Agricultural livelihoods, adaptation, and environmental migration in sub-Saharan drylands: a meta-analytical review

Drylands in sub-Saharan Africa are strongly affected by the impacts of climate change. Temperature increases, changes in rainfall patterns, and land degradation pose serious threats to food security, health, and water availability in the region. The increase in livelihood insecurity can in turn trigger migration as a way to adapt or cope with stress. Based on 89 original case studies, this study uses review and meta-analytical techniques to systematically explore the relationship between environmental change, adaptation, and migration in rural areas in sub-Saharan drylands. We show that households use a diverse range of strategies to respond to environmental hardships in different livelihood and ecological contexts. While migration is common in some communities, it is of less relevance to others, and it can take various forms. Our findings indicate that migration is often used as a complementary strategy to other forms of adaptation, which can vary depending on situational needs. We use cluster analysis to identify adaptation clusters and show how linked response strategies differ by socioeconomic conditions. We find that migration can serve as a last resort measure for highly vulnerable groups, or be used in combination with in-situ strategies for diversifying income and adapting agricultural practices. Our results have important implications highlighting the role of local conditions and complementary forms of coping and adaptation for understanding environmental migration.


Introduction
Drylands cover about 46.2% of the global terrestrial area and are home to over 2 billion people or around one third of the global population (IPCC 2019, Hoover et al 2020. They are ecosystems characterized by high temporal and spatial rainfall variability and aridity (FAO 2019). Climate change is expected to have major impacts on these areas with increasing levels of water stress, desertification, land degradation, and a continued loss of biodiversity. By the end of this century, drylands are projected to further expand by 11% under a medium (RCP4.5) and up to 23% under a high (RCP8.5) greenhouse gas emission scenario relative to a 1961-1990 baseline (Huang et al 2016).
In sub-Saharan Africa (SSA), drylands represent an important ecosystem accounting for 70% of the total cropland (Cervigni and Morris 2016). The region is highly exposed to climate change impacts, which come in form of slow-onset processes combined with the simultaneously increasing frequency and intensity of extreme events such as floods and droughts (IPCC 2018).
With a high share of its population depending on agriculture, high levels of poverty, and low adaptive capacities, the region is characterized by a high vulnerability (Hoffmann 2022, IPCC et al 2022. Climate change is expected to further contribute to this vulnerability by exacerbating existing development challenges and non-climatic stressors, such as conflicts (FAO 2018). With worsening climate impacts, limits may be reached beyond which adaptation becomes impossible as communities lack means and resources to adequately prepare against and tolerate hazards (Warner et al 2012, Dow et al 2013, Xu et al 2020. Under these circumstances, migration can represent a suitable strategy to cope with and adapt to the climate impacts (Foresight 2011, IPCC et al 2022. By having household members or relatives move to other locations, households can diversify income sources and spread risks, which in turn can increase their resilience (McLeman and Smit 2006, Black et al 2011b, Gemenne and Blocher 2017. Here, migration and in-situ adaptation can be two sides of the same coin, for example if households engage in migration and at the same time adapt their agricultural practices. Also the relocation of an entire household can lead to improved protection and adaptation, although researchers have also warned of maladaptation and potential detrimental effects for migrating households and their communities (Warner et al 2012, Jacobson et al 2019, Vinke et al 2020. Given that migration is just one of many coping and adaptation options, it has to be considered against the background of the whole range of options available to households (McLeman andSmit 2006, Piguet 2010).
Despite an increase in public interest in the links between climate change, adaptation, and human migration, empirical evidence on the relationships remains inconclusive (Hunter et al 2015. Many studies on the topic are localized, with a focus on one community or one country, making it difficult to assess the range of coping and adaptation options used by households to respond to environmental stress and the conditions under which they change their migration behavior. In particular, with regards to the special circumstances faced by rural populations in sub-Saharan drylands, limited comparative evidence exists on climate adaptation and the migration impacts of environmental drivers.
To this end, this review study synthesizes the evidence from 89 case studies from 22 countries on climate change coping and adaptation in SSA. Whereas coping refers to short-term, immediate activities to address, manage, and overcome adverse conditions, adaptation refers to sustained adjustments that are oriented toward longer-term livelihood security (IPCC 2018). As a clear distinction between the two is not always possible based on the reviewed case studies, we speak broadly of adaptation against the background of the wider literature on climate change and adaptation in developing countries (Mertz et al 2009a, Berrang-Ford et al 2011, Fankhauser 2017. Complementing previous reviews on the region (Wiederkehr et al 2018, Borderon et al 2019, we use a combination of review and quantitative meta-analytical tools to analyze the relationship between environmental change, adaptation, and migration based on a systematic screening of the empirical literature. The focus of our study is placed on rural populations of farmers and (agro-) pastoralists living in drylands south of the Sahara. The meta-data collected as part of our study allows us to effectively compare household behavior across different geographical contexts and populations and to analyze migration against the background of other adaptation activities reported in the case studies.
A novelty from a methodological standpoint is that we combine evidence from both quantitative and more qualitatively oriented case studies, including studies using inference from household and expert interviews or focus group discussions. Most of the considered case studies have a small sample size and provide in-depth insights in the local and contextual settings of the considered communities. This distinguishes our work from previous meta-analyses (Beine and Jeusette 2021 that typically consider the environmentmigration relationship at more aggregate levels.
Our findings highlight the diversity of adaptation strategies used by households, ranging from agricultural in-situ strategies, to the diversification of income sources, the support from social networks, or changes in food provision and consumption. Also, migration as a way of adapting to environmental stress is found to play an important role in some of the communities considered, but not in all of them, depending on the livelihood and ecological context. Considering the use of migration against the background of other strategies, we find that migration is closely linked to and complemented by other adaptation activities. Using cluster analysis, we identify three distinct migration clusters that are characterized by different socioeconomic profiles and levels of vulnerability, suggesting that migration in response to environmental change can take various forms along a continuum from coerced to more voluntary forms of migration.
The remainder of the paper is structured as follows. Section 2 provides further information on the context in SSA drylands and discusses some of the pertinent challenges faced by local populations. Section 3 introduces the research design and literature basis used and presents our screening and study selection process. Section 4 presents the main findings, which are discussed in section 5. Section 6 concludes.

The context: drylands in sub-Saharan Africa
Drylands are commonly defined on the basis of the aridity index (AI), which is the ratio of annual potential evaporation to precipitation (Safriel et al 2005). Drylands are areas with an AI of 0.65 or less. They can be further divided into hyper arid (AI < 0.05), arid (0.05 ⩽ AI < 0.20), semi-arid (0.20 ⩽ AI < 0.50), and dry sub-humid (0.50 ⩽ AI < 0.65) zones. Higher aridity within drylands is typically associated with lower population density, with hyper arid zones being only sparsely populated. Drylands are important for food production in SSA, accounting for 66% of cereal production and 82% of livestock holdings in the region (Cervigni and Morris 2016). Livelihood strategies are centered around agriculture, with populations in drier areas being mainly involved in pastoralism and those in less arid areas relying on rain-fed cultivation.
Drylands are characterized by challenging agroclimatic conditions, including high precipitation variability and chronic water scarcity (figure 1). They are exposed to a number of environmental hazards and changes and are frequently affected by weather extremes and droughts, which can result in widespread crop and livestock loss (Huho et al 2011, Middleton andSternberg 2013). Although dryland inhabitants are adapted to their environments, the frequency and severity of climatic shocks is increasing under climate change (Diffenbaugh et al 2017) putting additional pressure on local livelihoods (Thornton et al 2011).
For pastoralists and agro-pastoralists, the shrinkage of grazing areas is a major problem and these mobile populations respond by travelling further distances. Farmers are affected by delays in rainy season onset, mid-season heatwaves and heavy rains resulting in crop loss and low yields (Connolly-Boutin and Smit 2016). Frequent exposure to environmental hazards has been linked with the depletion of household assets, particularly among the poor who are not able to smooth consumption (Carter et al 2005, Ansah et al 2020. Moreover, population growth and environmental degradation are fueling disputes in the sub-Saharan drylands. An increase in conflict between farmers and semi-nomadic herders has been documented due to the increased competition for resources between the two groups (UNEP 2011, Brottem 2016. Certain population subgroups are also socially and politically marginalized, which further exacerbates their vulnerability to climate change and conflicts (Fjelde and von Uexkull 2012).
Besides climate change, human activity is also speeding up land degradation in the area, which is expected to have far-reaching consequences for food security, economic activity, and population health in the region (United Nations Economic and Social Council 2007). Land degradation is often the result of unsustainable use of land by smallholder farmers who lack the capacity to invest in more sustainable practices (Shiferaw et al 2014). Consequently, land degradation increases vulnerability to environmental stress even further, which leaves subsistence farmers in a vicious cycle.
Given that populations in sub-Saharan drylands are directly dependent on the natural environment for their livelihoods and are frequently affected by severe weather events, poverty is pervasive in the region (Middleton et al 2011). In fact, droughtsensitive countries in SSA have seen an increase in the prevalence of undernutrition from 17.7% in 2010 to 21.8% in 2018 (FAO, IFAD, UNICEF 2019). In the past years, extreme events triggered several million internal displacements, most notably in Eastern and Western Africa. Across all regions, an increase in internal and international migration was also observable (supplementary figure S1).

Literature search and screening
The literature search for this study was conducted in two stages-a first stage in June 2017 and a second stage in March 2020 when newer studies were added to the initial selection (Wiederkehr et al 2018). We followed a stepwise procedure to identify relevant studies for our review. In a first step, a systematic search of studies was carried out on the scientific search platform 'Web of Science' , which provides comprehensive citation data for various academic fields. We used relevant keywords, including the names of all countries in SSA, terms related to environmental change and associated hazards, and the terms 'adapt' and 'cope' . More information on the specific search terms can be found in supplementary material B.
The entire search process yielded a preliminary selection of 3625 papers, which were further examined and filtered (see supplementary figure S2 for a PRISMA diagram). The filtering involved two steps. First, the relevance of papers was assessed based on their titles, keywords, and abstracts. In a second step, the full texts were thoroughly assessed for eligibility based on predefined inclusion criteria. We only considered peer-reviewed articles published in English.
The following inclusion criteria were used to determine the eligibility of studies. A paper was considered eligible, if it (i) comprised primary data from a local case study, (ii) focused on populations in predominantly rural and (semi-)arid areas characterized by subsistence livelihoods or small-scale agricultural activities, (iii) reported an environmental change process 7 , (iv) presented data at the household  Since some studies considered populations, which use a combination of different livelihood strategies, the total percentage reported exceeds 100%. Panel (b) shows the distribution of the sample size and panel (c) the methods employed by the case studies. As some studies use multiple methods, the pie chart summarizes the share over all methods used across studies, resulting in a total of 100%. level, (v) provided information on multiple coping and adaptation strategies (excluding preferred or planned actions), and (vi) provided information on the percentage of households in a study population adopting specific strategies.
The search and selection procedure resulted in a total sample of 89 individual case studies covering more than 21 thousand households in 22 countries (see supplementary table S1 for a full list of studies). Information on the case studies were retrieved from 53 distinct articles, which were published between 2004 and 2019. The studies were conducted in different regions in SSA and used a variety of research designs and methods (figure 2(c)). Household surveys, focus groups, and key informant or expert interviews were the most common data collection methods, often complemented by semistructured interviews and field or participant observations. Sample sizes ranged from 15 to 623 participants, with most studies based on fewer than 200 participants ( figure 2(b)).
The dryland locations covered by our sample of case studies are exposed to a range of environmental changes and hazards. Most commonly, studies reported changes 8 in the amount of rainfall (71.9% of studies) as well as more unpredictable and erratic rainfall (48.3%), increasing temperature levels (67.4%), and drought (41.5%) as major sources of environmental stress. Also, risks related to the degradation of land (58.4%) and water bodies (31.5%) were frequently mentioned. These include issues related to soil erosion (10 study cases), lake drying (4 cases), and the desiccation of floodplains (1 case). Other mentions relate to risks of flooding (18.0%) and stress related to wind (10.1%), especially increasing wind speeds (7 study cases) or the occurrence of dust storms (4 cases).

Data and measurement
From each article, we extracted information about the study location(s), period of data collection, other study characteristics and results, which became the basis of our meta-data. We use this information to compare result patterns across studies and explore whether these were influenced by contextual factors at the household and community level. We are primarily interested in the adaptation strategies used by households, which we classify in 14 broad categories. In each category, we calculated the percentage of households among the study population who reported the adoption of a specific type of strategy. Table S1 in the supplementary materials gives examples of adaptation measures reported in each of the 14 categories. Typically, the households reported the use of several measures belonging to one category at a time. This means that in certain categories several percentage values were recorded, the sum of which may exceed the value of 100%.
Based on the information retrieved from the studies, we constructed two main outcome variables for each of the 14 adaptation categories: (i) A dummy variable taking the value of one if any measures belonging to a specific category were reported in the study population and zero otherwise, and (ii) a continuous variable which corresponds to the share of households using the most common measure in a category. The latter measure provides a conservative lower bound estimate of the share of households using measures in each of the 14 adaptation categories. For our main analyses, we rely on the continuous outcome. Results using the dummy variable are reported in the supplementary materials (supplementary figure S3). Importantly, both the dummy and the continuous outcome variables are comparable across categories and across studies, which allows us to investigate what the most common forms of adaptation are, and in which context they are most likely to be used.
In addition to the measures described above, we retrieved information from each article about the mode of data collection, characteristics of the interviewed households, indications of environmental change processes, and other environmental and nonenvironmental factors relevant for the local context.
One key piece of information which was retrieved from the studies is the predominant livelihood strategy among interviewed households. We used this information to categorize study populations in three types: sedentary farmers, (agro-)pastoralists, and mixed populations who use a combination of both farming and (agro-)pastoralism (figure 2(a)). Among the different livelihood strategies, farming was the most common across study populations, with nearly 80% of households on average involved in some form of farming activities.
We aimed to collect the precise global positioning system (GPS) coordinates (latitude and longitude) of the study populations. Some studies did not report GPS coordinates, but we managed to determine the approximate location of study sites based on available information, for example the name of the district in which the study was carried out. Once the GPS coordinates of each study location were determined, we were able to join the meta-data retrieved from the articles with external data sources. In particular, we added information about the overall socioeconomic conditions in the region, which may influence the adaptation options available to households.
We retrieved information about the overall socioeconomic conditions from Demographic and Health Surveys (DHS)-large and nationally representative household surveys conducted in over 90 low-and middle-income countries, including most countries in SSA. The DHS data are representative at the subnational level, which is usually based on the census enumeration areas in the respective country (for example region or district). Various contextual information were obtained from DHS and aggregated at the lowest representative enumeration area for the respective DHS survey. In particular, we calculated: (i) the percentage of households employed in agriculture, including both farmers and (agro-) pastoralists, as a proxy for agricultural dependence, (ii) the percentage of households with access to electricity and with a finished (non-natural) floor as wealth proxies, and (iii) the percentage of households with secondary education and higher as a proxy for adaptive capacity related to education. To assign accurate background information to the case studies, we used DHS data from the same period when the data for the original studies were collected. If studies were missing information about the exact data collection period, we used the publication year as reference point.

Analytical tools: identifying adaptation patterns across studies
We employed a range of analytical tools to identify adaptation patterns across the studies included in the analysis. The texts were comprehensively screened and relevant data concerning study context, methods, and findings were extracted into a spreadsheet. The distribution of adaptation patterns across studies was explored and described using quantitative methods and descriptive graphs for illustration. In addition, we reviewed and coded the texts to contextualize the findings from the studies and to provide an in-depth perspective and examples on how the study populations responded to environmental changes. Here, we primarily identified common protective and adaptive activities employed by households as well as challenges reported.
In this study, we are particularly interested in migration and how it is used in combination with other strategies depending on the ecological and socioeconomic context. The primary goal is not to consider migration in isolation, but to understand its links to adaptation in a community more broadly. To this end, we consider differences in response patterns conditional on characteristics of the local populations represented by the studies. Matching case study locations with external socioeconomic data derived from DHS allows us to complement the information provided in the studies with wider contextual measures and effectively compare the different study contexts with each other.
Cluster analysis is used to identify adaptation clusters within the case study sample, i.e. populations using similar strategies, and to determine the socioeconomic conditions associated with certain combinations of adaptation activities. For the clustering, we use a centroid-based k-means clustering approach, which partitions the observed study cases x into a set C of k different clusters so as to minimize the within-cluster sum of squares (WCSS), i.e. the difference of each observation from the mean µ i within each cluster C i .
The number of clusters k is defined prior to the clustering and we employ various tests to identify the optimal number in our application (see supplementary materials E). x represents a vector of the percentage of households in a study sample using the different adaptation strategies. The cluster analysis is performed in R using the cluster package (Maechler et al 2021) and the Hartigan and Wong algorithm (1979). The algorithm assigns data points to a randomly created initial set of clusters based on their difference to the cluster centroids and calculates changes in the WCSS (1) if the data point were to be included in another cluster. If the resulting WCSS is smaller than in the original assignment, the data point is assigned to this new cluster. This iteration continues until the assignment is stable, i.e. no more improvements in the WCSS criterion can be reached by changing cases between clusters. In this situation, any change would make the clusters more internally variable or more externally similar and thus increase the WCSS.

Households rely on diverse adaptation strategies
Households use a diverse range of strategies to cope with and adapt to the environmental changes observed (figure 3). Agricultural strategies, including measures related to crop cultivation (41.2%), livestock keeping (24.4%), and soil and water management (21.5%), are clearly the most common, likely reflecting the predominance of sedentary farmers in our sample of studies. This also indicates a tendency of households to adapt in-situ and continue with agriculture rather than completely abandon cropping or herding. The reliance on in-situ activities could be due to socio-cultural reasons that induce people to stay in their current place of residence, such as place attachment (Vinke et al 2020), or it could be due to barriers and constraints in mobility (Cattaneo and Peri 2016).
The agricultural strategies cover a broad range of activities further highlighting the diversity of adaptation measures used by rural households. As part of the most commonly reported category, crop management, households report changes in their use of fertilizer and compost (e.g. Barbier et al 2009, in Burkina Faso;Mertz et al 2009b, in Senegal), the use of more resistant crop varieties (e.g. Tambo and Abdoulaye 2013, in Nigeria;Okpara et al 2016, in Chad), the diversification of crops (e.g. Gebrehiwot and Van Der Veen 2013, in Ethiopia;Antwi-Agyei et al 2014, in Ghana), changes in crop rotation (e.g. Padonou et al 2014, in Benin;Yila and Resurreccion 2014, in Nigeria), and modifications in the farmland (e.g. Snorek et al 2014, in Niger;Tesfaye and Seifu 2016, in Ethiopia).
A broad variety of other (non-agricultural) strategies are also used by the households. These include short-term strategies, such as the search for aid (8.7%) or support within the social network (13.7%), or financial measures, such as taking loans or dissolving savings (7.6%). Changes in food consumption and provision are also often reported (15%). In many study contexts, these changes reflect adaptation challenges and heightened levels of vulnerability of affected populations. Some studies report changes in diets, such as an increase in the consumption of wild plants or fruits (Osbahr et al 2010, in South Africa;Bola et al 2014, in Zimbabwe) or low quality famine foods (McKune and Silva 2013, in Nigeria).
Others find that households reduce their food consumption to cope with environmental stress, e.g. by eating less (Silvestri et al 2012, in Kenya), eating In some communities, households were forced to sell their assets and livestock to buy food (Chianu et al 2004, in Nigeria;Yaffa 2013, in Gambia) or relied on external support and food aid (Silvestri et al 2012, in Kenya;Ariti et al 2015, in Ethiopia).
Of all households considered in the studies, on average 7.8% reported to not have used any adaptation strategy. While this might reflect lower levels of exposure to hazards, it could also indicate the presence of adaption barriers. For the case of Ethiopia, Gebrehiwot and Van Der Veen (2013) show that lack of information on adaptation measures and insufficient access to finance are two important factors inhibiting adaptation to climate change (see also Tambo andAbdoulaye 2013, for Ghana, Gbetibouo et al 2010, for South Africa).
The availability of information and experiences influence whether and how people perceive environmental risks and how they respond to them. At the same time, adaptation requires financial and other resources, which might not be accessible to all households, e.g. because of insufficient access to agricultural inputs, technologies or credit markets. To better understand household responses to environmental changes, there is a need to consider the wider context in which they take decisions, including their perception of environmental hazards and potential resource constraints (Koubi et al 2016).

Migration is common and used in a variety of ways
Besides agricultural strategies, migration is among the most common strategies reported by the interviewed households. Across all case studies, nearly one out of four households stated to have used some sort of migration as a means to deal with environmental change. As such, it constitutes an integral part of the adaptation options used by rural households in SSA. The in-situ diversification of income sources, which often goes hand in hand with migration (Wuepper et al 2018), was reported by 22.1% of households. In our classification, we treated in-situ diversification and migration as two separate categories to distinguish them in the analysis. Migration can serve the purpose to diversify income sources ex situ, which was also mentioned as an important motive for migration by several of the reviewed studies (Osbahr et al 2010, Dumenu andObeng 2016).
Of the 17 case studies with very high levels of migration (mentioned by >50% of all households), 12 (70.6%) were conducted in the Sahel region. Typically, the migration patterns reported by households in these studies were internal within the same country or region. For example, vulnerable communities in Northern Ghana had a long history of migration to Southern Ghana, especially by young men, to engage in farm or off-farm wage labor (Antwi-Agyei et al 2014, Ngwese et al 2018). A similar pattern was observed in a study from Northern Burkina Faso, where household members primarily migrated to the South of the country where land was still available or to neighboring Ivory Coast to work on cocoa plantations (Barbier et al 2009). In Niger, households reported the neighboring countries of Libya and Nigeria as primary migration destinations (McKune and Silva 2013).
Seasonal migration for a restricted time was very common among the study populations. In Chad, Botswana, and Burkina Faso, for example, households were found to travel to resource-abundant areas for collective livestock grazing or fishing (Motsholapheko et al 2012, Okpara et al 2016, highlighting the importance of mobility for communities depending on herding and transhumance (see also section 4.3). In Kenya, Turkana herders frequently moved across borders, especially to Uganda, South Sudan, or Ethiopia, to access resources and markets (Opiyo et al 2015). Similar patterns of livestock-related seasonal mobility were reported for Ethiopia (Berhanu and Beyene 2015) and Zimbabwe (Mubaya and Mafongoya 2017).
While temporal forms of mobility were found to be common in the considered sample of studies, also more permanent forms of outmigration were used. In Burkina Faso, for example, the majority of study participants considered permanent migration to other parts of Burkina Faso or to Ghana/Ivory coast a viable strategy if climate change impacts worsen (Zampaligré et al 2014). New international destinations are also becoming more attractive for migration from the region. For example, international labor migration from Ethiopia and other Eastern African countries to the Arab states of the Persian Gulf is a growing trend (Mersha and Van Laerhoven 2016).
If migration is used, it is typically only individual household members who migrate and not the entire household (Smucker and Wisner 2008, Barbier et al 2009, Dumenu and Obeng 2016. In the sub-Saharan context, migrants are usually young men who migrate for economic reasons to find work or to engage in seasonal jobs. In search for better opportunities, many migrate internally toward urban centers (Clement et al 2021), which is also reflected in several of the case studies (e.g. Yaffa 2013, in Gambia;Dumenu andObeng 2016, Kumasi et al 2019, in Ghana). This contributes to the rapid urban growth observed in SSA, which is the world region with some of the fastest growing cities worldwide (Hoffmann and Muttarak 2021).
Remittances play an important role for mobility in the area and migrants usually maintain close ties to their home regions. Even if migration in the considered studies is not explicitly framed as 'economic' , it is often mentioned in association with job search or remittances (Snorek et al 2014, Yila and Resurreccion 2014, Ngwese et al 2018. This also highlights the multicausal nature of mobility, which is typically influenced not only by one, but a range of different economic, sociopolitical, demographic, and environmental factors (Black et al 2011a).

Livelihoods and ecological conditions shape migration patterns
The relationship between environmental factors and migration is not uniform but depends on the local livelihood and ecological conditions (figure 4). While in-situ adaptation strategies, such as changes in cultivation practices or soil and water management, are most often reported in studies focusing on farmer populations, migration is clearly most common among (agro-)pastoralists. In those highly mobile communities, the role of environmental change processes has to be understood against the background of existing forms of mobility (e.g. transhumance), which are influenced by a range of factors (Boas et al 2019).
These populations are also more likely to report diversification (38.8%) and changes in food consumption (37.3%) as ways to cope and adapt when confronted with environmental changes and hazards (figure 4(a)). In many instances, migration in search for better economic opportunities were also reported. In Ngamiland, Botswana, for instance, high percentages of respondents stated to migrate to towns for wage labor in the context of pastoral rangeland degradation (Basupi et al 2019). Similarly, a study from Kenya mentions out-migration of agro-pastoralists in search of paid work (Smucker and Wisner 2008). There might be several reasons for the higher prevalence of migration among (agro-)pastoralist households, which we discuss in greater detail in section 5.
In rare cases, households reported the reduction of mobility (4.2%) as a way to deal with an environmental hardship (McKune and Silva 2013). This strategy was again most commonly reported by pastoralist populations, who might have had to settle temporarily if they were unable to sustain their livelihood with a nomadic lifestyle. This is also exemplified by Basupi and colleagues (Basupi et al 2019), who report an increasing dependency of pastoralists on social welfare programs due to the high uncertainty of pastoral income (resulting i.a. from livestock diseases and restricted access to productive rangelands) and lacking alternative income sources.
Whether or not migration is used by communities in the drylands is found to also depend on the reported environmental changes ( figure 4(b)). Most commonly, households listed migration as a strategy in the context of land degradation (28.4%), rainfall variability (27.8%), flooding (24.7%) and drought (23.8%). All of these stressors are directly linked to agricultural production and can be particularly destructive in dryland areas characterized by overall low precipitation and high temperatures (Falco et al 2019).
Typically, communities in the studies were exposed not only to one, but to multiple changes and hazards in their environments. Environmental impacts are not independent of each other, but are closely correlated and can co-occur . In additional analyses, we find that certain types of environmental stressors were often jointly reported by the case studies (supplementary table S4 and figure S4). For example, changes in rainfall amounts were often mentioned together with challenges arising from changing temperature levels, and drought hazards with changes in rainfall variability, land degradation, and flooding. Especially the cooccurrence of different stressors requires a combination of strategies to adapt to the changes (adaptation mix). With increasing levels of environmental stress, a threshold may be reached beyond which in-situ adaptation may no longer be sustainable for local populations (Warner et al 2012, Dow et al 2013, Xu et al 2020.

Households use a combination of adaptation strategies
The individual adaptation strategies are rarely used in isolation. Instead, households use a combination of different strategies to adapt to environmental changes and hazards ( In one community in Burkina Faso, for example, a large share of households used combinations of changes in cultivation practices, livestock, and soil and water management (Mertz et al 2012), whereas another community in Ghana more heavily relied on income diversification, social contacts, adjustments in food consumption and external aid in combination with migration (Antwi-Agyei et al 2014). Although the papers reviewed here indicate how many strategies were reported in total, they do not provide specific information on how many households actually applied several or all of them simultaneously. This still constitutes a missing piece of information to better understand the dynamics between different strategies, including migration (see also Wiederkehr et al 2018).
To understand households' responses to environmental stress, strategies have to be considered in a wider context of adaptation behaviors as these are closely linked to each other and correlated. In the literature, there is evidence that households tend to adopt several strategies simultaneously (Silvestri et al 2012, Mogotsi et al 2013, Hooli 2016, Tesfaye and Seifu 2016, Hermans and Garbe 2019. These strategies can either complement or substitute each other. Figure 5(a) shows the correlation of the share of households using different adaptation strategies across the sample of studies. Some of the strategies are highly positively correlated and seem to complement each other. For example, changes in crop management are often named in combination with soil and water conservation measures (but not livestock), and the search for social support is often named together with the diversification of income sources, changes in food consumption and provision, and external aid.
In our explorative analysis, strategies are found to be in most cases either uncorrelated or positively correlated suggesting positive complementarities. shows the percentage of households mentioning migration as an adaptation strategy given that also one of the other strategies (displayed in the rows) was reported to be used by the households in the case study. The bars hence indicate the percent of households using migration as a strategy if the other strategies were simultaneously used in the study population.
Likewise, based on data from surveyed households in Ghana, Tambo (2016) states that the majority of analyzed strategies act as complements, not substitutes. For Kenya, Tongruksawattana and Wainaina (2019) identified farm adjustments as a substitution strategy for selling assets and borrowing for drought adaptation, whereas a complementary reduction in consumption was observed for households who were forced to borrow money. Eriksen et al (2005) conclude that during periods of environmental distress households usually adopt one principal adaptation strategy that is complemented by other secondary measures.
Migration, like the other strategies, is only one of many potential responses to environmental stress and has to be analyzed against the background of other strategies and activities of the households (McLeman and Smit 2006, Perch-Nielsen et al 2008, Piguet 2010. Figure 5(b) shows the average percentage of households mentioning migration conditional on whether other strategies were used simultaneously by the households. Migration is commonly reported with other strategies, such as the search for social support, relying on humanitarian aid, changes in food consumption and provision, search for information, the diversification of income sources, and livestock management. On the other hand, we do not find a strong correlation between migration and the use of other more sedentary forms of agricultural adaptation, such as changes in cultivation and soil and water management. This also mirrors our findings on livelihood-specific differences in adaptation choices addressed in section 4.3, with migration being more common among (agro-)pastoralist households.

Identifying adaptation clusters
Focusing on the nine most common adaptation strategies, we conducted a cluster analysis to identify clusters of study populations using similar strategies when confronted with environmental change and hazards ( figure 6). For this, we follow a stepwise procedure to determine the optimal number of clusters and explore the resulting cluster assignment (supplementary materials E). Our cluster analysis reveals five distinct adaptation clusters from a total of 89 cases. The analysis allows us to explore cases with a similar adaptation profile and to reveal the interconnectedness of different strategies employed in the surveyed communities. It also illustrates patterns and differences in adaptation across the studied settings. Figure 6(a) shows the distribution of adaptation strategies across the clusters with higher values representing a greater share of households in a cluster using a strategy. Figure 6(b) shows the corresponding narrative and size of each cluster and figure 6(c) combines information about the location of the clusters with background information about the local socioeconomic conditions to explore under which conditions certain clusters are most likely to occur. Here, we consider the average percentage of the population in a cluster with at least secondary education, a finished floor, access to electricity, and an agricultural occupation. The latter measure accounts for both the share of farmers and (agro-) pastoralists in the population, which are not distinguished in the DHS.
Three of the considered clusters (1-3) are migration clusters with an increased share (median share > 25%) of households using migration as an Figure 6. Identifying adaptation clusters. Panel (a) shows differences in employed strategies across cluster groups identified with a k-means cluster analysis. Panel (b) describes the corresponding narrative of the common adaptation strategies used and the size (N) of each cluster, and panel (c) shows differences in socioeconomic conditions between the clusters. Here, the bars show the mean share of the population in a cluster with at least secondary education, a finished floor, access to electricity, and an agricultural occupation. The latter measures represent averages over all regions belonging to a cluster. adaptation strategy. The geographic location of the migration clusters is shown in figure 7. As can be seen from this figure, clusters with high levels of migration (filled symbols) can be primarily found in the western and central Sahel region as well as in selected locations in East and Southern Africa. While all three migration clusters can be characterized by increased levels of human mobility, they are distinct in their use of complementary strategies. A form of adaptation that is closely related to migration are activities related to the diversification of income sources. For all migration clusters, increased levels for this strategy can be observed, indicating a complementary relationship between both strategies.
The important role of context and socioeconomic influences is apparent in all results. Whereas migration is found in both areas with a higher and lower level of development, differences are visible in the combination of strategies used. In particular, high levels of poverty, limited access to education and information, lacking social protection, inequality and uneven development processes, weak institutions, and limited access to financial, labor, and agricultural markets have been identified in the case studies as major inhibitors to achieving climate resilience at large (Ng'ang'a et al 2016, Pauline et al 2017).
The first 'Mixed Strategies' cluster (n = 20 case study populations) uses migration in combination with a wide variety of other strategies, ranging from attempts to diversify income sources, to search for external support via social networks or aid organizations, or to engage in other adaptation activities. Also changes in food consumption, e.g. changes in diets or reductions of food intake, are reported by a considerable share of households in this cluster (median > 25%), which we consider as a sign of populations facing challenges in their adaptation, potentially contributing to increased vulnerability. Indeed, considering the socioeconomic profile of this cluster, it is most commonly found in more deprived areas where only a very low percentage of the population has access to basic infrastructures, such as education and electricity, and where livelihoods strongly depend on agricultural occupations.
The second 'Agriculture & Migration' cluster (n = 17) uses migration in combination with different agricultural strategies, both relating to changes in cultivation, soil and water management, and livestock keeping. Also attempts to diversify income sources play a role, but to a lesser extent. Unlike the first cluster, this cluster is characterized by a relatively higher socioeconomic status with on average more than 50% of populations in the regions having completed secondary education, having access to electricity, and having a finished floor. Also, only a comparably low share of the communities in this cluster directly depend on agriculture, suggesting greater opportunities to diversify income sources to adapt to environmental change. In this context, migration is often described as a complementary strategy used to support the household income, including mobility in search of wage labor (Smucker and Wisner 2008 in Kenya), seasonal mobility (Rasmussen 2018 in Burkina Faso), and herding-related mobility (Berhanu andBeyene 2015, Ng'ang'a et al 2016, in Ethiopia).
The third 'Mainly Migration' (n = 7) cluster represents populations where a large share of households report using primarily migration as a form of adaptation. While in some communities also agricultural strategies are employed, the majority of communities in this cluster do not use any of these strategies (median = 0). The only notable exception reported by the households are strategies related to food and nutrition. On average, more than 60% of households indicated that they changed their diets or made additional efforts to obtain food to react to environmental changes, which might reflect limited in-situ adaptive capacities and increased levels of stress. Like the first cluster, this cluster is characterized by a weaker socioeconomic profile, suggesting a close relationship between human development, vulnerability, and adaptation. Unlike the second cluster, agricultural dependence is high and migration and changing food consumption seem to serve as strategies of last resort here. While this observed pattern can be found in only few of the considered study populations, it hints toward a reduced ability of certain populations to withstand environmental pressures and a lack of viable adaptation options, making migration the only suitable strategy to deal with environmental hardships. While migration is described as an adaptation mechanism by study populations, it can also come with significant long-term repercussions for migrants and their families, for example negative health impacts (Yaffa 2013, in Gambia) or increased divorce rates (McKune and Silva 2013, in Niger).
The fourth 'Mainly Agriculture' (n = 35) cluster is the biggest of all clusters. Households in this group primarily rely on in-situ agricultural adaptation strategies, such as changes in cultivation and soil and water management. Migration, on the other hand, is not found to play a major role. The focus on agriculture in adaptation is also reflected in the background variables, which reveal that regions belonging to this cluster are characterized by a high agricultural dependence with on average more than 50% of the populations having an agricultural occupation. In general, the cluster is characterized by a weaker socioeconomic profile with less than 30% of the populations having completed secondary education and having access to electricity. Most commonly, households in this cluster use a combination of crop and soil and water management techniques to adapt to environmental changes. For example, households in a study in Burkina Faso (Mertz et al 2012) were found to use a combination of manure application and fertilization methods together with soil and water conservation, reforestation, improved irrigation, and vegetation protection (e.g. stone bunds, fallow periods) methods.
The final 'No Adaptation' cluster (n = 10) is a cluster where only very few households make use of the considered adaptation strategies. Communities belonging to this cluster are characterized by a relatively high level of education (>50% with secondary), comparably good access to electricity (>40%), and well-developed housing structures (>70% with finished floor). While all case studies included in this cluster were primarily focused on crop and livestock farmers, the considered communities were located in regions with an overall low agricultural dependence. All of the above suggests a relatively low vulnerability to environmental stressors and, hence, a reduced need to take action in terms of coping and adaptation. In addition, the environmental change processes reported in these studies were mostly related to perceived gradual changes in temperature and rainfall patterns, which were addressed by farmers-if at all-by using in-situ adaptation responses, such as modifications in farming practices (e.g. Gbetibouo et al 2010, in South Africa;Tambo and Abdoulaye 2013, in Nigeria). Migration as an adaptation strategy was rarely reported. While again only few of the considered communities show this adaptation pattern, it is worthwhile to emphasize that there are also communities in the surveyed sub-Saharan drylands that have no need to adapt to environmental changes, either because of a limited exposure to stressors or reduced susceptibility to their impacts.
The different combinations of strategies used by the five clusters shows the diversity of adaptation approaches in sub-Saharan drylands. Despite this diversity between clusters, similarities can be found for communities belonging to the same cluster. Common adaptation patterns exist, where households are found to employ a combination of certain complementary strategies to adapt to environmental changes. Our indicative findings suggest that these patterns vary by the socioeconomic conditions faced by the populations and possibly also the nature of the environmental change experienced. While a more rigorous analysis of contextual influences is challenging here given the limited scope of our data, we believe that a more careful exploration of these impacts and their role for the evolution of adaptation patterns would be a fruitful direction for future research.

Discussion
As our analysis underlines, migration is only one of many potential responses to environmental change and needs to be considered in the broader context of other coping and adaptation strategies, including potential in-situ adaptation alternatives (Kniveton et al 2011, Gharad et al 2014, Cattaneo et al 2019. The most common strategies adopted by rural households relate to crop and livestock management. Migration is relevant, but usually does not rank first. Furthermore, our results indicate that while migration is an important strategy in some of the considered communities, it is not common in others. There is no determinism linking environmental change and migration and not everyone affected is able to or desires to move (Wiederkehr et al 2018). Instead, a range of other factors come into play shaping adaptation processes and migration responses to environmental stress, including local agricultural conditions (Feng et al 2010, Schlenker and Lobell 2010, Bohra-Mishra et al 2017, adaptive capacities (Bardsley andHugo 2010, Vinke et al 2020) and possibilities for income diversification (Mertz et al 2009b, Garcia et al 2015. While some households in a community may change their mobility behavior, others remain immobile, either because they do not want to move, have no incentive or motivation to do so, or are restricted in their mobility. These diverse patterns also underline yet again the inaccuracy of simplistic and alarmist narratives of mass migration due to climate change (Boas et al 2019).
The cluster analysis results, which bring together a large amount of empirical case study insights and external demographic and socioeconomic data, indicate that households use migration as an adaptation strategy in situations of both high and low social and ecological vulnerability. This shows that migration may occur due to high migration pressure resulting from a high environmental risk level and lack of alternatives, but may also be a proactive choice by better-off and less exposed households. It is crucial to bear in mind that, depending on the respective context, migration can take different forms in terms of distance, time frame, level of agency and outcomes. While the case studies highlight the important role of temporary internal migration and seasonal moves as common mobility patterns, there is considerable diversity with households in some communities engaging in international and more permanent forms of migration (e.g. Mersha and Van Laerhoven 2016). In communities characterized by already mobile lifestyles, such as pastoralists, changes in mobility patterns are a common way to respond to environmental hazards.
Despite a range of potential benefits ensuing from migration (Tacoli 2009, Ng'ang'a et al 2016, scholars also address possible adverse and often unforeseeable consequences for the migrants themselves, their households, and their communities, such as a weakened local labor force and loss of solidarity (Mertz et al 2009a, McKune and Silva 2013, Yaffa 2013, Mersha and Van Laerhoven 2016. Tebboth et al (2019) suggest that the level of agency in the migration decision has a decisive impact on the resilience of households. However, there is still a need to better understand and conceptualize migration as an adaptation strategy to environmental change as well as to define what constitutes socially and ecologically sustainable migration and adaptation for whom (Gemenne and Blocher 2017, Ayeb-Karlsson et al 2019, Oakes 2019, Zickgraf 2019. The exploration of different types of migration, their drivers, and outcomes represents an important direction for future research. The descriptive results indicate that different types of mobility are particularly relevant for (agro-) pastoralist households in sub-Saharan drylands. This may be due to different reasons. Pastoralists are known to be among the most vulnerable population groups. Due to their mobile lifestyle, they are often marginalized, without secure land tenure, difficult to reach with food aid or extension services, and are viewed with suspicion by governments and farmers (Schilling et al 2012, Snorek et al 2014, López-i-Gelats et al 2016. Some studies show that they are generally well adapted to dryland environments and to highly variable rainfall conditions, with mobile herding and transhumance representing a long-term adaptation strategy (e.g. Opiyo et al 2015).
However, long agricultural droughts can deplete soil moisture and reduce the availability of natural pastures for grazing livestock, which can lead to livestock loss and can be devastating to pastoralist communities. Under these conditions, migrating beyond their usual grassing lands might be the only viable strategy to ensure survival and expose them to additional risks, including conflicts with other groups (Snorek et al 2014). Given their already mobile livelihoods, the threshold for engaging in migration rather than in-situ strategies might be lower for pastoralists than for farmers whose cultural identities are closely connected to their land and agricultural activities (Devine-Wright and Quinn 2020).
Different forms of hazards have differential implications for local livelihoods and require different forms of adaptation (Gemenne 2011, Black et al 2011a. Local contexts and conditions strongly shape what constitutes an environmental hazard and influence inter-dependencies with other factors. Migration impacts are typically non-linear with hazards affecting migration only after reaching a certain threshold beyond which the pressures become too strong for the system to resist or adapt (Schlenker and Roberts 2009, Burke et al 2015, McLeman 2018. Studying environmental impacts on migration hence requires a contextual understanding of local conditions and how they affect households' capabilities to respond to changing environmental conditions. As we show, migration appears to be particularly important in land degradation contexts, which are also often simultaneously affected by other hazards such as droughts or floods. However, why-or rather how-remains largely unclear (e.g. does it trigger certain migration types? Is it a strategy applied after in-situ measures to conserve and restore soils have failed?). In contrast to climatic and socio-economic migration drivers, the role of land degradation in migration processes still constitutes a major research gap (Hermans and McLeman 2021).
Our study faces different limitations which are important for the interpretation of our results. Given that this review is based on the published scientific and English-speaking literature, a certain level of publication bias needs to be acknowledged. Focusing on a narrower set of studies allowed us to manage the large scientific evidence base on this topic and to ensure a sufficient level of comparability and quality. The study areas are spatially unequally distributed, meaning that some countries such as Ghana, Ethiopia or South Africa figure prominently in our sample whereas Côte d'Ivoire, Mozambique, or Somalia, amongst others, are not included (see also Hendrix 2017, Piguet et al 2018. Moreover, the predominant focus on small-holder farmers in the literature on responses to climate change and variability in SSA means that fewer data are available on herder communities (see also Shackleton et al 2015).
Due to limitations in the reporting of the original studies, we were not able to draw a clear distinction between whether households in a community used a specific activity as a shorter-term coping response to a threat, or as a longer-term adaptation strategy (see also Wiederkehr et al 2018). Moreover, the majority of studies considered here does not report whether households were successful in addressing the respective challenges by employing a certain strategy. Also, while the surveyed studies focus on a range of coping and adaptation behaviors, several activities may have gone unnoticed by the researchers because of selected interests and prioritizations in the studies, a stigmatization of certain activities, and underreporting.
Our analysis was carried out at an aggregated meta-analytical level and hence does not allow us to further study the underlying mechanisms and to derive causal conclusions. A more careful exploration of some of the processes at more granular levels would be useful. For example, while we see patterns showing up in our cluster analysis pointing towards certain forms of migration being more dominant in a cluster, we are unable to clearly point to the concrete conditions and migration motivations faced by the households within each cluster. In addition, some of the used data are noisy, for example when it comes to determining the exact geographic location of communities, which is especially challenging for mobile nomadic communities. Despite of the explorative character of our analysis, it provides a comprehensive overview of central findings in the literature and yields important insights into the role environmental changes and adaptation play for migration and the importance of contextual influences in shaping the processes.

Conclusion
Drawing from comprehensive empirical data from 89 case studies covering more than 21 thousand rural households, our review study illustrates the diversity of ways communities in sub-Saharan drylands deal with environmental change and hazards. Measures related to agricultural and livestock management were identified as most common, followed by migration, income diversification, and soil and water conservation. Most importantly, our findings show that the relevance of migration as adaptation strategy depends on the respective livelihood and ecological context. Migration was found to be more prominent among (agro-)pastoralists than farming households, and was reported most often in the context of rainfall variability, land degradation, droughts and floods.
Our results indicate that migration is closely linked to and complemented by other strategies, in particular livestock-related forms of adaptation, the diversification of income sources, the search for social support, and changes in food provision and consumption. This corroborates that a holistic perspective, which considers migration in connection with other adaptation strategies, is indispensable to advance research on environmental migration. Our cluster analysis reveals three distinct migration clusters shaped by different socioeconomic profiles and varying levels of vulnerability. This confirms that environmental migration occurs in contexts of both low and high social and ecological vulnerability. In other words, depending on the circumstances, migration can be a strategy of last resort due to high environmental stress and lacking in-situ alternatives or a proactive choice of individuals and households with potential positive synergy effects with other strategies, increasing household resilience.
Policy interventions should generally be designed to increase the agency of vulnerable population groups and aim to facilitate movement of those who wish to migrate and to assist in-situ adaptation of those who wish to stay. Empowering households exposed to environmental risks to take selfdetermined decisions will be key to increase the adaptive and beneficial potential of migration. Moreover, our results underline the necessity of carefully tailoring policy measures to the respective livelihood and ecological context given the range of situational needs and household capacities. Migration as an adaptation strategy itself is embedded in a range of other activities carried out by households that should likewise be a focus of holistic policy responses to changing environmental conditions. Climate change will have major implications for drylands in SSA. The development of foresighted and anticipatory policy instruments and the strengthening of the adaptive capacities and capabilities of vulnerable populations are hence key to prevent and mitigate negative impacts in the future (Gemenne andBlocher 2017, Cattaneo et al 2019).

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors. The full list of studies considered in this review are displaced in supplementary table S1.