Spatial-Temporal Relations of Flood Risk and Its Potential Dynamic Pressures in Myanmar

Abstract

Despite multiple challenges, floods remain the most frequently occurring hazard in Myanmar. Current developments of political instability, multidimensional insecurity, and associated economic crisis have burdened the existing vulnerabilities and inequalities of the Burmese people and their ecosystems. Diminishing adaptive capacities of degraded ecosystems, poor infrastructure, and extreme poverty, together with major livelihood dependency on climate-sensitive agriculture, will further increase flood risk. Moreover, other hazards such as COVID-19, heatwaves, and droughts may exacerbate flood impacts leading to compound disasters. Understanding how and which factors drive flood risk, and where they distribute are important to reduce flood risk, address its root causes, and prevent future flood damages by lessening exposures, vulnerabilities, and even hazards. We aim to compare the spatial-temporal distributions between dynamic pressures and flood risk, and identify the spatial relations on a national scale and within floodplains. We draw on socio-ecological risk assessment, systematic review, time-series analysis and modified t-test after testing spatial auto-correlations of dynamic pressures and flood risk. Our results show that many socio-ecological dynamic pressures driven by economic- and governance-related root causes had positive spatial relationships with flood risks. We recommend effective land use and environmental governance that consider compound and cascaded flood risk and investment in public services and infrastructure such as health and education to reduce vulnerabilities and increase resilience of Myanmar people.

1. Introduction

The disproportionate rise in climate-related impacts, and losses and damage in countries of the Global South may attribute to differences in vulnerability to those climate and disaster impacts. Yet, the issue is treating disasters separately from their underlying root-causes in their disaster- and climate-related management and adaptation practices, especially in those vulnerable countries. For example, climate change adaptation through nature-based solutions is not possible without addressing root causes of deforestation and forest degradation in those nations. There is a clear call for transformative change in dealing with vulnerabilities in those countries.

Disaster risk and impacts are increasingly becoming the products of more compound and complex causal mechanisms in today’s interconnected world. Explanation of disaster risk needs to trace connections that join disaster impacts with a sequence of causal factors and processes. Cascading impacts are chains of disruptive disaster impacts on systems at different scales [1, 2]. This systemic nature of risk can be seen in the examples of the COVID-19 pandemic [3, 4, 5, 6]. In other words, a disaster at the same time can be a driving factor to cause a larger and more calamitous compound disaster. This type of compound disaster can be especially worse in the incidence of social, economic, financial, health, and political crises, which are outside of the climate and disaster risk framing.

Here, root causes are defined as “an interrelated set of widespread and general processes within a society and the world economy” [7]. They are economic, demographic, and political processes that affect the distribution and allocation of resources among different social groups. Those causal mechanisms are spatially and temporally distant from those disastrous situations (unsafe conditions of vulnerability) to be clearly understandable at a glance [7, 8, 9]. They are usually “invisible” and “taken for granted” as they are embedded in socio-cultural beliefs and assumptions. For example, inadequate natural resource governance that leads to inequalities and deforestation, which in turn increases flood risk, cannot easily be mapped or easily visible as a causal factor. Therefore, they are difficult to express spatially and temporally. As a result, those root causes cannot be easily incorporated into the disaster risk management.

Therefore, it is essential to understand the spatial and temporal relations of risk drivers (root causes) and the places where risk occurs. Dynamic pressures are “processes and activities that translate the effects of root causes both temporally and spatially into unsafe conditions” [7]. Therefore, these pressures can be visualized over space and trends. In this chapter, we will explain spatial-temporal relations of dynamic pressures exercised by root causes and disaster risk using case study of Myanmar.

Myanmar provides a comprehensive case study to understand compound cascading risk driven by the interaction of deepened root causes. Myanmar has been experiencing multiple exogenous compound stressors and endogenous dynamic pressures driven by several root causes currently and dynamically over history. Myanmar is one of the countries, which is mostly affected by and is highly vulnerable to extreme climate events, according to GAR Report, Inform Risk Index 2017, and Global Climate Risk Index 2021 [10, 11, 12]. Given consistent inadequate governance, the dependency of Myanmar’s economy on resource extraction and agriculture left Myanmar’s forests, land, water, and environment in a seriously degraded condition [13, 14, 15]. This inadequate governance also deepens socioeconomic and class inequalities [13, 14, 15]. The fact that the majority of populations depend on their livelihoods in those degraded ecosystems and climate sensitive sectors results in limited capacities to cope and adapt to extreme climate events due to the degraded ecosystems. Long history of conflicts and the current development of political instabilities render already poor Myanmar population to face multidimensional insecurities and inequalities including food, job, and safety, to name a few [15, 16]. These recent developments disrupted humanitarian assistance, international development aid, food production, environmental management, and climate actions, resulting in conditions which can be extremely catastrophic in the face of extreme disasters [16]. Considering the increasing impacts of climate change in the near future, those extreme disasters are likely to occur. Public services and infrastructure, such as education and health services, have received little priority and attention for investment in their development, which is now seriously disrupted due to the recent changes [15]. Last COVID-19 incidences and associated vast mortalities due to shortages in oxygen supplies and health facilities prove weakening health infrastructure. The associated economic and financial crisis put additional millions of Burmese people into poverty. Social security systems and climate insurance development are yet in the very early stage of development to support recovery from disasters and climate impacts and for the risk transfer [17]. Official Development Assistance from outside Myanmar can be difficult due to the ethical challenges, safety, and entry issues [17]. Therefore, in case of any disasters, already vulnerable Myanmar people can be helpless, seeing the examples of 2008 Nargis Cyclone.

Despite those multiple challenges in Myanmar, flood remains the most frequent and has the highest contribution to average annual loss compared to all other hazards in the country [18]. Winseminus et al. (2016) estimated that 1.24 billion out of 1.81 billion population exposed to 1 in 100-year floods live-in South East Asia [19]. In 2015, a riverine flood resulted in over US$1 billion in damages, affected over 1.6 million people, and caused at least 149 deaths [20]. With climate change, Myanmar is likely to continue to experience river flood events at the scale of the 2015 floods or worse [21]. In Myanmar, some river flooding occurs almost every year during the monsoon season and communities now rely on these floods for nutrient deposition in their agricultural fields [22]. Communities have learnt to cope with and have adapted to these annual floods, harnessing their value [22]. However, as demonstrated by the impact of the 2015 flood, which was characterized to be of a 20–50-year return period [23], Myanmar has little capacity to cope and adapt with major flood events alone given the overlapping effects of different pressures.

Transformational risk reduction and adaptation require addressing root causes of flood risk to reduce its vulnerabilities and increase resilience. If only symptoms are addressed and root causes are not considered in the risk management, those root causes will be intensified in the future, enhancing compound pressures in a climate-changing world. The Sendai Framework for Disaster Risk Reduction calls for disaster risk management that considers all dimensions of disaster risk, including hazard, exposure, and vulnerability. To reduce flood risk, decision makers and stakeholders must have a holistic understanding of the underlying factors.

Studies have focused only on drivers that drive individual components of risk and any of the assessments have barely explored risk drivers that contribute all components of risk [24]. Unlike the Pressure and Release (PAR) model, this chapter will consider that root causes and dynamic pressures will not only lead to the vulnerabilities of the people, but they can also lead to all components of risks which are hazard, exposure, and vulnerability of socio-ecological systems.

Several studies attempted to understand drivers of flood risk. Flood risk was assumed to be caused by various factors: including physical factors such as river levees and river incision [25], climate-related factors such as Climate variations associated with Oscillations such as Pacific Decadal Oscillation (PDO) and El Niño Southern Oscillation (ENSO) [26], sea level rise [24, 27] and storm surges, typhoons and cyclones [24], social factors such as urbanization [24, 25, 28, 29], soil sealing [27], poverty, and limited education [28], environmental factors such as land use changes and deforestation [24], degraded wetlands and decreased drainage capacities [30], land subsidence [24, 27], governance-related factors such as failures of top-down approaches in flood risk management, lack of risk communications, lack of local information, limited citizen involvement in flood risk management [31], and subjective factors such as risk perceptions in shaping flood risk in terms of developing insurance-based flood management and risk transfer [32]. Abbas and Routray (2014) identified exposure to vectors, inaccessibility, damaged, and non-operated health care facilities as major factors of flood-induced health risks [33]. Assessments have likely prioritized physical and environmental risk drivers over social, economic, and governance-related drivers [24].

However, spatial-temporal relations of disaster risk and driving factors were rarely explored. A few studies investigated spatial-temporal relations of flood-related losses and damages and their influencing factors [34, 35]. However, those studies have a focus on the hazard and impacts part with limited vulnerability considerations in assessing the spatial-temporal distribution of disaster risk. We will aim to fill this gap by exploring spatial-temporal relations of dynamic pressures and flood risk that is considered as a function of hazards, exposure, and vulnerability, driven by dynamic pressures and root causes. There are three main objectives of the chapter. First, this chapter will investigate root causes and driving pressures of flood risk in Myanmar, considering potential compound and cascading risk, while the focus is on the flood risk. The spatial and temporal evolutions of dynamic pressures in relation to flood plain areas will be assessed. Second, this analysis will show the spatial distribution of flood risk based on the IPCC framing of risk [36]. Third, spatial relations of dynamic pressures and flood risk will be examined. Understanding the spatial and temporal relations of dynamic pressures (that translate from root causes to unsafe conditions of flood vulnerability) and flood risk can be used for transformative change to effective flood risk reduction.

2. Methodology

2.1 Case study

Myanmar is located in South East Asia (Figure 1). Its land size is approximately 678,500 km2 and it has an estimated population of 42.5 million people [37]. The country has a complex multi-level governance system with 330 townships at the lowest level of government [38]. Much of the population in Myanmar lives in poverty (MMPF, 2017) and the country was ranked 145 out of 189 on the Human Development Index in 2019 [39]. There are four major rivers in Myanmar with populations reliant on each [22]. The Ayeyarwady is the largest and the most important used in commerce and daily life [22]. Most of the country has a tropical monsoon climate, with river floods commonly occurring between May and October [22].

Figure 1.

Myanmar with labeled states. Townships are the next administrative level down.

2.2 Conceptual framework

2.2.1 Understanding root causes and dynamic pressures

This chapter used the long-standing Pressure and Release Model (PAR) to understand how root causes were translated into unsafe conditions that are current vulnerabilities through dynamic pressures. As explained in the introduction, root causes are defined as “an interrelated set of widespread and general processes within a society and the world economy” [7]. Dynamic pressures are “processes and activities that translate the effects of root causes both temporally and spatially into unsafe conditions” [7]. Unsafe conditions are the specific conditions in which the vulnerability of the people and ecosystems is expressed in time and space when being exposed to a hazard. Unlike the original concept of the PAR model, this analysis will consider that root causes can lead to all components of risk: hazard, exposure, and vulnerability of socio-ecological systems.

2.2.2 Conceptual framework of flood risk

To define risk, this chapter built on the common risk framework from the IPCC 5th assessment report, where risk is the potential for adverse consequences and is a function of hazard, exposure, and vulnerability [36], and the PAR model where root causes and dynamic pressures are considered (Figure 2). In this analysis, we focused on present-day flood risk to people, agricultural, and ecosystems. Exposure was defined as the presence of people and ecosystems within the extent of the hazard [40]—a 100-year river flood and coastal flood event. Vulnerability was defined as the propensity or predisposition of the Myanmar people to be adversely affected and encompassed the components of susceptibility and coping capacity [40]. Root causes and dynamic pressures have impacts on all elements of risk. As mentioned in the above section, it is the difference from the PAR model, where root causes have only drive vulnerability. In this framework, those underlying causes and dynamic pressures drive all elements of risk (Figure 2).

Figure 2.

Conceptual framework of socio-ecological system of flood risk.

2.3 Workflow

Figure 3 outlines the methodological workflow of the quantitative index-based approach for the township level used in this analysis. All data were freely available, allowing for replication and validation of this study.

Figure 3.

Methodological workflow.

2.3.1 Identification of root causes and dynamic pressures of flood risk in Myanmar

To identify root causes and dynamic pressures, a systematic literature review was conducted to understand the main drivers and causes of flood risk in Myanmar. Two searches were conducted using Web of Science (WoS) and SCOPUS in November 2022. Search strings were constructed. The authors separately screened all titles and abstracts of the unique papers to determine relevance. Criteria included focus on Myanmar; risk and vulnerability to flood; and the exposure of people and ecosystems. Publications that met these criteria were selected. Papers that did not were excluded, including those that focused solely on flood hazard. Where one or both authors were uncertain, the paper was read entirely by the authors to determine selection. To find additional gray literature a Google search was conducted. After screening the first 10 pages of results two reports were added to the literature review. At the end of the process, 34 papers were included for final review. A summary of searches is provided in Table 1.

Searches	WoS	SCOPUS	Included
Search 1: (TITLE-ABS-KEY (flood) AND TITLE-ABS-KEY (risk OR vulner* OR resil* OR root AND cause* OR drive* OR pressur*) AND TITLE-ABS-KEY (eval* OR assess* OR profile OR index OR indic* OR multicriteria) AND TITLE-ABS-KEY (myanmar OR burma))	60	23	11
Search 2: (TITLE (myanmar OR burm*) AND TITLE-ABS-KEY (flood))	84	106	20
Google search “driving factors and root causes of flood risk in Myanmar” OR “Myanmar flood disaster risks and impacts”	—	—	3

Table 1.

Search terms used to capture papers to inform vulnerability indicator identification.

Search strings use SCOPUS operators. For WoS, TOPIC replaced TITLE-ABS-KEY. TITLE remained the same.

Root causes and dynamic pressures were extracted from each relevant publication. After identifying relevant dynamic pressures in Myanmar, associated datasets were collected and performed following analysis to understand their spatial-temporal evolution.

2.3.2 Time series analysis

Time series analysis was performed to understand temporal aspect of dynamic pressures. It started with acquisition of dynamic datasets for available years. Most of datasets associated with land use changes such as deforestation, agricultural expansion, urbanization, loss of wetland areas, and mining are acquired from Regional Land Cover Monitoring System developed by SERVIR-Mekong [41] from 1987 to 2018. Conflict data were attained from the Armed Conflict Location & Event Data (ACLED) for the available period from 2019 to 2022 [42]. Land Surface Temperature was computed from the Terra Moderate Resolution Imaging Spectroradiometer (MODIS) dataset using Google Earth Engine (GEE).

After data acquisition of relevant datasets, each dataset was preprocessed for time series analysis. Datasets for forest and agriculture were reclassified according to given land cover classes in the regional land cover monitoring systems. The classes that defined “forests” included mangrove, forest, flooded forest, ever green broadleaf, mixed forest, and orchard and plantation forest. The classes that defined “agriculture” included rice, cropland, and aquaculture. After extracting targeting land cover classes, respective values per townships were calculated from each raster datasets using “zonal statistics as table” in ArcGIS for the time period from 1987 to 2018 to see trends of values per townships. For flood plain calculations, raster data of land-cover related dynamic pressures were clipped with flood plain boundaries [43]. Conflict data were preprocessed by counting the number of conflict events for each township for the period from 2019 to 2022. Types of conflict events included in the analysis were battles, explosions/remote violence, protests, riots, strategic developments, and violence against civilians according to the ACLED definition [42]. For land surface temperature data, average annual land surface temperature was calculated from daily land surface temperature and emissivity values from 2000 to 2022 (until 23rd November 2022). Flood data related to losses and damages in terms of mortalities and affected populations were acquired from EM-DAT dataset from 1965 to 2021 [20]. Visualization of time-series trends was made in Excel.

2.3.3 Flood risk assessment

2.3.3.1 Flood hazard/exposure

Extent of river flood and coastal flood for a 100-year return period was obtained from the Joint Research Centre (JRC) of European Commission [44] and the World Bank Data Catalog [45], respectively. ArcGIS was used to crop the global dataset to the national boundaries of Myanmar and limit the extent of the flooding to modeled depth above 20 centimeters.

To calculate hazard exposure, we determined the percentage of exposed population per township ($e\_{\mathit{soc}}_i$) based on the modeled population distribution from WorldPop (WorldPop, 2020) ($p\_{\mathit{tot}}_i)$, and the population within the flood extent ($p\_{\mathit{exp}}_i$) (Eq. 1. i refers to each township).

$$e\_{\mathit{soc}}_i=\frac{p\_{\mathit{exp}}_i}{p\_{\mathit{tot}}_i}\times 100$$E1

The percentage of exposed population was normalized using linear min-max (Eq. 2) so that the range was reduced to between one (high exposure) and zero (no exposure) to create the exposure index (${\mathit{EI}}_i$). This is a very common normalization method for indicator-based assessments [45].

$${\mathit{EI}}_i=\frac{\left(E_i-E_{\mathit{min}}\right)}{\left(E_{\mathit{max}}-E_{\mathit{min}}\right)}\times 100$$E2

In Eq. 2, $E_i$ refers to the percentage of exposed population before transformation, $E_{\mathit{min}}$ refers to the minimum value of exposure, and $E_{\mathit{max}}$refers to the maximum value of exposure.

2.3.3.2 Flood vulnerability

The same systematic review explained above was used to understand driving factors of flood vulnerability and to identify relevant indicators in Myanmar. These were grouped into susceptibility and coping and adaptive capacity indicators.

Next, data at the township level were collected from multiple sources including the 2014 census and other surveys. Based on data availability, 26 indicators were used in the final assessment (Table 2). Appendix A provides an overview of all desired indicators from the literature review, and the data sources of those used.

Indicator	Code	Data Source	Direction
Susceptibility
Dependency ratio (people <15 and > 65 years / people between 16 and 64 years)	s_dep	[46]	+
Population below the poverty line (average annual income)	s_pov	[47]	—
Disability prevalence (population, %)	s_dis	[46]	+
Population with vector-borne diseases (malaria incidence per 10,000 people)	s_vec	[47]	+
Population with chronic illness (share of tuberculosis, dysentery, and hepatitis, %)	s_chr	[47]	+
Female-headed households (%)	s_fhh	[46]	+
Population affected by conflict in townships in the last year (number of conflict events per 10,000 people in the last year)	s_con	[42]	+
Households with access to safe sanitation (%)	s_san	[46]	—
Households with access to improved drinking water (%)	s_dri	[46]	—
Population living in poorly constructed housing (households with walls or floors constructed from leaf, bamboo, or earth, s_wal & s_flo average %)	s_wfl	[46]	+
Average travel time to the closest city (minutes)	s_ttc	[48]	+
Daily wage worker (%)	s_daw	[47]	+
Extent of Areas of Agricultural expansion in Flood plains (1987–2018) (%)	s_age	[41]	+
Extent of Areas of Urban extension in Flood plains (1987–2018) (%)	s_ure	[41]	+
Extent of Areas of Mining extension in Flood plains (1987–2018) (%)	s_mine	[47]	+
Extent of Areas of Forest Losses (1987–2018) (%)	c_ftl	[41]	+
Extent of Areas of Wetland Losses (1987–2018) (%)	c_wetl	[41]	+
Adaptive capacity
Households with access to Information (telephone, mobile, internet, or radio, s tel., s mob, s int., & s_rad max %)	c_ati	[46]	—
Literacy rate, (%)	c_lit	[46]	—
Density of roads (road kernel density: km road/km2, search radius = 5 km)	c_dens	[49]	—
Households owning a boat (%)	c_boa	[46]	—
Number of doctors per 10,000 people	c_doc	[47]	—
Number of hospital beds per 10,000 people	c_bed	[47]	—
Extent of Forest Areas (%)	c_ft	[41]	—
Extent of Wetland Areas (%)	c_wet	[41]	—
Households with access to alternative electricity source (households using solar energy or a generator as the main source of lighting, c_sol & c_gen total %)	c_aes	[46]	—

Table 2.

Final list of indicators including code, data source, and direction.

The indicators relating to access to telephone, internet, mobile, and radio were aggregated to one information indicator (c_ati), data for poorly constructed floors, and walls were aggregated to create one housing conditions indicator (s_wfl), and data for households with generators and solar energy were aggregated to an access to alternative electricity indicator (c_aes). Proxy variables were used for s_pov, s_vec, s_chr, s_con, c_ati, and s_wfl (for more information, see Appendix A).

As a third step, outliers were identified and treated following Damioli (2017) [50] using Microsoft Excel. Box plots based on the interquartile range and skewness and kurtosis were used to identify extreme values. Due to the paucity of data for triangulation to determine whether extreme values were errors, local expert knowledge of the authors was used. It was determined that only five indicators had outlying values that were errors (s_pov, s_vec, s_chr, s_fhh, and c_doc). These were treated using winzorization (see Appendix B).

Fourthly, the scale of missing data was assessed. Using findings from Downey and King (1998) on acceptable thresholds for missing data, no indicators were excluded as none had missing data above 20%. Ten townships¹ had missing data above 20% and were therefore marked as highly uncertain in the final risk and vulnerability assessments. Missing data were imputed using the IDW tool in ArcGIS and taking the mean from the output.

As a fifth step, multicollinearity analysis was conducted using Kendall’s Tau and two-tailed approach for statistical significance in SPSS (IBM SPSS Statistics). This is a commonly used technique for non-normal data [51] with r > 0.9 indicating highly correlated datasets [52]. No issue of collinearity was detected (Appendix B).

Sixthly, the final set of indicators were normalized to a range between zero and one using the linear min-max approach. For indicators where high scores increase vulnerability (direction is positive), Eq. 3 was applied. For indicators where high scores lower vulnerability (direction is negative), values were inverted using Eq. 4. In Eqs. 3 and 4, $X_i$ refers to the indicator value for a township (i) before transformation, $X_{\mathit{min}}$ refers to the minimum value of the indicator, $X_{\mathit{max}}$ refers to the maximum value of the indicator, and $X_i^{\prime }$ refers to the indicator value after transformation.

$$X_i^{\prime }=\frac{\left(X_i-X_{\mathit{min}}\right)}{\left(X_{\mathit{max}}-X_{\mathit{min}}\right)}$$E3

$$X_i^{\prime }=\frac{\left(X_i-X_{\mathit{max}}\right)}{\left(X_{\mathit{min}}-X_{\mathit{max}}\right)}$$E4

Finally, indicators were given equal weights due to the absence of knowledge on their relative importance and aggregated using additive arithmetic aggregation into the vulnerability index ${\mathit{VI}}_i$ (Eq. 5).

$${\mathit{VI}}_i=\frac{\sum X_i^{\prime })}{N}$$E5

In Eq. 5, $X_i^{\prime }$ refers to the normalized indicator values for the township and $N$ refers to the number of indicators.

2.3.3.3 Flood risk

The vulnerability and exposure indexes were aggregated using two methods. Arithmetic mean aggregation of hazard/exposure (${\mathit{EI}}_i$) and vulnerability (${\mathit{VI}}_i$) was conducted to determine relative risk for each township in a risk index (${\mathit{RI}}_i$) (Eq. 6). Natural break method was used to visualize spatial distribution of exposure, vulnerability, and risk indices.

$${\mathit{RI}}_i=\frac{{\mathit{EI}}_i+{\mathit{VI}}_i}{2}$$E6

2.3.4 Spatial analysis

Spatial differences of dynamic pressures were calculated in the raster calculator in ArcGIS to understand changes between the start and end of the available period. Resulting differences in terms of decrease, increase, and no changes were again calculated for each township using zonal statistics as Table tool in ArcGIS. For flood plain calculations, raster data of dynamic pressures were clipped with flood plain boundaries. Then, the similar processes of calculation of spatial differences were carried out for those dynamic pressures in the flood plains.

To understand the spatial relationships of dynamic pressures and flood risk, a modified-t test was carried out in R studio using SpatialPack library. Similarly, for the dynamic pressures on the flood plains, the same approach was conducted to understand their spatial relationships with flood risk index.

3. Results

3.1 Root causes and dynamic pressures

In response to the first research question, root causes and dynamic pressures of flood risk in Myanmar are explained as a result of the literature review. Social flood vulnerability in Myanmar is expressed in the form of multidimensional inequalities, and widespread poverty, while ecological flood vulnerability is expressed in the form of enormous environmental degradation. Multidimensional inequalities can be in terms of gender, ethnicities, class, and spatial divides such as urban, rural areas, bordered areas [53]. Those socio-economic and ecological vulnerabilities are connected with each other and also shape flood exposure and hazard.

Inadequate natural resource governance is a major root cause that deepens flood risk in Myanmar. Corruption at different levels allows most resource-based activities to happen in the black economy, resulting in widespread deforestation [54]. Particularly in sectors such as forestry and agribusiness, corruption can result in increased deforestation from land conversions [54]. Economic dependency on resource extraction can increase corruption [55]. Myanmar was ranked 136 out of 176 countries in terms of corruption index [56]. Widespread deforestation causes amplified erosion, limited water retention capacity and high sedimentation, worsening flood hazards. Studies proved that deforestation increases flood frequency [57, 58] and flood risk in nonlinear way [59]. As forests serve as a safety nets for forest-dependent communities, deforestation has limited their coping and adaptive capacities after floods. Biodiversity loss can have a cascading effect on the ecosystems and its regulation services for disasters.

An example can be seen in the Ayeyarwaddy delta, the rice bowl of the country, where wetlands and mangrove forests were massively transformed into rice fields. Mangrove forests were again degraded during insurgency periods [60]. Later, additional conversions into commercial fish and shrimp farming result in massive deforestation of mangrove forests in the delta [61]. As a result, risk reduction services and provision services of mangrove ecosystems were declined seriously affecting livelihood base and adaptive capacities of society and ecosystems to adapt to flooding and other coastal hazards such as cyclones, salt water intrusion, and sea level rise.

Poor resource governance contributes to environmental degradation and increased flood risk. Development projects such as natural gas extraction and dam construction are often carried out without proper environmental regulations or local participation [55]. Corruption results in such projects with violations of safeguards. These projects threaten the environmental security on which local communities depend [55]. One of the examples of inadequate natural resource governance which increases flood risk can be seen in the mining sector. The townships where many of mines existed, Hpa Khant, experience deterioration of its environment where mountains become valleys, and vice versa [62] increasing landslide risk and collapse of mining sites. Illegal and large-scale mining practices, such as exceeding permitted tailings height and dumping tailings near river banks, also contribute to high risk of collapse, resulting in loss of life and releasing large amounts of soil into rivers [62]. This in turn raises river bed levels, increasing flood hazards for nearby and downstream populations and villages, potentially washing them away. Mining also causes water pollution, producing toxic water that can be highly dangerous after flooding. Additionally, mining projects often result in land grabbing at the expense of local communities, weakening their ability to respond to flooding.

Inadequate governance of natural resources can lead to inequalities that exacerbate flood vulnerability [63]. This is because natural resources are often exploited in ways that benefit only a small group, while the pollution, deforestation, and other negative consequences affect the entire population [55]. It was worsened as foreign investment places immensely toward resource extractive sector [63]. For example, in the energy sector, even though energy resources are exported, the areas where the gas pipelines are located, Tanintharyi region and Rakhine State, have the lowest per capita electricity consumption in Myanmar [64]. Limited access to electricity relates to reliance of fuel wood on forests including mangroves and underdevelopment in those coastal regions. Deforestation, underdevelopment, and dependency on fuel wood contribute to flood vulnerability. Other injustices can also be witnessed in local protests against dam construction, which increase flood vulnerability. Moreover, dams in Myanmar are not designed for flood control [65].

Inadequate resource governance can also be witnessed in widespread land confiscations, which increases flood vulnerability through poverty and inequality. In 1999 and 2000, transition to market-based economy led to land appropriations, for the sake of agribusiness such as oil palm plantations and industrial production [55]. Therefore, millions of land acres were transferred to private companies. This made millions of people to depend only on labor wage alone [66]. By 2030, there was a national mandate to convert 10 million acres [67]. While violent land grabbing especially for agribusinesses makes many people landless, there is no systematic policy for those conflict resolutions of their concerns [15]. Those landless and displaced labors were less likely to be reabsorbed in the manufacturing industries [15] creating enormous inequalities. Finally, poor, landless, and jobless communities have limited capacities to cope and severely vulnerable to large-scale flooding after damaging their livelihood base.

Those flood vulnerabilities were exacerbated by continuous conflicts currently and dynamically. Myanmar has experienced a long history of civil war [16]. Conflicts in ethnic areas make developments of those areas seriously lacking behind. Conflicts with ethnic armed groups, and shared resource exploitation with private actors in those conflicted and ethnic people areas, caused marginalization of ethnic populations [68]. Current conflicts and instabilities in Myanmar have wider spill-over effects. Current instabilities disrupted humanitarian assistance, climate action, food production, employment, public services, and development activities. As many international projects and companies left Myanmar, many people faces difficulties and uncertainties to secure food, job, income, and safety. The problems are exacerbated by economic downturn of the political instability and the COVID-19 pandemic [16]. The consequences are the souring food and commodity prices, which make it difficult to be affordable by poor population. Despite some progress in the environmental governance in the recent years in terms of local recognition and transparency, now current political development has put all of those efforts to a halt.

A consequence of conflicts and land appropriations, displacement, is worth to mention for its contribution to flood risk. Besides increasing flood vulnerability, displacement due to the land grabbing, and conflicts, can lead to high exposures to flood-prone areas and flood plains and other sensitive areas such as urban slums. Displacement was used as a counter-insurgency measures, leaving many people vulnerable. Dramatically high figures of rural-to-rural migration [69] highlight that the displaced people are not absorbed by urban jobs [69]. Moreover, urban jobs are so inadequate that the income differences between urban and rural jobs were not apparent. In addition, out migration to Thailand has continued quickly [70], despite unsafe conditions there [71]. Increased migration and displacements disrupted traditional social networks of care and support, eroding community bonds [72]. Therefore, displacement can decrease social adaptive capacities to floods. Finally, displacement cycle begins when those people have to displace again from floods.

Despite those widespread inequalities and poverty in Myanmar, there is limited expenditure and funding on public services and infrastructure such as education and health facilities. Total government expenditure on education is still very low compared to the ASEAN average [73]. Moreover, there are huge disparities in infrastructure and provision of public services between rural and urban areas, especially lower in bordered areas [53]. Those regional differences become more apparent due to the conflicts faced in the bordered ethnic people’s regions [53]. Inadequate health facilities can lead to major fatalities in case of large-scale flooding and other disasters. Levels of education are directly related to risk awareness of flooding and preparatory actions. Deficient public services cannot facilitate well to vulnerable people to escape out of the underemployment cycle. Particularly in health and education sector, the poor people have to subsist with under-qualified and under-funded public options. Moreover, lack of road networks and communication infrastructure leave people in remote areas of Myanmar isolated until humanitarian aid could reach them.

In the disaster risk management, and climate change adaptation actions, there are certain factors that drive to insufficiency and ineffectiveness in flood risk reduction in Myanmar. Despite widespread deforestation, ecosystem-based climate change adaptation is still yet to be developed in Myanmar. Similarly, social protection is at its infancy, in terms of coverage and level of institutionalization, with remarkable exclusion errors [17]. The coverage on problems related to health, child, and elderly support is extremely inadequate. The share of government expenditure allocated to social protection is limited only at 0.8% of National GDP, making Myanmar the lowest in its spending for social protection in the region, compared to regional average of 2% [17]. Moreover, the current social protection system has focused on the wrong demographics, with only 0.1 percent of existing social protection spending emphasized on the most vulnerable and the poorest [17], making major population to depend on informal forms of safety nets. Community-based actions are crucial for the success of adaptation and risk reduction efforts; however, the current political climate makes it challenging for international actors to support these efforts [16]. Consequently, progress in climate action has been hindered. Moreover, development aid providers are struggling with issues of increasing domestic and regional inequalities [53]. Given these challenges, it is likely that international assistance for response and recovery efforts would be difficult in the event of a large-scale flooding.

Importantly, root causes were interconnected reinforcing each other. For example, Rustad et al. (2008) found a link between Myanmar’s forest resources and armed conflicts [74]. Woods (2019) also supported the relationship between Myanmar’s natural resources and history of armed conflicts in ethnic areas [75]. Additionally, countries affected by conflicts are at risk of corruption in their natural resource sector due to weakened and resource-deprived bureaucratic systems [76]. Poverty in Myanmar is related to low levels of education [73] as a result of inadequate funding for education infrastructure. Households head with primary or lower education are likely to be poor [73].

Based on the described root causes and their consequences on socio-ecological systems and associated flood risk, the following dynamic pressures were identified. They are deforestation, agricultural expansion, urban extension, mining, conflicts, and wetland loss. Other disasters such as heatwaves are also considered here as dynamic pressures in this study.

3.2 Temporal distribution of dynamic pressures

The figure shows changes in dynamic pressures of flood risk in comparison with flood impacts in Myanmar. Figure 4(a) and (b) shows consistently symmetric trends between deforestation and expansion of agricultural areas. While forests were constantly declined between the period between 1988s and 2000s (from 68 percent to 57 percent), agriculture was consistently flourished during that period. Ochards and planations (Figure 4(c)) mirrored the same trend of agriculture. In that period, forest areas of around 767,000 km² were declined while agriculture had increased by 38,000 km². When comparing to deforestation and agriculture increase with flood impacts during that period (Figure 4(g) and (h)), it shows that frequent relatively small-scale floods that affected up to 359,000 populations and brought up to 68 deaths during the same period. Later after years of 2000, rates of deforestation were slow down to further loss from around 404,000 km² to 387,00 km². The same rate applied for increase in agricultural and orchard plantation areas. At the same time after 2000s, flood events were more intense and rarer with less frequent smaller-scale floods. Therefore, the gaps in the intensities and frequencies of flood impacts become wider after 2000s.

Figure 4.

Temporal changes in flood risk and its dynamic pressures: (a) changes in forests areas (b), changes in agricultural areas, (c) changes in orchards and plantation areas, (d) changes in land surface temperature, (e) changes in urban, wetlands, and mining areas, (f) trends of conflict events, (g) trends of affected population due to floods, and (h) trends of mortalities due to floods in Myanmar.

When detecting at the figures in urban, wetland and mining areas (Figure 4(e)), there were substantial declines in wetland areas across Myanmar. Urbanization has increased steadily from 587 km² to 958 km² similar to sluggish increase in mining areas. Urban areas have also larger rate of increase before 2000s, similar to agricultural trends. Compared to agriculture, urbanization was less a dynamic pressure due to its minor spatial extent. Small fluctuation in mining areas might indicate that old mines were demolished and new mines were produced. Overall, looking at reduction trends in forests and wetland areas, the flood risk reduction capacities were diminished over time, while flood-sensitive and flood-induced and flood-sensitive agriculture, orchard plantations, and mining areas have proliferated over the period.

Average annual land surface temperature (Figure 4(d)) shows a fluctuated trend over the period from 2000 to 2022. Land surface temperature had its peaks in the years of 2005, 2010, 2013, 2015, 2019, and 2020. In 2010, there were extreme meteorological droughts occurred in Myanmar, which saw record-breaking temperatures of three times higher than normal summer months. When compared to floods, nationwide large-scale flooding occurred with 1.6 million affected populations and 149 deaths in 2015, which is also a peak year in average annual land surface temperature. However, in 2011 when there was a decline in land surface temperature, it coincided with a large-scale flood which caused 151 deaths of people.

When it comes to added pressures by conflicts (Figure 4(f)), there was a significant increase in the number of conflict events due to recent political changes in Myanmar. Despite existences of dynamic conflicts in Myanmar, data coverage was only limited to the last three years. Fortunately, only small-scale flood events were reported so far while experiencing the high nationwide conflicts. However, there was a deadly flood occurred in Mon State in 2019, while 33 conflicts were recorded. To understand the spatial overlap of those pressures, spatial dimension is required to considered. Therefore, the spatial distributions of dynamic pressures were discussed in Section 4.5.

Overall, the results show that significant changes in dynamic pressures over the period from 1987 to 2018. Forest cover reduced from 68 percent in 1987 to 57 percent in 2018. In 1987, coverage of agricultural area was 21 percent, which increases to 27 percent in 2018. For orchards and plantations, total plantation areas increase from 6 percent to 8 percent in 2018. Urban areas increase from 0.09 percent of the total country areas in 1987 to 0.14 percent in 2018. Mining areas increase from 0.08 to 0.09 percent, while wetland areas reduced from 0.32 percent to 0.26 percent in 2018 (Figure 5).

Figure 5.

Maps showing flood plain areas and ten highest (flood) risk townships in Myanmar.

3.3 Temporal evolution of dynamic pressures in highest risk townships

Figure 6 shows the temporal evolution of dynamic pressures in the 10 highest risk townships (See those township locations in Figure 5). It is important to note that the values are not relative values to township areas but rather absolute values in km². In trends of forest loss (Figure 6(a)), Paletwa, Myitkyina, and Minbya townships have substantially high forest areas compared to other high-risk townships, followed by Kyauktaw. Townships such as Myitkyina and Paletwa have highest forest loss of around 500 km² among them during the period from 1987 to 2018, while Myinbya and Kyauktaw have less loss of around 100 km². Most of those losses happened between 1999s and 2000s indicating a significant decrease during those periods.

Figure 6.

Temporal changes in dynamic pressures in high risk townships: (a) changes in forests areas, (b) changes in agricultural areas, (c) changes in orchards and plantation areas, (d) changes in urban settlement areas, (e) changes in wetlands areas, (f) changes in mining areas, (g) changes in land surface temperature, and (h) trends of conflicts in high-risk townships.

In trends of agricultural area increase (Figure 6(b)), extent of agricultural areas ranges from 51 km² to 1600 km². Townships with large agricultural extent are Bogale, Mawlamyinekyun, Wakema, Mrauk-U, Pauktaw, Minbya, Rathedaung, Kyauktaw, Myitkyina, and Palawe in ascending order. While most of the trends in agricultural extent are stable, Myitkyina has increase sharply from around 295 km2 to 579 km2 over the period, while it is assessed as the highest risk township in Myanmar. It sharply increased in years between 2000s and 2001s, and slowly increased in the later years until it saw again a sharp increase between 2017 and 2018. Bogale also experienced its increase in agricultural areas steadily after 2001.

In trends of orchard and plantation area increase (Figure 6(c)), there is a widespread range from around 1.1 km² to 900 km² in the year 2018. Townships with large plantation extent are Paletwa, Kyauktaw, Myinbya, Mrauk-U, Pauktaw, and Myitkyina in ascending order. Most townships with high plantation areas such as Myitkyina, Kyauktaw, Paletwa, and Minbya experienced a steep increase in 1994s till 2001s where the trend became stable one. Among them, Paletwa and Myitkyin experienced a steep rise during the whole period. Pauktaw and Mrauk-U experienced a level-off trend over the whole period from 1987 to 2018 with little raise between 1995s and 2000s. Wakema and Mawlamyinekyun have almost no plantation areas.

In trends of urban settlement areas (Figure 6(d)), all townships have similar urban extents and have little differences in area extents. Interestingly like other drivers, they saw similar changes during some points of times over the periods. Those sudden and slight increase happened in three steps of the times: in 1990, 2003, and 2017.

In trends of wetland areas (Figure 6(e)), only Myitkyina has relatively high wetland areas while other townships have only 0 to 3 km². Surprisingly, Myitkyina shows slight increase in wetland areas from approximately 25 km² to 34 km² over the period from 1987 to 2018. Similarly, only Myitkyina has mining areas (Figure 6(f)) among other high-risk townships. Mining areas of 0.4 km² have started to increase in 2000 till it has demolished in 2003, and it increased again in 2017.

Looking at the temperature trends in high-risk townships (Figure 6(g)), land surface temperature values range from 297 to 303 kelvin. Over the period from 2000 to 2022, temperature values saw fluctuated trends. It was found that temperatures have slightly downward trends during those critical years of large-scale and high-impacted floods in 2008, 2011, and 2015 except for Bogale township. That means temperature was slightly higher before those years except for the case in 2015, which has stable and lower temperature before.

When it comes to changes in conflicts over the year from 2019 to 2022 in high-risk townships (Figure 6(h)), it was uncovered that most of the townships (except Myitkyina) have high frequency of conflict events (54 events to 164 events) in 2020 and the numbers dropped in 2021 when the usurpation of national power took place. This connotes that those townships likely experienced pre-existing conflicts over their history. However, it required data from previous years to draw explicit conclusions. A reverse trend was occurred in Myitkyina, where there is stable trend before 2021, where it reaches its peak at 143 events.

3.4 Temporal evolution of dynamic pressures in flood plains

Figure 7 shows changes in dynamic pressures in the flood plains (See area distribution of flood plains in Figure [5]). It was apparent that many of the dynamic pressures experienced a remarkable drastic change (increase or decrease) in the period before 2001s followed by steady changes after 2001s except for mining. Among those drastic changes, it saw a steep slope especially in period between 1994s and 2001s indicating that some substantial root causes systematically existed to drive those dynamic pressures. Moreover, those temporal patterns also followed major trends of dynamic pressures in the whole Myanmar.

Figure 7.

Temporal changes of dynamic pressures in flood plains: (a) changes in forests areas, (b) changes in agricultural areas, (c) changes in orchards and plantation areas, (d) changes in urban settlement areas (e) changes in wetlands areas (f) changes in mining areas in flood plains.

The results underscore significant dynamic changes within the flood plain areas. Forest area coverage (Figure 7(a)) was 3 percent in the flood plains in 1987, which reduced to 2 percent in 2018. For agriculture (Figure 7(b)), the total percent of agricultural flood plains compared to the country areas increased from 7 percent to 8 percent in 2018. Plantations in the flood plains (Figure 7(c)) also rised double from 0.8 percent of the total country area to 1.16 percent. Urban settlements (Figure 7(d)) in the flood plain areas increased from 0.05 percent of the total country areas to 0.07 percent. However, 56 percent of total urban settlements within flood plains reduced to 53 percent of total urban settlements in 2018. Wetland areas (Figure 7(e)) also saw similar raise with similar figures. Mining in the flood plains (Figure 7(f)) increased from 0.03 percent of country areas to 0.04 percent. Surprisingly, 44 percent of total mining areas existed in the flood plains in both 1987 and 2018.

3.5 Spatial distribution of dynamic pressures in relation to flood risk

Figure 8 below shows spatial distribution of changes of dynamic pressures and flood risk. Figure 8(a) represents area amount of deforestation for the period from 1987 to 2018. Spatially, deforestation was mainly occurred in Shan States, with forest losses of more than 1000 square kilometers (Figure 8(j)). It can be seen that the majority of forest conversions were realized in the periphery areas of dry zone region and in the middle of Shan State. It was clear in the maps that forests were converted to agricultural areas except for the lower part of Bago region (Figure 8(a) and (c)). Highest forest loss occurred in Hsipaw, Kanbulu, Kyethi, Nansang, and Tanitharyi townships, where majority of those losses were converted to agricultural areas, except for the loss in Tanintharyi township which transformed to orchards and oil palm plantations. Forests were consistently declined in the bordered areas to the neighboring countries. Ayeyarwady delta experienced remarkable loss in mangrove forests compared to Tanintharyi Region which have higher proportion of orchards and plantations (Figure 8(e)). It is interesting to note that Myikyina Township is the township, which ranked as the second highest townships in terms of its forest loss and agricultural increase in flood plains (Figure 8(k) and (m)) while being the township with the highest flood risk of socio-ecological systems (Figure 8(aa)). Forest area loss within flood plains mainly occurred in the Kachin and upper part of Sagaing Region, which are mainly upland forests of Ayeyarwady River in Myanmar (Figure 8(k)). Spatial distribution of forest loss and flood risk may not match, except for mangrove forest loss in the Ayeyarwady delta (Figure 8(j) and (aa)).

Figure 8.

Spatial distribution of changes of dynamic pressures and flood risk: (a) changes in forest areas, (b) changes in forest areas within flood plains (FP), (c) changes in agricultural areas, (d) changes in agricultural areas within FP, (e) changes in orchards and plantation areas (f) changes in orchards and plantation areas within FP (g) changes in wetland areas, (h) changes in mining areas, (i) changes in urban areas, (j) amount of forest losses, (k) amount of forest losses within FP, (l) amount of agricultural areas increase, (m) amount of agricultural areas increase within FP, (n) amount of increase in plantation areas, (o) amount of plantation areas increase within FP, (p) amount of increase in urban areas, (q) amount of urban areas increase within FP, (r) amount of wetland losses, (s) amount of wetland losses within FP, (t) amount of mining area increase, (u) amount of mining areas increase within FP, (v) mean of average annual land surface temperature, (w) townships affected by conflict events, (x) hazard, (y) exposure, (z) vulnerability, and (aa) risk index.

Agricultural areas were mainly concentrated in the middle Myanmar: dry zone area, Ayeyarwaddy Delta, and Rakhine State which are highly prone to numerous climate risks such as floods, cyclones, and droughts (Figure 8(c) and (d)). However, increase in agriculture occurred likely in the form of a circle in the middle Myanmar during the period between 1987 and 2018 (Figure 8(c) and (i)). Over the period, the increase was the highest in Shan State and Sagaing Region with the largest in Hsipaw and Kanbulu of approximately 2000 km² each (Figure 8(i)). Many agricultural areas in the flood plains were mainly concentrated in the Ayeyarwady delta (Figure 8(d)). However, in the figure 8(m), increase in agriculture within flood plains mostly occurred in the Kachin, upper part of Sagaing region, and Shan State with amount up to 294 km² per township. Changes are insignificant in Ayeyarwady delta after 1987 compared to those regions. Areas with high agricultural extension were spatially coincident with the second highest risk classes (Figure 8(i) and (aa)).

Orchards and plantations areas were mainly concentrated in the Mon, Tanintharyi, Kayin, and Rakhine State (Figure 8(n)). Apart from those areas, orchards and plantation areas were limited in other states and regions. Mon State and Tanintharyi Region saw the highest increase in orchards and plantation during the period between 1987 and 2018, while the increase in the flood plain areas occurred mostly in Kachin State and Tanintharyi Region (at least 110 km² per townships) (Figure 8(o)). Some of spatial patterns of general increase of orchards (i.e., in Kachin and Rakhine state) were overlapped with flood exposure patterns (Figure 8(y)). When compared to the flood risk, increase in orchards and plantations in Mon State and Tanintharyi Region spatially matched with the second highest risk townships in those regions (Figure 8(n) and (aa)).

Urbanization rates were minor in Myanmar compared to other dynamic pressures such as deforestation and agricultural increase (Figure 8(i)). As shown in Figure 8(i), it is invisible at all at the National-scale map. In the zonal classification map that shows changes per townships (Figure 8(p)), it was apparent that urbanization mainly occurs near major cities such as Yangon and Mandalay, and in Kachin State to some extent. However, even though the extent of increase was small (at most 25 km² per township), the increase mainly happened within the floodplains (Figure 8(p) and (q)). Compared to flood risk, the distribution of urbanization was widespread throughout the country (Figure 8(p) and (aa)).

Unlike urbanization, mining and wetland areas were only limited to a few states and regions (Figure 8(r) and (t)). Decrease in wetland areas mainly occurred near Yangon and Ayeyarwady regions, while most of those losses located on flood plains (Figure 8(r) and (s)). Moreover, areas of wetland loss partly fall in some highest risk townships in Mon State. Similar to the wetland areas, major increase in mining areas realized in upper part of Myanmar (Figure 8(t)). Most of the increase (6.51 km² out of 7.26 km² in the highest class) occurred in the flood plains, indicating the potential unsafe conditions (Figure 8(u)). Some of the townships with highest mining increase fall in the second highest risk categories (e.g., Homalin, Tanai, Hkamti, Injangyang), in the highest exposure (e.g., Homalin) and in the township with the highest forest loss (e.g., Hsipaw).

Interestingly, average annual land surface temperature patterns match consistently with agricultural area patterns where the highest values exist in the central Myanmar (Figure 8(v) and (c)). This implies that almost all agricultural areas in Myanmar (1987 to 2018) face extreme land surface temperature (over 301 kelvin). When comparaed to flood risk with temperature values, some townships with the highest risk and those with second highest temperature matches each other (e.g., in Ayeyarwady delta and Rakhine State) (Figure 8(v) and (aa)). Townships with the second highest temperature values (over 301 kelvin) overlapped with some townships of highest flood vulnerability (e.g., in Ayeyarwady delta: Bogale, Maubin, Mawlamyinekyun) (Figure 8(z)) and highest flood exposure (e.g., in Rakhine State: Mrauk-U, Rathedaung, and Kyauktaw) (Figure 8(y)).

In terms of distribution of conflict events (Figure 8(w)), conflicts took place highly in Sagaing and Magway regions. It was followed by Kayah, Tanintharyi, Kachin, and Kayin states. Some townships with the highest conflicts such as Kale, Monywa, Yesagyo, Khin-U fall in the second highest risk classes (Figure 8(aa)), eventuating in compound risk. In addition, other townships that have the highest conflicts such as Demoso, Yinmarpin, Salingyi, and Myaing are also near to second highest flood risk with risk index of over 0.45. Similarly, those areas with high conflicts rates also have the second highest flood vulnerability indices, except for Kale.

The analysis detected that the 100-year return period river and coastal flood hazard predictably followed the contours of the rivers and coast lines in Myanmar (Figure 8(x)). The highly exposed populations and ecosystems appeared in the Ayeyarwady region, Rakhine, and Kachin states. Populations and ecosystems in 73 townships were not exposed (Figure 8(y)). Interestingly, two townships with the highest flood exposure (i.e., Myitkyina and Homalin) have high forest loss of 624 km² and 940 km², respectively. It was followed by Toungup, Pyapon, and Laputta with forest loss of at least 150 km² each where most of those forests are expected to be mangrove forests (Jones 2022).

Vulnerability was widespread through Myanmar (Figure 8(z)). Our analysis shows that some vulnerability indicators were more critical than others. For the six most critical indicators, most townships had an index score between 1.0 and 0.9. The total number of townships is 331 townships in Myanmar. Access to healthcare was poor with a maximum of one doctor and 51 hospital beds per 10,000 people in 222 and 312 townships, respectively. Extent of wetland areas for flood water storage and reduction capacities was limited below 2 km² for 307 townships. Accessibility was also low for 290 townships with the density of roads between 0 and 2.82 (road kernel density: km road/km2, search radius = 5 km). The percentage of households owning a boat, which are important for saving lives and transportation during flood events, was between zero and 4.86 percent in 242 townships. Poverty was also widespread. The average income for 312 townships was between 324,225 to 3,720,021 Kyat (US$ 251 to US$2884), and 14 townships had an average income below the 2015 poverty line of 475,595 Kyat (US$ 369) [23].

Among the remaining indicators, some contributed to vulnerability more than others. For access to alternative electricity sources such as solar panels and generators that are useful during the power shortages during floods, 199 townships had maximum of 20 percent of households with these facilities. Extent of forest areas has less than 9 km² for 127 townships. However, rate of loss in wetland areas, expansion rate in urban areas, and vector-borne diseases contributed less to vulnerability for most townships.

Spatially, the most vulnerable townships were concentrated in bordered areas of Myanmar (Figure 8(z)). The majority of townships in the highest vulnerable classes were located in the Rakhine state, followed by Shan, Chin, and Kachin states. Highly vulnerable areas overlapped with high forest loss areas and high agricultural increase areas to some extent. Logically, the least vulnerable townships were concentrated in Yangon region, where the former capital of Myanmar, Yangon, is located. This was followed by the Mandalay region which has the second largest city, Mandalay, and the Nay Pyi Taw which is the current capital city. Therefore, urban areas were less vulnerable to flood risk compared to rural bordered areas in Myanmar.

Figure 8(aa) shows the spatial distribution of risk. Risk was highly concentrated in townships in the Rakhine and Ayeyarwady regions, followed by Kachin State. The elements contributing to risk for townships in the highest class were multifaceted. Common critical elements included high levels of poverty, limited capacities of wetlands, and a low number of hospital beds and doctors per 10,000 people. Rural areas were mainly present in the highest risk class with the travel time to the nearest city ranging from 52 to 521 minutes.

3.6 Spatial relations of flood risk and its driving factors (dynamic pressures)

Table 3 below shows spatial relationships between dynamic pressures and flood risk. It shows some of the significant and non-significant spatial relations between dynamic pressures and flood risk. Agricultural areas’ increase in flood plains, wetland loss in flood plains, wetland loss, forest loss in flood plains, forest loss, increase in plantation areas, and increase in plantation areas in flood plains have statistically significant positive spatial relationships with flood risk. Therefore, many of the dynamic pressures, especially those in the flood plains, show positive spatial relationships, highlighting that their increase contributes to the flood risk. Among them, wetland loss, wetland loss in flood plains, and agricultural areas’ increase in flood plains have relatively stronger positive relations than other drivers. It was followed by drivers such as plantation areas’ increase in flood plains, forest loss, and forest loss within flood plains. Interestingly, forest loss over the whole of Myanmar has stronger positive relationships with flood risk than forest loss within flood plains, indicating the need to consider the impacts of scales and downstream effects on flood risk.

	ftl	ftlfp	agrg	agfp	wtl	wtlfp	ming	minfp	urb	urbfp	plt	pltfp	con	Lst5
Correlation Coefficient	0.1561*	0.1294*	0.1041	0.1907***	0.2061**	0.218***	0.0335	0.0627	−0.1137*	−0.0072	0.1437*	0.173**	0.0446	−0.1944
P values	0.0212	0.0405	0.1006	0.0012	0.0018	0.0008	0.5727	0.2798	0.041	0.8949	0.0296	0.0049	0.5048	0.001

Table 3.

Spatial relationships between dynamic pressures and flood risk.

*represent significant levels.

However, urban areas’ increase and land surface temperature have statistically significant negative spatial relationships with flood risk. Among them, land surface temperature has stronger negative relationships with flood risk. This means that low land surface temperature areas have strong flood risk to some degree. Urban areas’ increase within flood plains has almost neutral relationships with flood risk, while urban areas’ increase also has weak negative relationships. This highlights that urbanization is rather not a pressing problem for flood risk in Myanmar according to the provided data.

Non-significant but positive relationships with flood risk are detected in drivers such as agricultural areas’ increase, mining areas’ increase, mining areas’ increase within flood plains, and conflicts. Therefore, although not statistically significant, they are contributing to flood risk spatially. Here, it is again the questions of scales over development of flood risk.

4. Discussion

Our study aimed to analyze spatial-temporal relations of flood risk and its potential drivers. The main finding in our analysis is that dynamic pressures of flood risk such as agricultural increase in flood plains, forest loss, forest loss in flood plains, wetland loss, wetland loss in flood plains, increase in plantation areas, and increase in plantation areas in flood plains have statistically significant positive spatial relationships with flood risk. Therefore, their escalation contributes to the flood risk.

Reinforcing those statistical results, their spatial distribution shows that some townships with highest agricultural extension, with highest conflicts, highest mining, and highest plantations, were spatially coincident with the second highest risk classes, indicating compound risks of socio-ecological disasters. This kind of spatial overlap of dynamic pressures can be extremely dangerous for vulnerable populations of Myanmar.

These spatial overlaps and positive relations with flood risk are on account of poor governance in landuse and environment management. We found large proportion (44 percent) of total mining areas exist in the upper flood plains, which are linked to inadequate environmental governance. This can result in poor water qualities, and rising flood water levels deepen risk in downstream areas due to the rain-induced disposal of mine tailings in the upstream rivers. Moreover, deforestation in the upland flood plain areas of rivers can lead to high flood risk in the downstream areas. High deforestation in the upper peripheral areas of dry zone may increase flood and drought risk, especially to agrarian populations in the middle Myanmar. Despite the extent is small, wetland area loss is materialized within the flood plain areas. These dynamic increases can have serious implications in case of large-scale disasters.

More importantly, increase in urban settlements mostly occurs in flood plain areas although urbanization has shown weak negative spatial relationship. The fact that 53 percent of urban areas are within the flood plains in 2018, indicating high risk to those people. There might be many underlying reasons behind the existence on the high-risk flood plain areas. In Myanmar, most of the townships and villages are located near the rivers and streams. The reasons can also be some push and pull factors of migration or original existence in the alluvial areas. It may be voluntary relocations due to high fertility of soils in the flood plains. It may be forced displacements due to land grabbing, conflicts, or development activities. Under the conditions of structural constraints such as environmental degradation, land tenure insecurity, and limited access to the infrastructure, the agency of people has little choices to response and adapt to those pressures.

However, it is essential to notice the importance of scales in expressing the impacts. Impacts of mining and deforestation on flood risk are not just realized in the areas where mining and deforestation accommodate. Downstream effects, and spatial and temporal knock-on effects that cascade beyond the local areas should not be ignored. Although major increase in agriculture is not in the delta, they are mostly realized in the upland areas of Myanmar’s rivers, which are critical for flood risk of the downstream communities in the rest of the country. It is assumed that the delta already has limited space for the agricultural expansion. In addition, increase of plantations areas in Kachin State can exacerbate the flood risk for those downstream communities.

As a result, a history of poor environmental governance is related to flood risk in Myanmar. Ongoing deforestation, farmland expansion, and unstructured or poorly planned land use change can be linked back to natural resource mismanagement and state inefficiencies [37]. This has contributed to decreased flood plain areas and high levels of sedimentation and erosion in waterways, worsening flood hazard and increasing exposure of people and assets to floods [22, 77]. Moreover, the associated environmental degradation and poverty increase flood vulnerability of many Myanmar people, whose livelihood base and survival depend on the ecosystems, resulting reduced capacities of both society and ecosystems. The more the people are poor, and they depend more on ecosystems. Degraded ecosystems and environments make them more susceptible to flood risk.

Our results also highlighted the importance of recognition of compound and cascaded disasters in disaster risk management. The fact that over one-third (31 percent) of the current agricultural areas (in 2018) in Myanmar are within floodplains, showing potentially high financial agricultural losses in case of large-scale flood disasters. Ayeyarwady delta belongs to most agricultural areas within the flood plains. Being that Ayeyarwady delta produce more than 30 percent of country’s rice production and a biodiversity hotspot area, damaging this critical region, will have wider and long-term impacts millions of people in the rest of the country. In addition, almost all of agricultural area distributions spatially mirror with the highest land surface temperature areas. This indicates that agriculture in Myanmar is highly susceptible to heatwaves, droughts, and flooding.

Floods, heatwaves, and droughts are reinforcing each other in terms of increasing hazard intensities, even without considering vulnerabilities. Floods that follow droughts and heatwaves can be extreme due to damaged soil moisture and ability of soil to absorb water. This was epitomized in the case in Magway region where flooding occurs after serious drought and extreme temperature in the dry zone areas in 2011.

Overlap of disasters and their drivers in terms of time and space can lead to compound disasters. Another visible overlap can be seen in Myitkyina. Among other highest risk townships, Myitkyina experiences its peculiarities, representing an example of compound drivers. Among those townships, it has highest forest loss of 500 km², and experiences sharp increase in agriculture (2000–2001s), and a steep rise in orchards and plantation areas (1994–2001s). It also possessed mining areas and saw a reverse trend in conflicts which increases in 2021. Surprisingly, 100% of urban settlements of Myikityina in 2018 are in the flood plain area, which is critical for the residents to be aware of flood risk. Most importantly, Myitkyina has the highest socio-ecological flood risk in Myanmar.

Another important finding is that many dynamic pressures accelerated during the period between 1995s and 2000s. We also find that dynamic pressures of flood risk are temporally related to each other. Changes in pressures that contribute to flood risk (deforestation, agriculture and plantations increase, urbanization) were rapid and remarkable in the first-half period between 1987 and 2000s, especially between 1995s and 2000s. This highlighted that some socio-economic and political reforms during that critical period substantially and dramatically cause to increase flood risk. In the first-half period (1987–2000s), floods were more frequent but less intense, after 2000s, flood becomes more intense and less frequent.

It is alarming that dynamic pressures are increasing over time. On the one hand, global anthropogenic climate change is accelerating worryingly. On the other hand, forests and wetlands that will absorb and provide capacities to adapt climate impacts were decreasing while the agriculture, mining, and urban areas that increase climate vulnerabilities were increasing over time.

It is important to include indirect drivers that drive direct drivers of risk in the risk assessments. Indeed, those indirect drivers are root causes of risk. For example, while deforestation drives flood risk, drivers that lead to deforestation should be considered in flood risk assessments. If risk assessments only consider indicators that directly related to flooding, risk assessments will only recognize symptoms of flood risks, leading to prioritization of treatment of symptoms.

Similar to the identified root causes in the literature review, flood vulnerability is driven by poverty, limited health care such as insufficient doctors and hospitals, and wetland area loss. The consequence of mismanagement of natural resources, corruption in resource allocation, and social and economic insecurity is continued poverty and vulnerability in Myanmar [78]. Moreover, persistent high levels of poverty have resulted in a growing population living in unsafe areas, with poor infrastructure and housing conditions, contributing to compounding impacts. Widespread inadequate access to services that assist people in everyday life such as health care and electricity sources has also reduced the potential for economic participation that enables people to escape from poverty, while further limiting capacities to cope and adapt to flooding [79].

Spatially, the highest vulnerability occurs in ethnically dominated bordered areas such as Rakhine, Shan, Chin, and Kachin states where poverty and lack of access to services are especially severe due to the continuing conflicts and violence [15, 79]. This has resulted in a situation where it is difficult for people to pursue livelihoods and make it uncertain for their future due to the lack of security [15, 79].

One of our limitations is the resolution of the data, because most of the dynamic pressures were based on the datasets that have resolution of 618.74 m². Data paucity was also an issue for other datasets such as longer frame of conflicts, shrimp and fish farming, landslides, droughts, waste management, salt water intrusion, water and air pollution, and sea level rise, which might help to explain the bigger complex figure of flood risk in Myanmar. Other high-risk townships, for example, in Rakhine State, likely have long history of conflicts. However, due to inaccessibility of long-term data, it could not be proved.

Due to the differences in classification methods and aggregation methods, determination of risk for each township can be different. We purposely use natural-break method over quantile methods as it better reflects reality in terms of spatial differences. Moreira et al. (2021) also recommended the natural break method as the best performed classification approach in risk assessments [80] .

The results of our study are essential for disaster risk management and climate change adaptation in Myanmar because the study considers comprehensive socio-ecological flood risk that is compared to and incorporated its dynamic pressures and root causes. Only through addressing those underlying factors of flood risk, it is possible to reduce vulnerabilities and increase resilience of Myanmar people, while also facilitating flood exposure and hazard patterns. Our findings also stress the need of effective land use and environmental governance that consider compound and cascaded flood risk, and prioritized investment in public services and infrastructure such as health, and education, due to their links with factors increasing vulnerability and flood risk.

5. Conclusion

This chapter presents spatial temporal relation of flood risk and its dynamic pressures in Myanmar. It illustrates the first, conceptually sound risk assessment for major river and coastal flooding in Myanmar. The finding highlights many socio-ecological dynamic pressures driven by economic- and governance-related root causes were positively and spatially correlated with flood risks. We call attentions to critical concerns about existence of those dynamic pressures in the flood plains and their potential compound and complex disasters. Effective land use and environmental management is essential to prevent deforestation, agricultural extension, urban extension, mining, and wetland loses in the critical areas such as upstream areas and flood plain areas of Myanmar’s major rivers and to protect key ecosystems such as mangrove forests in the coastal areas.

Acknowledgments

The authors appreciate much to Ms. Alexandra Dudley for her contribution in the first version of river flood risk assessment, which was later further developed to socio-ecological risk assessment for river and coastal floods. We are very thankful to Dr. Michael Hagenlocher, Dr. Yvonne Walz, and Dr. Mostapha Harb for their guidance in the first version of river flood risk assessment.

Conflict of interest

The authors declare no conflict of interest.

A.

Table A1 provides details of all the relevant indicators for flooding in Myanmar. Indicators that were not included due to lack of data are highlighted red. Direction refers to the impact of the indicator on vulnerability and risk. A positive (+) sign indicates that a higher score increases vulnerability and risk. A negative (−) sign indicates that a higher score decreases vulnerability and risk.

Social Exposure
Category	Indicator	Code	Direction	References	Data Source	Justification	Comments
Exposure	The population exposed to floods (%)	e_pop	+	[40]	Hazard: JRC and the WorldBank Catalog; WorldPop	Without population exposure to a hazard there is no risk to the population.	Hazard return period: 100 years
	Exposed forests (%)	e_fl	+	[52]	Servir Mekong	Without forest exposure to a hazard there is no risk to ecosystems
	Exposed Agriculture (%)	e_agri	+	[52]	Servir Mekong	Without agriculture exposure to a hazard there is no risk to the ecosystems
Social susceptibility
Poverty and inequality	Dependency ratio (%)	s_dep	+	[81, 82, 83]	Census 2014	Young and old people are economically dependent on the state or their family. They may need assistance evacuating, and they are more prone to succumb to illnesses such as those from water borne diseases.	Calculated by summing the population under 15 years and over 65 years and dividing by the population 16–64 years.
	Average annual income	s_pov	+	[81, 84]	General Administration Department Survey 2016–2017	Low average income means that the township is likely to be deprived of goods, services, and opportunities. They have little buffer for shocks and will struggle to recover after a flood.	Proxy for: Population below the poverty line (%)
	Households with existing debt (%)	s_deb	+	[81, 82]	—	Existing debt restricts ability to access more credit after a flood. Can reflect financial stress.
	Minority population (%)	s_min	+	[85]	—	It is harder for minorities to access education, employment, credit, due to structural inequality.
Disability and health status	Proportion of population with a disability (%)	s_dis	+	[81]	Census 2014	High numbers of disabled people increase pressure on communities due to dependency, and more people will need assistance during flood.	Disability defined by MIMU (2018) as seeing, hearing, walking, remembering.
	Number of cases of malaria per 10,000 people	s_vec	+	[82, 83, 85, 86]	General Administration Department Survey 2016–2017	Vector borne diseases increase the health burden. Flooding increases breeding grounds for the vectors, increasing the likelihood further of contracting the disease.	Proxy for: Population with a vector borne disease (%)
	Percent share of the incidence of tuberculosis, dysentery, and hepatitis (%)	s_chr	+	[82]	General Administration Department Survey 2016–2017	Chronic diseases reduce the ability of people to pursue capital building activities and increase the need for assistance and specialized care during a flood.	Proxy for: Population with chronic illness (%)
	Population undernourished (%)	s_nou	+	[81]	—	Undernourishment impacts education, employment, and health outcomes.
Gender	Female-headed households (%)	s_fhh	+	[81, 87]	Census 2014	Females have less access to education, employment, land, and credit which affects the whole household here.
Stability/; conflict	Number of conflict events per 10,000 people in the last year	s_con	+	[81]	Conflict: The Armed Conflict Location & Event Data Project; Population: WorldPop	Conflict is extremely relevant for Myanmar. It affects people’s ability to achieve livelihood goals, and ongoing instability undermines development activities.	Proxy for: Population affected by conflict in townships (%); Time: June 1, 2019–June 1, 2020. Includes: battles, explosions, riots, and violence against civilians.
	Proportion of internally displaced persons (%)	s_idp	+	[81]	—	This population has no safety net, limited shelter, and emotional trauma.
WASH	Households with safe sanitation (%)	s_san	—	[81, 88]	Vulnerability assessment 2018 based on Census 2014	Poor sanitation leads to poor health outcomes. Unsealed sanitation will contaminate waterways during a flood.	Defined by MIMU (2018) as flush and improved pit latrine.
	Households access to improved drinking water supply (%)	s_dri	—	[81, 82, 85, 86, 89]	Vulnerability assessment 2018 based on Census 2014	Poor drinking water source leads to poor health outcomes. During and after a flood, water from non-improved sources is more likely to be contaminated.	Defined by MIMU (2018) as tap, tube well, protected well, and bottled water.
Settlement and Housing	Households with walls or floors made from leaf, bamboo or earth (%)	s_wfl	+	[81, 84, 87, 90, 91, 92]	Census 2014	These houses are likely to be damaged or washed away in a flood reducing quality shelter for populations.	Proxy for: population living in poorly constructed housing. Calculated by average the % of houses with walls made from leaf, bamboo, or earth and the % of houses with floors made from leaf, bamboo, or earth.
	Population living in informal settlements (%)	s_ins	+	[88]	—	Are less likely to be considered by aid agencies/government, have limited access to services.
	Population living in unofficial wards (%)	s_unw	+	[88]	—	Are less likely to be considered by aid agencies/government, have limited access to services.
Remoteness	Average travel time to the nearest city (minutes)	s_ttc	+	[83, 93]	The Malaria Atlas Project 2015	This defines how easily services normally only found in cities can be accessed, for example most health care. Also indicates access to markets, creditors etc.
Employment	Daily wage workers (% of working population)	s_daw	+	[81, 84, 94]	General Administration Department Survey 2016–2017	When disruption occurs, daily wage workers will be the first to lose employment and therefore income.
Environment and land use	Extent of forest loss (%)	s_ftl	+	[52]	Servir Mekong; 1987–2018	Loss of forest areas reduces capacities of ecosystems to reduce flood risk
	Extent of Wetland area loss (%)	s_wetl	+	[52]	Servir Mekong; 1987–2018	Loss of wetland areas reduces capacities of ecosystems to reduce flood risk
	Extent of urban extensions in the flood plains	s_ure	+	[52]	Servir Mekong 1987–2018	Urban extension in the flood plains increases risk for the people living in those settlements
	Extent of mining area extension in the flood plains (%)	s_age	+	[52]	Servir Mekong 1987–2018	Mining area extension can increase water pollution and sedimentation
	Extent of Agricultural extension in the flood plains (%)	s_mine	+	[52]	Servir Mekong 1987–2018	Agricultural extension in the flood plains reduces rooms for rivers and increases financial losses
Adaptative capacity
Information/early warning	Households with access to radio, television, internet, or mobile (max %)	c_ati	—	[81, 82, 89]	Census 2014	People can be warned, find more information on the flood and find services to help cope.	Proxy for: Households with access to information (%)
	Number of rain gauges per 10,000 km	c_rai	—	[95]	—	Provides accurate input to early warning systems.
Ecosystems	Extent of forest areas (%)	c_ft	+	[52]	Servir Mekong 1987–2018	Forest provides ecosystem services to reduce flood risk.
	Extent of wetland areas (%)	c_wet		[52]	Servir Mekong 1987–2018	Wetlands provide ecosystem services to reduce flood risk.
	Presence early warning system (yes/no)	c_ear	—	[86, 89, 91, 96]	—	Early warning allows timely evacuation.	Potential proxy: Households that received early warning in last event (%)
Social capital	Population that has lived in the township for more than 1 year (%)	c_ltr	—	[83, 87, 97]	—	Indicates knowledge of the area, evacuation routes, social capital that would help during and after a flood.
Education	Literacy rate (%)	c_lit	—	[83, 87, 97]	Census 2014	Literate people can read information and warning pamphlets. They can fill in forms to get aid.
Institutional capacity	Money spent on disaster risk reduction per person.	c_res		[86, 89, 91, 96, 98]	—	Indicates the institutional capacity to assist population and the systems in place to help populations cope.	Potential proxy: Number of disaster risk reduction projects per 10,000 people
Transportation	Households owning a boat (%)	c_boa	—	[87, 91, 94]	Census 2014	Boats are useful in floods to evacuate people and important goods.
Health	Number of doctors per 10,000 people	c_doc	—	[81, 82, 83, 86, 88, 89]	General Administration Department Survey 2016–2017	A population needs doctors to treat illnesses and injuries after a flood.	This was not aggregated with c_bed, as even if there are enough hospital beds, without doctors no treatment is available.
	Number of hospital beds per 10,000 people	c_bed	—	[81, 82, 83, 86, 88, 89]	General Administration Department Survey 2016–2017	Populations need access to hospital beds to access health care	This was not aggregated with c_doc as without hospital beds, it is challenging for populations to access doctors and treatment.
Access to energy	Households with access to alternative electricity sources (%)	c_aes	—	[82]	Census 2014	After floods it is likely that there will be damage to electricity transmission wires. Alternative sources can help communities stay safe with light, charge or use devices to maintain contact with others and authorities.	Solar energy and generators were considered alternative energy sources. The number of sources per township were summed and divided by the number of households.
Accessibility	Density of roads (road kernel density: km road/km2, search radius = 5 km)	c_dens	—	[81, 88, 94]	Open Street Map 2020	Presence of roads increases accessibility for emergency vehicles and provides paths of evacuation for populations.	Trunk, primary, secondary, and tertiary roads were considered. Data downloaded 6/6/2020.
Settlement and housing	Houses with more than one floor (%)	c_flo	—	[92]	—	Can move furniture out of flood, can evacuate or live on top floor.
Financial capital	Households with diverse incomes (%)	c_div	—	[81, 82, 87]	—	If one income is disrupted due to flooding, they can rely on the other.
	Households with savings (%)	c_sav	—	[81]	—	If income is disrupted, they can rely on savings.
	Households with insurance (%)	c_ins	—	[81, 87]	—	Provides financial security. Can rebuild after a flood.
Public infrastructure	Average distance to nearest shelter place.	c_dts	+	[81, 85, 89, 90, 96]	—	Determines how easily populations can evacuate and find safety.	Potential proxy: Number of shelter places per 10,000 people. Includes schools, monasteries, and evacuation shelters.
Food security	Households that save food (%)	c_sfo	—	[82, 86]	—	Flooding disrupts agriculture which impacts food availability. Saving food means families will be able to eat after a flood.

Table A1.

List of desired indicators with references, data sources (where applicable), and justification.

Datasets:

WorldPop: http://www.worldpop.org.ukCensus 2014: http://themimu.info/census-data (spreadsheet: BaselineData_Census Dataset - Sr., District & Township_MIMU 16Jun2016 ENG.xlsx).

General Administration Department Survey 2016–2017:

Income: https://data.opendevelopmentmekong.net/dataset/general-administration-department-gad-2016-17--module-6-individual-income-section-prod2?type=dataset

Population: https://data.opendevelopmentmekong.net/dataset/general-administration-department-gad-2016-2017---module-4-population-population-section-prod2?type=dataset

Malaria/chronic disease: https://data.opendevelopmentmekong.net/dataset/general-administration-department-gad-2016-2017-module-7-health-most-disease-in-region-section?type=datasetDoctors: https://data.opendevelopmentmekong.net/dataset/general-administration-department-gad-2016-2017-module-7-health-personal-health-care-section?type=dataset

Hospital beds: https://data.opendevelopmentmekong.net/dataset/general-administration-department-gad-2016-2017---module-7-health-hospital-section-prod2?type=dataset

The Armed Conflict Location & Event Data project (ACLED): https://acleddata.com/data-export-tool/

Vulnerability assessment 2018: http://themimu.info/vulnerability-in-myanmar (Spreadsheet: Datasets_Vulnerability Analysis in Myanmar_09Jul2018.xlsx).

The Malaria Atlas Project: https://malariaatlas.org/research-project/accessibility_to_cities/

Open Street Map: https://export.hotosm.org/en/v3/

Appendix B provides additional information on outlier analysis and treatment, and multicollinearity assessment.

Outliers and winzorization: Box plots based on the interquartile range and skewness and kurtosis were used to identify outliers (where skewness >1 and kurtosis >3.5 indicate potential outliers [52]. There was a high number of data points outside the interquartile range, and indicators with high skewness and kurtosis. However, only nine points in total were determined to be outliers and errors based on the local expert knowledge of one of the authors. These were treated with winzorization which is a common method to manage outliers [50]. As the data was not treated with the intention to create normally distributed data (as this would alter the characterization of vulnerability and risk beyond what is the true situation in Myanmar), the median as a measure of central tendency for each indicator. Table B1 outlines which data points were treated and which value they were given.

Indicator	Treated value	From	To	Value
s_pov	34,303,567	Seikkan	Yebyu	12,357,626
s_pov	26,322,559	Kyeemyindaing	Yebyu	12,357,626
s_pov	23,616,513	Kyauktada	Yebyu	12,357,626
s_pov	18,114,882	Magway	Yebyu	12,357,626
s_vec	509.32	Paletwa	Tanai	231.98
s_chr	2.281641	Palaw	Aunglan	1.558724
s_fhh	55.05	Thayetchaung	Amarapura	40.97
s_fhh	48.41	Patheingyi	Amarapura	40.97
c_doc	248.09	Oke Ta Ra Thi Ri	Cocokyun	8.97

Table B1.

Treated indicators and their values.

Multicollinearity assessment: A multicollinearity assessment was conducted to determine if there were any redundant indicators. Using Kendall’s Tau suitable for non-normal data (Puth et al. 2015), and two tailed significance, no issue of collinearity was detected based on a threshold of ±0.9. Table B2 outlines the results from the analysis.

Table B2.

Results of multicollinearity assessment using Kendall’s tau using SPSS (IBM SPSS statistics).