1 Introduction

In recent years, human displacement - arising from persecution, conflict, violence, human rights violations, climate change and natural disasters - has become a major development challenge, alongside being a humanitarian concern. The number of forcibly displaced people worldwide, including refugees, asylum-seekers, internally displaced individuals and others in need of international protection, reached an estimated 108.4 million by the end of 2022, an increase of 19 million compared to the end of 2021 [1]. This escalating displacement crisis has thrown affected populations into an intricate web of risks, uncertainties, and difficulties, including securing livelihoods, providing sustenance for families, and ensuring children’s education. Adding to the complexity, approximately 76% of the world’s displaced people are hosted by low- and middle-income countries, placing immense strain on their ability to deliver basic services and infrastructure [1]. Recognising displacement as a pivotal issue of our times prompts a call to action, pushing towards formulating strategies that extend beyond immediate relief to address the broader socio-economic implications of displacement on individuals and communities.

The Greater Horn of Africa has experienced pronounced displacement patterns in the past years, both within and across national borders. Climate change is indicated as one of the major triggers of mobility and displacement in the area [2]. In a matter of few weeks in 2023, devastating flash floods have caused the displacement of more than 770,000 people across the region [3]. Following the worst drought in 40 years - marked by five consecutive failed rainy seasons - the region now faces the aftermath of the El Niño weather phenomenon, characterized by unusually heavy rains, thunderstorms, and extreme floods. These protracted catastrophic events have led to the decimation of livestock and crops, pushing Somalia, Ethiopia, and Kenya to the brink of famine, and resulting in the displacement of 2.1 million individuals in 2022 alone [4,5]. Notably, the interaction of climate change with socioeconomic vulnerabilities and complex political dynamics in the region shapes mobility and displacement patterns in intricate and difficult-to-predict ways [6]. Thus, understanding how displaced people report and articulate their own experiences can provide deeper insight into their specific needs, challenges, and aspirations.

Mobility responses to climate change, in fact, can be highly heterogeneous depending on context-specific dynamics and individual circumstances [7]. For some, moving is a strategic choice for income diversification, risk management, and adaptation to the challenges posed by climate change [8–10]. Others experience forced displacement as they seek refuge from destruction and loss of livelihoods and opportunities, often exacerbated by climate change [11–13]. Importantly, the capacity to anticipate and recover from these shocks is not just a result of individual exposure, but is closely related to broader institutional and socio-political contexts. These include exposure to protection risks, legal recognition, and access to support networks, alongside demographic and economic conditions [14,15].

This study builds upon these lines of evidence, aiming to shed light on the complex relationship existing between climate variability and self-reported displacement in the Greater Horn of Africa, a region significantly affected by these two critical issues as well as by the compounding risks associated with them. Specifically, this study combines spatial climate information with a unique mobility dataset, the IOM Displacement Tracking Matrix (DTM) Flow Monitoring Survey (FMS), which tracks more than 200,000 moving individuals in 85 Flow Monitoring Points (FMPs) across the Greater Horn of Africa in the 2018-2022 time period. To ensure spatial precision in linking mobility histories with climate exposures, we develop a semi-automated geocoding procedure that converts respondents’ reported places of departure into GPS coordinates, enabling a direct match between individual-level mobility data and high-resolution climatic indicators. Guided by this rich dataset, an advanced spatial matching procedure, and well-established conceptual frameworks from the climate-mobility literature, namely the migration threshold theories (Bardsley and Hugo, 2010; McLeman, 2018), the environmentally induced migration typology (Renaud et al., 2011), and the aspiration-capability framework (De Haas, 2021; Carling and Schewel, 2020), this study attempts to unpack how climate variability shapes self-reported displacement in the Greater Horn of Africa. In particular, we employ linear probability models, supported by a battery of robustness checks to investigate: i) how individual exposure to climate shocks affects the likelihood of individuals reporting being displaced; ii) what are the pathways that underlie the relationship between climate change and self-reported displacement; iii) how individual socio-demographic characteristics and typology of movement (within and across borders) interact with climate shocks to shape displacement as reported by respondents.

In doing so, this work contributes to the existing literature in three main ways. First, it stands out as one of the first studies focusing specifically on self-reported displacement, departing from the broader scope of migration dynamics that prevails in most research within this domain (see, for instance, [16,17]). To the best of our knowledge, no studies have specifically investigated this patterns as most existing studies focus on the broader impacts of climate change, environmental degradation and socioeconomic factors on migration (most notable examples are [12,18–21]). Among existing literature, a study by [22] is particularly relevant as it explores climate variability and self-identification as environmental migrants in Kenya and Vietnam. However, our study stands out by using a larger sample and a broader geographic scope. Furthermore, instead of focusing on migrants’ self-identification, we examine self-reported displacement and include socio-economic characteristics, area attributes, and small-scale conflict variables. These factors provide a valuable and comprehensive view of climate-related displacement in an understudied region. While [22] offers valuable insights, analysis builds upon this foundation by enriching the empirical understanding of climate-related mobility in the region.

Second, the majority of empirical analyses in the African context consist of either micro-level country studies that focus predominantly on internal migration decisions [10,23,24] or macro-level studies investigating international migration flows [25,26]. Our dataset allows us to use information of mobile people interviewed in the FMPs to explore both internal and international movements from an individual perspective. In fact, while these data are particularly suited for capturing cross-border movements due to the FMPs’ proximity to borders, they also provide insights into internal movements.

Finally, our study contributes to understanding potential pathways linking climate variability and displacement. Climate-induced mobility is rarely driven by environmental factors alone; how people interpret and respond to these changes is shaped by local conditions and adaptation strategies [27]. Our findings suggest that climatic stressors, combined with other risks like conflict, influence whether individuals self-identify as displaced. This perspective helps contextualize statistical patterns and offers a more grounded view of climate-related displacement as experienced by affected populations.

We acknowledge the potential limitations inherent in our analysis. It is important to note that our data might not fully encompass the entire spectrum of mobile and displaced individuals originating from the Greater Horn of Africa. Specifically, our sample includes only mobile individuals identified in the FMPs, excluding non-mobile individuals and others who may follow distinct routes. This potential lack of representativeness poses challenges to the external validity of our results, urging caution in extrapolating our findings to the broader context of the Greater Horn of Africa. In addition to this inherent bias, other factors such as measurement error and social desirability bias could introduce nuances that impact our results. For instance, individuals might perceive potential benefits, such as receiving humanitarian assistance or legal protection, in self-reporting as displaced. All these limits constrain the causal implications of this study. Despite this limitation, this study represents a crucial attempt to empirically examine the link between climate variability and individual reporting of displacement, with a focus on identifying existing gaps in knowledge. Furthermore, our thorough analysis of the demographic groups most affected by the climate-displacement nexus and the underlying mechanisms involved can provide valuable insights for the formulation and targeted implementation of policies related to climate adaptation, displacement issues, and conflict risks. This study makes a unique contribution to the climate-mobility literature by exploiting a large multi-country dataset on mobile populations in the Greater Horn of Africa and an advanced spatial matching procedure that links individual mobility histories to high-resolution climatic indicators. Together, these features allow for a detailed examination of self-reported displacement, of differences between internal and cross-border movements, and of the mechanisms through which climate variability shapes displacement outcomes.

The paper proceeds as follows. Sect 2 closely looks at the nuances of climate-related mobility. Sect 3 presents our data and variables used. Sect 4 explores displacement patterns in the Greater Horn of Africa at the country-level and individual-level from a descriptive perspective. Sect 5 presents our analysis and results, including main findings, analysis of the mechanisms, heterogeneity analysis and robustness checks. Finally, Sect 6 concludes.

2 Theoretical framework on climate mobility

Understanding climate-related displacement patterns requires first distinguishing between various forms of human mobility and clarifying their definitions. These distinctions are not merely semantic: they reflect different drivers, levels of agency, legal implications, and policy responsibilities. Human mobility is defined as an umbrella concept that includes all forms of human movement, whether voluntary or involuntary, internal or cross-border [28]. Recently, the term human mobility has increasingly been used to describe the wide spectrum of movement types occurring within the climate change context. Given our focus on individuals on the move, it is important to highlight that mobility responses to climate change vary widely depending on individual circumstances and contextual factors [7]. According with [28], the term human mobility is usually understood as also encompassing tourists that are generally considered as not engaging in migration. Migration, a somewhat narrower concept compared to mobility, is defined as “the movement of persons away from their place of usual residence” [28, p. 150]. Like mobility, migration covers a broad spectrum of forms, ranging from voluntary to forced, with many intermediate situations along that continuum. Voluntary migration typically involves strategic decisions made to diversify income, manage risk, or adapt to environmental stress [8–10]. In contrast, forced migration results from coercion, compulsion, or an inability to remain in place due to external pressures [28, p. 77]. Forced migration is used to describe a large variety of coerced movements, including forced labor, international refugees, and displaced persons [28].

Our work deals with forced displacement, defined by [28, p. 55] as the “movement of persons who have been forced or obliged to flee or to leave their homes or places of habitual residence, in particular as a result of or in order to avoid the effects of armed conflict, situations of generalized violence, violations of human rights or natural or human-made disasters”. Displacement can be considered a more specific term within the broader forced migration category, occurring when a shock or disaster undermines the possibility of remaining, leading to the loss of home or habitual residence. Importantly, the ability to avoid or recover from such shocks is mediated by broader structural conditions, such as access to land, markets, governance, and historical inequalities, which shape both exposure and response [14,15,29,30].

In our study, however, displacement is identified through a survey-based question that asks whether individuals consider themselves forcibly displaced. This yields a self-reported indicator of displacement, which does not necessarily align with official definitions but reflects respondents’ own understanding and experience of their mobility. While this approach may introduce variation due to differing individual interpretations of what constitutes forced displacement, it allows for a more nuanced capture of personal experiences, including situations that might not be officially recognized. It also makes it possible to detect less visible or informal displacement dynamics that often go unrecorded in formal datasets. However, the lack of standardization may limit comparability across settings and blur the lines between voluntary and forced mobility. Despite these challenges, we argue that self-reported measures provide valuable insights into how affected individuals frame their own displacement experiences, especially in contexts where institutional recognition or legal categorization is limited or absent. Although this measure can complement more formal classifications by adding granularity to lived realities and policy responses, it also calls for a conceptual lens that can account for heterogeneity in how people report and relate to displacement.

To allow a clear interpretation of the dynamics around self-reported displacement in relation to climate shocks and variability, we build on three climate-related conceptual frameworks to guide our hypothesis development and interpretation of results: (1) the migration threshold theories [31,32], (2) the environmentally induced migration typology by [33], and (3) the aspiration-capability framework [34,35]. Migration threshold theories help us conceptualize how different exposures to climate may lead people to report their mobility as displacement. According to these theories, mobility emerges as a forced response once environmental stress crosses certain thresholds, making local adaptation no longer viable [31,32]. In our context, migration threshold theories help hypothesize why some mobile individuals report their movement as displacement when faced with sudden-onset events that are more likely to overwhelm coping capacity and trigger displacement reporting, while slow-onset events may allow for more adaptive responses. This interpretation is supported by empirical findings. For instance, [36] and [22] show that people exposed to rapid-onset shocks are more likely to identify climate as a major driver of mobility, while those facing gradual change often describe their movement in economic or adaptive terms [29,30].

In line with these theoretical developments, [33] distinguish between environmentally forced migrants as those compelled to move due to severe environmental degradation and lack of viable alternatives and environmentally motivated migrants, who move preemptively or adaptively, often when financially or socially equipped to do so. Following their decision framework, this distinction is based primarily on two main factors, i) the occurrence of rapid onset hazards (e.g. floods) and the gradual loss of ecosystem services (e.g. recurring droughts) and ii) presence of rapid and effective social, economic and physical recovery of impacted areas [33]. Finally, the aspiration-capability framework, which conceptualizes mobility as a function of people’s aspirations (desires or intentions to move) and capabilities (the means and freedoms to do so), provides important insights into the variability of displacement outcomes [34,35]. Climate events shape both dimensions, as sudden and destructive events may erode the capability to remain, while intensifying the aspiration to move. This helps explaining variation in self-reported displacement across social groups, as capabilities are mediated by factors like wealth, gender, education, and institutional access [34].

Bringing these frameworks together strengthens the conceptual foundation of our study. This integrated conceptual-analytical structure, in fact, enables us to interpret self-reported displacement as an outcome shaped by exposure to climate stress, individual characteristics and contextual vulnerability, allowing for greater analytical nuance in how displacement is expressed among mobile populations.

3 Data and variables

This study uses secondary data collected by the International Organization for Migration (IOM) through a voluntary survey conducted with migrants in transit at designated Flow Monitoring Points in key transit areas. Participation was entirely voluntary, with all respondents providing written informed consent prior to participation. The data collection and processing adhered to IOM’s Data Protection Principles, which are informed by international standards and aim to respect the privacy, dignity, and well-being of migrants. These principles ensure that personal data are obtained lawfully and fairly, used only for specified and legitimate purposes, and kept secure to prevent unauthorized access or disclosure. The dataset was anonymized by IOM before being shared with the authors under a data-sharing agreement signed on November 17, 2020. The authors accessed the anonymized dataset for research purposes on August 7, 2023. At no point did the authors have access to personally identifiable information. Given that this study involved the analysis of anonymized secondary data, additional ethical approval was not required.

Building on these principles, this study utilizes the IOM’s Displacement Tracking Matrix Flow Monitoring Surveys (FMSs) to capture information on mobility [37]. The FMSs collect information about people on the move in the period 2018-2022 interviewed in different transit locations, the Flow Monitoring Points (FMPs). The majority of FMPs are located near countries’ borders, or in critical transit locations to efficiently capture inter- and intra-regional migration routes, flows, and trends. Although FMSs are implemented in many countries worldwide as part of IOM’s global DTM system, in this study we focus on the network of FMPs located within the Greater Horn of Africa. We selected this region because it is one of the areas most affected by recurrent climate shocks and climate-related displacement and because the FMS provides the most complete and temporally consistent coverage there over the period 2018-2022. All observations in our raw dataset are therefore collected at FMPs within this regional corridor, however, as a result, some respondents may originate from or travel to locations just beyond some definitions of the Greater Horn of Africa, but they were already moving through this corridor at the time of interview. While this setup enables consistent data collection across high-volume corridors, it inherently excludes individuals using informal, less-monitored, or off-route pathways. As a result, the sample may not fully capture the diversity of displacement experiences in the region and may over represent migrants who move along more visible or accessible routes. Questions in the FMS are asked in a standardized format, with most requiring either single-choice or multiple-choice responses, the questionnaire includes a wide range of information, including various demographic and socio-economic information, along with details about migrants’ journeys such as the date of departure, reasons for migrating, the village or city they departed from, and their intended destination.

Using a R based semi-automated geocoding process based on the Google Maps API, we exctracted precise GPS coordinates from cleaned village and city names, pre-processed through a reverse-engineering routine developed by IOM. S1 Fig shows the resulting geocoded locations of migrants’ reported places of departure, together with the existing FMPs in the region. This approach minimizes the risk of spatial misclassification in linking reported departure locations to external climate and conflict data. Climate data comes from TerraClimate, a monthly climatic dataset with a spatial resolution of about 4-km (1/24th degree), available from 1958 to 2022 [38]. Conflict data comes from the Armed Conflict Location and Event Data Project (ACLED), a disjointed dataset that reports real-time geo-coded information on location, fatalities, and actors involved in various conflict events around the globe [39]. Finally, we use the Copernicus Global Land Cover Layers dataset to capture the land cover typology of the location of departure [40]. This high-resolution data (100 ms) allows us to control whether the location of departure of the migrants can be classified as land covered with temporary crops followed by harvest and a bare soil period and thus be more exposed to climatic anomalies. After merging all sources, we exclude approximately 120,000 observations with missing or incomplete information on key variables, such as the place of departure, departure date, displacement status, or core socio-demographic characteristics. Our final dataset comprises 54,131 individuals interviewed in 85 FMPs and departed from Ethiopia, South Sudan, Uganda, the Democratic Republic of the Congo and the United Republic of Tanzania between 2018 and 2022.

Our outcome of interest is a self-reported displacement dummy, equal to 1 if an individual responds affirmatively to the question “Have you been forcibly displaced?”. The excluded observations include around 25,000 individuals that did not answer to this question. As the preferred climate indicator, we employ the Palmer Drought Severity Index (PDSI, henceforth), which is a widely utilized index in the climate literature [18,41,42], that systematically measures long-term water balance within a given location. Specifically, the PDSI considers various hydroclimatic factors, including precipitation, evaporation, temperature, and soil characteristics over a defined temporal interval to measure the deviation from the long-term average moisture conditions [43,44]. The benefit of employing an index lies in its ability to encompass the interplay between both measures, enabling the identification of exceptional events. To illustrate, the adverse effects of subpar rainfall on agricultural yields might be heightened by higher-than-average temperatures [45]. This is important also from a statistical point of view because temperature and precipitation tend to be correlated and it is preferable to control for both measures [46]. Positive values of PDSI indicate wetter-than-average conditions and negative values of PDSI denote drier-than-average conditions.

In our sample, there exists a strong and positive correlation between the PDSI and the self-reported environmental reason for migration (see S1 Table). This correlation provides conceptual support for using the PDSI as the key climatic indicator in our analysis. We utilise two primary definitions for our variable of interest: (i) a dummy variable equal to one when the PDSI is greater than zero to capture wetter than usual conditions; (ii) a set of four dummies denoting whether the village where the migrant lived encountered extreme dry (PDSI <–3), dry (–3< PDSI 0), wet (0< PDSI <3), or extreme wet conditions (PDSI >3), as opposed to the long-run normal conditions. The –3 and +3 thresholds represent more than severe dry and wet conditions, respectively. The choice of the PDSI thresholds to identify climatic conditions relies on previous literature that employed the index (see [47] among others). Following [48], we compute such indicators in the three months before the departure date. This is commonly considered a reasonable time frame to capture short-lived mechanisms occurring immediately after a climatic shock and before the departure [7].

We also perform a number of robustness checks to enhance the reliability of our findings, using a PDSI computed over alternative time periods, as well as different climate indicators, namely the Standardised Precipitation-Evapotranspiration Index (SPEI), the temperature anomalies and the rainfall anomalies. The SPEI is computed as the temporal difference between rainfall accumulation and potential evapotranspiration to assess changes in the climatic water balance. Temperature anomalies and rainfall anomalies capture how specific time and location maximum temperature and total rainfall deviate from the long-run condition and are computed respectively as:

where T_i,m,y and R_i,m,y is the monthly average climatic variable (temperature or rainfall), and is the long-term monthly mean, and and is the standard deviation. i is the village, m is the month, y is the year, while n comprises the number of months that identify the selected time period over which the variable is computed.

In order to take into account the impact of conflicts, we create a variable that represents the number of violent conflicts that occur within a buffer area of 50km radius around the location from where the migrants depart in the month of their departure. This geographical scale is consistent with previous research. For instance, a study conducted by [49] shows that infants born within 50km of an armed conflict have a higher risk of dying before their first birthday. To distinguish between regions heavily reliant on seasonal rainfall for crop yield we create a dummy equal to 1 whether the individual comes from a village where most of the land is cultivated for crops, as opposed to the other possible land cover types. Lastly, we employ a set of control variables to account for relevant characteristics of the individuals in the sample. Specifically, we employ a dummy equal to 1 if the individual is male, a dummy equal to 1 if the individual is young (less than 30 years old), a dummy equal to 1 if the individual is married, a categorical variable that details the level of education and a dummy variable equal to 1 if the individual is employed. We impute the missing values of the control variables by replacing them with the mode of the region. Most variables have less than 2% missing data, except for the level of education, which has around 16%. This method was selected because it is a standard approach for addressing missing data when missing values are not related to the key traits being studied. Importantly, the missing values appear to be randomly distributed due to enumerator errors rather than being systematically linked to the respondents’ characteristics, mitigating concerns about bias or measurement error. The results without imputing missing values are consistent with our main analysis and are available upon request.

4 Displacement in the Greater Horn of Africa

Our analysis starts with a descriptive examination of migration and self-reported displacement, considering both individual-level and country-level perspectives. Subsequently, we delve into the exploration of the influence of climate variability on the individual report of displacement. As previously discussed, individuals who self-report as displaced in our sample might not always be universally acknowledged as such by the formal definitions. However, to simplify the narrative in our results, we will commonly refer to individuals in our sample as “displaced” if they report as such, and as “non-displaced” if they do not.

4.1 Individual characteristics and displacement patterns

Summary statistics of our sample, as reported in S2 Table, indicate that 10,494 individuals, or 20 percent of the full mobility sample, self-report as displaced. As shown, there is substantial variation across countries in the share of individuals who self-report as displaced. For example, the proportion of displaced individuals is notably higher among respondents from South Sudan and Tanzania compared to those from Ethiopia or the DRC. These differences may reflect not only variations in displacement dynamics but also broader institutional and discursive contexts that shape how individuals interpret and report their movement. The majority of migrants are male (68% and 66% for displaced and non-displaced, respectively) and young (around 30 years for both categories). Migrants who consider themselves displaced are more likely to be married (45% vs. 35% of non-displaced), more educated (65% of non-displaced have no more than primary education vs. the 50% of displaced people with at least secondary education), and employed before leaving (35% vs. 26% of non-displaced). This suggests that, in our sample, individuals who report themselves as displaced often possess a more established socio-economic status before moving from their original location. Such circumstances imply a compelling need to migrate, driven, for instance, by security concerns such as climate-related shocks or violence. Consistent with this trend, recent evidence has highlighted the role of conflicts - alongside natural disasters - in the “brain drain” phenomenon, forcing highly skilled and qualified individuals to displace [50]. This argument gains further credibility when considering that a substantial proportion (50%) of displaced individuals are attempting to move for the first time (30% more than non-displaced). Also, the majority of the sample were interviewed within 2 weeks to 3 months after their departure date and relied on their savings or those of family and friends, showcasing self-reliance as a prevalent funding source.

Individuals’ travel experience is marked by a number of challenges and difficulties.Notably, not all of the 10,494 self-reported displaced people provided information about the difficulties of their journey. As a result, we report the responses of the over 6,000 people who responded to these questions. More than half of the respondents among people who self-report as displaced endure hunger and lack shelter during the journey, while almost 20% of them suffer a form of sickness. A lower, yet still critical, share of displaced individuals experience injuries and attacks and have been deported or arrested by the authorities. Additionally, it is common for displaced people to encounter financial challenges throughout their journey, lack information about their destination, and often lack of an identity document, making the process of travel even more challenging (see S2 Fig).

To further understand the distinction between people who report themselves as displaced and those who do not, in Table 1 we conduct a descriptive analysis on the reasons to leave, categorising factors into push and pull drivers. Looking at the push drivers, displaced individuals are significantly more likely to leave their origin areas for environmental reasons (11%) and conflict (38%), as opposed to non-displaced (2% and 4%, respectively). This emphasises that environmental and conflict-driven migration is more prevalent among those who self-report as displaced. However, more than half of both displaced and non-displaced individuals attribute their departure to economic reasons, with a significant difference between displaced and non-displaced (46% and 70% respectively). In this preliminary analysis, it remains challenging to discern if the economic reasons are intertwined with conflict or environmental factors; this will be further explored in the subsequent sections. Turning to the pull factors of mobility, i.e., security reasons, job opportunities or improved living conditions at destination, among others, the difference between displaced and non-displaced are less pronounced in terms of magnitude, though still statistically significant. This shows that the primary differences in the drivers of mobility between displaced and non-displaced people mainly pertain to the challenges, obstacles and lack of opportunities of the areas of origin rather than the appeal of favorable conditions in intended destinations.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Summary statistics of displaced and non-displaced for reasons to leave (2018–2022).

https://doi.org/10.1371/journal.pclm.0000594.t001

Lastly, we investigate origin-intended destination flows of displaced individuals (see Fig 1), and compare these patterns with those of non-displaced people (see S3 Fig). The majority of both displaced and non-displaced people on the move originated from Ethiopia, accounting for approximately 67% and 80% of the total sample, respectively. This country is not only the most populous of the region, but also a major origin and host country for IDPs, refugees, and asylum seekers [51]. Moreover, Ethiopia has been severely impacted by extreme weather events and violence, with implications for agricultural inputs, safety, assistance, and thus human mobility. This dual vulnerability, both demographic and environmental, positions the country at the nexus of complex migration and displacement patterns within the Greater Horn of Africa.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Circular mobility flows: Self-reported displacement across the Greater horn of Africa (2018–2022).

Notes: This figure presents the mobility flows of 10,494 self-reported displaced individuals across the Greater horn of Africa between 2018 and 2022. We follow the strategy and R codes developed by [56]. Origin and destination countries are indicated in the segments around the figure. Flow values are divided by 1000.

https://doi.org/10.1371/journal.pclm.0000594.g001

In 2020, for instance, a severe conflict erupted in the North of the country, specifically in Tigray and subsequently in Afar and Amhara, leading to the displacement of at least 2 million people [52]. Additionally, Ethiopia exhibits a strong migration culture, often turning to it as a strategic means to improve living conditions [53]. According to the [54], in 2021, excluding the Tigray region, approximately 16% of the population was estimated to be internal migrants, with 36% considered recent migrants. Furthermore, the rate of international migration has increased significantly from 4.9% in 2010/11 to 14.1% in 2018/19 [54,55]. The interplay of these environmental, political, and socio-economic factors contributes to shaping Ethiopia’s role as a complex hub of both forced displacement and voluntary migration, primarily pursuing better economic opportunities abroad.

South Sudan and Uganda also emerge in the sample as frequent origin countries for self-reported displaced individuals. This can be attributed to the large-scale violence in South Sudan since 2016 as well as the environmental degradation processes that have resulted, especially among pastoral communities, in outflows of people from certain regions of Uganda [52].

In terms of destinations, the majority of those who provided information on their intended destination intend to settle in countries other than their country of origin. More precisely, 78% of individuals who report being displaced intend to relocate abroad, mainly intra-regionally - to Uganda, South Sudan or Ethiopia - or to move along the Eastern Route towards the Middle East. Looking at S3 Fig, it is evident that the latter destination is also predominant among non-displaced people (39% of the sample). Overall, this trend underscores the region’s appeal as a dynamic hub, providing significant opportunities and employment prospects, particularly relevant for economic migrants.

Despite these patterns, unidentified locations seem to play a prominent role, particularly for displaced individuals (55% for displaced vs. 34% for non-displaced). The large percentage of individuals who do not specify any intended destination raises questions about whether their movement has occurred under time and information constraints. These dynamics would be in line with displacement movements, which, in fact, add an extra layer of uncertainty to individuals attempting to relocate, as it often unfolds suddenly and unexpectedly, compelling individuals to move primarily based on proximity and security without adequate planning [57]. Nonetheless, the fact that almost 50% did report an intended destination suggests that there may be variations in the preparation and voluntariness of movement. This variability in movement planning can be further observed by examining non-displaced individuals. Although unknown destinations are significantly less prevalent compared to displaced people (around 20% less), a large number of migrants do not report their intended final location. We acknowledge that this could be due to missing information, but it might also suggests that the overall captured mobility flows may be more spontaneous than planned, or individuals may have been compelled to alter their destination while in transit.

4.2 The role of climate variability

The descriptive analysis has shown a more prominent role of climatic factors as main reasons to move among self-reported displaced individuals as compared to non-displaced. In this section, we delve further into exploring the correlation between climate and self-reported displacement, laying the foundation for the development of our analysis.

We begin by examining descriptive evidence on whether displaced and non-displaced individuals were exposed to different climatic conditions prior to departure. Fig 2 reports kernel density estimates of the main climatic indicators used in our analysis, namely PDSI, SPEI, and precipitation and temperature anomalies, separately for self-reported displaced and non-displaced individuals. Visually, the curves for displaced individuals are shifted towards wetter-than-usual conditions for precipitation, PDSI and SPEI, and towards slightly cooler values for the temperature anomaly, relative to the curves for non-displaced individuals. These patterns suggest that, in the three months before departure, displaced people may have been exposed to wetter climatic conditions (higher precipitation, PDSI and SPEI) and lower temperatures compared with non-displaced individuals. A Kolmogorov-Smirnov (K-S) test, also reported in Fig 2, confirms that the cumulative distribution functions of these variables are statistically different. Additionally, the raw scores of these climate indicators are summarized in S3 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Density plot: Climate indicators by displaced and non-displaced individuals (2018–2022).

Notes: This Fig shows two density plots of disparities in climate indicators between self-reported displaced (straight line) and non-displaced individuals (dotted line). The sample includes individuals surveyed between 2018–2022. The result of the Kolmogorovov–Smirnov (K–S) test is reported on the top-left of each panel.

https://doi.org/10.1371/journal.pclm.0000594.g002

We then plot our main climatic variables, i.e., the PDSI, the SPEI, the precipitation and temperature anomalies, averaged at country level, against the percentage of people that report themselves as displaced coming from the main five origin countries of our sample, i.e. Ethiopia, South Sudan, Uganda, the Democratic Republic of the Congo and the United Republic of Tanzania. Ethiopia is presented as a red dot on the scatter plot for its importance in our sample (80% of the interviewees originate from there). The plots, shown in Fig 3, again suggest higher levels of self reported displacement in correspondence of wetter climatic conditions (higher precipitations, PDSI and SPEI) and lower temperatures. The reduced levels of self-reported displacement associated with drier conditions will be part of our investigation in the subsequent sections.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Scatter plot: Climate indicators and displacement in the horn of Africa (2018–2022).

Notes: This figure presents the scatter plot of country average rainfall anomalies (top left), PDSI (top right), SPEI (bottom left) and temperature anomalies (bottom right) against the percentage of self-reported displaced people from South Sudan, Uganda, the Democratic Republic of the Congo, the United Republic of Tanzania and Ethiopia (in red). The sample includes individuals surveyed between 2018-2022.

https://doi.org/10.1371/journal.pclm.0000594.g003

5 Self-reported displacement under climate variability

In what follows, we attempt to assess the impact of climate variability on self-reported displacement, more precisely, through a regression analysis. We implement a pooled data approach in which our main estimating equation uses a linear probability model (LPM, henceforth) that is equal to:

(1)

where y_it denotes our self-reported displacement dummy, equal to 1 if the respondent i self-report as displaced, and 0 otherwise, in each year t. C_i,t−1 is the individual exposure to a climate variability three months before moving, at the village level, in each year t, compared to the long-run mean. X_i is a set of conservative individual observable characteristics. The coefficient represents the impact of our climate indicator on reported displacement, taking into account other individual characteristics that influence the self-reported displacement. Equation 1 incorporates fixed effects for regions and years to account for unobservable characteristics at the regional and temporal levels that could affect various aspects of self-reported displacement. Standard errors are clustered at the regional level r.

We first analyze the relationship between our PDSI dummy and the probability of report being displaced. Note that a PDSI whose value is equal to 1, i.e. positive, here is defined as a water surplus relative to the long run mean in the village from where the respondent lived. Then, it could be that individual A experienced lower rainfall relative to the rainfall received by individual B. However, the PDSI is positive for individual A because the rainfall received by individual A is more than the long-term average received in the locality in which A resides. Similarly, it is negative for B because the rainfall they received was lower than usual. If the self-reported displacement is higher in individuals experiencing wetter climates as compared to individuals with drier climates, the coefficient on the dummy variable is positive. Subsequently, we showcase results for the categorical variable of PSDI that distinguishes extremely dry, dry, wet, and extremely wet conditions. We then explore the pathways underlying the nexus between climate and self-reported displacement, using sub-sample analysis and mediation analysis. Finally, we analyze how climate affects the likelihood of self-reported displacement across various socio-economic characteristics. To do so, we interact our climatic indicator C_i,t−1 with a specific characteristic, such as gender, age, education, and within or outside border movements, and compute the marginal effects for the sub-groups of interest to capture the related specificity.

It is important to note that this quantitative analysis cannot be interpreted as fully causal for two main reasons. First, the use of a pooled cross-sectional relationship depicts a long-run equilibrium, potentially overlooking slow processes that unfold gradually, making predictions challenging for the future impact of climate variability on self-reported displacement. Essentially, the historical equilibrium represented in the cross-sectional analysis may depend on mechanisms that no longer operate similarly. Second, studying the impact of climate on self-reported displacement within a sample limited to a mobile population rather than encompassing a representative mix of both mobile and non-mobile individuals raises concerns about selection bias. Focusing solely on mobile people may result in an incomplete understanding of the broader population dynamics affected by climate-induced displacement. This approach might overlook factors influencing the decision to move, leading to an overemphasis on characteristics specific to migrants, potentially skewing the assessment of climate-related effects. A representative sample, inclusive of both mobile and non-mobile individuals, would provide a more comprehensive basis for evaluating the nuanced relationship between climate conditions and displacement. Finally, while we acknowledge the limitations associated with the use of LPMs, most notably the potential for predicted probabilities outside the [0,1] range and inherent heteroskedasticity, we opt for this model due to its straightforward interpretation of marginal effects and computational efficiency. Furthermore, clustering standard errors at the regional level helps account for intra-cluster correlation and heterogeneity in the error structure. LPMs are frequently used in empirical work involving binary outcomes due to their straightforward interpretation and the fact that, in many cases, their marginal effects closely approximate those obtained from logistic regression models [58–61] Additional tests using logistic models are also included in the robustness checks to ensure consistency among the two models.

Despite these limitations, the analysis seeks to provide meaningful insights into the potential connections between climate conditions and self-reported displacement. The subsequent examination of the main results sheds light on such relationship within the scope of the identified constraints.

5.1 Main results

Our initial findings, reported in Table 2, suggest a positive association of the PDSI dummy with the probability of self-reported displacement between 2018 and 2022. Notably, the results are positive and significant across various specifications. In particular, the effect of the PDSI is twice as large when controlling for both region and year fixed effect (col. 4) as compared to our base model, which only includes our variable of interest (col. 1). This shows the relevance of using a pooled dataset in this econometric analysis as PDSI values (as well as climate anomalies) matter mostly for differences in displacement when specifically geo-localized within a certain region and within a certain year. Our most comprehensive model suggests that an individual exposed to a positive PDSI compared to the long-run mean in year t and region r is 8.7% more likely to report themselves as displaced than someone living in the same region in the same year who is exposed to a negative PDSI relative to the long-run mean.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. LPM: The effect of PDSI on displacement (2018–2022).

https://doi.org/10.1371/journal.pclm.0000594.t002

These results are in line with other empirical studies in East Africa. For example, [20] find a positive correlation between precipitation and displacement in East Africa, a result that is also found specularly in the same region by [62], who show that in Ethiopia the Mursi (a small cultivating pastoralist community in the Lower Omo Valley) responded to severe droughts through mobility as a spontaneous and strategic choice rather than forced displacement. This strategy allowed them to adapt to new environmental and climatic conditions and maintain their subsistence, despite the challenges posed by the drought. On the other hand, [18] finds that the PDSI is not statistically significant on displacement, but this happens only when accounting for potential mechanisms, such as conflict, a result we also find in Sect 5.2.2.

Despite our analysis focuses on the likelihood of self-reported displacement within a mobile population, it is important to note that our findings are also consistent with some prior evidence from the broader mobility literature. In low-income countries, where a significant portion of the population relies on agriculture, heavy rains and floods can disrupt livelihoods unexpectedly, forcing vulnerable groups to relocate immediately. In contrast, the effects of slow-onset climate changes, such as dry spells, droughts, or increasing heat stress, do not automatically translate into immediate relocation [26]. Additional findings support our hypothesis that decreasing rainfall has gradually worsened living conditions in rural areas, leading people to voluntarily relocate to urban centers in search of better opportunities and improved living circumstances. This movement is motivated by economic necessity and survival strategies, rather than being directly triggered by immediate disasters [63]. Conversely, other studies have shown conflicting results to ours, indicating that wetter conditions were either associated with immobility or did not significantly affect migration [64].

In order to delve into the pathways behind the climate-displacement nexus emerging from our specific findings and to better understand the role of climatic shocks as induced by extreme events, we regress a set of models where we alternatively employ a dummy capturing extremely dry (PDSI < –3), dry (–3 < PDSI < 0), wet (0 < PDSI < 3) or extreme wet conditions (PDSI > 3), as compared to the long-run mean. In comparative terms this means, for example, comparing an individual that experienced extremely dry climate in year t and region r compared to the long-run mean, with individuals that experienced, dry, wet and extreme wet climates in the same year t and in the same region r compared to the long-run mean. Results, depicted in Fig 4, show that extremely dry conditions reduce the probability of reporting as displaced by 8%. On the other hand, both wetter and extreme wetter conditions act as catalysts for self-reported displacement. Compared to the long-run mean, wet conditions increase the probability of displacement by 5.7%. Similarly, extreme wet conditions are associated to higher probability of being displaced by 10.55%. Normal dry conditions seem to have instead a null effect. In S4 Fig we present the estimates at different time spans ranging from the past three months to the past twelve months. Results remain consistent.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Coefficient plot: The effect of positive and negative PDSI on displacement (2018–2022).

Notes: This Fig shows the correlation between categories of the PDSI and self-reported displacement. The sample includes individuals surveyed between 2018–2022. The dependent variable is a dummy equal to 1 if individuals self-report as displaced. The main variable of interest is a set of dummy variables that take the value 1 if the condition is satisfied (e.g., extreme dry) and 0 otherwise. The PDSI is categorised as extremely dry (PDSI <–3), dry (–3< PDSI <0), wet (0< PDSI <3), or extremely wet (PDSI >3), relative to the long-run mean. Each row corresponds to a separate OLS regression in which the dependent variable is self-reported displacement and the independent variable is the category shown on the y-axis. Variable definitions are provided in Sect 3. Each regression includes control variables, region fixed effects, and year fixed effects. Confidence intervals are 90% and robust standard errors are clustered at the regional level.

https://doi.org/10.1371/journal.pclm.0000594.g004

We conjecture two plausible explanations for these results, which will be further investigated in Sect 5.2. First, wetter and extremely wetter conditions such as heavy rains and floods trigger immediate forced displacement due to security risks and livelihood collapse, whereas slow-onset events like droughts reshape migration as an adaptation strategy rather than a crisis response, particularly for individuals reliant on agriculture [33,65]. Second, excessive wet climate could promptly trigger displacement due to sudden disasters as well as increasing conflict risks. The urgency for individuals to evacuate the area arises from floods directly influencing the escalation of conflict dynamics. In this scenario, the severity of floods may trigger rapid and urgent movements of populations as a response to the exacerbation of conflict, thus delineating a specific pathway through which extreme wet weather can impact patterns of displacement [66,67].

5.2 Pathways analysis

Climate adaptation, forced immediate responses, and conflict are the main pathways hypothesized to be involved in the climate-displacement nexus. In this section, we test empirically (when possible) whether these are significant mechanisms within our sample. It is important to note, however, that the exploration of these pathways does not intend to rule out the possibility that additional (unexplored) mechanisms may simultaneously contribute to shape this complex relationship.

5.2.1 First pathway: Climate extremes and displacement in different agricultural zones.

In our interpretation, the impacts of climate variability on displacement are mediated by the nature and speed of climatic events, with slow-onset events, such as droughts, and rapid-onset events, such as heavy rains and flooding, influencing mobility through distinct pathways.

Examining our findings, we assert that individuals who self-identify as displaced are more inclined to have been forced to move rather than moving out of motivation. Specifically, we conjecture that rapid-onset events like heavy rains and flooding have a quite immediate and disruptive effect on human livelihoods, forcing people to move due to security concerns and/or a sudden collapse of livelihood systems, thus increasing the self-reported displacement among those on the move. On the contrary, drought and desertification have a more gradual impact on human mobility compared to sudden events, as they may reduce financial means for mobility while also providing time for adaptation, which can support the categorization of the mobility response as a voluntary, and proactive decision to seek better opportunities elsewhere rather than a forced displacement.

Arguably, individuals who rely on agricultural activities are more vulnerable to climatic-related disasters. Increased exposure may be linked to people fleeing an area not only due to direct threats to human security but also due to the complete loss of their livelihoods. To empirically test this hypothesis, we assess if displacement occurs in those regions heavily reliant on crop production. By distinguishing whether the individual comes from a village where the land is cultivated or covered by managed vegetation, as opposed to the other possible land cover types. Other possible land cover types include shrubs, herbaceous vegetation, urban area, bare vegetation, permanent water bodies, herbaceous wetland, closed forest and open forest. We acknowledge that our use of cropland land-cover as a proxy assumes, rather than directly observes, that individuals originate from those areas rely on agricultural activities. Unfortunately, the survey occupation data are too coarse to precisely identify individuals’ agricultural engagement, limiting our ability to validate this assumption. However, general literature in the area of study supports our assumptions as more than 80% of the population in East Africa is dependent on agriculture, and agricultural income accounts for roughly 40% of regional GDP [68,69]. In Ethiopia specifically, approximately 80% of the rural workforce is engaged in agriculture, further strengthening our proxy and underlying assumption [70]. Table 3 shows our findings for cropland areas, where the effect of PDSI wet and extremely wet is strongly and positively correlated with displacement with an increase of 9.3% and 8%, respectively, while the effects are statistically not significant for the areas without cropland. This suggests that one primary mechanism by which wetter weather conditions influence self-report displacement may be by disrupting livelihoods and agricultural output, leaving individuals with no other choice than to flee. The coefficient’s magnitude of the PDSI wet for individuals leaving agricultural areas is significantly higher than that from the pooled model, increasing from 5.7% to 9.3%. In contrast, the coefficient for extremely wet conditions is slightly lower, indicating that some of the effects of extremely wet conditions are also felt in non-agricultural areas. This seems reasonable, as more severe weather events can also significantly impact individuals who do not rely on agriculture. Furthermore, as shown in S10 Table, dry and extremely dry conditions do not show a positive correlation with self-reported displacement in both agricultural and non-agricultural areas. These findings support our interpretation that slow-onset climatic extremes are less likely linked to self-reported displacement, even among individuals who are heavily reliant on agriculture.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. LPM: Sub-sample analysis to test floods impacts on displacement in agricultural areas (2018–2022).

https://doi.org/10.1371/journal.pclm.0000594.t003

This interpretation aligns well with existing studies. [71], for example, links droughts with gradual, long-term movements, with individuals moving from dry, low-lying areas to regions with more favorable rainfall and arable land, or to other areas for alternative livelihood opportunities. Using advanced remote sensing techniques, [72] demonstrate notable differences in the effects of droughts and floods on displacement in Somalia. The study reveals that the effects of drought-induced movements start with a delay and then increase gradually. While individuals may require time to perceive the impact of extremely dry conditions and adapt accordingly, it is also possible that, upon moving, they may respond negatively when asked whether they consider themselves displaced. This response could be attributed to the relatively longer-term impact on human livelihoods and agriculture, as well as the relatively less abrupt and disruptive nature of drought compared to floods. Consequently, drought-induced movements may not explicitly identify as displacements, while flood-induced movements do. Suggestive evidence of this was found by [73] that emphasizes the role of individual’s perceptions of climate anomalies in migration decisions. [72] show that floods often lead to immediate displacement, with people relocating shortly before or during the floods [72]. Furthermore, [72] also recognize agricultural areas as highly vulnerable to both droughts and floods. Droughts can lead to soil moisture deficits and reduced vegetation productivity, resulting in significant crop loss and food insecurity. Conversely, floods can inundate fields, destroying crops and disrupting agricultural activities.

5.2.2 Second pathway: Compounded extreme wet climate and conflict effects.

This mechanism highlights a reaction to climate variability in the form of conflicts, which then may lead to displacement. Recent evidence, particularly in Africa, indicates that wetter climate is likely to trigger civil conflict and insurgencies [74]. Variations in precipitation can significantly impact various forms of political conflict, with the most pronounced association observed with violent events, which are particularly impacted by abundant rainfall. In the Greater Horn of Africa, [67] find that humanitarian crises happened because of extensive flooding, when this event leads to government instability, crop pest outbreaks, and ethnic conflicts, ultimately increasing the likelihood of displacement. Additional evidence from [75] demonstrates that in Kenya, periods of increased precipitation are associated with a rise in violence. Finally, [76] find that in Africa, Asia and the Middle East, flood-related political unrest occurs within two months after 24% of the 92 large flooding events recorded in their sample. In our sample, we find that 40% of the interviewees have experienced at least one conflict during the departure period, with an average of five violent conflicts within 50 km radius from their location.

To empirically test whether conflict is actually a mediating factor between floods and displacement in our sample, we conduct two distinct analyses. The first one is a mediation analysis that examines the relationship between PDSI (extreme wet), conflict, and displacement (see Table 4 for details). Such analysis provides estimates of the effect of the PDSI (extreme wet) on conflict (col.1); the effect of the PDSI (extreme wet) on self-reported displacement, which represents the total impact of climate exposure on self-reported displacement (col.2); the effects of PDSI (extreme wet) and conflict (respectively and , in one single specification) on self-reported displacement (col.3). Following the approach of [77], the mediating percentage effect of the PDSI on self-reported displacement through the conflict mechanism is then computed as , while the residual effect induced by other mechanisms is (col.4). In this analysis, the conflict variable is standardised, with mean 0 and standard deviation 1, for each year. The mediating effect is normalised to 100 and presented in percentage terms.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Linear Probability Model: Mediation analysis (2018–2022).

https://doi.org/10.1371/journal.pclm.0000594.t004

Our findings indicate that extreme wet conditions increase by 0.441 standard deviation the conflict within 50 KMs of the village of the migrant. Overall, living in areas affected by extreme wet conditions is associated to an increase in the probability of displacement of 10.6%. When we incorporate conflict in the displacement model, however, climate loses economic and statistical significance, which is partially captured by conflict. More precisely, the mediation analysis shows that conflict accounts for 9.31 percent of the total effect of climate on displacement. It is noteworthy to emphasize that a substantial, unaccounted-for 90 percent mediating effect persists, encapsulated by the PDSI (extreme wet). This effect inherently signifies the diverse array of potential mechanisms that have not been integrated into the model but are effectively represented by this variable. These may include macro-factors such as political instability, but also micro-factors such as the impact extreme wet climate may have on housing, income and employment, health, access to resources, and education [6]. Additionally, we acknowledge the limitations inherent in interpreting the quantitative mediation analysis as entirely causal. The incorporation of a mechanism can introduce bias, especially when the potential mechanism is endogenous, as discussed by [77] - a circumstance applicable to conflict in this context.

As a second empirical test, we show two distinct pieces of evidence, reported in S6 Fig. First, as a falsification test, we show that only wet conditions increase the probability of conflict, in particular when extremely wet, excluding the suggestive idea that also dry or extremely dry conditions are conducive to conflict and, therefore, displacement.

Second, we show that the extreme wet conditions at different time lags affect the probability of conflict within 50 km from the village of the interviewee only when it is very proximate to the time of departure, i.e., three months before, while the effect is null at six, nine and twelve months. As highlighted above, these results align with some recent evidence on the climate and conflict interplay, suggesting that periods characterized by the presence of heavy rains are linked with an increased risk of conflict and violent events, compared to drier or non-deviation periods [74,78].

5.3 Heterogeneity analysis

Our primary findings suggest that, on average, we are likely to observe effects related to heterogeneity. A detailed analysis of how the Palmer Drought Severity Index (PDSI) influences the likelihood of self-reported displacement across various socio-demographic groups and types of movement reveals significant differences. Despite these variations, the overall trend of our main findings remains consistent. Specifically, extreme dry conditions diminish the likelihood of reporting displacement, while dry conditions show no noticeable impact. In contrast, wet and extremely wet conditions contribute to an increased self-reported displacement (see Fig 5 for details). Full results of the heterogeneity analysis are presented in S4 Table.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Coefficient plot: The heterogeneous effect of the PDSI on displacement (2018–2022).

Notes: This Fig shows heterogeneous associations between the PDSI and self-reported displacement. The sample includes individuals surveyed between 2018–2022. The dependent variable is a dummy equal to 1 if an individual self-reports as displaced. The main variable of interest is the PDSI index interacted with the characteristic shown on the y-axis (each row corresponds to a separate OLS regression). Variable definitions are provided in Sect 3. Each regression includes control variables, region fixed effects, and year fixed effects. Confidence intervals are 90% and robust standard errors are clustered at the regional level. The PDSI categories are extremely dry (PDSI <–3), dry (–3< PDSI <0), wet (0< PDSI <3), or extremely wet (PDSI >3), relative to the long-run mean.

https://doi.org/10.1371/journal.pclm.0000594.g005

More in detail, the analysis reveals that extremely dry conditions significantly reduce self-reported displacement for both below and above 30-years old individuals, with a more pronounced effect in favour of the elder cohort (–0.13 for the elder compared to –0.06 for the younger). This indicates that, although self-reported displacement decreases for all age groups during extremely dry conditions, younger individuals are relatively more likely to report displacement during this condition compared to older counterparts. The relatively smaller decrease in reporting displacement among the youth is consistent with existing literature focused on East Africa, suggesting that the youth are more vulnerable to drought conditions, likely due to higher unemployment and less stable lives [24,79]. It is important to note that, while we have previously shown that more socioeconomically settled individuals tend to report higher levels of displacement in our sample, the relative responsiveness to extreme dry conditions is different among diverse age and socioeconomic groups, highlighting the importance of a disaggregated analysis. Moreover, our main findings underscore that men and individuals with relatively higher educational levels (at least primary education) are less likely to report displacement in the face of extremely dry conditions. In contrast, the decrease in the self-reported displacement observed among women and those with lower educational attainment (no primary school) does not reach statistical significance. Consistently with our interpretation relying on adaptation mechanisms occurring during extreme dryness, this additional insight can be attributed to the fewer social and economic obstacles to voluntary migrate faced by men and individuals with higher education levels compared to women and those with lower education levels. These hurdles include barriers to accessing productive resources, participating in decision-making processes, or being subject to social and cultural norms. This advantage allows them to more effectively implement adaptation strategies, such as voluntary migration, thereby reducing the likelihood of reporting displacement during droughts [80].

The heterogeneous impact of wet climate, characterized by a significant rise in self-reported displacement among individuals aged 30 and above, females, and those with lower educational attainment, supports our core findings on the abrupt nature of wetter-than-usual conditions as displacement triggers. The results indicate that sudden climatic shocks, such as extreme precipitation and flooding, disproportionately heighten the reporting of displacement among sub-groups who may already face heightened vulnerability due to limited adaptive capacity and socioeconomic constraints. Unlike gradual climatic stressors, which may provide room for adjustment, wet and extreme wet conditions create immediate disruptions in livelihoods, infrastructure, and security, increasing the likelihood that individuals report themselves as displaced. This effect is particularly pronounced among those reliant on agricultural livelihoods, as sudden excess rainfall can lead to crop destruction and loss of income, compelling movement (Momeni et al., 2024). The stronger association observed among females and less-educated individuals suggests that these groups, who may have fewer resources to mitigate the impact of climate shocks, are more likely to report themselves as forcibly displaced.

Additionally, our findings indicate that wet conditions heighten the likelihood of self-reported displacement among individuals interviewed outside their country of origin. This pattern may partly reflect the influence of the interview context, particularly at border points where institutional presence is more pronounced and individuals might perceive potential benefits in identifying as displaced. At the same time, this could also suggest that the increasing vulnerabilities related to crossing national borders, particularly in response to sudden climatic shocks, reinforces the subjective experience of being displaced.

Ultimately, the occurrence of extreme wet climate suggests that older migrants confronting such conditions are more susceptible to displacement than their younger counterparts. This vulnerability could be linked to their limited adaptive capacity, making it challenging for them to improve their current conditions on-site [81]. Simultaneously, their dependency on women [82] implies that if women move due to worsening climatic conditions, the elderly are likely to follow suit.

6 Discussion and conclusions

Understanding the impacts of climate anomalies on displacement is crucial, particularly within the Greater Horn of Africa. In this context, escalating climatic challenges, coupled with heightened insecurity and substantial displacement, have culminated in a pronounced humanitarian and development crisis. Addressing such crisis demands urgent attention and concerted efforts to enhance solutions that alleviate the broader socio-economic implications of climate-induced displacement.

This study represents an important attempt to delve into the drivers of displacement in the region and to empirically examine how climate variability impacts individual reporting of displacement, as well as the mechanisms connecting the two. The study resulted in three main findings. First, there is a positive correlation between wet and extreme wet conditions and the likelihood of displacement, implying that people tend to report themselves as displaced after experiencing wetter or flooded circumstances. Conversely, dry and extremely dry conditions are associated with a non-significant and reduced likelihood of self-reported displacement within the sample of individuals traced in the FMPs. This trend is consistent with some recent evidence that underscores how wetter climatic conditions, due to their sudden nature, can have a much more immediate effect on displacement than a drought, particularly in the Greater Horn of Africa.

Second, in unpacking the displacement responses to wet, extreme wet, and extremely dry climate conditions, we identify two distinct pathways through which climate variability shapes the self-reporting of displacement. Firstly, slow-onset stressors, such as droughts, may lead individuals to gradually adapt and have a higher degree of choice in timing, route, and destination, reducing the likelihood of self-reported displacement. In contrast, sudden climatic shocks, especially extreme wet conditions, increase the likelihood of self-reported displacement among the interviewed migrants due to their abrupt impact on livelihoods and security risks. Secondly, we observe that extreme wet conditions contribute to displacement by heightening the risk of conflicts. Wetter periods are evidently more susceptible to conflict and violent events, in contrast to drier or non-deviation periods. These findings highlight how the speed and intensity of climate shocks shape the way individuals report displacement. Recent climate-model assessments for East Africa indicate a robust, scenario-independent increase in temperatures throughout the twenty-first century, combined with a tendency toward more frequent and prolonged hydrologic extremes and rapid oscillations between severe droughts and flood events, even though projections of mean rainfall remain highly uncertain [84]. In such a warmer future, characterized by more intense and possibly faster-onset droughts alongside flood hazards, it is plausible that severe drought episodes could act as more acute triggers of displacement, more closely resembling the role of extreme wet conditions documented in our analysis and potentially altering the balance of mechanisms we observe in the current study.

Lastly, as we investigate the connection between climate variability and self-reported displacement and consider the individuals’ socio-demographic characteristics and typologies of movement, our results reveal different patterns. Men and individuals with relatively higher education (at least primary education) show lower likelihood of reporting themselves as displaced in the face of extremely dry conditions, likely due to greater adaptive capacity and alternative livelihood strategies. Wet and extremely wet conditions significantly increase self-reported displacement among people aged 30 and older, particularly women and individuals with lower levels of education. This reinforces the idea that sudden climate shocks act as triggers for displacement, especially among more vulnerable groups. Additionally, our results suggest that individuals interviewed outside their country of origin are more likely to report as displaced. While this result could be due to heightened vulnerabilities associated with cross-border movements, particularly following wet climate shocks, it could also be influenced by the context of the interview, which may affect how people frame and report their experience. These dynamics underscore age, gender, education, the nature of the movement (within or outside the country of origin) and context of interview as critical factors shaping the self-reporting of displacement under different climate conditions.

It is important to note that these findings do not present conclusive evidence regarding the climate-displacement nexus in the Greater Horn of Africa. As previously highlighted, we acknowledge the potential biases in our analysis, arising from selection bias, measurement error, and social desirability bias, which could impact our results. Untangling these empirical challenges necessitates further research and access to appropriate data. The data collection on mobility, in particular, poses significant challenges in tracing individuals and households across different points in time and identifying samples of migrants that are plausibly representative of the areas of origin. Despite its novelty and originality, the dataset used in this analysis is not representative of the Greater Horn of Africa, and is a cross-sectional dataset, which limits the assessment of climate-displacement dynamics over time. In particular, we acknowledge that the use of FMP-based data could introduce a spatial selection bias, limiting the generalization of our findings to the broader population of displaced individuals. Mobile individuals who avoid formal monitoring points, due to insecurity, informality, or route preferences, are likely underrepresented in our sample, and this may influence the observed demographic and socio-economic patterns. From this perspective, enhancing the synergies between data collection efforts, research, and policy action can significantly improve the quality of research outputs. Moreover, future research on mobility would benefit from a more comprehensive integration of climate components into mobility data. Analyses such as the one presented here can help inform the design of more tailored mobility and displacement data collection, regarding sampling strategies, longitudinal follow-up and the systematic inclusion of climate and environmental modules, thereby creating the conditions for more rigorous regression analyses that can better identify causal relationships and elucidate the mechanisms underpinning the climate-displacement nexus.

Despite the limitations of our findings, this work calls for the development of ad-hoc policies aimed at addressing the enduring crisis related to climate variability, displacement, and insecurity. Firstly, it is evident that, in the face of extreme climatic conditions, effective anticipatory action schemes are necessary to prepare and support affected areas against diverse climate shocks. In particular, wetter periods and related flooding hazards act as acute triggers of displacement. Community-level investments in early warning systems, anticipatory cash transfers, and flood preparedness may help to reduce displacement by empowering people to act before vulnerability thresholds are reached. These measures should be aligned with existing regional forecasting tools such as the ICPAC seasonal forecast platform, which provides early warning information that can be used to trigger anticipatory action in flood-prone areas [85]. Simultaneously, there is a pressing need to invest in infrastructure resilience, particularly in flood-prone areas, to mitigate the impact of extreme wet conditions. This involves the fortification of physical structures and the adoption of comprehensive floodplain management strategies that constrain both natural and human-induced factors contributing to flooding. Climate adaptation and climate adaptive social protection strategies, as well as development measures, should also be tailored to the specific groups of people who are identified as the most vulnerable to the climate-displacement dynamics in order to efficiently locate resources where they are most in need. Our analysis indicates that self-reported displacement patterns are influenced by geographic location, demographic and socio-economic status, highlighting the significance of disaggregated targeting. This will ensure that funds for climate adaptation, livelihood assistance, and social protection programs are allocated to groups who have highest exposure and vulnerabilities and lowest response capacity. Yet these needs arise at a time when major bilateral development programmes, including USAID, are being dismantled or significantly reduced, potentially narrowing the fiscal space for climate adaptation, livelihood assistance and social protection in the region. This makes it even more important to safeguard and diversify funding streams for displacement monitoring, climate-adaptive social protection and anticipatory action, so that policies can be grounded in reliable evidence and remain operational despite shifts in the aid landscape. Social protection systems that are adaptable and climate-sensitive are critical in addressing the structural vulnerabilities that influence mobility choices as well as the immediate effects of climate stress. Lastly, addressing the climate-displacement issue in its multidimensional nature requires efforts to coordinate climate-resilient policies, the development of sustainable economic opportunities in the areas of origin, and institutional measures of conflict prevention and peace-building initiatives.

Supporting information

Acknowledgments

We are very thankful to the International Organization for Migration (IOM) for providing the DTM Flow Monitoring Surveys (FMS) data. We are particularly thankful to Laura Nistri, Maxwell Aderoh and Daniel Ibañez Campos from the Regional Data Hub team at the IOM Regional Office for the East and Horn of Africa for the administrative and logistical support on the usage of the data. Furthermore, we are grateful to our colleague Bina Desai for her support. We are particularly indebted to all individuals who participated in this study and the enumerators who helped collect the data. All views expressed and mistakes are our own.