Evaluating flood hazards in data-sparse coastal lowlands: highlighting the Ayeyarwady Delta (Myanmar)

With ongoing climate change, consequently rising sea level and shifting patterns of atmospheric circulation and extreme weather events, coastal lowlands are increasingly exposed to flood hazards. Globally, 2.23 million km² and up to 290 million people were affected by floods between 2000 and 2018 (Tellman et al 2021). Especially large and densely populated river deltas, many of them located in tropical Southeast Asia, are among the most threatened and constitute areas of high exposure and vulnerability (Woodruff et al 2013, Tessler et al 2015, Hill et al 2020, Hooijer and Vernimmen 2021). According to Edmonds et al (2020), 89% of the global delta population (estimated to 302 million of in total 340 million of 2017 population) lives in regions with tropical cyclone (TC) activity and from the 76 million people which would be directly affected by a 100 year flood, 41% (∼31 million) live in deltas (Edmonds et al 2020). These numbers reflect lower-end estimates as estimations of global delta populations count to ∼500 million or even more (Syvitski et al 2022, Schmitt and Minderhoud 2023, Becker et al in review). Relative sea-level rise (RSLR) adds on this exposure, driving and increasing coastal flood frequency and shoreline retreat also in areas where cyclone activity is comparably low (Woodruff et al 2013). As already moderate RSLR of 1 m would increase the present global population at risk to 410 million, ∼60% living in tropical Asia (Hooijer and Vernimmen 2021), Kulp and Strauss (2019) estimate up to 630 million people (of 2010) to be affected by flooding by 2100 due to extreme water levels on top of SLR scenarios of high representative concentration pathways. However, as population growth is projected for many coastal lowlands in the world, particularly in Asia and Africa, these numbers provide only minimum estimates and might be exceeded already by the mid of the 21st century (Neumann et al 2015). This holds especially true as these studies did not account for the contribution of vertical land motion, while land subsidence currently affects ∼20% of the world's population (Herrera-García et al 2021), many of them living in densely populated deltaic areas where highest RSLR occurs (Nicholls et al 2021). Unlike extreme water levels due to storm surge and RSLR that—in terms of cyclones—may occur on annual time scales, or—in terms of RSLR—will become effective over decades, tides constitute a coastal flood type of daily frequency. High tides in coastal lowlands regularly lead to flooding, often up to several kilometres inland (e.g. Marfai and King 2008, Kuenzer et al 2013), and dominate many of the world's deltas (Goodbred and Saito 2012), where they add on elevated water levels from surges and waves while their constituents being altered by RSLR (Pickering et al 2017). Furthermore, tsunamis also pose a coastal hazard as they constitute high-energy wave events caused by submarine earthquakes, submarine failures or flank collapses, or meteorological forces (Röbke and Vött 2017). However, their timing occurs on much larger recurrencies, i.e. in the range of centuries (e.g. Brill et al 2011), and their assessment involves large uncertainties stemming from uncertainties in underlying scenario parameters (e.g. moment magnitudes of such low-frequency events; Davies et al 2018 and references therein).

Besides these coastal flood hazards, river deltas are exposed to flooding from the hinterland, which is induced by high discharges, thereby increasing inundation extents in case these inland floods occur simultaneously to other flood types. Particularly river basins characterised by a tropical monsoon climate show high annual discharge variabilities that follow the seasonal monsoon precipitation pattern (Hansford et al 2020) and have experienced serious flooding (e.g. Islam et al 2010, Syvitski and Brakenridge 2013). Livelihoods of 60% of the world's population are directly affected by the Asian monsoon (Plink-Björklund 2015). In this region, most of the population concentrates along large South- and Southeast-Asian rivers and their deltas, such as the Ganges–Brahmaputra–Meghna, the Chao Phraya, Mekong and Ayeyarwady (WorldPop and CIESIN, 2018), and all of them were affected by catastrophic flooding in the recent past (Cosslett and Cosslett 2014, Liew et al 2016, Brakenridge et al 2017, Krien et al 2017).

The Ayeyarwady Delta in Myanmar was severely affected by the storm surge of category-4 TC Nargis in 2008 that caused >138 000 fatalities (Webster 2008, Fritz et al 2009), and the exceptional monsoon flooding in 2015 that affected >530 000 ha agricultural and aquacultural land (Government of the Union of Myanmar 2015). However, information on flood exposure and hazards is still limited (e.g. table S1; Brakenridge et al 2017, Phongsaphan et al 2019) and spatio-temporal patterns of floods have been only addressed by emergency mappings from international initiatives (ITHACA et al 2008, Copernicus Emergency Management Service 2015, UNOSAT 2015) or national institutions (e.g. Myanmar Information Management Unit (MIMU) 2015, 2020), rather than being systematically studied by scientific endeavours and other efforts using decentralised information and following standardised data organisation pathways. So far, flood hazard and risk maps only exist on township level but do not provide information on local-scale characteristics, nor distinguish different types of flood hazards (Phongsaphan et al 2019) or they only delineate zones in specific urbanised areas that will be inundated in case of a 'Nargis'-like flood and a 100 yr probability (ICHARM et al 2016a, 2016b, 2016c, 2016d). Similarly, areas, people, and assets at risk of both inland and coastal flooding have been only partially addressed so far (Brakenridge et al 2017, Heinkel et al 2022), mainly by large-scale models (Dottori et al 2016, Tierolf et al 2021), while on local scale, RSLR has been shown to cause major parts of the delta to fall below sea level in the coming decades (Seeger et al 2023a) and thus constitutes a substantial determinant of future flood risk.

For other data-sparse coastal lowlands, where comparably more information is available, some regional-scale flood hazard and risk assessments were generated for entire river basins. For example, Sriariyawat et al (2022) identified areas at risk of flooding from rainfall and runoff in the Chao Phraya Basin based on simulated flood inundation and duration while Shakti et al (2020) addressed pluvial flooding in the Chao Phraya Basin based on 100 yr, 200 yr, and 500 yr return periods estimated from rain gauge measurements. Although indicating a flood risk assessment, neither societal nor socio-economic variables were considered, thereby underscoring the wide range of how exposure, hazard, vulnerability, and risk are defined and consequently handled in related research and assessments. For the Ganges–Brahmaputra–Meghna Delta, efforts have been undertaken to investigate the impact of multiple flood hazards on the delta as Rahman et al (2019) modelled storm surge inundation in combination with SLR scenarios to report the effects on population and the Sundarban mangroves. Still for several local to regional hazard assessments in data-sparse coastal lowlands, the linkages between flood exposure and environmental characteristics such as local elevation and geomorphological features have not been adequately addressed. These physical characteristics have rarely been quantified or qualitatively assessed, which is partly due to the insufficient spatial resolution and focus of those assessments.

The above examples show that existing flood hazard assessments in data-sparse coastal lowlands are often incomplete as they (1) do not distinguish different flood types and their combination in terms of compound flooding, (2) do not cover entire, hydrologically-connected low-lying landscapes (such as deltas), (3) are not sufficiently spatially resolved, (4) are not up to date and miss new available information and data, or (5) require measurement data for validation (e.g. ICHARM 2016a, 2016b, 2016c, 2016d, Hein Min Htet et al 2017, Ikeuchi et al 2017). Global flood hazard assessments cannot fill these gaps as they are comparably coarse and typically involve data processing that, if applied on smaller scales, would introduce substantial errors and therefore should not be applied on regional or local scale (Jongman et al 2012) but requires validation with observational data (e.g. Ward et al 2018, Maranzoni et al 2023). They also tend to exclude data-sparse coastal lowlands since these are often not covered by the input datasets used (Eilander et al 2020). As a result, information on flood exposure and hazards is limited in many densely populated coastal areas in the world. Here, high-quality data either does not exist or is not accessible which often constitutes a main barrier for generating sound assessments and developing risk mitigation strategies, especially for scientific and public communities in the so-called Global South.

In this study, we present a standardised, integrative approach to generate a first-order assessment of both single and multiple flood-type hazards for data-sparse, poorly accessible and inaccessible coastal lowlands by combining various open and publicly available datasets and software. The approach yields information on spatial patterns as well as population and assets at risk. It thereby forms the basis for follow-up focus studies and enables stakeholders from all fields of action to generate flood hazard assessments as a basis for risk mitigation strategies. We highlight the applicability of our approach by using the example of the Ayeyarwady Delta in Myanmar as this region suffers from data paucity and severe difficulties in accessibility due to political restrictions and recent conflicts. We demonstrate the generation of a thorough understanding of the exposure of areas and population to flood hazards by (a) unravelling monthly flooding activity in the Ayeyarwady Delta in the era of Sentinel-1 imagery; (b) investigating single-type flood hazards in the delta such as monsoon and storm surge flooding, and RSLR; (c) identifying and characterising areas prone to both inland and coastal flooding, as well as RSLR; and (d) placing the findings from (a) to (c) into the broader context with implications for other data-sparse coastal areas. Therewith, our workflow proves successful in attributing different types of flood hazards, ranking exposed areas, and offers valuable insights for vulnerability and risk assessments as base for risk-informed management decisions.

The extents of recent flood events were mapped using free-available Sentinel-1 data which offers actively sensed earth observation imagery at high spatio-temporal resolution, being highly proficient to capture flood events independent of daylight and weather conditions. Flooding in the Ayeyarwady Delta during the last ∼10 years was mainly associated with monsoon precipitation and river discharge. For this reason, our investigations focused on months of the monsoon season, i.e. from May to October. Monthly flood extents were mapped for the years of 2015–2023 using the cloud computing platform of Google Earth Engine (GEE; Gorelick et al 2017) and a pixel-based image rationing approach following the open accessible GEE script of the United Nations Platform for Space-based Information for Disaster Management and Emergency Response (UN-SPIDER) program (UN-SPIDER, no date). A detailed description of the applied GEE script settings and data processing to obtain monthly flood extents in the Ayeyarwady Delta is given in the supplementary material.

The individual monthly flood maps were summed (i) for all months of individual years of the investigation period (2015–2023), (ii) for individual months of all years, and (iii) for all months of the entire investigation period (2015–2023). To better understand the spatio-temporal pattern of the floodings mapped, we supplemented our investigations by open-source freely available precipitation data from the GPM IMERG Final Precipitation L3 v.6.0 product (0.1° × 0.1°) and runoff data from River and Reservoir Watch version 4.5 (figures 1 and S1–S9; Huffman et al 2019, Brakenridge et al 2023a, 2023b, 2023c).

Zoom In Zoom Out Reset image size

To exemplify the suitability of the approach for assessing the hazard of multiple flood types and providing a glimpse towards their impacts in the future, we studied both storm surge flooding and RSLR in the Ayeyarwady Delta separately and investigated their combination with monsoon flooding. To determine areas at risk of storm surge, we used the flooding extent of the storm surge of TC Nargis 2008 which directly hit the delta. The related data was obtained via the freely accessible global flood database from Tellman et al (2021). To consider RSLR in the hazard assessment, we used a RSLR scenario of 1 m to map areas falling below sea level based on a locally improved FABDEM from Seeger et al (2023b). The exposure to multiple flooding was assessed through two scenarios: The short-term scenario included monsoon flood frequency and storm surge flooding while the mid- to long-term scenario also integrated the impact of RSLR. To characterise assets that nowadays would be affected by those hazards, we used the local land-use land cover (LULC) dataset of Vogel et al (2022) for the year 2021. Latest available WorldPop data (i.e. for the year 2020; Bondarenko et al 2020) was used to estimate population exposed to flooding. All post-processing investigations were performed in GIS on all single-type flood maps (i.e. monsoon flood frequencies, storm surge, and RSLR) and two scenarios of multiple flood hazards (i.e. the combination of monsoon and storm surge, and the combination of monsoon, storm surge and RSLR; figure 1).

3.1. Flood exposure of the Ayeyarwady Delta

Our systematic mapping approach delivers detailed information about the flooding history of the last ∼10 years and characterises areas in the delta that are frequently flooded in terms of their exposure by today's population and land use. Except of few outcrops and parts in the western and eastern delta margins, the entire delta is prone to (regular) flooding.

3.1.1. Exposure to single flood-type hazards

Monsoon flooding constitutes the most frequent flood hazard in the Ayeyarwady Delta, affecting ∼60% of the delta where ∼43% (∼5.4 million people) of the current delta population live (figures 2, 3 and S1–S12). Flood frequency is highest in Pathein District as more than 5 250 people of ∼786 000 people exposed in the district are prone to flooding every month of the monsoon season (figure S14). Especially Hinthada, Thayarwady, and Ma-ubin Districts in the upper and central delta as well as Pyapon District are among the most frequently and extensively flooded (∼64%–71%) whereas population counts indicate that the majority is not exposed (figure S14).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Though ∼73%–77% of the population in the districts of Yangon seem to be unaffected by monsoon flooding, generally higher population densities in these districts compared to other deltaic ones result in substantially higher absolute population amounts of ∼1.8 million people, most of them being based in Yangon North. However, relatively, the administrative districts of Labutta and Myaungmya show the highest exposure as most people live in areas being flooded several months a year (figures 2 and S14).

Assets at risk are mainly agricultural crops (figure 3). With increasing flood frequency, mainly irrigated areas are flooded. However, as our mapping approach does not allow for the differentiation between intended vs. unintended flooding, and cropping seasons and patterns vary annually, we cannot further characterise whether the exposed irrigated crops are assets with potential loss due to flooding. To a lesser extent, monsoon flooding also affects other economic sectors such as aquaculture and mineral extraction through brine ponds as well as shrubland (figures 3 and S14). Given only ∼2% of the entire delta are covered by urbanised areas, and several small settlements in the delta are located outside areas flooded during the monsoon season (figure S13(e)), they are relatively less exposed. However, built-up areas are particularly at risk in Yangon East and West where they constitute up to ∼50% of frequently flooded areas (figure S14).

Monthly and interannual flood frequency and duration vary, ranging from water-saturated soils to extensively flooded depressions and reactivated cut-off meanders, being subject to precipitation amount and intensity, runoff and discharge as well as infiltration capability of the soils (figure S13; Brakenridge et al 2017). Between years, variability in flood duration and/or frequency depend on the combination of onset, intensity, and duration of the monsoon season whereas the spatial flood pattern varies less being dominated by the effects of local topography where depressions are longer/more often flooded (figure S13), and properties of underlying soils (mainly gleysols and fluvisols that will be waterlogged; FAO and UNESCO 1976). In contrast, exposure to monsoon flooding also varies within the monsoon season due to a shift of control from a precipitation-dominated pattern in the first months (May to July), affecting the entire delta and causing delta-wide flooding and water saturation of soils, to a runoff-dominated pattern (July to October) where water flow results in flooding of particular areas such as topographic depressions (figures S1–S13). Consequently, this dynamic variability of flooding needs to be taken into account when considering the annually occurring, and therefore well-assessable monsoonal hazard in flood risk assessments.

Compared to monsoon flooding, affecting both inland and coastal areas, storm surge and RSLR are coastal hazards that may propagate inland depending—in case of storm surge—on incidence angle during landfall, wind direction, surge and wave heights, and—especially in terms of RSLR—on land elevation with respect to the sea. In this sense, the storm surge of TC Nargis constitutes a worst-case scenario as the 5 m surge compounded with 2 m storm waves led, together with tides, up to 50 km inland propagation of water masses, ∼50–100 m coastline retreat, and ∼1 m of vertical erosion (figure 2, Fritz et al 2009, Tasnim et al 2015, Besset et al 2017). In total, >6 200 km² were flooded, highlighting the exposure of ∼20% of the delta and ∼658 000 people currently living there (∼5% of the entire delta population), of which Labutta, Pyapon and Yangon South show the highest susceptibility with up to ∼50% of the districts' area and ∼462 700 people being exposed (figures 2, 3 and S15). Properties at risk are more diverse than for the monsoon hazard which is mainly due to the fact that agricultural production is adapted to regular monsoon flooding (that serves for intended flooding in contrast to extreme monsoon floods that will cause damage). A storm surge will largely hit parts of the delta used to produce non-irrigated crops (∼67%), and to lesser extent irrigated crops (∼15%), aquaculture (∼5%), brine ponds and urban areas (∼1%), respectively (figures 3 and S15). While the coincidence of storm surge flooding with water bodies in inland districts reflect the upstream migration of the surge in the river, those LULC types constitute major sources of income for the delta population and have implications for economic development in the region. The economic sectors show different susceptibilities on district scale, thus calling for local to regional measures to reduce the vulnerability of urban spaces in Yangon West and East, cropland and brine ponds in Pathein, Pyapon, Yangon South and Labutta, as well as aquaculture in Ma-ubin, Yangon North and South (figure S15).

RSLR constitutes a hazard to coastal lowlands like the Ayeyarwady Delta that is often not considered but is projected to lead to permanent loss of land and consequently challenge sustaining housing and livelihoods of the population. Using 1 m RSLR is a realistic scenario backed up by latest IPCC AR6 projections for 2100 (Fox-Kemper et al 2021, Garner et al 2021, 2022). However, as the delta is subsiding up to several centimetres per year (van der Horst et al 2018, Seeger et al 2023c), RSLR in sinking areas is higher than projected by the IPCC and impacts may already be noticeable sooner than 2100, i.e. over years to decade(s). We here focus on one of the currently most reliable elevation datasets and document ∼30% of the delta to fall below 1 m RSLR (figures 2 and 3, Seeger et al 2023a, Seeger et al 2023b). Both coastal and inland areas are exposed, mainly in the districts of Myaungmya (∼65%), Pyapon (∼63%) and Labutta (∼57%; figures 3 and S16). About 14% of today's delta population (∼1.8 million people) will be directly affected; however, population in highly exposed districts already amounts up to ∼50%, respectively (together accounting for >1 million people; figure S16), and slightly higher RSLR scenarios will pose substantially more people at risk (Seeger et al 2023a).

The land-use pattern in the low-lying areas at risk of 1 m RSLR differs only slightly from those exposed to storm surge as the spatial extent of both flood types largely overlap (see section 3.1.2). About 83% constitute non-irrigated and irrigated crops (figure 3). In most districts, the lowly-elevated areas prone to RSLR impacts are characterised by more built-up areas than those exposed to Nargis storm surge or monsoon flooding (figures S14–S16), thereby implying the increasing risk in case urban growth towards small cities in the southern delta continues as observed for the past decades (Vogel et al 2022).

3.1.2. Exposure to multiple flood-type hazards

To generate a holistic understanding of flood hazards in the delta and to characterise the overall exposure of deltaic areas, a synergistic perspective combining individual flood hazards and information about them is required. About 64% of the area and ∼32% of the population are exposed to multiple, short-term flooding events (figures 4 and 5).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Approximately 70% of the area prone to storm surge is regularly flooded during the monsoon season, an area currently inhabited by ∼294 400 people (∼42% of the delta population) and mainly used for agricultural production (figure S17). In relative terms, multiple flood risk is highest for Pyapon, Labutta and Yangon South, of which Yangon South has the highest population exposed while Pyapon denotes the largest area with land-use properties at risk (figure S18).

Including 1 m RSLR allows for a rough estimate of mid- to long-term flood exposure towards the future and highlights the susceptibility of >36% of the deltaic area to both coastal and inland flood hazards, especially in the southernmost districts in the delta (figures 4, 6 and S20). From ∼2 million people currently living in this zone (∼16% of 2020's total delta population), ∼1.4 million live within the highly exposed districts of Pyapon (∼53%; ∼498 800), Labutta (∼59%; 335 700), Myaungmya (∼47%; 334 300) and Pathein (∼18%; 241 600).

Zoom In Zoom Out Reset image size

The zone exposed to all three hazard types covers an area of ∼7% in the delta currently inhabited by ∼137 600 people (∼2%; figure 6). Pyapon and Labutta would be the most affected districts, both in terms of flood extent (∼26% of Pyapon; ∼16% of Labutta) and exposed population (∼57 700 in Pyapon; ∼30 900 in Labutta), followed by Pathein and Myaungmya (figures S19 and S20). Recent studies in the delta revealed between 10 and >30 available at: of land subsidence in these parts (2017–2023; Seeger et al 2023c), thereby implying an increasing risk in the future if not adequately mitigated.

3.2. Implications for coping with flood hazards in the Ayeyarwady Delta

3.3. Implications for generating first-order flood hazard assessments in data-sparse coastal lowlands in the world

We presented a standardised, integrative approach to conduct a first-order assessment of flood hazards in data-sparse coastal lowlands. We characterised present-day flood exposure due to monsoon and storm surge flooding as well as RSLR by combining only open and freely available datasets in open-source software, using the megadelta of the Ayeyarwady River in Myanmar, a region marked by data scarcity and restricted accessibility. The flood extent mappings were supplemented with information on precipitation and runoff, elevation, land use, and population, thereby allowing to explain observed flood extents and to characterise flood-prone areas in detail.

While >70% of the Ayeyarwady Delta and ∼40% of its present population is prone to flooding, we demonstrated that the southernmost districts, especially Pyapon and Labutta, show the highest exposure, and agricultural and aquacultural products range among the most exposed assets, stressing the need for strategies of sustaining food security and livelihoods relying on those economic sectors. Our work underlines the necessity of long-term adaptation measures and risk-sensitive development (which may include the diversification of livelihoods) and overall to strengthen the people's awareness and preparedness (not only) in respect to flood-related risks.

We have shown the general applicability of this approach for data-sparse coastal lowlands in the world in order to generate a first-order flood hazard assessment by highlighting its effectiveness in terms of workflow set up, data availability and collection, as well as computing capacity. The flexibility of the approach to be extended by integrating various available datasets and combining them in manifold ways supports to describe and evaluate flood-prone areas on regional and local scale, thereby allowing for attribution and complementing concepts of vulnerability and risk, and supporting risk-informed decision making and development of effective multi-flooding adaptation strategies.

This research was funded by the German Research Foundation (DFG Project Numbers 411257639; BR 5023/4-1; BR 877/37-1; KR 1764/28-1).

The data that support the findings of this study are openly available at the following URL/DOI: https://doi.org/10.5281/zenodo.10889566 (Seeger et al 2024).