Introduction

Exposure to natural hazards and poverty are interconnected in complex and reinforcing ways. Low-income and marginalised populations are typically more likely to live in hazard-prone areas, often with limited capacity to reduce their exposure by moving to safer locations or to a strongly built settlement1,2. Thus, multiple studies have acknowledged that people living in poverty are particularly vulnerable to natural hazards3,4,5,6,7, but that poverty can also be a key driver in increasing the vulnerability of households to such external shocks7,8,9. These mutually reinforcing factors can result in poverty traps, where the continuous setbacks can prevent individuals or whole populations from achieving their development potential, or result in households fluctuating across the poverty line over the course of their lifetimes6,7,10,11,12,13,14,15. These poverty traps are exacerbated when individuals have climate-sensitive, resource-dependent livelihoods and limited access to up-to-date information on climate hazards and protective infrastructure4,9,15.

The interactions between different natural hazards and poverty have been widely assessed, linking poverty with flooding4,5,7,16, riverbank erosion17,18,19,20, droughts6,7,12,13, cyclones1,21, extreme heat6,22, saline intrusion23 and sea level rise24. However, the interplay between hydrogeomorphic hazards and poverty have not yet been quantified. Hydrogeomorphic hazards may take several forms, including coastal erosion and deposition, river erosion and deposition, subsidence, siltation, landslides, and river avulsions25. In this paper, the focus is on water-induced landscape changes that affect human and environmental processes in deltaic systems8,26. These changes are expressed through four predominant processes: (i) riverine erosion and deposition; (ii) conversion of deltaic land to water bodies; (iii) wetting and drying of land due to land-use changes; and (iv) coastal erosion and accretion. These dynamics take place due to the complex interactions between background natural processes and extensive human interventions. It is important to note that hydrogeomorphic hazards are not inherently bad; they are natural processes of landscape evolution, and attempts to interrupt them are usually short-lived and have consequences elsewhere, but they can generate risks where they intersect with dense populations with limited capacity to adapt27. Importantly, unlike other hazards, such as flooding, there are limited options for coping with hydrogeomorphic hazards. For example, if a household’s land is removed by a river or the sea, temporary or permanent relocation is inevitable27,28,29. In such circumstances, hydrogeomorphic processes have been acknowledged to be key drivers of socio-economic deprivation25, particularly in densely populated and geomorphically complex landscapes.

Bangladesh is a prime example of this type of environment. The country is situated in a global hotspot for natural hazards and is also home to an exceptionally geomorphically complex and dynamic landscape: the Ganges-Brahmaputra-Meghna (GBM) delta. More than 80% of Bangladesh consists of floodplain land5 exposed to frequent and severe flooding and widespread erosion and hydrogeomorphic change25. It is also one of the world’s most densely populated countries30, with over 165 million people living on just under 150,000 km2 of land. Of the 165 million people, approximately 20.5% live below the national poverty line of less than US$1.90 (United States Dollar) per day in purchasing power parity (as of 2019)31. Given the densely populated fertile floodplain lands, the migration of river channels due to sudden and gradual channel shifting, as well as increasingly exposed coastlines to erosion and inundation, means that valuable cultivable land is lost, settlements are destroyed, and hundreds of thousands of people are displaced annually20,28.

Consequently, key policy documents for the GBM delta, such as the Bangladesh Delta Plan 2100, have acknowledged that hydrogeomorphic hazards, particularly river and coastal erosion, can act as key drivers in furthering poverty32. Although previous work in Bangladesh has assessed fluvially-induced channel changes18,33,34,35,36, as well as the vulnerability to erosion18,19,20,29,37,38,39, and multiple coastal hazards1,40,41,42, these studies did not focus on different hydrogeomorphic hazards at the national scale and did not explicitly combine these hazards with exposed populations and their associated levels of wealth.

This study addresses this gap by presenting a framework to quantify hydrogeomorphic risk that intersects hazard modelling with population exposure across different wealth groups (see Supplementary Fig. 1), with an application to Bangladesh. This is achieved through the (i) identification of hotspots of hydrogeomorphic changes at the national scale; (ii) assessment of population exposed to hydrogeomorphic hazards; and (iii) assessment of wealth levels of the exposed population. This study, therefore, presents a national-scale understanding of disparities in exposure to hydrogeomorphic hazards. The approach taken in this Bangladesh case study is applicable and relevant to other geomorphically dynamic deltaic nations in the world, where more than 500 million people live and where these interactions have similarly not been assessed43,44,45.

Results

Hotspots of hydrogeomorphic change in Bangladesh

Across all of Bangladesh, 28,300 km2 of land has undergone hydrogeomorphic change over the last three and a half decades, which equates to 20% of the country’s land mass (Fig. 1). Of the 28,300 km2, 38% (10,800 km2) has been exposed to erosion or wetting of land (e.g., waterlogging or extended inundation) (blue), while 62% (17,500 km2) has primarily experienced accretion or drying up of surface water bodies (brown). The total land area that is severely and consistently affected is 6568 km2, 2049 km2, and 782 km2, for the 25%, 10% and 5% most extreme ends of the Channelised Response Variance (CRV) spectrum (see “Methods”), respectively, which, in all cases, seems to be predominantly situated within and along the main river systems of the country.

Fig. 1: Hydrogeomorphic change across Bangladesh from 1987–2022.
Fig. 1: Hydrogeomorphic change across Bangladesh from 1987–2022.
Full size image

Brown colours indicate decreases in channel presence (accretion or drying up of land) and blue areas represent increases in channel presence (erosion or wetting of land). The intensity of the colour represents the frequency of change over the 35-year period. Inset maps illustrate the key hydrogeomorphic processes taking place: (a) riverine erosion and accretion; (b) conversion of deltaic land to water bodies; (c) wetting and drying of land; and (d) coastal erosion and accretion. Panel (e) shows the areas for all hydrogeomorphic hotspots and for the extreme ends (25%, 10%, and 5%) of the CRV spectrum. Note, the scale bar is for the main country-wide map. CRV = Channelised Response Variance.

As evident in Fig. 1, the CRV metric does not only identify where hydrogeomorphic changes are occurring, but also enables the distinction of different types of hydrogeomorphic processes taking place46. The different processes evident in Bangladesh can broadly be categorised into four groups: (i) riverine erosion and accretion (Fig. 1a); (ii) conversion of deltaic land to water bodies (Fig. 1b); (iii) complex patterns of wetting and drying of land (Fig. 1c); and (iv) coastal erosion and accretion (Fig. 1d).

The river corridors of Bangladesh's major river systems (Fig. 1a), experience substantial hydrogeomorphic changes, particularly within the Jamuna River corridor. Approximately 3430 km2 of land has eroded within the Ganges, Jamuna, and Padma river corridors, equivalent to 98 km2 annually, whilst 5050 km2 of land has been created (144 km2 annually). Of this, approximately 55% of erosion and accretion has been taking place within the Jamuna River corridor. Along the Ganges River, the model results illustrate that over recent decades, the river eroded 767 km2 of land, while accreting 1050 km2. Finally, along the Padma River, approximately 1210 km2 of land has been created in the last three and a half decades, whilst 788 km2 has been eroded away by the river. It’s evident from these figures that all three major river systems are currently creating more land than they erode; however, the magnitude of land loss is still profound.

In south-west Bangladesh (Fig. 1b), the modelling results illustrate an extensive area of waterlogging. Given that the model relies on dry season satellite imagery, it is unlikely that this is linked to seasonal inundation, which is also an issue in this part of Bangladesh47,48. Instead, this waterlogging is most likely caused by the polderisation and widespread aquaculture development in the form of brackish water shrimp and fish cultivation4,46,49,50,51. Particularly during the dry season, farmers pump brackish and saline water from surrounding rivers into their land for aquaculture production52.

In addition, the Sylhet basin in the north-eastern part of Bangladesh (Fig. 1c) is experiencing a complex combination of both wetting and drying up of land. In this region, land use has changed in recent years in the form of dry-season Boro rice cultivation, intentionally drying up natural wetlands53, as well as embanked road construction54, and localised subsidence55,56. The interactions between these multiple different driving factors have resulted in a highly hydrogeomorphically complex landscape.

Finally, in the Meghna estuary, the modelling results present deposition rates as almost three times higher than erosion rates (Fig. 1d). This finding agrees with multiple previous studies, which have highlighted that this region experiences the greatest rates of land creation, with coastal erosion being more prevalent further away from the river mouth25,57,58. The results also illustrate erosion along the western coastline of Bangladesh, reaching up to 2 km of erosion over the 35-year modelling period, equivalent to 57 m per year in parts of the western delta. Along this western coastline, the tides play a crucial role in pumping fluvial sediment from the Meghna estuary back towards south-western Bangladesh, ensuring the coastline has remained relatively stable despite the dramatic reduction in fluvial sediment delivery to this region since the construction of the Farakka barrage in 197525,59,60,61.

Population exposed to hydrogeomorphic change

At present, just over 22 million people (~13% of the total population) live within hydrogeomorphically unstable regions in Bangladesh, a figure that has grown by 5 million since 2000. Of the 22 million people, 4.3 million people are living within highly hydrogeomorphically unstable regions (25% of the CRV spectrum), and 1.7 million and 860,000 people live within the 10% and 5% most extreme cases of hydrogeomorphic instability, respectively. Fig. 2 illustrates the spatial distribution of population exposure to hydrogeomorphic hazards for the year 2020, with percentages in administrative divisions showing that the Sylhet, Dhaka and Chittagong divisions have the greatest increases in exposure since 2000, whilst the southern coastal divisions of Khulna and Barisal have the lowest. This can be attributed to greater background population growth in Dhaka and Chittagong, and more widespread recent hydrogeomorphic changes observed in the Sylhet basin. The main localised hotspots of greatest population exposure to hydrogeomorphic change can also be delineated. The three key areas with the most densely populated land susceptible to hydrogeomorphic change are along the Teesta River in the Rangpur Division, along the lower Jamuna River, near the city of Sirajganj, and on the outskirts of Dhaka city, highlighted by brown circles in Fig. 2.

Fig. 2: Spatial distribution of population exposed to hydrogeomorphic hazards in Bangladesh.
Fig. 2: Spatial distribution of population exposed to hydrogeomorphic hazards in Bangladesh.
Full size image

The population within hydrogeomorphic hotspots is highlighted with greater colour intensity. The percentages highlight the overall increase in population exposed per administrative division over the last two decades (2000–2020). The brown circles indicate the key regions of highest population density within areas of hydrogeomorphic instability.

Hydrogeomorphic change and wealth

Of the 22 million people exposed to hydrogeomorphic hazards in Bangladesh, 86% have low levels of wealth (Wealth Index (WI) ≤ 0), with 40% in the lowest wealth index group (WI ≤ − 1) (see Table 1). When this is compared to Bangladesh as a whole (see Table 1), it is evident that the lowest wealth index group (WI ≤ − 1) makes up 20–22% of the population, which is half of that observed within the hydrogeomorphically unstable parts of the country. This finding suggests an exposure bias of populations with lower wealth levels living in hydrogeomorphically unstable areas. This difference in exposure is diagrammatically represented in Fig. 3, which clearly illustrates the shift in the population’s socio-economic structure between the population within hydrogeomorphically unstable areas compared to Bangladesh as a whole. In order to test whether this exposure bias is statistically significant, the Mann-Whitney U test was performed. The resulting test statistic and p-value were found to be U = 6116.5 and p = 0.022, respectively, which implies that the shift observed in Fig. 3 is statistically significant.

Fig. 3: Exposure bias of population with low wealth.
Fig. 3: Exposure bias of population with low wealth.
Full size image

Normalised histogram showing the distribution of the population’s socio-economic structure for 2020, comparing the population exposed to hydrogeomorphic hazards to Bangladesh as a whole. The population count (y-axis) has been normalised for visualisation purposes.

Table 1 Wealth distribution of populations exposed to hydrogeomorphic hazards compared to the population of Bangladesh as a whole
Full size table

In addition, Table 1 illustrates that population growth rates (detailed in the final column) for the two lower wealth groups are relatively similar between populations exposed to hydrogeomorphic hazards and nationwide; however, for the higher wealth groups, there is a more substantial difference. The slower growth rate of wealthier populations observed within hydrogeomorphically hazardous regions could suggest that wealthier people may choose not to live in these unstable areas and move away, whilst the population that remains may face more challenges finding resources to migrate to safer locations. A similar trend is evident in the more extreme cases of hydrogeomorphic instability. For the 25%, 10% and 5% most extreme areas of hydrogeomorphic instability, the relative percentage change of population exposed in the lower WI groups (WI ≤ 0) has been increasing, whereas for the higher WI groups (WI > 0), the percentage change over the 20-year period has been negative (see Table 1 in the Supplementary Information). This suggests that, in areas with greater hydrogeomorphic risks, people with lower wealth levels continue to move towards highly unstable areas, while wealthier populations are moving away.

When assessing the spatial distribution of wealth within the hydrogeomorphic hotspots across the eight administrative divisions (Fig. 4), it becomes evident that the northern divisions of Rangpur and Mymensingh have the highest proportion of exposed populations with lowest relative levels of wealth. This is likely linked to these divisions generally having low levels of wealth (see wealth index map in Fig. 2 of the Supplementary Information), rather than disproportionate impacts of hydrogeomorphic hazards62. Interestingly, the spatial trends are also related to the location of cities relative to the hydrogeomorphic hazard areas (Dhaka and Chittagong), as cities are areas with better access to basic services and a greater concentration of assets, thus skewing the WI for the rest of the division. The poverty map utilised in this study63 includes highly granular data for urban centres, to ensure that urban poverty (e.g., informal settlements in and around Dhaka) is captured. Figs. 2 and 4 demonstrate that populations with lower wealth levels in both rural and urban areas, as well as wealthier populations in large urban centres such as Dhaka and Chittagong, are exposed to hydrogeomorphic instability.

Fig. 4: Wealth distribution of population exposed to hydrogeomorphic hazards.
Fig. 4: Wealth distribution of population exposed to hydrogeomorphic hazards.
Full size image

The wealth index of populations exposed to hydrogeomorphic hazards is highlighted with greater colour intensity (national wealth index more transparent in the background). Red colours represent lower levels of wealth. The bar charts illustrate the proportional percentage of the population within hydrogeomorphic hotspots for each administrative division for the year 2020. WI = Wealth Index.

Future population exposure to hydrogeomorphic hazards

In order to explore how future exposure to hydrogeomorphic hazards may unfold, the 2UP population growth estimates for Shared Socio-economic Pathway 2 (SSP2)64 are combined with the hydrogeomorphic hotspots map. Unfortunately, dynamical modelling of future hydrogeomorphic change has not been undertaken, which results in the assumption here that future hydrogeomorphic changes will continue to take place where there are current instabilities. Although this assumption introduces uncertainties, exploring future population exposure within these hotspots can help to identify areas that may experience increased hazard exposure due to population growth or migration.

By taking the population projections from the 2UP model, the findings show that future exposure to hydrogeomorphic hazards may increase by up to 29% by 2050 as a result of population growth (which is lower than the 38% increase in population growth expected in Bangladesh by 2050, according to the 2UP model), of which 14% is likely to occur in urban areas and 86% is expected to increase in rural areas. This finding emphasises that the currently disproportionate exposure to hydrogeomorphic hazards experienced by rural populations with lower levels of wealth is likely to continue to grow over the coming decades.

Discussion

The hydrogeomorphic hotspot modelling results in this study have shown that 20% of Bangladesh’s land area is prone to continuous and persistent hydrogeomorphic changes. The erosional processes along riverbanks and coastlines can degrade or completely consume valuable floodplain and coastal lands, leading to losses of livelihoods, reduced agricultural productivity and income, food insecurity, unemployment, and ultimately social unrest and internal displacement19,20,23,38,65. Oftentimes, in response to these issues, rural resource-dependent populations are forced to sell their productive assets, reduce their food consumption, stop children’s education, and take informal loans, which all deepen their vulnerability to these hazards9,66,67. Unlike other hazards, riverbank and coastal erosion often result in temporary or permanent migration27,29. Given the dynamic river systems in Bangladesh, land is constantly lost and created. Thus, riverbank erosion differs from coastal erosion in that land may reappear after seasonal submergence or may be created in the vicinity of where it was eroded27,68,69. Temporary displacement is therefore more common within and along riverbanks, whilst coastal erosion leads predominantly to permanent relocation70.

By providing a national-scale spatial depiction of where hydrogeomorphic hazards intersect with densely populated populations with low levels of wealth, this study identifies the key areas in Bangladesh that are likely experiencing, or have experienced, hardships in the face of erosion hazards. These areas are predominantly along the upstream reaches of the Meghna River, as well as north of Sirajganj along the Jamuna River, along the Teesta and Darla rivers in the Rangpur Division, and in some coastal areas west of the river mouth. It is in these areas where targeted action to support populations affected by erosion hazards is needed in order to diminish or prevent potential livelihood disruptions, socio-economic setbacks, and possible poverty traps.

Extended waterlogging, such as evident in both south-west and north-eastern Bangladesh, also results in the loss of valuable floodplain land. This waterlogging is due to human modifications to the landscape that have maximised agricultural and aquacultural outputs71,72. Poorly maintained infrastructure systems have resulted in drainage congestion, subsidence of embanked lands, saline intrusion in the south-west of Bangladesh, and environmental degradation48,71,73. This extensive area of permanent waterlogging has rendered land unproductive, affecting millions of resource-dependent livelihoods each year, undermining poverty alleviation efforts, and exacerbating vulnerabilities to other hazards4.

Contrastingly, land accretion may provide multiple benefits associated with newly created fertile land. Given the widespread pressures for land to accommodate the continuously growing population of Bangladesh, the ability for fluvial sediment to accumulate and stabilise to form new land is of great importance and value to the country. In fact, the Bangladesh Delta Plan 2100 proposes to construct 11 cross-dams in the Meghna estuary area to encourage and accelerate land reclamation in the river mouth, as well as extract sediment from the river systems to raise the elevation of floodplain lands32. The high-resolution spatial information provided in this study on sediment deposition trends supports the Government of Bangladesh in identifying where these natural sedimentation processes can be tapped into to enhance land generation.

In addition, this study also sheds light on the areas where recently accreted land coincides with populations with lower levels of wealth. Newly accreted land takes some years to stabilise. However, due to widespread erosion and flood hazards within river corridors, newly accreted land is often inhabited by settlers from newly eroded land when it is not yet stable9,37,39,68,74,75. Here, people-centric development, including formalised land tenure, is required to support the stabilisation of livelihoods and enhance the resilience of communities at the front lines of riverine hazards20,27,39.

These existing challenges are likely to intensify in the future in the face of both population growth and climate change. Although not assessed in detail in this study, climate change is likely to intensify hydrogeomorphic processes, particularly fluvial dynamics, potentially expanding the extent and modifying the characteristics of identified hotspots and shifting the size and location of populations. This could lead to the emergence of new areas prone to hydrogeomorphic hazards or increased susceptibility in existing ones. The mobility of communities residing in hydrogeomorphically unstable regions may become increasingly constrained, as surrounding areas might become unsuitable for relocation due to climate-related factors like sea-level rise or heightened flood risk. Looking ahead, future work should investigate the influence of climate change on population exposure to hydrogeomorphic hazards in the intermediate (2050) and long-term (2080–2100), through, for example, predictive coupled human and hydrogeomorphic dynamical systems models. This would provide invaluable insights into the complex interplay between climate change, socio-economic vulnerabilities, and the spatial distribution of these hidden hazards.

A focus on the long-term evolution of hydrogeomorphic hazards and exposure would support the Government of Bangladesh in its ambitious plan to eradicate most poverty by 2033, and completely by 2050, in line with the United Nations Sustainable Development Goals (SDGs)24,32. The Government also plans for the country to be resilient and prosperous in the face of climate change, with substantial mobilisation and allocation of resources towards disaster risk reduction32,76. Previous studies have argued that disaster risk management and poverty alleviation go hand in hand7,76. Here, we demonstrate that the joint efforts in disaster risk and poverty reduction ought to also consider and incorporate hydrogeomorphic hazards.

To conclude, this study has found that over 22 million people are currently exposed to hydrogeomorphic hazards in Bangladesh, of which 86% (18 million people) have low levels of wealth and 40% (8.5 million people) are within the lowest wealth population group. The statistically significant disparity suggests that populations with lower levels of wealth are disproportionately exposed to hydrogeomorphic hazards. Bangladesh’s increasing pressures of continuous and rapid population growth, combined with a shortage of available land, and the growing frequency and magnitude of climatic shocks, will likely result in more people being exposed to hydrogeomorphic hazards in the future, with low-income and rural populations expected to be more severely affected. This calls for disaster risk reduction and poverty eradication efforts to be addressed in unison77, with the incorporation of hydrogeomorphic hazards, in order to set deltaic nations like Bangladesh towards a path of sustainable growth.

Methods

The concept of hazard exposure refers to the presence of people, livelihoods, species or ecosystems, environmental functions, infrastructure, or assets in places and settings that could be adversely affected by hazards68,74,78,79. This study examines whether there is an exposure bias of populations with lower levels of wealth towards hydrogeomorphic instabilities in Bangladesh by intersecting high-resolution hydrogeomorphic modelling with population and poverty data. An exposure bias is measured as the number of exposed populations to hydrogeomorphic hazards with lower wealth levels compared to all people exposed across the country7,80. Previous studies have shown that populations with lower wealth levels are subject to exposure bias from flood hazards7, and cyclones1, but the extent of exposure bias to hydrogeomorphic hazards has not been quantified.

The methodology is broken down into four main steps, as outlined in the flowchart below (Fig. 5), with quantitative statistical analyses undertaken in each step, including: (i) generating a national hydrogeomorphic hotspot map for Bangladesh; (ii) assessing population exposed to hydrogeomorphic changes nationally and regionally; (iii) assessing how hydrogeomorphic changes may impact different wealth groups; and (iv) exploring how population exposure to hydrogeomorphic instabilities may change in the future. Each of these steps is discussed in more detail in the following sections.

Fig. 5: Methodological flowchart to map disparities in exposure to hydrogeomorphic hazards.
Fig. 5: Methodological flowchart to map disparities in exposure to hydrogeomorphic hazards.
Full size image

Dark grey boxes indicate the aim of the step, light grey boxes indicate data inputs, blue boxes represent methodological processes, and maps illustrate the key outputs. CRV = channelised response variance.

Hydrogeomorphic hotspot analysis

Until recently, there had not been an automated approach of mapping hydrogeomorphic changes at a large scale. However, Jarriel et al.46 developed the DeepWaterMap model, which is entirely based on remotely sensed imagery and automatically distinguishes water from land, clouds, and shadows within each satellite image, producing an almost binary representation of channel presence46. Their model builds on previous work by Isikdogan et al.81,82 and Passalacqua83, who developed the first automatic extraction of channel networks from satellite imagery. This approach eliminates the laborious process of manual inspections and delineation, and enables monitoring of near-live changes in complex channel networks at large spatial scales81,82.

Here, we apply the openly available DeepWaterMap model to Bangladesh from 1987 to 2022. Nation-wide satellite imagery is extracted from Landsat 5 TM, Landsat 7 ETM +, and Landsat 8 OLI using Google Earth Engine, at a resolution of 1 arc second (approximately 30 m x 30 m at the equator). Images from the dry season (October until March) are chosen, as there is less cloud cover and channels are at their lowest stage46. For each of the 35 years assessed, one composite image is created by taking the median value of all cloud-free pixels46,81. This approach of averaging the input satellite data across the dry season constrains the effects of changes in channel levels, avoids misinterpreting channel changes due to flooding, and dampens the unavoidable fluctuations in tidal levels46.

Once 35 new images are created – one composite image per dry season per year – they are incorporated into the DeepWaterMap model. The model uses a convolutional neural network approach to produce maps of water presence, where, after normalisation, land surface pixels are given a value of zero, channels are given a value of 255, and non-channelised features (flooded polders or shrimp ponds) are given values in between46. The resulting DeepWaterMap channel system is then used to map the Channelised Response Variance (CRV), a metric developed to track changes in channel morphodynamics over space and time46. High positive CRV values illustrate hotspots of increasing water presence (river or coastal erosion or longer-term wetting or inundation of land) over time, whilst high negative CRV values represent key areas that are decreasing in water presence (river or coastal deposition or drying up of land). The two extremes on the CRV spectrum, therefore, provide an indication of where the most frequent and severe hydrogeomorphic changes are observed. The lowest 5% of values from the CRV assessment have been removed to minimise potential errors introduced from image classification, but all other pixels were considered as experiencing hydrogeomorphic change. In this study, we also explore the 25, 10 and 5% most extreme areas of hydrogeomorphic instability in Bangladesh.

The raster images of tracked hydrogeomorphic changes (CRV) over the 35-year period can then be further analysed in a Geographic Information System (GIS). Here, GIS is used to assess and visualise the outputs of the CRV results and calculate the overall hydrogeomorphically unstable land area, as well as for areas undergoing specific hydrogeomorphic changes.

Population exposure to hydrogeomorphic hazards

WorldPop84 data for Bangladesh is used to map the population exposed to hydrogeomorphic hazards. This population data is simulated using a semi-automated dasymetric modelling approach that combines census data with remotely-sensed and secondary data within a random forest estimation technique85. This open-access dataset provides annual spatial population data from 2000 until 2020, at a resolution of 3 arc seconds (approximately 100 m x 100 m), available at the country level. Using GIS, this spatial population layer for each year is intersected with the hydrogeomorphic hotspot map from the above CRV model to assess the proportion of the Bangladeshi population living within these hydrogeomorphically unstable areas. Where hydrogeomorphic changes cut partially across a population grid (due to differences in spatial resolution), the proportion of the population grid impacted is calculated, assuming uniform population density across the population grid cell. The analyses are then repeated for each of the eight administrative divisions (Rangpur, Rajshahi, Mymensingh, Sylhet, Dhaka, Khulna, Barisal, and Chittagong) to understand the regional trends in changing population exposure from 2000 until 2020.

Exposure bias

The exposure bias is estimated by comparing the wealth levels of exposed populations to hydrogeomorphic changes to wealth levels across all of Bangladesh, using high-resolution wealth distribution maps developed by Steele et al.63. These maps were created through the combination of remote sensing, mobile phone operator call detail records, and traditional survey-based data from Bangladesh to provide highly granular maps of wealth for commonly used indicators of living standards, such as the Demographic and Health Surveys (DHS) Wealth Index (WI)4,63. Their study selects suitable predictors for modelling wealth by transforming and preprocessing mobile phone data and remote sensing covariate data. They use statistical modelling techniques, considering different combinations of covariates, employ non-spatial generalised linear models, and apply hierarchical Bayesian Geostatistical Models to predict wealth metrics at unsampled locations. These models account for the spatial autocorrelation and incorporate a spatially varying random effect, based on Voronoi polygons63. Model performance is evaluated using out-of-sample validation statistics, and prediction maps with associated uncertainties are generated. It should be noted that this is a static dataset of wealth – a concept that is highly dynamic – with a basis from census data captured in 2011, and surveys undertaken in 201463. Nevertheless, this dataset still currently provides the most granular and detailed information on wealth distribution across Bangladesh.

In this study, the DHS WI is extracted, as it measures household welfare, which is based on asset ownership (e.g., refrigerator, radio, or bicycle), dwelling characteristics, and access to basic services, such as clean water, electricity, healthcare, and education4,63. This type of information more robustly captures the multidimensional nature of wealth, compared to income-based data23,86, and better captures the wider welfare impacts of hydrogeomorphic hazards, such as asset losses, a disruption in educational and health services, and reduced consumption3,21,87,88. The WI values can be positive, negative, or zero (ranging between − 1.165 and 2.185 for Bangladesh), with greater positive values indicating higher socio-economic status4,63. The WI is based on relative wealth and takes many factors into account that go beyond income-based indicators; therefore, it cannot be mapped onto traditional definitions of poverty thresholds (e.g., less than US$1.90 per day in purchasing power parity). Consequently, this study uses higher and lower wealth population groups, defined by the WI, to assess disparities in exposure to hydrogeomorphic hazards.

By intersecting all three layers (i.e., hydrogeomorphic hotspots, population, and wealth) in GIS, the number of people living within hydrogeomorphic hotspots, and their levels of wealth, can be estimated. To fill in data gaps and areas without direct WI measurements, interpolation techniques were employed to estimate levels of wealth between the WI classifications. The interpolated wealth levels were then normalised by adjusting them to the highest-level frequency observed within the corresponding WI population group. This normalisation ensures that the estimated wealth distribution better reflects the wealth of the population across different wealth index levels, with the group having the highest frequency serving as the reference. The resulting wealth distribution of people exposed to hydrogeomorphic hazards is then compared to the wealth distribution of the country to estimate whether there is an exposure bias of populations with lower levels of wealth towards hydrogeomorphic hazards. In order to assess whether the exposure bias is statistically significant (p-value < 0.05), the non-parametric Mann-Whitney U statistical test is performed, as the underlying distributions are non-normal.

Future population exposure

Population projection models can be used to explore how the exposure to hydrogeomorphic hazards may change in the future. In this study, future population projections are taken from Van Huijstee et al. 64, who developed the 2UP model to simulate future population distribution on a global scale from 2010 until 2080, also distinguishing between rural and urban population changes. These spatially-explicit population projections have a relatively high resolution of 1 km near the equator, providing detailed future population scenarios for all SSPs64. For the purpose of this study, SSP2 (Middle of the Road) is used, reflecting the middle-ground socio-economic development trajectory for Bangladesh. Using this scenario, Bangladesh’s population is projected to continue to grow to just under 200 million people by 2050 and then gradually reduce thereafter64,89, with the greatest population growth predicted to occur in urban centres.

Future population exposure is assessed following the same approach as current population exposure. The assessment is undertaken at the country-wide scale, as well as within the identified hydrogeomorphic hotspots, for the short-term (2030), medium-term (2050) and long-term (2080), aligning with standard planning horizons in development frameworks and demographic projection intervals. Future trends in hydrogeomorphic change have not yet been projected; hence, this exploration of future exposure assumes that hydrogeomorphic hotspots remain the same as observed over the past 35 years. While in the future we can expect the precise location of hydrogeomorphic changes to evolve, the general hotspots of hydrogeomorphic change are likely to remain active over the coming decades25. Although this assumption introduces uncertainties, exploring future population exposure within these hotspots can help identify areas that may experience increased hazard exposure due to population growth or migration.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.