Over the last 20 years, there has been growing interest in the relationship between climate change and migration. Findings from this research show that shifts in precipitation and temperature are linked to migration decisions, although the patterns they identify vary and depend on the method, theory and geographic area of interest applied (Hunter et al., 2013). Migration itself has also changed, with the motives for leaving becoming more complex and the numbers of forced migrants, including refugees and internally displaced persons, growing (Donato and Ferris, 2020; Sharpe, 2018). Thus, international migration has become increasingly associated with evading threats related to war, conflict, natural disasters and environmental degradation (Donato and Massey, 2016; Betts, 2010).

Studies have suggested that agricultural productivity is a primary pathway through which climate change influences the likelihood of migration (Cai et al., 2016; Cattaneo and Peri, 2016). A recent macroeconomic study of 108 countries covering the period from 1960 to 2010 reported that the effects of weather shocks on migration occur mostly through changes to agricultural productivity (Falco et al., 2019). Such disruptions may directly affect households that cultivate crops, as well as agricultural laborers and those with businesses tied to the agricultural sector. Examining the effects of these disruptions on agriculture is central to understanding the migration-climate change relationship in low-lying rural Bangladesh, where, despite a shift away from agricultural employment, more than half of rural households remain engaged in agriculture (Lobell et al., 2011; Sen et al., 2021).

Building on prior studies, this paper examines whether and how tidal and river channel erosion (or land loss) intensifies or reduces the risks of making internal and international migrant trips. Bangladesh’s location in the floodplain of the Ganges-Brahmaputra-Meghna river delta makes it prone to seasonal flooding, channel erosion and accretion as water and sediment move across the landscape (Rahman et al., 2020). Riverbank erosion is one of the most costly and disruptive environmental stressors Bangladeshi residents face each year (ICCHL, 2016). Sediment starvation from large infrastructure projects (e.g., dams, water diversions, embankments) is modifying natural sediment transport processes, intensifying erosion in some locations. Climate change further compounds these impacts through sea level rise and changes in rainfall, flooding and cyclone activity (Islam et al., 2021; Yang et al., 2017; Dewan et al., 2017; Moors et al., 2011). Although it is well-established that erosion often temporarily displaces families who lose their homes (Barua et al., 2019; Hutton and Haque, 2003), we know relatively little about the effects of erosion on individuals’ decisions to migrate from their origin communities. Using data collected from 3,600 households in 18 communities in southwest Bangladesh, we investigate whether and how erosion is associated with the likelihood of making a first domestic or international migrant trip. Using an objective measure of erosion derived from Landsat satellite images and controlling for river and tidal flooding, we model the initial internal or international migrant trip as a process that unfolds over time, and consider whether and how the effects of erosion on household heads vary by agricultural landownership and livelihood.

It is widely known that Bangladesh experiences many natural disasters, including flooding, cyclones and riverbank erosion (Karim, 1995). Its flat topography, low-lying elevation, high population density and weak infrastructure combine to increase the exposure and the vulnerability of coastal Bangladesh to natural hazards. ¹ Seasonal flooding along with erosion and land deposition are natural processes in the Ganges-Brahmaputra-Meghna delta that have long shaped local practices and livelihoods in the region. To protect and expand agricultural production, embankments were constructed in many coastal communities from the 1960s through the 1990s. Originally designed to provide flood control, the embankments (also known as polders) have weakened in recent decades because of lack of maintenance. These polders have altered the natural fluid dynamics and sedimentation processes, intensifying erosion in some areas (Brammer, 2014; Auerbach et al., 2015; Valentine et al., 2021; Valentine and Wilson, 2023). Evidence suggests that upstream damming also impacts sediment processes, exacerbating land loss downstream (Syvitski and Milliman, 2007). Climate change compounds these risks through changes in the amount and the intensity of rainfall, as well as in the water levels and flows in river and tidal channels (Islam et al., 2021; Yang et al., 2017; Dewan et al., 2017). Land lost to erosion frequently causes loss of and damage to homes, road infrastructure, agricultural land, buildings and other community infrastructure (e.g., irrigation works).

These natural and human forces combine to create to a dynamic and complex physical landscape in Bangladesh (Arefin et al., 2021; Edmunds et al., 2015). Using GIS and remote sensing imagery, Arefin et al. (2021) assessed changes in land use/land cover along Bangladesh’s Padma River between 1955 and 2016. They found that, during this period, rates of riverbank erosion and vegetation loss related to settlement expansion increased, as did oscillations in the ratio of cultivated land to sand bars and water. Moreover, in recent decades, salinity intrusion has increased along the coast as higher water levels expand tidal inundation and groundwater reservoirs become more saline (Edmunds et al., 2015). Higher salinity in the soil and the water affects agricultural yields (Islam et al., 2021), raising food prices and poverty risks, especially for those with agricultural livelihoods (Ahmed et al., 2009). In addition, surface flooding is associated with exposure to drinking water from wells with high arsenic concentrations (Connolly et al., 2022). ²

Specific environmental conditions are differentially associated with out-migration in Bangladesh. For example, Chen et al. (2017) reported that extreme flooding is negatively associated with out-migration. In addition to identifying effects of weather, Chen and Mueller (2018) found that gradual increases in soil salinity reduce agricultural income, which is associated with more internal migration. Carrico and Donato (2019) reported that dry spells are most consistently associated with increased migration of household heads with agricultural livelihoods, whereas heavy rainfall is less robustly related to out-migration. ³ However, warm spells were also found to be strongly associated with internal migration for those employed in agricultural livelihoods – signaling that some environmental conditions influence migration by disrupting agricultural livelihoods (Call et al., 2017; Nawrotzki et al., 2015).

Given that riverine Bangladesh is prone to tidal channel and riverbank erosion, it is not surprising that the country, along with India, has been the focus of many studies analyzing the relationship between human migration and erosion (Zaman, 1989; Das et al., 2014, 2017; Mili et al., 2013; Majumdar et al., 2023). In an early publication, Zaman (1989) described patterns of dislocation and resettlement related to riverbank erosion and accretion by Bangladeshi residents living in the Brahmaputra-Jamuna floodplain. Zaber et al.’s (2018) ethnographic study of four districts of Bangladesh described how riverbank erosion leads to forced migration and social displacement. Analyzing data collected from 420 internally displaced persons residing in two sub-districts in southwest Bangladesh, Chowdhury et al. (2020) found that approximately 40 percent of respondents reported riverbank erosion as the most significant reason for their displacement. Islam et al. (2020) looked at internal migration from the coast to the upland areas in the Chittagong Hill Tracts, and showed how erosion and climate events are associated with decisions to migrate.

Relying on survey data from 4,456 residents in 1,188 households in rural embanked communities in five districts in coastal Bangladesh, Bernzen et al. (2019) examined whether and how erosion and other measures of environmental stress are related to out-migration. They found that erosion, measured by self-reports of arable land, was positively associated with internal migration but not with international migration, controlling for other relevant factors. In addition, human capital variables, such as not being employed in core agricultural/aquaculture industries, were both shown to be positively associated with migration and more important covariates than erosion and environmental stressors. Ali et al. (2021) analyzed a study area in Sirajganj Sadar, an upazila in North Bengal prone to erosion, and reported that after 10 years, more than 20 percent of the area’s population had migrated (mostly internally) as a result of river bank erosion. Islam et al. (2021), who collected survey and focus group data in a small administrative unit in the district of Bhola, concluded that river bank erosion leads to successive and gradual losses among the affected population and serves as a push factor for migration. Abdullah Kaiser (2023) also described how riverbank erosion negatively affects internally displaced persons in Lakshmipur, where they face physical and psychological risks due to displacement.

Therefore, prior studies offer empirical evidence that riverbank erosion is associated with human suffering, material loss, displacement and migration. Many of these studies are based on small samples that focus on one or two areas of the country, illustrating how the relationship between land loss and migration is complex, difficult to measure and not well understood. Yet, this is not surprising given that the country’s physical landscape is highly dynamic and not uniform (Brammer, 2014). Across space, land loss occurs at different rates and human interventions contribute to these processes.

In this paper, we integrate data from remotely sensed imagery and household migration histories between 1998 and 2017 to assess the extent to which land loss and migration occur over time at a fine spatial scale. Thus, our analysis differs from prior studies in three important ways. First, we measure erosion using objective indicators derived from satellite data, rather than relying on subjective measures about perceptions of erosion obtained from survey respondents. Second, we capture erosion as a process that unfolds over time, rather than examining the effects of self-reported arable land loss on migration at a single point in time. To do this, we estimate net land loss on an annual time-step and integrate the data with household migration histories from 1998 to 2017. Third, in contrast to Bernzen et al. (2019), we use a household sample that includes randomly selected households in coastal and non-coastal communities, which may reveal more variation than has been observed to date.

For this analysis, we use data collected as part of the 2019 Bangladesh Environment and Migration Survey (BEMS) in the country’s southwest region. Building on a pilot study of 1,695 households in 2014 (Donato et al., 2016; Carrico and Donato 2019), the 2019 BEMS collected detailed migration history data, including information about the first, last and second-to-last trips to internal and international destinations of household heads, their spouses and other adults. We use these data to estimate discrete-time event history models that assess relationships between livelihood and agricultural land in a given person-year and the risk of making a first internal or international migration trip in the subsequent year. We separately model initial internal and international migration trips, following King and Skeldon (2010).

The BEMS was collected in 20 mouzas, which are small administrative areas that usually contain 1–3 villages that vary in size from a few hundred to thousands of households. ⁴ In each mouza, we conducted a census of all households and then randomly selected a sample of 200 households. The 2019 BEMS data include a total of 4,000 households in communities that broadly cover the southwest region. From household heads and their spouses, we obtained information about the current and prior migrant trips they and other adults living in the household had taken. Thus, the BEMS data include information on prior and current migrants and non-migrants, even if the household members had not yet returned to the surveyed communities. ⁵ Another strength of these data is that, consistent with findings from the 2014 pilot and other surveys (see, for example, the Bangladesh Multiple Indicator Cluster Survey 2012–13, 2015; NIPORT, 2017), the BEMS had a very high completion rate (95%). Most of the non-complete households were unavailable during the time of data collection, and these households were replaced with a randomly selected alternate to meet the targeted sample size. ⁶

The analytic sample focuses on household heads because they play an especially important role. Compared to other adult household members, heads are more likely to migrate; to send money to their origin household; and to remit larger amounts (Tonassi and Donato, 2023). Other adult household members are not included in the analysis because we lack data about the extent to which they had engaged with or supported the origin household. ⁷

Finally, the analysis below relies on data from 18 of the 20 research sites that are in close proximity to a river or tidal channel, ⁸ and it is restricted to the years 1998 to 2017, for which we have consistent satellite imagery to construct measures of erosion. The final analytical sample includes 40,939 person-years generated from 2,717 respondents in 18 of the 20 research sites.

We begin by describing the variation among household heads in the probability of taking a first internal or international migration trip. We then present findings from the discrete-time event history models to assess whether and how tidal and river channel erosion are associated with the risk of taking a first internal or international migration trip in the subsequent year.

Table 2 describes the spatial variation in the household heads’ internal and international migration experiences. Across the entire sample, approximately 42.1 percent of household heads reported having made a first domestic trip, and 9.6 percent reported having taken a first international trip. However, substantial variation exists in patterns of first trip migration across the research sites. With respect to domestic migration, the share of household heads who reported having made a first trip range from 2.5 percent in community 220 to 63.2 percent in community 214. Although the variation is smaller for initial international migration trips, it is still sizeable, ranging from approximately zero in community 216 to 27.9 percent in community 212.

In this paper, we have combined household data collected in 2019 with data from remote sensing imagery to examine how erosion is associated with the likelihood making a first domestic or international migrant trip among Bangladeshi household heads. Findings from our models suggest that erosion is tied to internal out-migration, but the effect varies by livelihood and landownership. As expected, we observe that the effect of erosion on the likelihood of making a first domestic trip is strong and positive for non-agricultural non-landowners: high levels of erosion mean higher odds of making a first domestic trip. In contrast, with respect to the likelihood of making a first international trip, high levels of erosion are associated with lower odds for non-agricultural landowners, and, somewhat surprisingly, with higher odds for agricultural landowners.

Our finding that for agricultural landowners in high erosion areas, the chances of making a first international trip are higher than the chances of making a first domestic trip is challenging to explain. Table 1 shows that landowners whose livelihoods depend on agriculture are less wealthy than landowners whose livelihoods do not depend on agriculture. Our models nonetheless reveal that, controlling for wealth differences, erosion is related to lower odds of making an international move for non-agricultural landowners and to higher odds for agricultural landowners. This suggests that the role of landownership is complicated. Thus, landowning heads without an agricultural livelihood may adapt by selling at least some of their land, or they may be supporting their land through migrant remittances sent from other household members – which are strategies that landowners with an agricultural livelihood may be unable to employ. This is a research question we will address in the future using BEMS data on remittances received in the last 12 months.

Similar to most studies, this study has limitations. One is that the BEMS data do not include entire households that have migrated from the sampled research sites and have not returned. Without these households in the analytic sample, we may underestimate migration and are unable to capture covariates leading to permanent out-migration (Hamilton and Savinar, 2015). In addition, the BEMS data may be selective in that in some households, certain individuals, such as the elder brother, may have migrated and not returned, leading the younger brother to assume the role of household head at the time of the interview. There are likely differences between migrants who have returned and migrants have not, especially in terms of livelihood characteristics and the effects of erosion. For example, for migrants who have returned, the effects of erosion may be more robust in facilitating additional migrant trips. We therefore urge readers to keep in mind that the findings reported here do not capture these types of household selectivity. However, our future work will do so. With new National Science Foundation funding, we recently returned to a subset of households to collect panel data that will enable us to investigate the extent to which entire households have left BEMS research sites since 2019 and the onset of the COVID-19 pandemic.

For some readers, another limitation of this study may be that the analysis does not permit a deeper consideration of the social underpinnings of migration. Our focus on household heads and on agro-ecological factors does not provide a deep understanding of intra-household dynamics, such as strategies that involve sending a child or a sibling to migrate domestically or abroad. Nonetheless, the analysis captures the livelihood and migration activities of heads, who play very important roles in their households. Thus, while it does not fully interrogate the social foundations of Bangladeshi migration, the paper’s main purpose is to investigate whether and how erosion is related to the migration of household heads – a question that few studies have addressed up until now. In other work currently in development, we are using BEMS data to examine intra-household migration strategies. A final limitation of the analysis may be related to its emphasis on first, and not subsequent, migrant trips. Although preliminary evidence from robustness checks described in footnote 13 suggests that processes for first and later trips are similar, future research should explore how the processes for subsequent trips vary from those for first trips.

Despite these limitations, the findings presented here suggest that measuring a process such as erosion at a fine spatial scale offers new insights into how physical landscape changes are linked to migration. On the one hand, our findings are consistent with prior work that found a positive relationship between erosion and domestic (but not international) migration. On the other hand, the findings describe the different ways in which landownership and having an agricultural livelihood moderate the relationship between erosion and the risk of taking a first migrant trip. Thus, the analysis suggests that attachment to land moderates how erosion affects the risk of migration. That is, erosion increases the odds of making a domestic trip only among household heads who have a non-agricultural livelihood and do not own land. Among heads who own land but whose livelihood does not depend on agriculture, erosion lowers the likelihood of making an international trip; while among agricultural landowners whose livelihood depends on agriculture, erosion raises the likelihood.

Taken together, the results of the analysis point to a complex set of relationships. Although there are strong associations among shifting physical landscapes of migration and land attachment through agricultural livelihoods and landownership, the findings do not support causal interferences. Investigating the causal links among changing physical landscapes, differences in agricultural and land attachment and out-migration requires prospective panel data and enhanced research designs that can be used to disentangle the pathways among variables (Falco et al., 2019). However, it appears that landowners – whether agricultural or non-agricultural – are better able to finance migration. International migrants are diverse, and migration to some countries, such as those in the Middle East, may be lucrative enough to encourage people both to migrate more than once and to remain in Bangladesh after returning and thus not to migrate again. Therefore, it may be the case that erosion pushes households to send at least one household member to seek employment elsewhere and then eventually return. In addition, future studies must identify and disentangle the observed relationships reported in prior studies. For example, given our findings about the relationship between land erosion and migration, we call upon future studies to integrate migration history data with other physical landscape data, including data on the salinity of surface and ground water, as well as on the volumes of sediment and water discharge moving through the river channels.

The impacts of climate change on erosion have been documented for decades (Li and Fang, 2016). Among the direct impacts are shifts in the amounts and the intensity of rainfall. The indirect impacts include rising temperatures, which also affect erosion by changing vegetation cover. Alongside anthropogenic changes, such as upstream damming and the creation of polders, erosion influences people’s agricultural landholdings and livelihoods, which, as we show, moderate the relationship between erosion and migration. Given the scale and the impacts of erosion in southwest Bangladesh and the growing threat of climate change in the region, there is an urgent need for more research on how households are impacted; how they are adapting; and how their well-being can be supported despite these risks.

Supplementary file 1. Tables S.1 and S.2.