Observed climatic change, independent of the cause of such change, human-induced climate change, and various forms of human mobility such as migration are discussed widely, and connect discussants from the public, policymakers, and academic researchers across disciplines (Gray and Mueller, 2012; Randall, 2015). Leading, in part, to a perceived causal chain between anthropogenic climate change, extreme events—often labelled as “climate change event,” and associated mobility suggested in the migration literature portray an over-simplified picture of climate-related human mobility (Hastrup and Olwig, 2012; Hoffmann et al., 2020). Most of the debate, or at least a noisy part, has been framed by apocalyptic thinking or anti-immigrant sentiments (Haas, 2008; White et al., 2011), which construe particularly migration as a problem that needs to be solved. International organisations outline that migration will increase in so-called climate change hot-spots, or places where communities are already vulnerable to the physical and ecological effects of climate change (Rigaud et al., 2018). Despite their prevalence in climate migration rhetoric, such perceptions that problematise climate change-induced human mobility have been criticised for their insufficient scientific robustness, for example, in the context of conflict research (Abel et al., 2019; Ide et al., 2020). The migration-development literature provides a more nuanced understanding of human mobility (Klepp, 2017). Scholars in social science and development studies (e.g., Bakewell, 2008, 2010; Haas, 2008, 2010; Flahaux and De Haas, 2016) break down and clarify the often-misunderstood debate on migration and development: because human mobility requires financial capital, poverty reduces out-migration; and in turn, economic development and increased financial means lead to more migration (Bastia and Skeldon, 2020). With currently little consensus on appropriate research methods to identify whether and to what extent climate change affect human mobility, it is not clear what role human induced climate change plays in this nexus at present and in the future. Research that attempts to juxtapose the vast heterogeneity of topical perspectives have resulted in different concepts and terminologies (Piguet, 2010; Borderon et al., 2019; Hoffmann et al., 2020).

This Review is guided by the question: How sound are the lines of evidence for a causal link between human-mobility and a changing climate in East Africa? Here, we provide a dual-method evidence approach, consisting of a qualitative case study analysis and a quantitative topic model that are set to disentangle the complex drivers of climate-related human mobility. Through the dual-method, we can reveal connexions between climate mobilities and contextual factors (e.g., food security and livelihoods) which a single-method approach would have otherwise masked.

To this aim, we systematically review the literature and explore the lines of evidence on the link between weather and climate-related events and human mobility (climate mobilities). We focus on the methodological compatibility (or lack thereof) from climate mobilities research stemming from different disciplinary spheres in a potential climate change hot-spot region (Byers et al., 2018). We show that climate data finds little application in existing climate mobilities literature in East Africa. This might lead to an oversimplified picture of human responses to weather and climate-related events. From the analysis, we can infer there is not yet a framework that allows for a disentangling of weather and climate-related human mobility drivers from non-climatic, contextual drivers, assessing the probability of climate change impacts, and outcomes. Our key takeaway is that the problematic use of climate and weather data is not limited to East Africa, but points to a wider problem in this field of climate change impact research which requires truly interdisciplinary research. Therefore, we outline the applicability of quantitative methods from econometrics and extreme event attribution to support the development of a more comprehensive understanding on human responses to weather and climate-related events and attributable links to climate change in future research. This analysis contributes to the ongoing effort of assessing the probability of climate change impacts and human responses to a warming climate.

East Africa (Figure 1) is vulnerable to climate change and variability and the impacts from extreme weather and compound events (Marthews et al., 2019). The region is characterised by a traditionally mobile population, consisting of pastoralists and herders. Human mobility is part of many livelihoods across East Africa (Collier et al., 2008; Shongwe et al., 2011; Uhe et al., 2018; Funk, 2020). In recent years, a number of extreme weather events have had significant social and economic impacts, including displacement, food insecurity and conflict. To illustrate, the 2015 drought in Ethiopia put an estimated 22 million people in need of food aid (Philip et al., 2017). In Somalia, 26 different flood, drought, and storm events have been recorded between 2010 and 2020 by the disaster database Em-DAT (Guha-Sapir et al., 2016).

The analytical framework consists of three elements: a systematic literature review, a resulting case study analysis and a topic model evaluation (Figure 2). While setting out to qualitatively analyse the case studies, we use a topic modelling approach to augment the understanding on the range and significance of contextual and climatic influences on human mobility in the study region. This dual-method approach allows us to synthesise and quantify the current evidence in the field.

In a broad and systematic search of the literature (see Supplementary Section A for a detailed description of the systemic literature review), we identified qualitative and quantitative studies that investigate the link between a variety of climatic change factors and human mobility in East Africa. Applying the inclusion-exclusion criteria, the systematic literature review yields in 14 case studies published between 2000 and 2019 (see Supplementary Figure 1). For the analysis, we focus on country studies to cover large geographical areas in East Africa (Supplementary Table 1).

We apply a set of categories to the selected case studies, well-established in climate mobility research to systematically categorise the studies (Piguet, 2010). We synthesised evidence on socio-economic and political drivers which we hereafter refer to as contextual drivers as well as weather and climate-related drivers of human mobility (section Result). We then categorised the case studies according to the type of weather and climate-related event studied. To better understand the spatio-temporal dynamics of human mobility, we separately list the departure and arrival regions, and the analytical time horizon covered in the case studies. We also separately marked case studies which refer to multi-causality as a theoretical concept of human mobility (Supplementary Table 2).

In a second analytical step, we apply a topic model approach on the selected case studies to classify them according to their key topics and word frequencies. The systematic literature review serves as statistical base for our model. Topic models are unsupervised machine learning techniques aiming at exploring text collections thematically. Our approach allows to generate topics through a set of statistical models that depict common occurrences of words within and across the case studies. From the systematic literature review, we gather that all selected case studies are broadly about the topic of climate mobilities. Using the topic model approach, it is generally assumed that a text collection consists of different topics with a topic being understood to be a cluster of words; in this case words such as “climate,” “migration,” and “household” frequently occur together. Broadly, a topic can be regarded as a statistical phenomenon and therefore some correspondence, but not exactly the same as a (content-defined) theme. However, topic modelling methods are typically used to analyse large text collections (Supplementary Section C). We overcome the challenge of working with a smaller sample size by using contextual clues to connect words with similar meanings and distinguish between uses of words with multiple meanings. Topic models are therefore well-suited to explore the thematic contexts of a smaller sample size (Griffiths and Steyvers, 2004). Expanding on the results from the case study approach, topic models can help to discover hidden topical patterns that may be present across the case studies and might have been hidden otherwise. Hidden topical patterns can be revealed by annotating documents according to key topics; results of which we show in Figure 3. Using these annotations, topic modelling helps to organise, search and summarise the case studies. The method illustrates both the quantity and relevance of key terms. Thus, it is a well-suited approach to expand our understanding on the vast parameters contributing to climate mobilities.

As a region, East Africa is considered a hot-spot of compound events due to increased population exposure resulting from extreme weather and population growth (Weber et al., 2020). Although parts of the region are well-represented in the literature, the evidence on drivers of climate-related human mobility beyond voluntary migration, direct and indirect factors, and their interlinkages remain inconclusive and context-specific (van der Land et al., 2018; Groth et al., 2020).

Apart from the very diverse temporal (including permanent migration) and geographical scales on which human mobility takes place, there are different forms human mobility can take. We find that the case studies distinguish between voluntary forms of mobility (i) migration and (ii) pastoralism, and involuntary mobility (iii) displacement and iv) refugees, (v) asylum seekers and (vi) resettlement (Table 1).

Conversely, we find large differences on how studies analyse and describe a country or regional context (macro-level studies), e.g., employing models to control for specific socio-economic and contextual political variables to isolate the specific impact of climate on migration or investigating climate change perceptions of affected populations on the move through interviews. GM12 for example, investigates drought effects on local populations in Ethiopia through multivariate regressions (Gray and Mueller, 2012). AG12 (Afifi et al., 2012) provides evidence on climate change perceptions and vulnerability from the perspective of refugees based in camps in Ethiopia and Uganda.

Our topic model allows to disentangle weather and climate-related drivers in human mobility from contextual drivers. By adding this approach, we can further infer that the studies document and discuss various types, drivers and impacts (sub-themes) of human mobility in the study region (Figure 3, Supplementary Figure 2). The topic model reveals structures highlighting that a range of contextual factors regarding food security, economy, conflict, and social norms influence human mobility in addition to climate factors. Within these contextual factors, armed conflict, social norms, and livelihoods, e.g., pastoralism, and community stand out across all case studies (Figure 3).

Interestingly, the studies OP13 (Opondo, 2013), ST11 (Stal, 2011), AB17 (Alemayehu and Bewket, 2017) focus on a single factor influencing human mobility, whereas the other case studies use a framework proposed by Black et al. (2011), acknowledging the multi-causality in human mobility drivers. A multi-causal framework proves to be well-suited to conceptualise different geographical and temporal dimension of direct and indirect factors influencing human mobility (Groth et al., 2020). Several case studies used the framework as theoretical baseline to study the links between human mobility and weather and climate-related events, often in the context of loss and damages or economic vulnerability, e.g., EK01 (Ezra and Kiros, 2001), TA11 (Tacoli, 2011), and HG19 (Hermans and Garbe, 2019).

In light of limitations on human mobility data (Hauer et al., 2020), widely-available climate data takes a profound role in drawing conclusion on the effects of weather and climate-related events on human mobility and the potential links to anthropogenic climate change.

We find that at minimum, if at all, studies only use basic climate indicators to investigate climatic determinants of mobility. For example, publications around the Same-Kilimanjaro region in Tanzania (Afifi et al., 2014: AL14) or in (Nyaoro et al., 2016: NS16) use annual rainfall trends but forego to take the complexity of seasonality into account. One third of the case studies include climate data in their analysis or rely on previous publications of climate data assessments in the region, for example GF14 (Ginnetti and Franck, 2014), AG12 (Afifi et al., 2012), and AB17 (see Table 2). Of those, several use a single aspect of climatic change, such as rainfall anomalies (e.g., HG19) or temperature change (e.g., AB17) while others use historical precipitation data on different temporal scales such as monthly rainfall or minimum and maximum temperature (e.g., AB17, data from 1981–2013).

One case, Afifi et al. (AL14), uses monthly mean precipitation observations from the Tanzania Meteorological Agency (TMA) to describe changes in climate variability. They refer to annual mean temperature of 1992–2011 as a drought indicator of the 2011 East African drought. Alemayehu and Bewket (AB17) describe rainfall and temperature variability observations with results from their previous study that analysed average annual maximum and minimum temperatures from 1981 to 2013 (Alemayehu and Bewket, 2016). Interestingly, neither of these publications justify how temperature anomalies are linked to the occurrence of drought in the area. Instead, by analysing temperatures, the studies seem to imply that changing temperatures are a defining factor of drought. Yet, in an arid region it is usually rainfall that determines drought (Manning et al., 2018; Coughlan de Perez et al., 2019). Recent scientific studies suggest that temperature changes play a negligible role compared to other drivers of drought in East Africa (Philip et al., 2017; Kew et al., 2021).

Instead of analysing climate data, three studies use country-level records from the disaster statistics database EM-DAT (Guha-Sapir et al., 2016) to define an extreme weather event, for example, the 2002 and 2008 Ethiopian drought (GM12). Based on these records, NS16 lists the impacts of drought and flood disasters on affected populations from 1964 to 2015, whereas Stal (ST11) uses the database to define the two flood events investigated in the Zambezi River Valley. By incorporating satellite imagery of the 2001 flood and a risk assessment map for the 2007 flood, ST11 supports the characterisation of the event through physical characteristics of an extreme weather event (Stal, 2011).

Hermans and Garbe (HG19) investigate direct livelihood strategies, migration experiences and perceptions of the 2015 Ethiopian drought with household-level and individual information from a household survey and focus group discussions in early 2016, shortly after the drought ended. Despite framing the start and end of the drought by using a time series of 30-year daily rainfall observations (a triplet of 10-year blocks from 1985 to 1994, 1995 to 2004, and 2005 to 2014), the authors use the rainfall data to illustrate an increasingly variable start of the Belg rainy season across the time series through basic descriptive statistics.

Gridded climate datasets are used by a single case study of our document corpus. Gray and Wise (GW16) use observed monthly temperature and rainfall data from the Climatic Research Unit (CRU) in combination with NASA's MERRA reanalysis data on land surface temperature to address limitations identified in previous studies (see Table 2). They criticise the usage of a single climate dataset to explore links between climate variability and human migration (Gray and Wise, 2016).

All case studies focus on historical weather and human mobility; only GF14 also models future displacement with a systems dynamics model that aims to improve the understanding of root causes of lost livelihoods. Along livestock and displacement statistics, Ginnetti and Franck (GF14) compare 2008 to 2013 mean annual rainfall over Somalia, Kenya and Ethiopia with anecdotal, national-level monthly displacement records. This data serves as a proxy for drought and is mapped with socio-economic data such as agriculture productivity. On this basis, the resulting model produces estimates of drought-related displacement by 2040 that can be compared to empirical evidence of past displacement due to drought. In terms of extreme event definition, the study relies on existing reports for defining drought impacts on livestock during 1991 to 2010 rather than using drought indicators. However, it remains unclear why such data was chosen instead of metrics such as the Standartized Precipitation Evaporation Index (SPEI) or precipitation anomalies for which the Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) dataset is well-suited (Dinku et al., 2018).

All studies use a single mobility dataset for their analysis (Supplementary Table 2). A fact primarily resulting from the very limited availability of datasets on combined mobility types or simply due to time and budget constraints (e.g., ST11, EK01, AG12). Ezra and Kiros for example, list a range of data limitations to their household survey implementation, such as spatial and temporal elements of their survey design (EK01).

In terms of data sources, the four grey literature studies rely on locally collected mobility data over the span of a few weeks. NS16 use UNHCR-collected country-level data from 2016 on refugees from various East African countries, most of them originating from Somalia (71%) and South Sudan (16%). Similarly, Jayawardhan (JA17) uses the same data source for annual displacement from 2009 to 2012 to show a peak in drought-related displacement during the 2011 drought in Somalia (Jayawardhan, 2017). Ten case studies collected empirical mobility data through household surveys and interviews with government officials and population affected by extreme weather using common combinations of sample surveys and qualitative research methods (Supplementary Table 2) at either the source or destination area (e.g., OP13, AG12, AL14, and GM12). Only Tacoli (TA11) interviewed Maasai men, an increasingly mobile group, in both the origin (rural) and destination (urban) across Tanzania. AL14 uses a combination of empirical data that includes a survey of 165 households, expert interviews and focus group discussions. Herman and Garbe (HG19) differentiate between participants who return after a month or longer (temporary migrants), and those unlikely to return (permanent migrants) in their dataset and contextualises migration decisions of male household heads with focus group discussions of farmers in northern Ethiopia. The empirical studies use a combination of complementing methodologies to support the time frame of the analysis. The combination of sample surveys and qualitative research methods is common. AG12 use a mixed-method approach of expert interviews, household survey together with six participatory research methods to understand the interrelation between rainfall and mobility. Among affected populations, reported perceptions of climate stressors are fundamental for the definition and scope of extreme weather and help understand climate impacts experienced. These qualitative studies do not incorporate observational or reanalysis climate data, solely relying on perceived, and self-reported observations on climatic changes (see Table 2). This use of complimentary datasets on human mobility in AG12 is unique and sets apart the analytical validation. In contrast, three papers solely rely on secondary statistics from international organisations or non-governmental organisations such as UNHCR or FEWS Net (JA17, NS16, Achour and Lacan, 2012: AL12).

To some extent, the studies base their analyses on the assumption that human-induced climate change has already affected mobility patterns in case study locations. It remains unclear whether case studies refer to human-induced climate change by defined by the UNFCCC (1992), or the evolution of local climate independent drivers which represents the Intergovernmental Panel on Climate Change IPCC usage of the term 2018. Climate change is often discussed in the context of drought (e.g., JA17, AG12, and NS16). In particular, grey literature contextualises adverse impacts of climate change on sectors such as forestry (for example, ST11), agriculture, and livelihoods, like JA17 (Figure 3).

From a theoretical standpoint, the studies refer to indirect and direct drivers of human mobility as suggested by Black et al. (2011) in the context of climate change (Supplementary Table 2; Figure 3). For example, GF14 uses intervening facilitators for drought-related displacement in a dynamic model of livestock and market trends and estimates future dynamics of such displacement under climate change. AG12 specifically add the climate change component to migration in the initial data collection by including climate change perceptions across different groups of interviewees to understand its (perceived) role in out-migration. They find that refugees affected by climate variability directly link changes in climate to increases in political conflict and violence occurrences. GF14 make it clear that the study only links to trends in the climate independent of the cause by defining the term according to the definition provided by the IPCC. Synthesising the results of their systems dynamics model, emphasis is put on the multi-causality of human mobility drivers, particularly that drought and climate change (and/or variability) are not the only drivers of displacement.