Transcript Comparison of landcover patterns in Taipei, Kyoto, and Tokyo

Comparing landcover patterns in
Tokyo, Kyoto, and Taipei using
ALOS multispectral images
Ke-Sheng Cheng, Wei-Chun Hung
Department of Bioenvironmental Systems Engineering
National Taiwan University
Yen-Ching Chen
Department of Landscape Architecture
Fu-Jen Catholic University
4/13/2017
2011 ACRS Conference, Taipei
1
Introduction
Urbanization is a process of transforming
natural landscapes to built environments
which typically contain large amounts of
impervious surfaces.
As the process of urbanization evolves,
landcover pattern, namely the types of
landcover and their proportions, present in a
city and its proximity may also change over
time.
4/13/2017
2011 ACRS Conference, Taipei
2
Many studies have found that changes in
landcover pattern can have significant
impacts on urban ecosystems.
Not only are landcover changes the most
apparent effect of urbanization, but also the
driving force of many ecological
consequences of urbanization. Thus,
characterizing the landcover pattern in a
region is crucial for assessing the effects of
urbanization.
4/13/2017
2011 ACRS Conference, Taipei
3
This study aims to
Explore the feasibility of characterizing
landcover patterns using landscape metrics
and other indices derived from remote
sensing images and,
Compare landcover patterns in cities of
different degrees of urbanization.
Specifically, we seek to develop an index
which can provide a quantitative
representation of the degree of urbanization
for different cities.
4/13/2017
2011 ACRS Conference, Taipei
4
Study Areas and Data
Three regions (Tokyo and Kyoto in Japan
and Taipei in Taiwan) of different degrees of
urbanization were chosen for this study.
A subregion centered at the Tokyo Metropolis
Government
Kyoto city and its vicinity, and
Taipei city and its vicinity.
4/13/2017
2011 ACRS Conference, Taipei
5
Kyoto study area
Tokyo study area
The Tokyo metropolis is composed of 23
more urbanized special zones in the east
and also a huge suburban area including
several local cities in the west. Our Tokyo
study area only encompasses some of the
specialized zones in the east.
Taipei study area
Each study area has a spatial
coverage of 15km x 15km.
4/13/2017
2011 ACRS Conference, Taipei
6
ALOS satellite multispectral images of these
studies areas were collected. These images
were acquired by the AVNIR2 sensor
onboard the ALOS satellite with a spatial
resolution of 10 m x 10 m.
AVNIR2 sensor acquires images in four
spectral bands
blue (B, 0.42 – 0.50 m)
green (G, 0.52 – 0.60 m)
red (R, 0.61 – 0.69 m)
near infrared (IR, 0.76 – 0.89 m).
4/13/2017
2011 ACRS Conference, Taipei
7
Landuse/landcover classification
Priori to landcover pattern analysis,
landuse/landcover (LULC) classification
needs to be conducted for each study area.
In this study an indicator kriging approach
for LULC classification was adopted.
The indicator kriging (IK) classification is a
nonparametric supervised classification
approach that can achieve high classification
accuracy for training data set.
4/13/2017
2011 ACRS Conference, Taipei
8
Indicator kriging for LULC classification
n
P( s ( x0 )  C j )  I j ( x0 )   (i j ) I j ( xi ), j  1, 2, , k.
i 1
11/16/2010
9
Two-dimensional three-class feature space
n
P( s ( x0 )  C j )  I j ( x0 )   (i j ) I j ( xi ), j  1, 2, , k.
i 1
Indicator
variogram
Class 1
11/16/2010
Class 2
10
Class 3
Four landcover classes
Woods (land surface covered by forest, trees,
and shrubs)
grass/crop (areas covered by grass, paddy field,
and vegetable crops)
built-up (buildings, paved surface, and bare soil)
water bodies
4/13/2017
2011 ACRS Conference, Taipei
11
Confusion matrices (Training pixels)
4/13/2017
2011 ACRS Conference, Taipei
12
Area percentages of individual
landcover types
Tokyo study area has a dominant area percentage
of built-up coverage, whereas Kyoto has a more
balanced built-up and vegetation (including woods
and grass/crop) coverage. Comparing to Tokyo,
Taipei has a less dominant percentage of built-up
coverage and a significantly higher percentage of
vegetation coverage.
4/13/2017
2011 ACRS Conference, Taipei
13
Kyoto study area
Tokyo study area
Landcover images of the
three study areas.
Water
4/13/2017
Built-up
Woods
Taipei study area
Grass/Crop
2011 ACRS Conference, Taipei
14
Landcover pattern analysis
Landscape metrics
Landcover patterns in coverage-ratio
space
NDVI-based landcover pattern analysis
4/13/2017
2011 ACRS Conference, Taipei
15
(1) Landcover pattern analysis using
landscape metrics
Landcover pattern is an integrated reflection
of the available natural resources, dynamic
natural processes, and anthropogenic
activities in a region.
Generally speaking, it can be characterized
by three key aspects
number of different landcover types,
relative proportions of landcover types, and
spatial distribution of these landcover types.
4/13/2017
2011 ACRS Conference, Taipei
16
In this study landcover pattern analysis is
achieved by assessing two aspects
landcover heterogeneity within individual local
cells (1 km x 1 km or 10,000 pixels), and
variation of such landcover heterogeneity over
the whole study area.
Landcover heterogeneity within a cell is
expressed by the Shannon diversity index
(SHDI) defined as
k
SHDI   pi ln( pi )
4/13/2017
i 1
2011 ACRS Conference, Taipei
17
SHDI values are influenced by richness and
evenness of landcover types.
Richness (total number of landcover types
present in a cell) and evenness (relative areal
proportion among different landcover types)
respectively represent the compositional and
structural components of diversity.
SHDI equals zero when a cell contains only
one landcover type (i.e., no diversity). A cell
with a dominant landcover type has a very
low or near zero SHDI. SHDI increases as
the number of different landcover types
increases and/or the area percentages among
landcover types becomes more equitable.
4/13/2017
2011 ACRS Conference, Taipei
18
There are a number of cells within each
study area and each cell corresponds to an
SHDI value. Thus, variation of cell-level
landcover heterogeneity over the study area
is characterized by the empirical cumulative
distribution function (ECDF) of SHDI.
Using SHDI at cell scale and ECDF of SHDI,
the three key aspects of landcover pattern
can be investigated.
4/13/2017
2011 ACRS Conference, Taipei
19
Spatial variation of cell-level SHDI in
different study areas
4/13/2017
2011 ACRS Conference, Taipei
20
ECDF of cell-level SHDI
4/13/2017
2011 ACRS Conference, Taipei
21
Cell-level SHDI values of the Tokyo study area are
more homogeneous in space. It indicates that both
the landcover type richness and evenness are low in
most of the Tokyo study area.
The Tokyo study area is highly urbanized and
built-up landcover accounts for 86 % of its spatial
coverage. Built-up is the single dominant
landcover type in almost all cells.
By contrast, Kyoto and Taipei study areas have
more significant spatial variation of cell-level SHDI.
4/13/2017
2011 ACRS Conference, Taipei
22
Landcover patterns of Kyoto and Taipei
show more diversified landcover distribution,
i.e. in addition to cells for which woods and
built-up are dominant landcover types (with
very low SHDI), there are also significant
amount of high-SHDI cells which have
mixed and non-dominant landcover types.
4/13/2017
2011 ACRS Conference, Taipei
23
(2) Landcover patterns in coverageratio space
Each cell is associated with four area
percentages (i.e. coverage ratios)
corresponding to landcover types of woods,
crop/grass, built-up, and water bodies.
4/13/2017
2011 ACRS Conference, Taipei
24
Neglecting the coverage ratio of water bodies,
all cells should fall on a plane (hereafter
referred to as the coverage-ratio plane)
defined by three vertices (points A, B, and C)
of dominant landcover types.
4/13/2017
2011 ACRS Conference, Taipei
25
Urbanization can be viewed as a process of
cells on the coverage-ratio plane merging
from points B (woods dominant) and C
(crop/grass dominant) towards the
confluence point A (built-up dominant).
The more concentrated these cells near point
A, the higher degree of urbanization the city
is.
4/13/2017
2011 ACRS Conference, Taipei
26
Scatter plot of cells in a coverage-ratio space
4/13/2017
2011 ACRS Conference, Taipei
27
The Tokyo study area is most urbanized with very
low coverage ratio of woods or grass/crop
landcover type.
The Taipei study area has most of its cells
concentrated near built-dominant point A,
although there are also many cells falling along a
curved path near the AB line. Such pattern is a
reflection of the fact that the Taipei study area
encompasses mountainous regions at its
northeastern and southeastern corners.
For the Kyoto study area, cells are more widely
spread over the coverage-ratio plane, indicating a
good mixture of different landcover types. It also
indicates that the Kyoto study area is least
urbanized among the three study areas.
4/13/2017
2011 ACRS Conference, Taipei
28
(3) NDVI-based landcover pattern
analysis
Cell-average NDVI ( NDVI cell ) was calculated
for individual cells of the three study areas.
NDVI pixel
4/13/2017
( LIR  LIR ,min )  ( LR  LR ,min )

( LIR  LIR ,min )  ( LR  LR ,min )
2011 ACRS Conference, Taipei
29
11/16/2010
30
Relationship between cell-average
NDVI and cell-level SHDI
4/13/2017
2011 ACRS Conference, Taipei
31
Criteria for delineation of areas with
different dominant landcover types
NDVI cell  0 , built - up dominant
0  NDVI cell  0.4 , diversifie d landcover
0.4  NDVI cell , woods dominant
Implementation of such method of
dominant-landcover-areas delineation does
not require an a priori LULC classification,
and thus is particularly useful when good
training data for LULC classification are not
available.
4/13/2017
2011 ACRS Conference, Taipei
32
Areas of different dominant landcover types in the
three study areas delineated using cell-average NDVI.
4/13/2017
2011 ACRS Conference, Taipei
33
Assessing the degree of urbanization
using an urbanization index
Although the spatial variation and ECDF of celllevel SHDI can reveal differences in landcover
patterns of different study areas, they do not
provide an explicit and direct measurement of
urbanization.
Similarly, the cell distribution in the coverage-ratio
space can only serve as a visually and qualitative
indicator of the degree of urbanization.
An urbanization index (UI) which not only reflects
the characteristics of urbanization, but also
provides a comparable scale is worth pursuing.
4/13/2017
2011 ACRS Conference, Taipei
34
The most apparent effect of urbanization is the
increase of built-up landcover. As the process of
urbanization continues, the built-up dominant
areas expand and the coverage ratio of built-up
landcover type increases.
Modern urban planning and zoning may also
require establishing urban forestry or city parks
and planting of road trees to alleviate the adverse
effect of urbanization. Presence of trees, parks, and
forestry in a neighborhood can be reflected by
higher cell-average NDVI values.
4/13/2017
2011 ACRS Conference, Taipei
35
We propose to develop a cell-specific
urbanization index by integrating the
coverage-ratio of built-up landcover type
and the cell-average NDVI:
UI cell  pb  (1  NDVI cell )
Theoretically, the value of cell-specific
urbanization index can vary between 0 and 2.
In reality, most UI cell values fall in between 0
and 1, and higher UIcell values indicate more
significant effect of urbanization.
4/13/2017
2011 ACRS Conference, Taipei
36
Relationship between cell-specific urbanization index
(UIcell) and coverage-ratio of built-up landcover type (pb).
4/13/2017
2011 ACRS Conference, Taipei
37
For an overall comparison of the degrees of
urbanization of the three study areas,
average urbanization index UI for each study
area was calculated.
Value of UI for the Tokyo, Kyoto, and Taipei
study areas are 0.91, 0.55, and 0.72,
respectively. These values are consistent
with the qualitative evaluation of the degree
of urbanization as described earlier.
4/13/2017
2011 ACRS Conference, Taipei
38
Gradient analysis of landscape spatial pattern
using cell-specific urbanization index
4/13/2017
2011 ACRS Conference, Taipei
39
Two-dimensional gradient analysis of the cell-specific
urbanization index of the Tokyo, Kyoto, and Taipei study areas.
4/13/2017
2011 ACRS Conference, Taipei
40
4/13/2017
2011 ACRS Conference, Taipei
41
11/16/2010
42
11/16/2010
43
11/16/2010
44
4/13/2017
2011 ACRS Conference, Taipei
45
Conclusions
Both the landcover type richness and evenness are
low in most of the Tokyo study area. The Tokyo
study area is highly urbanized and built-up is the
single dominant landcover type in almost all cells.
Landcover patterns of the Kyoto and Taipei study
areas show more diversified landcover distribution.
In addition to cells for which woods and built-up
are dominant landcover types, there are also
significant amount of cells with mixed and nondominant landcover types.
4/13/2017
2011 ACRS Conference, Taipei
46
Comparing to the Tokyo and Taipei study areas,
Kyoto is least urbanized and enjoys a good mixture
of different landcover types, based on the analysis
of landcover pattern in coverage-ratio space.
Area-average urbanization index for the Tokyo,
Kyoto, and Taipei study areas are 0.91, 0.55, and
0.72, respectively. Such results are consistent with
the results of qualitative evaluation using different
landscape metrics.
The cell-specific urbanization index can also be
used for two-dimensional gradient analyses to show
the spatial variation of the degree of urbanization
within each study area.
4/13/2017
2011 ACRS Conference, Taipei
47
Thanks for your attention!
4/13/2017
2011 ACRS Conference, Taipei
48