Analyzing urban green space pattern and eco network in hanoi, vietnam

This study aims at quantifying the landscape patterns and ecological processes or clearly linking pattern to process to identify green space changes and their driving forces, based on gr

O R I G I N A L P A P E R

Analyzing urban green space pattern and eco-network

in Hanoi, Vietnam

Pham Duc Uy Æ Nobukazu Nakagoshi

Received: 24 May 2007 / Revised: 20 August 2007 / Accepted: 7 September 2007 / Published online: 12 October 2007

International Consortium of Landscape and Ecological Engineering and Springer 2007

Abstract In Hanoi, the capital city of Vietnam, there has

recently been a growing awareness about the roles and

benefits of greening in urbanized areas As a result,

plan-ners and decision-makers propose a combination of water

bodies and green areas, using cultural as well as historic

values, in a strategic concept for city planning in Hanoi

This study aims at quantifying the landscape patterns and

ecological processes or clearly linking pattern to process to

identify green space changes and their driving forces, based

on gradient analysis combined with landscape metrics, GIS

support, and FRAGSTATS 3.3, from 1996 to 2003 The

results of gradient analysis taken four directions show that

green spaces have been become more fragmented in this

period, especially in the south and west directions These

changes could be caused by land use change, economic

growth, population increase, urbanization, and weakness in

planning and managing the urban development From this

context, graph theory was also applied to find any

eco-networking, by mitigating the fragmentation and enhancing

the green space connectivity, as a biodiversity conservation

strategy for the city Analyzing the green network based on

graph theory indicates that among six different network

scenarios which were produced from several models

(Traveling Salesman, Paul Revere, Least Cost to User),

network F with 37 links, and gamma (0.07), beta (0.62),

cost ratio (0.606), circuitry (0.098) and connectivity

(0.398) is the best option for ecological restoration in the Hanoi city This will be a basis for the 2020 Green Space Planning in Hanoi

Keywords Urban green spaces
Gradient analysis
Graph theory
Connectivity
Landscape metrics

Introduction

Urbanization is a vital process and one necessary for human development; and has been occurring much faster in developing countries than in developed countries How-ever, it also had a negative impact on city dwellers, the environment, and biodiversity To reduce these impacts, it

is found that the conservation and development of green areas are a good solution Therefore, recently, human beings over the world are paying attention to the roles and functions of them more and more Previous urban green space studies mention many cases where methods of landscape ecology are especially suitable for the urban process

Gradient analysis originated from vegetation analysis, and it is found that gradient analysis based on landscape metrics is useful and effective for studying the urbanization process (Luck and Wu 2002; Ma et al 2005; Yu and Ng

2007; Zhu et al 2006) Kong and Nakagoshi (2006) find that this method is useful for studying urban green spaces because the results of gradient analysis show changes in the spatio-temporal pattern and give light to the driving forces behind the process as well Luck and Wu (2002) also show that quantifying the urbanization gradient is an important first step to linking pattern with process in urban ecological studies because they found spatial pattern undoubtedly affects physical, ecological and socioeconomic processes

P D Uy (&)
N Nakagoshi

Graduate School for International Development

and Cooperation, Hiroshima University,

1-5-1 Kagamiyama, Higashi-Hiroshima 739-8259, Japan

e-mail: ducanhy2000@yahoo.com

N Nakagoshi

e-mail: nobu@hiroshima-u.ac.jp

DOI 10.1007/s11355-007-0030-3

How to conserve the pre-urban natural remnants and

create urban green spaces will be the most important task in

any effort to mitigate the potential impacts of urbanization

Linking gradient analysis with urban dynamics can help

detect such spatially explicit urban green space patterns,

and improve the ability of planners to integrate ecological

considerations in urban planning (Yu and Ng2007) Also,

applying graph theory, which is a useful tool in researching

landscape connectivity especially ecological network

research (Bunn et al 2000; Forman and Godron 1986;

Gross and Yellen 1999; Linehan et al 1995; Rudd et al

2002; Zhang and Wang 2006), helps to organize green

space networks for ecological restoration in terms of

reducing fragmentation impact and enhancing the

connec-tivity Because, in graph theory, like island biogeography

theory, gravity model is used to express the interaction of

habitat areas, which shows the greater area and number of

patches, the closer they are, the higher biodiversity and

colonization Graph theory used here represents through

green nodes, their interactions, and links used to connect

these nodes The root purpose of graph theory in ecological

restoration is to identify the most optimal network or flow

which satisfies both ‘‘least cost to builder’’ and ‘‘least cost

to user’’ as the best potential network for conserving

bio-diversity, especially in the urban context, where number

and area of green spaces are usually constrained Moreover,

in biodiversity, landscape connectivity has a special

sig-nificance for seed dispersal and wildlife movement, which

play a decisive role in determining the survival of a

metapopulation Rudd et al (2002) have showed that

connectivity analysis in urban green spaces, based on graph

theory presented here, explores the numbers and patterns of

corridors required to connect urban green spaces as part of

an overall biodiversity conservation strategy

The objectives of this study are to assess spatio-temporal

changes in green spaces, as well as identify their driving

forces; and examine the most effective network for

biodi-versity conservation based on graph theory In addition,

this study will research how to apply graph theory and

landscape metrics in organizing green spaces and

eco-networking, in order to optimize the benefits of urban green

spaces for biodiversity

Methods

Data and study area

Study area: Hanoi—the capital of the Socialist Republic of

Vietnam, is the political, economic, cultural, scientific and

technological center of the whole country with latitude

from 20530to 21230north, and longitude from 105440to

106020 east Hanoi is an ancient city with nine urban

districts and five rural districts, which has been developing for almost 1,000 years, viz since establishment in 1010 It

is located in the center of the Northern Delta with a pop-ulation of 3,055,300 (2004), and an area of 920.97 km2 (within downtown: 150 km2) The downtown area of Hanoi city was selected for this study (Fig 1)

Data sources: the primary data was obtained from satellite images including those from the 1996 Spot3 BW taken in September with a resolution of 10 m, band 1; and

2003 Quickbird taken in November with a resolution of 0.7 m, three bands A 2005 topographic map of 1:25000 was used for geo-referencing In addition, secondary data includes that from the 2020 Hanoi Master Plan, from the Hanoi Department of Planning and Architecture, and other sources

Analysis methods All satellite images were rectified, processed, and geo-referenced to the Universal Transverse Mercator (WGS_1984_UTM_Zone_48N) coordinate system, using the ERDAS image system (Version 8.5, ERDAS, Atlanta, GA, USA) The geo-referencing process was car-ried out with the necessary information from labeled latitude and longitude and distinct ground control points through field verification with a GPS-model Garmin-12 (Global Positioning System) and then these images were interpreted manually based on the ArcGIS 9 (Arc/Info, release version 9.1, ESRI, Redlands, CA, USA) platform

Fig 1 Hanoi (left down) and the studied urban area of Hanoi, Vietnam

Because the different resolution of the 1996 and 2003

satellite images caused difficulties in interpretation, we

used not only the ERDAS system to perform a resolution

merge but also the 1992 aerial photos, historical data and

reports combined with field surveys and ground-truthing

taken in August 2006 as referencing sources This allowed

for referencing, merging and validating of the necessary

data to make them more reliable and accurate Urban green

spaces in Hanoi were reclassified into seven types

includ-ing real green spaces or evergreen (parks, public green

spaces, roadside green spaces, riverside green spaces,

attached green spaces), and non-real green spaces called

open green spaces (agricultural land and cultivated alluvial

land) using Vietnamese standards and regulations as shown

in Table1 This allowed vector green maps for 1996 and

2003 to be created, and then converted into raster format

with a pixel size of 10 m · 10 m with the support of Arc/

Map Spatial Analysis (version 9.1, ESRI)

To analyze urban green space pattern change, only

landscape metrics, which is sensitive to landscape change,

was chosen since it includes compositional and

configu-rational metrics including: class area (CA), percent of

landscape (PLAND), patch density (PD), largest patch

index (LPI), landscape shape index (LSI), mean patch size

(MPS), and a weighted mean shape index (AWSI), number

of patches (NP), and mean shape index (MSI) by using the

raster version of FRAFSTATS 3.3 (McGarigal et al.2002)

(Table2) Firstly, an analysis of green space change at

class level metrics (CA, PLAND, PD, LPI, LSI, MPS, NP,

AWSI) over the entire area was implemented to capture synoptic features Then, to detect the urban green space gradient change, samples were taken along two transects: west–east and south–north, cutting across the Hanoi downtown area The center area is identified as the ancient quarter and shown in Fig.2 The west–east and south– north transects were composed of eight and seven

2 km · 2 km zones respectively Landscape level metrics were computed using an overlapping moving window across transects with the support of FRAGSTATS 3.3 The window moved over the whole landscape and calculated the selected metrics inside the window As shown by Kong and Nakagoshi (2006), although this method can cause over-sampling in the center and under-sampling in the periphery, it does not affect the final conclusion Moreover,

it can describe the landscape pattern better; and the moving window analysis supported by FRAGSTATS combined with landscape metrics is a suitable approach for such analysis, Luck and Wu (2002), Yu and Ng (2007), Zhu

et al (2006)

Network analysis for organizing green space systems, with the purpose of ecological restoration based on graph theory, is done in terms of nodes (non-linear elements) and links (linear elements) Nodes in this study refer to green patches or habitat areas with an area of more than 10 ha Ten hectares was chosen as a hypothetical minimum area because it can encompass a wider range of species Hanoi areas are home of a variety of species such as insects (595 species, 395 genera, 101 families and 13 orders), reptilia

Table 1 Reclassification of urban green spaces

Circular Number 20 2005 TCXDVN 362: 2005

central park \15 ha and

£ 50 ha, multiple functional park [10 ha and £15 ha, small park

planted vegetation and higher bio-diversity

recreational areas such as flower gardens, squares, historical sites and others

Roadside green space (linear element)

Roadside green spaces RoSP Trees planted beside transportation

routes, creeks, canals to prevent dust, noise, add beauty and create corridors Riverside green spaces RiSP

hospitals, factories, temples and other organizations

places sometimes cultivated in the year, grassland, and aquatic plants

cultivated activities

Table 2 Definitions of landscape metrics (adopted from McGarigal et al 2002)

Compositional measures

patches of the corresponding patch type divided by 10.000 (to convert to hectares).

corresponding patch type (class).

particular patch type; a measure of landscape composition and dominance of patch type.

100 hectares

[ 0 without limit

divided by the number of patches of that type.

the corresponding patch type divided by total landscape area (m 2 ), multiplied by 100 (to convert to a percentage).

Configurational measures

corresponding class divided by the maximum length of class edge for a maximally aggregated class, a measure of class aggregation or clumpiness.

divided by the square root of patch area (m 2 ) for each patch of the corresponding patch type, divided by the number of patches of the same type or MSI equals to the average shape index of patches of the corresponding patch type.

Area weighted mean shape index AWMSI AWMSI equals the sum, across all patches of the

corresponding patch type, of each patch perimeter (m) divided the square root of patch area (m2), multiplied by the patch area (m2), divided by total class area or AWMSI equals

to the average shape index of patch of the corresponding patch type, weighted by each area.

limit

Fig 2 The 1996 and 2003

green transects for gradient

analysis

(33 species, 12 families, 3 orders), mammalian (38 species,

16 families, 6 orders) etc Especially, there are many

threatened species (9 reptiles), (3 insects), (7 small

mam-mals) (Yen2005) Almost all these species have a habitat

area smaller than 10 ha, for example the musk shrew

(Suncus murinus) and tree shrew (Tupaia glis) with habitat

ranges 240–1,200 m2 (0.024–0.12 ha), Chinese

ferret-badger (Melogale moschata) with habitat ranges 4–9 ha

etc The green patches left were considered as links acting

as corridors or stepping stones In graph theory and gravity

models for analyzing networks, node weight was calculated

as follows: Na= {X (ha)/S (ha)} · 10 (Linehan et al

1995) Where: Na= the node weight for the green space,

X= the area of the green space measured in hectares,

S= the minimum area required for the indicator species,

and multiplying by a factor of 10 normalizes the data

Connectivity analysis is based on the interaction between

pairs of nodes in the gravity model as shown by Linehan

et al (1995) Gab= {Na· Nb}/Dab2 (km) and Gab= Gba;

where Gabthe level of interaction between nodes a and b;

Nathe weight of node a; Nbthe weight of node b; and Dabis

the distance between the centroid of node a and the

cen-troid of node b Then, network generation was carried out

based on the concept of ‘‘least cost to user’’ and ‘‘least cost

to builder’’ There are two major groups of network

mod-els: branching and circuit, producing three graphs (Fig.3)

Branching networks, for example Paul Revere model-the

simplest network, are formed based on connecting all

nodes but visiting once, and there are no extraneous

seg-ments (Linehan et al 1995; Rudd et al 2002) Thus, no

circle is created While circuit models are established based

on the form of closed loops, for instances Traveling

Salesman-the simplest circuit network where each node is

connected only to two other nodes, and Least Cost to

User-the most complex circuit network where all nodes are

connected each other (Linehan et al 1995; Rudd et al

2002) Connectivity analysis, which is tested following the

above network models, shows the level of interaction

between each of the green spaces in the study area Next, it

is necessary to evaluate the circuit network and branching

network approaches This evaluation is based on gamma,

beta, and cost ratio indices (Forman and Godron 1986;

Linehan et al.1995; Rudd et al.2002) where:

Gamma¼ ðnumber of linksÞ=ðmaximum number of linksÞ; Beta¼ ðnumber of linksÞ=ðnumber of nodesÞ; and the Cost ratio ¼ 1  ðnumber of linksÞ=ðdistance of linksÞ:

To analyze networks here, the formulae of circuitry and connectivity (Forman and Godron 1986) were also used, where L and V are links and nodes respectively

Circuitry: a = L – V + 1/2V – 5 where zero means no circuitry, and positive values mean more circuitry Connectivity: c = L/3(V – 2) in that greater values mean more connectivity

Results

Synoptic characteristics of urban green spaces in Hanoi

A study of the synoptic characteristics using landscape metrics over the entire study area will provide general information on urban green space patterns in Hanoi In the year 1996, there were 357 green patches totalling 8449.6 ha; and in the year 2003, there were 669 green patches totalling 7139.4 ha Comparing these two years, there was a reduction in green space area of 1310.2 ha and

an increase in the number of patches by 312 The reduction

in the whole area: parks, attached green spaces, and agri-cultural land was 2.2, 3.4, 2.7 and 3.1% per year The patches increased at about 12.5% per year Likewise, the increase rate of patches for P, PGS, AGS, AA, CAL, RiSP, RoSP were 14.3, 23.8, 11.6, 11.1, 5.3, 14.3, 20.95% (Table 3a, i) respectively The increase in the fragmenta-tion index, such as in the number of patches (NP) and patch density (PD), indicates that the landscape was highly fragmented providing less connectivity, greater isolation and a higher percentage of edge area in patches McGarigal

et al (2002), Luck and Wu (2002) have shown that NP and

PD are two important metrics, which are usually used for assessing the landscape fragmentation As expressed in Table 3a, b, agricultural land (AA), attached green spaces (AGS) and parks (P) had a reduction of area of 1,170 ha,

247 ha, and 20.5 ha, respectively This suggests that the urban sprawl process is occurring strongly in the peri-urban areas, and the city became more compact However, public green spaces (PGS) and roadside green spaces (RoGS) showed a remarkable increase PLAND (percent of land) of real green spaces (parks, public green spaces, riverside green spaces, roadside green spaces) showed a slight increase from 18% in 1996 to 19% in 2003 However, non-real green spaces or open-green spaces (agricultural land) reduced from 63 to 58% in the period 1996–2003 This reflects the dominance of this green space type AA exists

at the periphery of urban areas Thus, a decrease of its

Paul reserve Traveling salesman Least cost to user

Where Node: and Link:

Fig 3 Examples of branching and circuit networks

PLAND suggested an increase in the urban sprawl process.

The ranking of PLAND for urban green spaces is

AA[CAL[AGS[RoSP[PGS[P[RiSP for both of the

years mentioned The density of all types of green spaces

increased from 1996 to 2000 (Table3c) This index

indi-cated a higher fragmentation of all green space types and

could be confirmed by the decrease in mean patch size

index (MPS) of all green space types (Table3f)

The largest patch index (LPI) of AA reduced from 12.8

to 10.34 indicating that AA patches became smaller

(Table3d) An increase in LSI (landscape shape index)

showed that the total length of edges within the landscape

increased, and shape become more irregular as these

green spaces suffered more impact from surrounds The

AWMSI (area weighted mean shape index) of almost all

green space types increased also, indicating that the patch

shape became more irregular However, the decrease of

AWMSI for RiGS (river green spaces) combined with an

increase of CA and PLAND indicated an improvement of

this green space type over that of other green spaces In

general, fragmentation of green patches increased from

1996 to 2003 Green patches became smaller and more isolated

Gradient analysis of landscape level metrics Gradient analysis of landscape level metrics is shown in Figs 4a–g,5a–g By comparing NP and PD (Fig.4a, b) in the west–east transect, there was a shift in peak position, as well as an increase of NP and PD in going from the center

in 1996 to 4 km west in 2003 Fluctuation in NP and PD in the east was smaller than that in the west Both indicators suggest that the dynamic for this variation might be the urbanization process The above judgment was confirmed

by considering Mean Patch Size (MPS), where the lowest values were distributed from 4 km west to the center, the closer to the center the higher the MPS The MPS peaked at

a distance 4 km to the east A decline of NP from 1996 to

2003 indicated that green patches became smaller This is obvious since they are under pressure from human impact more and more The LPI in 1996 at 4–2 km west was

Table 3 Class level metrics of green spaces

(a) Class area (CA)

(b) Percent of landscape (PLAND)

(c) Patch density (PD)

(d) Largest patch index (LPI)

(e) Landscape shape index (LSI)

(f) Mean patch size (MPS)

(g) Area weighted mean shape index (AWMSI)

(i) Number of Patches

higher than that of the year 2003 showing that green spaces

at this distance became more fragmented and smaller

except other distances This may indicate that some green

spaces were preserved as core areas while other green

spaces were reducing in area Combining this result with

configurational metrics, we can quantify and understand

better the variation in urban green space patterns As shown

in the Fig.4e, LSI peaked at a distance around 4 km west

and in the transect center, suggesting that at these distances

the shape of urban green spaces is the most complex This

seems to reflect different stages in urban development The

center area is the old quarter and is very compact; the

neighboring areas belong to the government and French

colonial towns; and outside these are new urbanized areas

and urban fringes However, the Mean Shape Index (MSI)

was stable along the transect and over time While there was a big fluctuation of AWMSI in the year 2003, espe-cially in the center area to 3 km west, it then decreased slightly on going eastward

Like the west–east transect, the peak position of NP in the south–north transect varied from near center (1996) to

4 km south (2003) and then reduced in both directions The

NP of 2003 was much larger than that of 1996 and its fluctuation in the south was stronger as well (Fig.5a) Together with NP, PD is one of the most important frag-mentation indices, the PD of 1996 and 2003 peaked at

4 km south and its change in the north was lower than that during 2003 The LPI for urban green spaces varied irregularly with multiple peaks At 4 km south, the varia-tion of NP and PD was the strongest, but the fluctuavaria-tion of

0 20 40 60 80

-8 -6 -4 -2 0

Number of Patches (NP) Patch density (PD)

0 30 60 90 120

Mean Patch Size (MPS) (ha)

0 30 60 90

Largest Patch Index (LPI)

0 20 40 60 80 100

Landscape Shape Index (LSI)

0 4 8 12 16

Mean Shape Index (MSI)

0 1 2 3 4 5

Area weighted mean shape index (AWSI)

0 1 2 3 4 5 6 7

Distance to center (km)

1996 2003

d c

f e

g

6 4 2

Fig 4 Gradient changes in

landscape level metrics of

Hanoi urban green spaces, from

west to east in the period

1996–2003

LPI and MPS was lowest For MPS closer to the center,

there was a remarkable decrease comparing 2003–1996,

especially from a distance of 6 km southward This is

evidence that these green patches here suffered more

pressure from surrounds A decrease of LSI at 4 km south

suggested that the shape of green patches at this distance

became more complex, while in the center area there was

an improvement The MSI showed no big changes along

the transect and a slight increase toward the center

Com-pared to the west–east transect, variation in the MSI of the

south–east transect was bigger The AWMSI seems to be

similar to the LSI, with the highest values being found at a

distance of 2 km south where the AWMSI then showing a

reduction at 4 km south This was consistent with an

increase in NP and PD The AWMSI then slightly

increased again at a distance of 6 km south However, it

decreased toward the center when comparing 2003 and

1996 In general, the variation in landscape metrics of

urban green spaces in the south was stronger than that of

the north The peak change was at around 4 km south

indicating that land use change at this distance was

greatest Moreover one of the more interesting results, in terms of configurational metrics, was found at the center where the LSI, MSI and AWMSI for urban green spaces declined on comparing 2003 and 1996 This revealed an improvement in green patch shape

Network analysis The result of the node interaction (gravity model) of the 33 existing green patches with an area larger than 10 ha (Table 4) and the common network types (Fig.3) have produced six different network scenarios from A to F (Fig.6) Specifically, the theory maximum expresses all nodes connected each other including unfeasible links and feasible links Feasible links to connect these nodes are identified based on the existing land use including corridors (road green ways, etc.), open spaces, or other small green spaces, and unfeasible links are virtual links or do not exist

in the reality (business areas, busy highways, etc.) (Linehan

et al 1995) The network A based on the network model

Number of patches (NP)

0 10 20 30 40 50 60

Patch Density (PD)

0 50 100 150 200

Largest Patch Index (LPI)

0 25 50 75 100

Mean Patch Size (MPS) (ha)

0 10 20 30 40 50

Landscape Shape Index (LSI)

0 4 8 12 16

Mean Shape Index (MSI)

0 0.5 1 1.5 2 2.5

Area weighted mean shape index (AWSI)

0 2 4 6

Distance to center (km)

1996 2003

a

d c

f e

g

2

b

Fig 5 Gradient changes in

landscape level metrics of

Hanoi urban green spaces, from

south to north in the period

1996–2003

‘‘Least Cost to User’’, namely project max, expresses the

highest connectivity or connects all green spaces with all

feasible links The network B, based on circle networking,

represents the connection of all largest nodes only The

network C was built based on the network model ‘‘Paul

Revere’’ or branching network The Network D was

developed following the network type ‘‘Traveling

Sales-man’’ or circle networking The network E represents the

connection of the closest green patches as its name

‘‘Minimum Spanning Tree’’ Finally, the network F, based

on the ‘‘Least Cost to User’’, expressed the connection of

selected groups of green patches The gamma, beta and

cost ratio were used to evaluate each graph model or

net-work scenario (Table5) In addition to using gamma, beta

and cost ratio scenarios to evaluate networks, the circuitry

(a) and connectivity (c) indices were also used to analyze

network structure These formulae were adopted by

For-man and Godron (1986), Hagget et al (1977) In analyzing

networks, these indices are not as sensitive as the other

mentioned indices but they support connectivity analysis

more efficiently and clearly (Table5)

Discussion

What is the driving force of green space change

in Hanoi?

Analyzing green space patterns over the entire landscape,

and analyzing gradients based on landscape metrics along

two transects, showed that green spaces have changed at

different distances and in different directions, from 1996 to

2003 However, analyzing synoptic characteristics of

landscapes as traditional ways that the averaging of

land-scape metrics over an entire study area may lead to

incorrect interpretation of the causal dynamics in the

region As shown by Kong and Nakagoshi (2006, p 12),

‘‘It is difficult to link changes in green space patterns in

local areas accurately with the processes that produced

these changes’’ This difficulty can be solved by using

gradient analysis or the ‘‘moving window’’ method

com-bined with spatially explicit landscape metrics This

method can provide adequate quantitative information

about the structure and pattern of urban green spaces

Therefore, a better link between pattern and process, and a

more effective capture of the dynamic changes can result

Generally, there are two main driving forces causing the

urbanization process: population and economy (Ma et al

2005) In addition, Luck and Wu (2002) recognize

urbanization as one of the most important driving forces

for land use and land cover change When studying the

spatio-temporal green space change in Jinan City (China),

Kong and Nakagoshi (2006) found that the driving forces Table

Table 4 continued

1 116.9 344.2 184.2 1407.1 171.5 1111.4 1959.9 430 57.3 1066.2 806.9 1485.9 9039.6 1040.4 1072.6 1320

2 151.5 400.9 142.6 1192 92.7 826.3 1562.1 300.2 39.3 713.6 473.8 783.9 3063.6 543.7 332.9 256.3

4 240.3 606.6 225.5 1194.8 55.7 455.2 839.8 156.3 20.9 332 205.7 330.4 930.5 225.3 97.9 62.2

Because Gab= Gbaso that this table is symmetrical and it is unnecessary to calculate both values (Linehan et al 1995 )