Graduate School of Life and Earth Sciences
Master Thesis
42 EC
River flood risk in the Greater Mekong
Subregion under different urban
development scenarios
Master Earth Sciences
Future Planet Ecosystem Science Track
by
Lars Tierolf
April 2020
Supervisor and assessor
dr. ir. Jasper van Vliet
Vrije Universiteit
Second assessor
prof. dr. Marc Davidson
Universiteit van Amsterdam
Acknowledgment
This research project was supervised by dr. ir. Jasper van Vliet, assistant professor in the Environmental Geography group the Institute for Environmental Studies of the Vrije Universiteit. The R scripts applied in the sampling procedure and regression analysis were kindly provided by Niels Debonne. I thank Jasper van Vliet for his supervision, feedback and discussions during both the research proposal and project. I would furthermore like to thank Thomas Hofman for proofreading parts of my thesis, and my other friends for supporting me in many ways during this project.
Abstract
The share of the global human population residing in an urbanized environment is pro-jected to grow from 50 percent in 2008 to 68 percent in 2050. This trend is especially relevant in Southeast Asia, where developing nations transition from predominantly rural to predominantly urban societies in the coming decades. Much urban development here occurs in river deltas and floodplains, exposing humans and human assets to flood hazard. However, how urban areas will develop we do not know. This study aimed to assess future flood risk in five countries of the Greater Mekong Subregion under two contrasting ur-ban development scenarios. Urur-ban land was mapped on a rural-urur-ban gradient and land change was projected to the year 2035 in an urban expansion and an urban densification scenario. Flood risk curves were constructed for inundation levels with 10 to 500 year return periods. Results indicated flood damages to increase in all countries of the study area in both scenarios. Optimal urban development pathways for mitigating future flood risk were shown to be context dependent and resulted in lowest damages for all return periods. Flood risk estimates indicate that choices in urban development affect flood risk in severe events with low exceedance probabilities.
Contents
1 Introduction 9 2 Theoretical background 11 2.1 Urbanization . . . 11 2.1.1 Defining urban . . . 11 2.1.2 Urban development . . . 12 2.2 Flooding . . . 132.3 Land system modeling . . . 15
3 Methodology 17 3.1 Study area . . . 17
3.2 Classifying settlement systems . . . 18
3.2.1 Mapping settlement systems . . . 20
3.2.2 Recording the built environment . . . 21
3.3 Modeling land system change . . . 22
3.4 Flood risk analysis . . . 26
4 Results 29 4.1 Land system change projections . . . 29
4.2 Flood risk . . . 36
4.3 Sensitivity analysis . . . 42
5 Discussion 44 5.1 Methodology . . . 44
5.1.1 Land system mapping . . . 44
5.1.2 Land-system modeling . . . 45
5.1.3 Flood damage calculations . . . 45
5.2 Results . . . 46
5.2.1 Projecting urban development . . . 46
5.2.2 Flood risk . . . 47
CONTENTS CONTENTS
6 Conclusions 49
References 50
Appendices 55
A Land system mapping 56
B Recording the built environment 58
C CLUMondo model settings 61
D Land system maps 64
List of Figures
1 Conceptual framework of risk as operationalized in this study following definitions of risk by Kron (2002); UN-SPIDER (2019). . . 14 2 Conceptual framework of land systems. Figure reprinted from “The
emer-gence of land change science for global environmental change and sustain-ability,” by B. L. Turner, E. F. Lambin, & A. Reenberg, 2007, Proceedings of the National Academy of Sciences, 104(52), p. 20667. Copyright © National Academy of Sciences. . . 16 3 The percentage of the total population residing in an urban environment.
Data provided by United Nations (2018). . . 18 4 Boxes 1-4 provide brief descriptions of the settlement systems mapped in
the initial land system map. . . 19 5 Distribution of building material found in each settlement class. . . 21 6 Conceptualization and parameterization of the CLUMondo model for
ap-plication in the Greater Mekong Subregion (GMS). Figure adapted and modified from Van Asselen and Verburg (2013). . . 23 7 Flood damage curve for residential buildings provided by Huizinga, De Moel,
Szewczyk, et al. (2017). A damage factor of 1 is reached at an inundation depth of 6 meters. . . 28 8 The amount of settlement pixels present in the beginning and end of the
simulation period. . . 30 9 Snapshot of Battambang, Siem Raep (1) and Phnom Penh (2) of 2015 and
2035 in the settlement expansion (a) and intensification scenario (b). . . . 31 10 Snapshot of Vientiane (1) and Pakse (2) of 2015 and 2035 in the settlement
expansion (a) and intensification scenario (b). . . 32 11 Snapshot of Mandalay (1) and Yangon (2) of 2015 and 2035 in the
settle-ment expansion (a) and intensification scenario (b). . . 33 12 Snapshot of Phuket (1) and Bangkok (2) of 2015 and 2035 in the urban
LIST OF FIGURES LIST OF FIGURES
13 Snapshot of the Red River Delta near Hanoi (1) and the Mekong River delta near Ho Chi Minh City (2) of 2015 and 2035 in the settlement expansion (a) and intensification scenario (b). . . 35 14 Flood damage map of Phnom Penh in both scenarios with inundation levels
of 10 year return periods. Inundation map provided by Dottori et al. (2016). 37 15 Flood damage map of Vientiane in both scenarios with inundation levels
of 10 year return periods. Inundation map provided by Dottori et al. (2016). 37 16 Flood damage map of Mandalay in both scenarios with inundation levels
of 10 year return periods. Inundation map provided by Dottori et al. (2016). 39 17 Flood damage map of Bangkok in both scenarios with inundation levels of
10 year return periods. Inundation map provided by Dottori et al. (2016). . 39 18 Flood damage map of Mekong River delta near Can Tho iin both scenarios
with inundation levels of 10 year return periods. Inundation map provided by Dottori et al. (2016). . . 40 19 Flood risk curves plotted with modeled flood damage in settlement area
for return periods of 10, 20, 50, 100 and 500 years. . . 41 B1 Random village pixel near Pak Thang in Thailand. Imagery obtained from
Google Earth (2014a). . . 59 B2 Random urban dense pixel near Thon Buri, Thailand. Imagery obtained
from Google Earth (2014b). . . 59 C3 Conversion matrix . . . 63
List of Tables
1 Explanatory variables included in the regression models . . . 24
2 Building classes with corresponding adjustment factors from Huizinga et al. (2017). . . 28
3 GDP per capita in 2015 and construction cost . . . 28
4 Net increase in population, settlement area and built-up land in 2035 com-pared to 2015. . . 29
5 Damages to settlement systems in flooding with 10 year return periods. . . 36
6 Sensitivity of annual flood damage estimates (in billion 2010 EUR) to un-certainty in maximum damage and settlement building composition . . . . 43
7 Observed average annual growth rate (AGR) of built-up land compared projected AGR. . . 46
A1 Geospatial data sets used in the classification of the initial land system map. 56 A2 Population density and built-up land thresholds used in settlement classi-fication. . . 56
A3 Reclassification scheme of land cover to land systems. . . 57
B1 Recording of the built environment of settlement grid cells. Standard de-viation indicates a variance between observations. . . 58
B2 Average built-up density extracted from the GHSL. . . 60
C3 Suitability factors assessed in the regression analysis . . . 61
C4 Country specific demand for population . . . 62
C5 Land system services (population) . . . 62
Chapter 1
Introduction
As of 2008, the majority of the human population is residing in an urban environment (United Nations, 2018). This portion is projected to increase to 68 percent by 2050. Most of this urban growth is expected to occur in Africa and Asia, where rural- urban migration comprises a large component of the urbanization process (Jones, 2002). Due to the emergence of motorized mobility, urbanization is no longer considered a diffusion process from established urban cores (Antrop, 2004). Rapid economic development results in transformation of rural villages into cities and cities leapfrogging into traditionally agricultural areas, both accompanied by gradual urbanization of the peri-urban landscape. This trend is especially relevant in the developing world, where employment opportunities continue to attract rural migrants to its growing urban settlements (Jones, 2002). An increase of urban population is accompanied by an even larger increase in urban area, however how this urban area develops in future remains unknown (Angel, Parent, Civco, Blei, & Potere, 2011).
Urbanization processes thus create a heterogeneous landscape of different built envi-ronments, ranging from villages and extensive suburban landscapes to dense metropolitan area (Li, van Vliet, Ke, & Verburg, 2019). This development is considered a major driver of increased future flood risk (Jonkman, 2005; Winsemius et al., 2016). However, large scale flood risk estimates are sensitive to uncertainties in asset exposure and vulnerability of affected urban area (de Moel et al., 2015). Representation of urban area in a single class fails to capture a diversity of urban environments and limits assessment of the impact of urban development on future flood risk. Projecting two different urban scenarios enables assessment differences in flood risk under different development pathways while taking into account the heterogeneity of the urban landscape. In this study two urban develop-ment scenarios were projected to year 2035 to analyze the effect of urban expansion and densification on future flood risk.
CHAPTER 1. INTRODUCTION
Research aim and questions
Urban development in flood-prone areas is a major driver exacerbating future impacts of flooding on socio-economic development (Winsemius et al., 2016). How urban areas develop in the future is unknown. However, two main development pathways can be distinguished: 1) urban densification and 2) urban expansion. Both pathways result in different urban landscapes, ranging from multistory apartment buildings to spread out single household housing. Differences in built environment affect flood damage estima-tions, since the exposed asset value is higher in a dense urban core compared to a sparsely built-up suburb. Understanding how different urban development trajectories affect future flood risk addresses uncertainty in urban development and contributes to the advance-ment of sustainable urban developadvance-ment strategies in areas coping with enhanced flood risk.
The aim of this study was to assess future river flood risk in two urban development scenarios. A land system map aimed at differentiating settlements on a rural-urban gradi-ent was constructed, two urban developmgradi-ent scenarios were projected to the year 2035 and flood risk curves for a range of return periods were constructed. The two main research questions addressed in this study therefore were:
1. How does the urban landscape change considering two different urban development trajectories?
Chapter 2
Theoretical background
In this chapter the various concepts applied in the proposed study are discussed. Defini-tions and drivers of urban development are presented, flood risk is discussed and concep-tual framework to land use modeling is introduced.
2.1
Urbanization
2.1.1
Defining urban
Urbanization is defined as a shift of human residence from rural to urban area and ex-pressed as the percentage of the total population residing in urban areas. Iimi (2005) identifies three factors of urbanization: natural increase in urban population, rural-urban migration and the transformation and reclassification of rural to urban areas due to a pop-ulation increase. Defining urban area however proves a difficult task, with countries often applying different definitions. The UN World Urbanization Prospects thus determines urban area with varying population density thresholds, resulting in areas being classified urban in one country, and rural in another (United Nations, 2018). Furthermore, official national statistics may reflect political motives rather than actual conditions, and when peri-urban regions are included cities may have drastically larger population sizes than stated in official documents (Jones, 2008; Sheng, 2011).
The simplified rural-urban dichotomy fails to accurately describe urbanization pro-cesses, as many areas considered rural have through increased connectivity, immigration and economic development adopted urban land use patterns (Simon, McGregor, & Nsiah-Gyabaah, 2004). These transitional zones between distinctly urban and rural areas are often defined rural-urban fringes/ transition zones, peri-urban zones/ areas or interfaces (Simon, 2008). A uniform definition of this rural- urban fringe does not exists, resulting in a lack of comparative datasets. Simon (2008) argues it to be more useful to conceive these interfaces as an urban-rural continuum or gradient. Illustrating the erratic growth of Southeast Asia’s cities, Friedmann (2011) defines this peri-urban region as a “zone
2.1. URBANIZATION CHAPTER 2. THEORETICAL BACKGROUND
of encounter, conflict, and transformation surrounding large cities” (p. 426), indicating different land-use patterns than observed in the urban center.
Wandl, Nadin, Zonneveld, and Rooij (2014) aimed to improve classification meth-ods for representation of land that was neither completely rural or urban, providing a conceptual framework describing the different functions and land-use intensities of these ’territories-in-between’. The approach to urban land in this thesis aims to account for this heterogeneity in urban area by representing urban land in settlement systems on a rural-urban gradient. Urban development is considered a consequence of urbanization processes, and represented by incremental steps on a rural- urban gradient (Li et al., 2019).
2.1.2
Urban development
Urban development in general is characterized by the intensification of land use for set-tlement and transportation purposes (Storch, Eckert, & Pfaffenbichler, 2008). In most cases this development comes at a cost of former agricultural land. Urban development in the developing world is largely driven by rural-urban migration resulting from employ-ment opportunities in a growing manufacturing industry (Webster, Cai, & Muller, 2014). As regions develop other factors determining the need for urban area emerge. Demand for shopping complexes, international airports, or larger living spaces become important drivers of urban land take. Although population growth is slowing nearing the end of this decade, decreasing household size as a result of improved living standards and aging populations continue to increase demand for urban area. Policy decisions, economic de-velopment and demographic change results in various trajectories of urban dede-velopment. Two main urban development trajectories are discussed in this study, each affected by demographic, socio-economic changes and spatial planning. Development can result in 1) urban densification, when a city or town accommodates a growing number of house-holds and activities on already urbanized land resulting in a higher functional density with limited expansion, and 2) urban expansion or sprawl, when a city or town accom-modates a growing number of households by expansion resulting in dispersed residential and commercial development in the urban-rural fringe, or 3) combinations of both.
Densification of urban area, also termed compact city development, aims to accom-modate a high number of households and urban functions while reducing areal expansion, often enforced through spatial planning. This intensification of urban land use is realized by transformation of already built-up land into multifunctional high-rise building and infill development, potentially replacing green urban infrastructure. Storch et al. (2008) emphasize that the variety of urban land uses, a mix of housing, employment and facilities, increases the efficiency of land use in compact cities. While densification has the poten-tial to reduce carbon footprint by shortening of commute and increased energy efficiency,
2.2. FLOODING CHAPTER 2. THEORETICAL BACKGROUND
loss of urban green infrastructure due to infill development however proves detrimental to urban ecosystem services, such as flood retention (Haaland & van Den Bosch, 2015). Urban development without spatial planning often results in urban sprawl. Features of urban sprawl include low population density, segregated land uses and a lack of significant centers (Ewing, Pendall, & Chen, 2002).
2.2
Flooding
UN-SPIDER (2019) defines flood: A general and temporary condition of partial or com-plete inundation of normally dry land areas from overflow of inland or tidal water from the unusual and rapid accumulation or runoff of surface waters from any source. Flooding is one of the most disruptive and frequent of natural hazards, resulting in damage to the built environment, temporal displacement and loss of life and livelihood. The worldwide impact of river flooding is expected to increase, mainly due to increasing population den-sities and economic development in flood prone areas (Jonkman, 2005; Winsemius et al., 2016). Asia is the most affected, resulting in societal disruption and loss of life (Jonkman, 2005).
Flood risk
Flood risk is conceptualized as a product of hazard, exposure and vulnerability (Kron, 2002). Hazard in the context of flood risk is perceived as the possible future occurrence and severity of a flood event. Exposure refers to buildings and infrastructure present in flood prone areas. In this study vulnerability refers to the structural vulnerability of buildings when exposed to flooding. An overview of the operationalization of flood risk in this study is presented in figure 1.
Hazard
An essential component in assessing flood risk is flood hazard mapping. Flood hazard mapping was traditionally done with historical inundation observations from past flood events. However, satellite images of these events are of limited availability and observed time series are likely to be too short to detect rare destructive events (Jongman, Win-semius, Fraser, Muis, & Ward, 2018). Computer models allow researchers to assess flood hazard based on discharge events with a certain estimated return period, also termed flood waves, facilitating assessment of downstream flood prone areas beyond the range of observed discharge events. Depending on the scale of the assessment, precipitation com-pounding the flood wave discharge or climatic data from an atmospheric model accounting for climate change may be included in the model. In this study hazard is operationalized with modeled global inundation maps with return periods of 10, 20, 50, 100 and 500 years
2.2. FLOODING CHAPTER 2. THEORETICAL BACKGROUND
(Dottori et al., 2016). Return periods represent exceedance probabilities, indicting the probability of an inundation event occurring in a given year.
Exposure
Jongman, Ward, and Aerts (2012) distinguished two approaches to modeling exposure in flood risk assessment, namely a population method and a land-use method. This study applies a land-use based method, operationalizing exposure as the reconstruction value of residential buildings in areas subjected to flooding. Settlement systems are assigned economic values based on GDP, building composition and building density following the methodology proposed by Ward et al. (2013); Huizinga et al. (2017).
Vulnerability
Vulnerability has many and operational definitions and biophysical and societal aspects, and broadly entails the potential for loss and the capacity of a society to deal with the flood event (Cutter, 1996; Koks, Jongman, Husby, & Botzen, 2015). As this study addresses tangible flood impact, vulnerability is operationalized with a depth-damage curve indicating the fraction of a building damaged considering different water depths (Merz, Kreibich, Schwarze, & Thieken, 2010). Huizinga et al. (2017) developed a suite of depth-damage curves accounting for regional differences in building integrity inferred from local data on flood damages. This study makes use of the depth-damage curves developed by Huizinga et al. (2017).
Figure 1. Conceptual framework of risk as operationalized in this study following definitions of risk by Kron (2002); UN-SPIDER (2019).
2.3. LAND SYSTEM MODELING CHAPTER 2. THEORETICAL BACKGROUND
2.3
Land system modeling
Modeling land systems provides possibilities for exploring human-environmental interac-tions. Modeling is often regarded solely as a tool for predicting system behavior. However, models serve a multitude of purposes essential for scientific conduct. The rigorous qualita-tive and quantitaqualita-tive analysis required for modeling land-use change can aid in explaining land systems behavior, guide data collection and lead to the identification of knowledge gaps (Verburg, 2006; Epstein, 2008). Conceptualizing the various biophysical, economic, cultural and institutional factors interacting in a land system furthermore provides a framework on which to base future scientific research. Applications of the land system model framework include assessments of agricultural abandonment, agricultural expan-sion, carbon sequestration, biochemical cycles and urbanization (Lawrence et al., 2007; Verburg, van Berkel, van Doorn, van Eupen, & van den Heiligenberg, 2010; Weng et al., 2002). The land system framework thus facilities a broad variety of research possibilities, deeming it suitable for projecting urban development pathways.
A land system is here defined as a system comprised of a human and biophysical subsystem, with human decision-making being the dominant driver of change (figure 2). The human subsystem interacts with the biophysical subsystem through land use, the biophysical subsystem feeds back through delivering goods and services (Turner, Lambin, & Reenberg, 2007). Examples of land use are rice cultivation, grazing by livestock and residential land use. Goods and services provided by these land uses would be rice, livestock produce and population. Land-system modeling aims to capture interactions between the two subsystems based on driving forces and processes.
Settlement systems are considered systems of mixed land use. Land in the more urban classes such as dense urban and suburban settlement systems is used primarily for residential and commercial purposes, while in the sparse suburban and village settlements a larger fraction of land is used for agricultural practices. Since this study aims to project urban development based on demand for population only residential land use is considered in the model.
2.3. LAND SYSTEM MODELING CHAPTER 2. THEORETICAL BACKGROUND
Figure 2. Conceptual framework of land systems. Figure reprinted from “The emergence of land change science for global environmental change and sustainability,” by B. L. Turner, E. F. Lambin, & A. Reenberg, 2007, Proceedings of the National Academy of Sciences, 104(52), p.
Chapter 3
Methodology
In this study urban area was represented by settlement systems and mapped on a rural-urban gradient. The CLUMondo land-use model developed by Van Asselen and Verburg (2013) was applied to create two contrasting settlement change scenarios for the year 2035 in Cambodia, Laos, Myanmar, Thailand and Vietnam. An initial land use map for the year 2015 was constructed, demand for services in terms of population provided by settlement systems was determined and local suitability was calculated based on a regional regression analysis. Flood risk was then quantified as asset damage and flood risk curvers were constructed for inundation levels with 10 to 500 year return periods. In this chapter the study area, settlement classification procedure, CLUMondo application and flood risk calculations are presented. Model settings and scrips applied in the flood risk estimations are made available on https://doi.org/10.6084/m9.figshare.12164007. The CLUMondo model was made publicly available by the Institute for Environmental Sciences of the Vrije Universiteit (Amsterdam).
3.1
Study area
A study area in mainland Southeast Asia including Vietnam, Thailand, Laos, Cambo-dia and Myanmar was selected. These countries together with the Yunnan province and Guangxi Zhuang Autonomous Region in China make up the Greater Mekong Subregion (GMS). Countries cooperate in the Greater Mekong Subregion Economic Cooperation Program through investment in infrastructure and urban development (Greater Mekong Subregion, 2020). Differences exist in between countries in terms of economic development and governmental structures. However, all have experienced rapid urbanization recently (see figure 3). Cambodia, Laos, Vietnam and Myanmar, nations with a socialist/ commu-nist history, have adopted more liberal policies and market-oriented reforms in the 1980s and 1990, resulting in recent economic development (Shultz & Pecotich, 1997). In Thai-land the process of urbanization has been longer ongoing, however decreasing household size continuous to increase the need for residential land use.
3.2. CLASSIFYING SETTLEMENT SYSTEMS CHAPTER 3. METHODOLOGY
Three major river systems flow through the study area, the Irrawaddy (Myanmar), the Chao Phraya (Thailand) and the Mekong River (Thailand, Laos, Cambodia and Vietnam). Although functioning as vital commercial waterways, supplying hydroelectric power and facilitating agriculture on their floodplains, these rivers often in inundate sig-nificant amounts of urban land resulting in economic and social damages. Most notably the 2011 flooding of the Chao Phraya river resulted in an estimated damage to housing of approximately 2.7 billion US dollar (The World Bank, 2011).
Figure 3. The percentage of the total population residing in an urban environment. Data provided by United Nations (2018).
3.2
Classifying settlement systems
A classification system aimed to map settlement systems on a rural-urban gradient in-ferred from population density and land cover was constructed. Settlement mapping along this gradient allows for modeling gradual urban densification, simulating the increase of land use intensity in terms of building and population density. Since this modeling exer-cise precedes the application of depth-damage curves to various building types, the choice was made to distinguish settlement systems based on building composition. Four settle-ment system types were formulated, namely village, sparse suburban, dense suburban and dense urban systems. It is assumed that the built-up environment of each settlement class has a characteristic composition of informal, wooden, masonry and concrete structures, with wood being the main construction material applied in villages and concrete being the dominant construction material used in a dense urban environment. A dense urban environment is furthermore associated with a combination of both commercial and resi-dential functions, with for example shops being located on the ground floor of high-rise residential structures. A brief description of the built-up environment associated with each settlement class is provided in figure 4.
3.2. CLASSIFYING SETTLEMENT SYSTEMS CHAPTER 3. METHODOLOGY
Figure 4. Boxes 1-4 provide brief descriptions of the settlement systems mapped in the initial land system map.
3.2. CLASSIFYING SETTLEMENT SYSTEMS CHAPTER 3. METHODOLOGY
3.2.1
Mapping settlement systems
Settlement systems were mapped with a GIS approach in Esri ArcGIS Pro 2.4.0. Built-up area and population in a square kilometer cell were used as proxies to infer the presence of different settlement systems. Prior to the classification procedure all maps were projected in an Asia South Albers Equal Area Conic projection and clipped by a polygon covering the study area compiled by administrative data included in the Eurostat GISCO (2019) countries of 2016 dataset. Built-up area and population densities were retrieved from the Global Human Settlement Layer (GHSL) (Corbane, Florczyk, Pesaresi, Politis, & Syrris, 2018). In this data package built-up area derived from Sentinel imagery is combined with municipal census data to create a population density grid. Classification results were continuously compared with Google Earth imagery and population and built-up density thresholds were adjusted until the resulting settlement system map corresponded with the characterization of the built-up environment as described in figure 4. Datasets applied in the classification procedure are listed in table A1, thresholds applied in the classification of settlement systems are shown in table A2.
Natural and agricultural land systems were inferred from a detailed land cover map of the Mekong region provided by SERVIR. Land cover was reclassified into categories presented in table A3 and aggregated by majority rule in a square kilometer grid by a zonal statistics procedure. Since this study focuses on settlement systems transitions, all forested land cover was reclassified to a single class. Classes from the SERVIS land cover map that remain static or for which land system dynamics were assumed to have a negligible effect on future flood damage were reclassified as ’other’. This included the land cover ’aquaculture’, which was mostly found in coastal Thailand and Vietnam. The settlement land system map was mosaicked over this aggregated land cover map. Since built-up land cover in the GHSL slightly differs from the SERVIR land cover map, areas that were determined built-up by SERVIR and were blank in the GHSL were classified as barren.
3.2. CLASSIFYING SETTLEMENT SYSTEMS CHAPTER 3. METHODOLOGY
3.2.2
Recording the built environment
Random samples of settlement grid cells were visually inspected to determine whether the settlement classification procedure distinguished built environments as described in figure 4. Buildings were categorized in four classes based on their construction material, namely: informal/ slum structures, wooden structures, masonry structures and concrete structures. For each country ten random grid cells per settlement class were examined using imagery available in Google Earth. A description and example of this procedure is presented in appendix B. Since sample sizes were small and numbers based on rough estimates no statistical analysis was performed. Estimates of building composition based on the average of ten settlement class cells are shown in figure 5. Estimates were later used in the flood risk calculations described on page 26.
(a) Cambodia (b) Laos
(c)
Myanmar (d) Thailand
(e) Vietnam
Figure 5. Distribution of building material found in each settlement class.
3.3. MODELING LAND SYSTEM CHANGE CHAPTER 3. METHODOLOGY
3.3
Modeling land system change
Two urban development scenarios were created based on literature and visual inspec-tion of local historical expansion of built-up land over the past 20 years. Two possible development pathways were hypothesized: 1) settlement expansion and 2) settlement in-tensification. The expansion or urban sprawl scenario gives rise to an extensive suburban landscape characterized by single story residential structures with a relatively low building and population density. Settlement intensification represents intensification of residential land use, resulting in further densification of existing urban area through infill develop-ment and construction of multistory apartdevelop-ment buildings. The CLUMondo model was run for each country and parameterized to create two possible future settlement change scenarios for the year 2035.
CLUMondo land use model
CLUMondo is a spatially explicit model that allows for an integrated top-down and bottom-up modeling approach of land system behavior (Van Asselen & Verburg, 2013). Next to simulating transitions of land system to mutually exclusive classes, the model is designed to simulate changes in land use intensity. Three modules make up the model, namely a macro-economic demand module representing demand of land system products and services, a spatially explicit allocation module and a settings module governing land-allocation. The allocation module designates land system change fulfilling this demand in a raster format based on location suitability, neighborhood influence and conversion resistance as defined in the settings module. A schematic overview of the model as applied in this study is shown in figure 6.
Demand and land system services
Settlement system change is driven by demand for population. Population growth was interpolated from the mean five-yearly projections of population growth provided by the World Population Prospects 2019 for each country (United Nations, 2019b). Settlement systems respond to this increased demand for population either by expansion of intensifi-cation of residential land use. Intensifiintensifi-cation is simulated by transitions of settlement on a rural-urban gradient. The mean population residing in each grid cell of each settlement system was determined for each country by use of a zonal statistics method and gridded population data from the GHSL (Schiavina, Freire, & MacManus, 2019). Is is assumed that each pixel of a settlement class provides housing to the same amount of residents. Country specific tables of demand for population and population residing in settlement systems are provided in appendix C.
3.3. MODELING LAND SYSTEM CHANGE CHAPTER 3. METHODOLOGY
Figure 6. Conceptualization and parameterization of the CLUMondo model for application in the Greater Mekong Subregion (GMS). Figure adapted and modified from Van Asselen and Verburg (2013).
Data from the UN Household Size and Composition 2019 database indicated a sharp decline in household sizes throughout the study region (United Nations, 2019a). A con-servative estimate of a ten percent increase of settlement area demand per capita at the end of the simulation is assumed for all countries. Decrease in household size only affects land system services in settlement classes, the average household size of the population residing in grid cells of other land systems is assumed to remain constant. Thailand and Vietnam experience declining population sizes near the end of the simulation period, however population decline is smaller than the decrease in household size so backward development is not required.
3.3. MODELING LAND SYSTEM CHANGE CHAPTER 3. METHODOLOGY
Land system allocation
The CLUMondo model constructs different land systems through an iterative procedure, which compete to optimize the allocation likelihood while meeting yearly demand for land system services. The model allocates the land system with the highest total probability for each location in each timestep. Total probabilities for land systems in each cell are calculated with formula 1:
P toti,t,lu = P loci,t,lu+ P nbhi,t,lu+ ConvReslu+ Compt,lu (1)
where Pi,t,lu is the total probability of land system lu at location i at time t as a sum
of local suitability (Ploc), neighborhood influence (Pnbh), resistance to conversion (Con-vRes) and the competitive advantage of the land system (Comp) (Verburg & Overmars, 2009).
Local suitability and neighborhood influence
Local suitability for each settlement system was determined with a logistic regression model of socio-economic and biophysical explanatory variables. Regression models were made only for settlement system suitability, since natural and agricultural systems re-main static during the simulation. A sample dataset of all settlement system cells in the study area was used in the regression analysis. Suitability maps were checked for multi-collinearity and correlations higher than 0.7 were dropped from the analysis. An overview of suitability maps assessed in the regression analysis is presented in table C3. Suitability maps included in the regression model are presented in table 1.
Although the presence of settlement systems was significantly correlated with temper-ature, precipitation, sand and clay content, inclusion of these explanatory variables in the CLUMondo model resulted in the emergence of patches of settlement systems away from existing urban cores and thus were dropped from the analysis. Significant and logical regressor maps included in the regression models were a map indicating the travel time to cities by Weiss et al. (2018), an euclidean distance map to highways, primary and secondary roads based on the global road dataset by Meijer, Huijbregts, Schotten, and Schipper (2018), elevation data from the Shuttle Radar Topography missions (Rabus, Eineder, Roth, & Bamler, 2003) and hillslope derived from this elevation data. Local suitability Ploc was calculated for each settlement system with formula 2:
Table 1. Explanatory variables included in the regression models
Settlement system Explanatory variable AUC Village Traveltime to city(-) + Distance to roads (-) + Slope (-) + Elevation (-) 0.89 Suburban sparse Traveltime to city(-) + Distance to roads (-) + Slope (-) + Elevation (-) 0.93 Suburban dense Traveltime to city(-) + Distance to roads (-) + Slope (-) + Elevation (-) 0.95 Urban dense Traveltime to city(-) + Distance to roads (-) 0.97
3.3. MODELING LAND SYSTEM CHANGE CHAPTER 3. METHODOLOGY
log( Pi 1 − Pi
) = β0+ β1 ∗ X1,i+ β2∗ X2,i+ ...βn∗ Xn,i (2)
where Piis the probability of finding that settlement system at location i, β0the intercept,
Xnthe value of explanatory factor n in cell i and βnthe regression coefficient indicating the
relative effect of explanatory factor n on the presence of the settlement system type. The goodness of model fit in the logistic regression was determined with the ROC method and expressed as area under the receiver operating characteristic curve (AUC). AUC values between 0.8 and 1 indicated good to excellent model fits (see table 1).
An estimated neighborhood influence simulating the effect of land systems on land system transitions in their vicinity is added to the empirically determined local suitabil-ity. It is assumed that new settlement systems are more likely to emerge nearby existing ones due to the presence of infrastructure. Suburban sparse-, dense and urban dense settlement systems received neighborhood additions based on the presence of settlement systems in their neighboring cells.
Conversion and area restrictions
Settlement systems transition unidirectional through the various settlement types as de-scribed in figure 4, with each incremental step on this rural-urban gradient representing intensification of urban land use. Since settlement systems require time to develop, sub-urban sparse may transition to subsub-urban dense in 2 years and subsub-urban dense to sub-urban dense in 2 years. Agricultural land may transition into villages and suburban sparse set-tlement systems in a single year. A random number generator in CLUMondo was used to compile an initial age map with ages ranging from 0 to 5 years. A raster indicating areas where land system conversion is not allowed was retrieved from the UN Environmental Program protected planet database (UNEP-WCMC, 2020). Restricted areas include cul-tural and historical sites, nature reserves and wildlife sanctuaries. Land systems present in the beginning of the simulation period contribute to fulfilling demand.
Conversion resistance and model parameterization
The conversion resistance parameter represents the relative ease of land system conversion and was adjusted to develop two different settlement change scenarios. In an urban expansion scenario development of agricultural land into a residential area may, as a result of policy decisions, increased mobility or lower land prices, be more profitable to developers then further development of existing urban area. In this modeling exercise these processes are simulated by reducing the conversion resistance of agricultural and natural land-use classes, favoring settlement expansion. Settlement densification may on the other hand be enforced through spatial planning simulated by increasing the conversion resistance of agricultural and natural land while reducing the resistance within settlement classes.
3.4. FLOOD RISK ANALYSIS CHAPTER 3. METHODOLOGY
In order to create two settlement change scenarios, the model was run two times for each country with different conversion resistance values. Conversion resistance values were parameterized based on visual comparison of resulting land system maps with satellite imagery of the twenty-year period between 1995 and 2015 obtained from Google Earth. This meant no formal validation data was used in the model parameterization. Since the stylistically opposite settlement change scenarios were created for exploration and com-parison only, this methodological weakness was accepted. Differences in model behavior required that conversion resistances were parameterized for each country separately. In this procedure conversion resistance values and settlement change patterns were kept as similar over countries. First the settlement expansion was created , after which the resis-tance value of agriculture and forested land systems was raised by 0.03 for all countries to create a settlement intensification scenario. The model was run again, after which the resistance values of settlement systems were lowered to create the settlement intensifica-tion scenarios. The aim of this procedure was to realize a magnitude of settlement change roughly corresponding between nations in both scenarios. Conversion resistances applied in the models are listed in table C6. Model outcomes were compared to historical urban growth by assessing the increase in built-up area. The average built-up area in a grid cell of a settlement system in a country was multiplied by the amount of grid cells present in the land-use map. This was done with land-use maps at the beginning and the end of the simulation period, after which the average annual growth was calculated and compared with observations from Mertes, Schneider, Sulla-Menashe, Tatem, and Tan (2015).
3.4
Flood risk analysis
Based on construction material settlement system cells were assigned maximum poten-tial damage values which in combination with inundation maps and a depth-damage curve were used to estimate flood damages. Buildings were categorized in four damage classes based on their construction material, namely: informal/ slum structures, wooden structures, masonry structures and concrete structures. The building composition charac-teristic for a settlement system in each country was estimated based on satellite, Google street view and ground level imagery available in Google Earth (see figure 5). Residential structures were assumed to cover 20 percent of the built-up environment as reported by Huizinga et al. (2017). The average built-up density of each settlement system of a coun-try was extracted from the GHS Built-up dataset and is presented in table B2 (Corbane et al., 2018).
3.4. FLOOD RISK ANALYSIS CHAPTER 3. METHODOLOGY
Flood risk calculations
In the next step maximum potential damage to all buildings in a settlement type cell was calculated based on construction cost with methods adapted from Englhardt et al. (2019) and Huizinga et al. (2017). However, where Englhardt et al. (2019) developed separate depth-damage curves for each building class, here a single curve is applied with different exposure values. The calculations consisted of applying the following formulas:
C = a ∗ GDP ∗ b (3)
where C is construction cost per square meter in Euro’s, GDP is the gross domestic product per capita, and a and b parameters provided by Huizinga et al. (2017). The adjusted reconstruction cost for building class k is then calculated by:
Sk = C ∗ Cdepricated∗ M axAdjk∗ (1 − U ndamk) (4)
where Sk is building cost per square meter for a building class k ; C is the construction
cost per square meter; Cdepreciated is a conversion factor to calculate the depreciated value
of the construction material; MaxAdjk is the building type specific adjustment factor to
account for the use of less expensive building material in building class k ; Undamk is the
fraction undamageable of building type k. Next the maximum damage to building content and inventory per square meter is calculated with the following formula:
Invk = Sk∗ M axInvk (5)
Invk is the maximum potential damage to building inventory of class k as a function of
building cost and conversion factor MaxInvkfor building type k. In the next step maximum
potential damage for each cell of settlement class j is calculated by the following formula adapted and modified to incorporate building footprint without building floor size from Englhardt et al. (2019) Dj = k X k=1 (Sk+ Invk) ∗ Nj,k∗ Aj ∗ CellSize (6)
Djis herein the maximum potential damage for a grid cell of settlement system type j as a
sum of the potential damage to all building of types present; Nj,kis the fraction of built-up
area consisting of building type k in a cell of settlement class j ; A is the percentage of a cell covered by building footprint in settlement type j ; CellSize is the cell size in square meters.
After calculating maximum potential damage for cells of each settlement class, a single depth damage curve for residential buildings provided by Huizinga et al. (2017) (see figure 7) was applied with inundation maps with 10, 20, 50, 100 and 500 year return periods
3.4. FLOOD RISK ANALYSIS CHAPTER 3. METHODOLOGY
provided by Dottori et al. (2016). The modeled flood maps represent probabilistic inun-dation maps for the entirety of the region and thus do not represent a single events. Flood damage to settlement systems was summed per country over the initial land-use map of 2015 and the resulting maps in the settlement expansion and intensification scenarios. .
Table 2.
Building classes with corresponding adjustment factors from
Huizinga et al. (2017).
Building class Max damage
adjustment Undamaged part Max inventory
Informal/ suburb 0.125 0 0.2
Wood 0.33 0.1 0.5
Masonry 0.8 0.4 0.5
Concrete 1 0.4 0.5
Table 3. GDP per capita in 2015 and construction cost
Country GDP per capita in 2015[USD 2010] Construction cost [EUR 2010] Depreciated value factor Cambodia 1024 348 0.5 Laos 1539 407 0.5 Myanmar 1335 385 0.5 Thailand 5741 675 0.5 Vietnam 1667 419 0.5
Figure 7. Flood damage curve for residential buildings provided by Huizinga et al. (2017). A damage factor of 1 is reached at an inundation depth of 6 meters.
Chapter 4
Results
In this chapter land system modeling outcomes and flood calculations for each country are presented. The results of this study are intended for comparison between the scenarios and to provide inside into relative change in flood risk considering different urbanization trajectories. Land system maps created in this study are presented in appendix D.
4.1
Land system change projections
The CLUMondo model was run for a 20 year period to create two possible settlement change scenarios. A comparison of increase in total settlement system area and built-up land based on the average built-up area of each land system is shown in table 4. Settlement area increased in all countries of the study area, and in all countries the expansion of built-up land was greater than the increase in population. Built-built-up area increased the least in the land-use intensification scenarios, indicating vertical building development. Net settlement change in each country is presented in figure 8. In this section settlement projections are presented for each country in the study area.
Table 4. Net increase in population, settlement area and built-up land in
2035 compared to 2015.
Settlement area Built-up area
Country Population Expansion Intensification Expansion Intensification
Cambodia 27% 258% 74% 129% 82% Laos 27% 121% 43% 95% 81% Myanmar 15% 76% 22% 40% 26% Thailand 2% 7% 1% 8% 5% Vietnam 15% 34% 4% 31% 15% 29
4.1. LAND SYSTEM CHANGE PROJECTIONS CHAPTER 4. RESULTS
(a) Cambodia (b) Laos
(c) Myanmar (d) Thailand
(e) Vietnam
Figure 8. The amount of settlement pixels present in the beginning and end of the simulation period.
4.1. LAND SYSTEM CHANGE PROJECTIONS CHAPTER 4. RESULTS
Cambodia
Settlement area increased in both urban development scenarios, shown in figure 8a. Total area of settlement systems increased with 258 percent in a settlement expansion and 74 percent in an intensification scenario. Most settlement change occurred in the vicinity of Battambang, Siem Raep and Phnom Penh. Snapshots of these areas are shown here in figure 9. Built-up land increased by 129 percent and 82 percent in an settlement expansion and intensification scenario, respectively. This is a result of the low population density in sparse suburban systems, which require a larger extent of settlement area to fulfill population demand while building density remains low. In the intensification scenario built-up land increased more than total settlement area, representing further densification of already urbanized land though infill development.
Figure 9. Snapshot of Battambang, Siem Raep (1) and Phnom Penh (2) of 2015 and 2035 in the settlement expansion (a) and intensification scenario (b).
4.1. LAND SYSTEM CHANGE PROJECTIONS CHAPTER 4. RESULTS
Laos
The total settlement area increased by 121 percent in a settlement expansion and 43 percent in a settlement intensification scenario. Settlement change mostly occurred near Luang Prabang, Vientiane and Pakse. Dense urban settlement systems emerged only in Luang Prabang. Built-up land increased by 95 percent in the expansion and 81 percent in the intensification scenario. The increase of built-up land was a factor two of areal expansion in the intensification scenario, indicating further densification of already ur-banized land. A land system map of of Laos in 2015 and in both urban development scenarios is shown in figure 10. Snapshots of Vientiane and Pakse show the emergence of a suburban landscape in the expansion scenario, while dense suburban landscape emerged in the settlement intensification scenario.
4.1. LAND SYSTEM CHANGE PROJECTIONS CHAPTER 4. RESULTS
Myanmar
Settlement area in Myanmar increased by 76 percent in the settlement expansion and 22 percent in a settlement intensification scenario. Built-up land increased by 40 and 26 percent, respectively. This result indicated less densification of already urbanized land compared to other countries in the study area (table 4). As a result of low road densities, settlement expansion in Myanmar resulted in settlement ’corridors’ expanding from urban cores along roads. Suburban dense settlement systems increased in both of the development scenarios. In the settlement intensification scenario densely clustered patches of the urban dense settlement type emerged in Mandalay and Yangon, shown here in figure 11.
Figure 11. Snapshot of Mandalay (1) and Yangon (2) of 2015 and 2035 in the settlement expansion (a) and intensification scenario (b).
4.1. LAND SYSTEM CHANGE PROJECTIONS CHAPTER 4. RESULTS
Thailand
Settlement area increased by 7 percent in a settlement expansion and 1 percent in a settlement intensification scenario. The increase in built-up land was 8 and 5 percent, respectively. The increase in built-up land was greater than the increase in settlement area in both the expansion and intensification scenario. Settlement area increased less than population in the intensification scenario, indicating a general increase of population density. Compared to other countries changes in settlement area were small (see figure 17). This is attributed to the relative small net population growth. Nearing the end of the simulation period population growth reversed, however decreasing household size resulted in a yearly increase of settlement area.
4.1. LAND SYSTEM CHANGE PROJECTIONS CHAPTER 4. RESULTS
Vietnam
In Vietnam the total area of settlement systems increased 34 percent in an expansion and 4 percent in a settlement intensification scenario. Built-up area increased by 31 and 15 percent, respectively. The increase of built-up area was greater than the increase of settlement area, indicating further urban development on already urbanized land. The in-crease in settlement area was smaller than the inin-crease in population in the intensification scenario, indicating increasing population densities as a result of settlement transitions. Most settlement expansion occurred in the Mekong and Red River deltas, shown in figure 13. Settlement systems expanded outward of existing cores along access roads. Denser land systems emerged mostly near existing urban cores, although some land transitions occurred along the coast.
Figure 13. Snapshot of the Red River Delta near Hanoi (1) and the Mekong River delta near Ho Chi Minh City (2) of 2015 and 2035 in the settlement expansion (a) and intensification scenario (b).
4.2. FLOOD RISK CHAPTER 4. RESULTS
4.2
Flood risk
In this section the results of the flood risk calculations are presented. Flood risk was quantified as damage estimates given inundation levels with 10, 20, 50, 100 and 200 year return periods. Changes in flood risk for inundation events with a 10 year return period are shown in table 5. An overview of flood risk curves for return period of 10- to 500 years is presented in figure 19. Flood damage maps of major cities are shown in figures 14 to 18.
Cambodia
Flood risk to flooding with 10 year return periods increased with 146 percent in a settle-ment expansion and 154 percent in a settlesettle-ment intensification scenario, indicating that settlement expansion is the development pathway resulting in the lowest flood damages for inundation events with all exceedance probabilities (shown in figure 19a). Phnom Penh was most affected by changes in flood risk as result of its location on the Mekong and Tonle Sap River. Inundation levels higher than 6 meters resulted in maximum damage to dense urban land systems. A flood damage map of Phnom Penh considering inundation levels with a ten year return period is shown is figure 14.
Laos
Risk to flooding with 10 year return periods increased with 119 percent in a settlement expansion and 143 percent in a settlement intensification scenario. In Laos settlement ex-pansion resulted in lower flood damages compared to settlement intensification in flooding with all exceedance propabilities (see figure 19b). Vientiane and Pakse, both located on the Mekong River, were affected most by increased flood risk as a result of urban devel-opment. A flood damage map of Vientiane is shown in figure 15.
Table 5. Damages to settlement systems in flooding with 10 year return
periods.
Damages [billion Euro] Increase (%)
Country Initial Expansion Intensification Expansion Intensification
Cambodia 1.1 2.8 2.9 146 154
Laos 1.2 2.7 3.0 119 143
Myanmar 3.3 4.9 4.8 50 48
Thailand 43.3 48.6 46.1 12 7
4.2. FLOOD RISK CHAPTER 4. RESULTS
Figure 14. Flood damage map of Phnom Penh in both scenarios with inundation levels of 10 year return periods. Inundation map provided by Dottori et al. (2016).
Figure 15. Flood damage map of Vientiane in both scenarios with inundation levels of 10 year return periods. Inundation map provided by Dottori et al. (2016).
4.2. FLOOD RISK CHAPTER 4. RESULTS
Myanmar
In Myanmar flood risk increased by 50 percent in a settlement expansion scenario and 48 percent in a settlement intensification scenario in flood events with 10 year return peri-ods. Differences in flood damage between the two urban development scenarios slightly diverged when with decreasing exceeding probabilities, shown in figure 19c. Since differ-ences in flood damage between both urban scenarios are small, determining an optimal development trajectory may prove difficult. However, estimates indicate urban expansion the safest development pathway in flooding considering all exceedance probabilities. A damage map of Mandalay is shown in figure 16.
Thailand
Damage in flood events with 10 year return periods increased by 12 percent in a settlement expansion and 7 percent in a settlement scenario. As shown in figure 19d, settlement expansion remained a safer urban development pathway in flooding with lower exceedance probabilities. Most flood damages occurred in the Chao Phraya River Delta, affecting urban land near Bangkok. A flood damage map of this area in both urban development scenarios is shown in figure 17.
Vietnam
Flood damage in flooding events with a 10 year return period increased by 46 percent in a settlement expansion and 16 percent in a settlement intensification scenario. Expansion of urban land thus resulted in higher flood damages compared to urban densification, and remained a safer pathway considering flooding with lower exceedance probabilities (see figure 19e). This can be attributed to the highly urbanized Mekong and Red River deltas, where urban expansion onto land not subjected to frequent flooding is not possible. A flood damage map of Can Tho in the Mekong River delta is presented in figure 18.
4.2. FLOOD RISK CHAPTER 4. RESULTS
Figure 16. Flood damage map of Mandalay in both scenarios with inundation levels of 10 year return periods. Inundation map provided by Dottori et al. (2016).
Figure 17. Flood damage map of Bangkok in both scenarios with inundation levels of 10 year return periods. Inundation map provided by Dottori et al. (2016).
4.2. FLOOD RISK CHAPTER 4. RESULTS
Figure 18. Flood damage map of Mekong River delta near Can Tho iin both scenarios with inundation levels of 10 year return periods. Inundation map provided by Dottori et al. (2016).
4.2. FLOOD RISK CHAPTER 4. RESULTS
(a) Cambodia (b) Laos
(c) Myanmar (d) Thailand
(e) Vietnam
Figure 19. Flood risk curves plotted with modeled flood damage in settlement area for return periods of 10, 20, 50, 100 and 500 years.
4.3. SENSITIVITY ANALYSIS CHAPTER 4. RESULTS
4.3
Sensitivity analysis
A one-at-a-time sensitivity analysis was performed to assess the sensitivity of calculated flood risk to uncertainty in maximum damage and settlement system building composi-tion. Uncertainties in land system classification and projections, inundation maps and the depth-damage curve were not tested in this sensitivity analysis. However, since the same method was applied to all countries in the study area it assumed that flood damage estimates still provide a basis to compare model outcomes between scenarios.
Huizinga et al. (2017) report a 90 percent confidence interval of maximum damage values between 28 percent lower and 53 percent higher than calculated maximum damage values. To assess the effect of differences in built environment, flood calculations were also performed with a homogenized building stock for all settlement systems. As a result, in this run only differences building density influence changes in flood damage estimates between settlements. Results of the sensitivity analysis are reported in table 6.
Results of the sensitivity analysis shows flood risk estimates scale linearly with changes in both maximum damage estimates. This relationship is trivial considering the formulas applied in the flood damage calculations, though highlights the need for accurate estima-tions of reconstruction cost when estimating flood risk. By homogenizing the composition of building types for each settlement systems flood damage estimates were lower for all countries. Except for Thailand relative differences in estimated flood damage between the two scenarios remain, indicating that differences in flood damage within settlement sys-tems are mostly explained by building density rather than applied construction material. In Thailand the urban development trajectory resulting in the lowest flood risk switches from intensification to urban expansion.
4.3. SENSITIVITY ANALYSIS CHAPTER 4. RESULTS
Table 6. Sensitivity of annual flood damage estimates (in billion 2010 EUR)
to uncertainty in maximum damage and settlement building composition
Normal run Maximum damage 90% confidence
No differences in settlement building composition
Country Scenario Damage 10yr flood Lower Upper Damage 10yr flood
Cambodia Initial 1.14 0.81 1.74 0.88 Expansion 2.81 2.02 4.30 2.22 Intensification 2.90 2.09 4.44 2.27 Laos Initial 1.22 0.88 1.87 1.13 Expansion 2.67 1.92 4.09 2.08 Intensification 2.96 2.13 4.53 2.44 Myanmar Initial 3.25 2.34 4.97 2.71 Expansion 4.86 3.50 7.44 3.84 Intensification 4.81 3.46 7.36 3.72 Thailand Initial 43.26 31.15 66.19 34.75 Expansion 48.58 34.98 74.33 39.33 Intensification 46.10 33.19 70.53 39.72 Vietnam Initial 45.61 32.84 69.78 32.18 Expansion 66.79 48.09 102.19 48.77 Intensification 52.87 38.00 80.75 42.90
Chapter 5
Discussion
5.1
Methodology
5.1.1
Land system mapping
Settlement systems were mapped based on population and built-up land in a square kilometer grid cell. The same thresholds were determined for all countries in the study area in an expert based assessment of classification results. A visual inspection of random settlement pixels was hampered by a lack of available ground level imagery in Vietnam, Myanmar and rural Laos. Although this assessment resulted in crude estimations of building stock composition, it was determined that the settlement system map succeeded in distinguishing settlement systems on a rural-urban gradient, hereby capturing different stages of urban development. Inclusion of local data on building material has the potential to greatly improve representation of the built environment in settlement system types.
Natural and agricultural land systems were inferred from a detailed land cover map. All forested land cover was aggregated into a single land system class, merging both es-sentially agricultural plantation forest with natural land systems. Aggregation of land cover furthermore resulted in a lack of land systems predominantly used for rice produc-tion in Laos. This could be result of rotaproduc-tional crop regimes or classificaproduc-tion errors in the SERVIR land cover map, since rice makes up a large portion of Lao’s agricultural pro-duction (International Rice Research Institute, 2012). This generalization was accepted, as this study focused on settlement change and agricultural land systems did not respond to demand for population .
The classification procedure aimed to further distinguish industrial systems from set-tlement systems. Since industrial land systems did not change in both urban development scenarios, the land system was dropped from further analysis. For better representation of land use in flood risk assessments uncertainties induced by this land system need to be addressed (Merz, Kreibich, Thieken, & Schmidtke, 2004).
5.1. METHODOLOGY CHAPTER 5. DISCUSSION
5.1.2
Land-system modeling
Differences in model behavior between countries required different parameter sets to achieve comparable land-system change results. Although the same conversion resistances were applied in Vietnam and Thailand, Cambodia, Laos and Myanmar each required a different parameterization of conversion resistances to result in similar urban development scenarios. This may be due to differences in accessibility, reducing the overall suitabil-ity for settlement systems in countries where road denssuitabil-ity is low. Furthermore, as only settlement systems respond to increasing demand for population, countries that have a greater fraction of their population residing in rural land systems can expect a greater increase in settlement area when compared to countries that are already more urbanized. The relative amount of settlement change in a single time step required to fulfill demand thus varied between countries, affecting the maximum likelihood of land systems fulfilling demand.
A decrease in household size was included in the model as a yearly reduction of people residing in a settlement system grid cell. In this study a decrease of ten percent in pop-ulation density compared to the beginning of the simpop-ulation is assumed for all countries. However, temporal household data was available only for Cambodia, Thailand and Viet-nam and indicated a greater decrease over a period of twenty years than the ten percent assumed in this study (United Nations, 2019a). As population growth slows and reverses, decreasing household size becomes a dominant driver affecting the need for urban develop-ment. Due to the large implications of decreasing household size on human-environmental interactions, addressing this process when projecting urban development may improve the land system model (Bradbury, Peterson, & Liu, 2014).
5.1.3
Flood damage calculations
Flood damages were estimated based on exposure, inundation depth and a single depth-damage curve. Other flood characteristics, such as stream velocity and the duration of inundation greatly affect flood damage estimates (Middelmann-Fernandes, 2010). As this study aims to explore differences in flood risk considering different urban development pathways rather than providing realistic damage estimates this simplification of flood hazard was accepted. Since the approach to modeling future flood risk in this study enables differentiating vulnerability of different building classes, damage estimates will be improved by use of building-type specific depth-damage curves (Englhardt et al., 2019).
5.2. RESULTS CHAPTER 5. DISCUSSION
5.2
Results
5.2.1
Projecting urban development
Projection of urban development in a settlement expansion and settlement intensification scenario yielded two very different land system maps. As we do not know how urban land develops in the future, addressing the diversity of possible urban environments in large scale studies related to human-environmental interactions is paramount to account for uncertainties in urban development.
Since literature on urban change with a similar settlement classification system in the Great Mekong Region is sparse, comparisons to studies modeling past settlement change provides a challenge. However, studies on the expansion of built-up land are readily available. In table 7 results from Mertes et al. (2015) are cited for comparison between modeled and observed settlement system change in terms of built-up area. The average annual growth rate of built-up area Laos roughly corresponds with expansion of built-up area observed by Mertes et al. (2015). Expansion of built-up land in Thailand and Vietnam is lower than observed by Mertes et al. (2015). This can be attributed to a decreased rate of urban development as a result of less population growth and the higher level of urbanization when compared to the year 2000. Rural-urban migration is not included as such in the model, possibly resulting in an underestimation of settlement change in countries where population growth is slowing or reversing (in Thailand and Vietnam). The averaged annual growth of built-up area in Cambodia and Myanmar was higher in model projections than observed change. Since the rate of urban development is likely lower in this simulation period compared to the period between 2000-2010, this may indicate an overestimation of urban development.
Table 7. Observed average annual growth rate (AGR) of built-up land
compared projected AGR.
Mertes et al. (2015) Expansion Intensification Country 2000 [km2] 2010 [km2] AGR [%] 2015 [km2] 2035 [km2] AGR [%] 2015 [km2] 2035 [km2] AGR [%] Cambodia 218.47 291.05 2.91 321.61 738.80 4.25 321.61 584.88 3.04 Laos 162.17 222.72 3.22 162.33 317.62 3.41 162.33 293.81 3.01 Myanmar 1822.69 2005.08 0.96 1349.71 1883.37 1.68 1349.71 1705.68 1.18 Thailand 4617.17 5366.56 1.52 6609.04 7124.73 0.38 6609.04 6931.60 0.24 Vietnam 4201.72 5099.06 1.95 5710.55 7466.88 1.35 5710.55 6561.35 0.70
5.2. SUSTAINABLE DEVELOPMENT CHAPTER 5. DISCUSSION
5.2.2
Flood risk
Flood damage estimations indicated that river flood risk increased in all countries of the study region. As flood protection infrastructure was not addressed in this study, flood damage estimations likely overestimate actual flood damage (Ward et al., 2013). Estimated flood damage to settlement systems in Cambodia and Laos was lowest in the settlement expansion scenario in all flood return periods. In Myanmar, Thailand and Viet-nam damages to settlement systems was lowest in an settlement intensification scenario. This result indicated that a single optimal urban development pathway for reducing fu-ture flood risk does not exists. A sensitivity analysis showed that flood damage estimates based on a homogenized representation of urban land resulted in lower flood damage esti-mates. In Thailand the optimal development pathway reducing future flood risk changed from urban land expansion to urban intensification. Although further research is required to better represent urban land, results indicate underestimations of future flood risk when urban area is represented in a single class. Results of the sensitivity analysis of damage estimates in Thailand furthermore demonstrate value of the approach to future flood risk presented in this study, as it affects the optimal urban development trajectory.
5.3
Toward sustainable urban development
Projecting urban change in the Greater Mekong Subregion indicated that further urban development is associated with large increases in river flood risk. Most notably in Cam-bodia and Laos, where estimated flood damages more than doubled in 20 years’ time. Flood risk calculations furthermore indicated that the most favorable urban development pathway for mitigating future flood risk differed among countries. This optimal pathway resulted in lower flood risk estimates under all exceedance probabilities, indicating that choices in urban development will affect urban flood damages in severe flood events with long return periods.
In Cambodia and Laos urban expansion resulted in lower estimated flood damages than urban densification. This result indicates that compact city development in these regions enhances river flood risk through increased exposed asset value. While compact city development addresses excessive land take and inefficient energy consumption, policy makers should be aware that increased flood protection standards are required to mitigate trade-offs on river flood risk. In Myanmar, Thailand and Vietnam urban densification resulted in lower flood risk in all return periods. From this result it can be inferred that compact city development in these regions results in a lower total value of assets exposed to flooding compared to urban sprawl, even in severe flood events. This provides an additional argument in favor of urban strategies aimed to limit settlement expansion in these regions.
