the pattern and mode
of the urban renewal between 1997 and 2000 by the
way of raster calculation supported by ARCGIS/GRID.
In Fig. 12, red, green and blue colors represent renewal
areas of urban renewal. Red color refers to new buildings;
green color refers to vegetation; blue color refers
to both new buildings and vegetation.
From the result, we found that the main renewal areas
were changing from CBD in central urban (between
Fig. 12. The pattern and mode of urban renewal at pixel scale from
1997 to 2000.
Nanjing Road and Huaihai Road and around Yuyuan
Park) to the inner ring road. Besides Yangpu District
and part of Hongkou District near Huangpu River,
other areas which were mainly rebuilt with new construction
distributed in Putuo District, Changning District,
Xuhui District and part of Luwan District near
inner ring road. Another characteristic in this urban-renewal
was the recovery of vegetation and the
constructing of large parks. There were two districts
with greatly increased vegetation: one was the communities
of Hunan Road and Tianping Road on the north
of Xuhui District. These areas used to be French settlement
and the main construction was villas in old
style. In the recent years, more trees were planted
among these villas resulting in the increased abundance
of vegetation. Other areas were large greenlands in
Hongkou District, Changning District and the north of
www.scichina.com www.springerlink.com 983
ARTICLES
Luwan District, e.g. the large green land in Tianshan
Park where Yan’an Elevated Road and inner ring road
met, Luxun Park and Heping Park etc. Between CBD
and those areas above, most old residential areas were
reconstructed to new residential administration areas.
The reconstruction of these areas included increasing
both new construction and vegetation to reduce the
building density and increase the abundance of vegeta-
tion.
The first round urban renewal of Shanghai City
(1992―2000), “365 Urban Renewal Project”, was to
reconstruct the old building with areas of 3.65×106
m2
.
Statistically, 3.8×107
m2
of old city areas was reconstructed
including more than 4×106
m2
of risky houses
and 1×106
residents moved to new houses. Between
1995 and 2000, about 3×107
m2
of aged construction
was removed1)
. According to the result of LSMA, between
1997 and 2000, 19.93% areas of urban renewal
inside inner ring road on the west of Huangpu River
were 1.576×107
m2
and took up nearly half of the total
areas of reconstruction; 7.9×106
m2
new buildings were
reconstructed and took up 50.13% of urban renewal;
the renewal of vegetation recovery and greenland constructing
was about 2.72×106
m2
and took up 17.26% of
the total renewal area; The reconstruction of both took
up 32.61%. Thus it can be seen that the scale of urban
renewal is very big. The renewal of residential and administration
area is the mainstream accompanied by
Fig. 13. The pattern of urban renewal on street community scale.
1) Wang, W., Fu, G. L., It is perfect for urban renewal in Shanghai, Wen Hui Daily (in Chinese), July 17, 2003.
984 Chinese Science Bulletin Vol. 51 No. 8 April 2006
ARTICLES
www.scichina.com www.springerlink.com 985
cosmically-built infrastructure. Inner Ring Road,
Yan’an Elevated Road, North-South Elevated Road,
subway line No.1 and No.2 and elevated road No. 3
have been built or are being built. Green vegetation
cover is more and more increasing for its importance to
urban environment. Results from LSMA accord with
statistical data to some degree except temporal difference
that one image is photoed in April, 1997 and the
other is in June, 2000, which leads to the vegetation
fraction in 2000 much higher than that in 1997.
2.3 The pattern and mode of urban renewal at the
street community scale
We analyzed the urban-renewal results above at the
pixel scale. Due to the low resolution at the pixel scale,
areas of urban-renewal and its trend were distributed to
each community to further analyze the spatial pattern
and its characters in detail from 1997 to 2000.
On the community scale (Fig. 13), spatial pattern of
urban-renewal from 1997 to 2000 was more obvious.
Most evident areas spread out along inner ring road
including street communities of Siping Road, Jiangpu
Road and Pingliang Road in Yangpu District, Xingang
Road in Hongkou District, Changshou Road in Putuo
District, Huayang Road and Xinhua Road in Changning
District, Xujiahui, Hunan Road and Tianping Road in
Xuhui District from north to south, while the core areas
to the east of People Square are lightly renewal. The
most renewed communities were Xujiahui Road and
Tianping Road in Xuhui District. When coming to the
renewal patterns, Jiangpu Road and Pingliang Road in
Yangpu District in the north, Changshou Road in Putuo
District in the west, and Xujiahui Road in Xuhui District
and Dapuqiao Road in Luwan District in the south
were mainly changed by newly-built construction and
formed a semi-circularity facing to the east. The areas
most changed by vegetation included: Siping Road in
the north, Tianping Road, Hunan Road and Huayang
Road in the South and formed two highly centralized
areas in south and north. Among them, the most greatly
changed areas were where Xuhui, Jing’an, Changning
and Luwan Districts matched, while for the high land
value in the middle areas along Suzhou River and in the
east of Huangpu District, the vegetation fraction increased
little[17,18]
. On the periphery mainly covered
with vegetation, the newly-built vegetation on the prevenient
basis had led to relatively high vegetation
abundance; there was rarely newly-built vegetation
among old construction and most of them were distributed
around the crossroads; the vegetation among
newly-built residential construction was mainly small
patches but had a broad distribution, which contributed
to the improvement of ecological environment in center
of the city.
3 Conclusions and discussion
This paper, based on the Landsat TM/ETM+ imagery,
uses improved LSMA and discusses the method and
significance of the urban renewal in the field of remote
sensing. The conclusions and discussion were shown as
follows:
(1) Aiming at the restricted factors in LSMA, PPI
was introduced to help to select endmembers and reduce
the extent of endmember selection by 95%. The
first three MNF components are selected to form
three-dimensional mixing space. By expanding the traditional
two-dimensional mixing space to three-dimensional
mixing space, the third component is considered,
and during the process, we compared the original and
high-resolution imagery with the results of PPI, which
will obtain more accurate results.
(2) Spatio-temporal changing of high and low albedo
and endmembers of vegetation was used to reflect the
pattern of urban-renewal, which provides a new way to
study the urban landscape changing in the field of
urban geography. The precondition of this idea is based
on the strict spectral test and field test. With the test of
spectral reflectivity in field and random-chosen
samples in high resolution color aerial photograph, the
pattern and mode of urban-renewal can be explained by
the changing abundance of high and low albedo.
(3) The quantitative calculation of LSMA of the period
reveals the pattern of urban renewal. At the scale
of pixel, the focus of urban renewal is extended from
CBD to the inner ring road and forms a region of
semi-circularity; the main mode of renewal was the
change from old construction to new construction accompanied
by recovered and newly built vegetation;
nearly 1/3 of the newly built construction was residential
and administration area with reduction of building
density and increase of vegetation. When coming to the
scale of street community, the communities of Siping
Road, Jiangpu Road, Changshou Road, Xujiahui Road
and Tianping Road show the most obvious changing;
light reconstruction was detected along the Suzhou
River and Huangpu District; newly built construction
took up most on the northeast, west and southwest of
the region while Tianping Road and Siping Road were
ARTICLES
986 Chinese Science Bulletin Vol. 51 No. 8 April 2006
mainly reconstructed by the cover of vegetation.
(4) Although comparing to the current research, accuracy
of LSMA is increased by the improvement of
endmember selection, in essence, evaluating the quantity
and quality of endmembers directly from image
still is lack of quantitative criterion. Besides, the complex
landscape of urban and temporal difference of imagery,
some low albedo still can be detected in the
newly built construction and that may lead to some errors
in the results, e.g. over increased abundance of
vegetation due to the temporal difference of imagery.
Confirmation of criterion of endmember evaluation,
application of automatic selection by computer and
finding the typical objects with different reflectivity in
both old and new construction by spectral analysis all
need to be further studied.
Acknowledgements The authors would like to thank Editor
Li Yawen, Prof. Wei Ji and two anonymous referees for their
valuable comments on an earlier draft and pains for revising
the paper, and especially Prof. Wei Ji and the two referees
who gave much important suggestion to the paper. This work
was supported by the National Natural Science Foundation of
China (Grant No. 40371092).
References
1 Tao Q X, Tao H X, Zhang L P. Study on applying linear mixing spectral
model in vegetation classification using hyperspectral remote
sensing. Survey Technology (in Chinese), 2004, (1): 21―24
2 Shimabukuro Y E, Smith J A. Least squares mixing models to generate
fraction images derived from remote sensing multispectral
data. IEEE Transactions on Geoscience and Remote Sensing, 1991,
29: 16―20
3 Emma U, Susan U, Deanne D. Mapping nonnative plants using
hyperspectral imagery. Remote Sensing of Environment, 2003, 86:
150―161
4 Small C. Estimation of urban vegetation abundance by spectral
mixture analysis. International Journal of Remote Sensing, 2001,
22(7): 1305―1334
5 Small C. Multitemporal analysis of urban reflectance. Remote
Sensing of Environment, 2002, 81: 427―442
6 Lü C C, Wang Z W, Qian S M. A review of pixel unmixing model,
Remote Sensing Information (in Chinese), 2003, 3: 55―60
7 Small C. The Landsat ETM+ spectral mixing space. Remote Sensing
of Environment, 2004, 93: 1―17
8 Ridd M K. Exploring a V-I-S (vegetation-impervious surface-soil)
model for urban ecosystem analysis through remote sensing: comparative
anatomy for cities. International Journal of Remote Sensing,
1995, 16: 2165―2185
9 Wu C S. Estimating impervious surface distribution by spectral
mixture analysis. Remote Sensing of Environment, 2003, 84:
493―505
10 Yao K. The diagnosis for Shanghai urban renewal. Beijing Plan and
Construct (in Chinese), 2004, 1: 151―153
11 Ma J W, Zhao Z M. Buheaosier, Remote Sensing Data Model and
Processing Method (in Chinese). Beijing: Chinese Science and
Technology Press, 2001, 62―70
12 Smith M O, Johnson P E, Adams J B. Quantitative determination of
mineral types and abundances from reflectance spectra using principal
components analysis. Journal of Geophysical Research, 1985,
90: 797―804
13 Gillespie A R, Smith M O, Adams J B, et al. Interpretation of residual
images: spectral mixture analysis of AVIRIS. In: Proceedings
of the 2nd AVIRIS workshop, 4―5 June 1990 (Pasadena, CA:
JPL): 243―270
14 Van Der Meer F, De Jong S M. Improving the results of spectral
unmixing of Landsat Thematic Mapper imagery by enhancing the
orthogonality of end-members. International Journal of Remote
Sensing, 2000, 21: 2781―2797
15 Boardman J W, Kruse F A, Green R O. Mapping target signatures
via partial unmixing of AVIRIS data. In: Summaries, Fifth JPL
Airborne Earth Science Workshop. JPL Publication 95-1, 1995,1:
23―26
16 Weng Q, Lu D, Schubring J. Estimation of land surface temperature-vegetation
abundance relationship for urban heat island studies.
Remote Sensing of Environment, 2004, 89: 467―483
17 Xu J H, Yue W Z, Tan W Q. Spatial stat. law for the effect of urban
sight structure. Acta Geographica Sinica (in Chinese), 2004, 59(6):
1058―1065
18 Yue W Z, Xu J H, Tan W Q. Spatial scale analysis of the diversities
of urban landscape. Acta Ecologica Sinica (in Chinese), 2005,
25(1): 122―128
(Received September 23, 2005; accepted November 7, 2005)