We selected ten study sites in mountain regions (both mountains and highlands, as defined by ref. 13) covering a wide range of ecological contexts (for example, different elevation or annual rainfall), socioeconomic contexts (for example, different livelihood strategy or market access) and political contexts (different countries). Site selection was also affected by security situation on the ground (for example, ongoing conflict in Anglophone Cameroon) and previous engagement in the area by local partners facilitating fieldwork. In each study site (Extended Data Fig. 1), four villages located at different elevations were selected. These villages were selected by local partners facilitating fieldwork based on accessibility, given the limited resources available for this research. In each village, we first conducted exploratory FGDs with four or five elders. After we explained the aim of the study to the village chief, he explained it to the elders (mostly male, typically >60 years of age), and some decided to participate on a voluntary basis. These FGDs were used to adapt a common semistructured questionnaire to each study context and to build trust among community members. The common (for all ten sites) semistructured questionnaire included a long list of potential (1) climatic changes observed, (2) impacts in the biophysical domain and (3) adaptation responses (from ref. 50) which were narrowed down to those relevant for each study area, according to FGDs participants. During the FGDs we also gathered information on agents of change promoting adaptation responses in the village (the government, NGOs or local communities without external support) and on perceived constraints for further adaptation.
Then, in the same villages, we conducted semistructured questionnaires to 37 or 38 randomly selected households aiming to interview about 50% males and 50% females of the main decision-making couple (if more than one generation lived together) (n = 150 in total per study site). In each village, households were selected by walking the main road (or footpath as defined by local inhabitants) and selecting every third household to the right. If the household head was not available, the next-door neighbour was targeted. We first interviewed the household head who opened the door (male or female), until we reached the targeted sex quota for that village, and then we asked to interview the other sex in the subsequent households. We acknowledge that there are preferred methods for selecting households (for example, randomly from a list), but a register of households was unavailable in several study sites. The ‘main road’ approach might have led to interviewing richer households in more market-integrated contexts (for example, in Mt Kilimanjaro). As the main focus of our research was on differences across sites (and not within sites), we consider this a minor issue, but future research should investigate differences across households within study sites.
The questionnaires used addressed household characteristics and assets, climatic changes observed, impacts in the biophysical domain, adaptation responses used to cope with or adapt to observed changes and impacts (Supplementary Information). They also included climate change literacy, defined as a combination of climate change awareness (having heard of the concept of climate change) and the knowledge and acceptance of its anthropogenic cause. Climate change literacy, combined with climate information services that are demand driven and context specific (for example, for agriculture) can be the difference between coping and informed adaptation responses51.
The methodological approach and the questionnaire used follow the guidelines of the project ‘Local Indicator of Climate Change Impacts’, a project focused on providing data on the contribution of local and indigenous knowledge to climate change research50. We adjusted the framework proposed by ref. 52, in which changes in the climate itself and the effects of climate change observed (in the physical, biological and social systems) are differentiated. We adhere to the Framework Convention on Climate Change14 and use ‘climate change’ to refer to a change in the state of the climate that can be identified by changes in the mean and/or the variability of its properties, and that persists for an extended period. Similar to ref. 8, we use the term ‘local perceptions of climate change‘ to refer to reports provided by local peoples about changes in the climatic system (temperature, precipitation and wind).
The exploratory FGDs and the household questionnaires were carried out in the languages Ngombale (Bamboutos), Rukiga (Kigezi Highlands), Kinyarwanda (Nyungwe), Kirundi (Kibira), Oromo (Bale Mountains), Swahili (Itombwe, Mount Kenya, Aberdare, Mount Kilimanjaro and Udzungwa) and were facilitated by some co-authors between November 2020 and January 2022. All study participants (FGDs and household questionnaires) were selected on a voluntary basis and were first informed that the study aimed to better understand climate change impacts and adaptation practices. Free, prior and informed consent was orally secured after reading a consent form in the local language, which clarified the study aim, voluntary participation, confidentiality and procedure for withdrawal from the study.
In each study site, data gathering was led by a researcher from the same ethnic group studied, who had previously worked in the study area targeted: someone who could be considered an insider. Because of this, and also because of the use of a standardized questionnaire and the engagement in reflexive practice during eight webinars used to coordinate results interpretation across sites, we consider that researchers’ positionality across sites was rather uniform. Owing to the predominance of agriculture-based livelihoods and historically sedentary settlements and culture, throughout the paper we refer to our study respondents as farmers, but we acknowledge multiple livelihood strategies. We also refer to our study respondents as subsistence-oriented farmers, because even if some cultivate cash crops (coffee; Table 1), the proportion of their farms allocated to coffee is smaller than the proportion allocated to staple crops.
To investigate differences across study sites, the main unit of analysis was percentage of respondents per study site. Initially, we explored differences in the responses within one study site related to sex of the respondent using paired t-tests but these were non-significant, probably because most of the females interviewed were married and were not female-headed households (those without a husband or adult male relative living with them). Thus, we do not include sex-based analysis in this manuscript. We also investigated if: (1) perceiving more climatic changes or (2) household wealth, influenced adaptation responses, using mixed-effects models. For each study site and respondent, we calculated the proportion of potential climatic changes, impacts and adaptation responses reported. Changes, impacts and adaptations that did not apply to a site (for example, reduction in frost in sites that would not normally experience frost) were excluded from the calculation of proportions. We used hierarchical models to evaluate within and between site variation in adaptation responses. To do this, we fitted linear mixed-effects models using the lme4 R package v.1.1-31 (ref. 53) which modelled the proportion of adaptations as a function of the proportion of climatic changes and household wealth category as fixed effects, study site as a random effect, with both proportion of climatic changes and household wealth allowed to vary among random effect levels (fitting a random slope model). This treatment was especially important for wealth, as it is a relative index for each site so categories differ more in less equal societies, but it also allowed the effect of climatic changes observed to vary between sites. The response variable was the proportion of possible adaptations observed in a household (that is, varying from zero to one). We used a Gaussian error distribution for the hierarchical model as the response variable was approximately normally distributed, and reviewed diagnostic plots to ensure that model assumptions of normality and homoscedasticity of residuals were met. Confidence intervals for linear model coefficients were obtained through parametric bootstrapping.
In each study site, households were classified into three wealth categories (poor, average and wealthy) on the basis of a wealth index created from ten asset indicators specific to each study site54,55, identified during the FGDs. For a list of assets used in each site, see Supplementary Information, section B. Assets that varied most across the households in that site (>25% of households did not own them) were weighted 0.25 greater than those more commonly found.
Constraints and opportunities
Throughout the 18 month research project, bimonthly webinars were organized with all co-authors (including at least one with long-term expertise in each site), to share findings and reflections across study sites. During the eighth webinar, we realized that some constraints identified at some sites, could be considered opportunities in other sites. Therefore, we reframed our approach to also consider opportunities. First, study site leaders (both student who led the fieldwork and the professor with years of experience working on that site) used the information on constraints mentioned during the FGDs to identify the top three constraints at their site (those cited most often), according to the list provided in ref. 19, which groups constraints into broad categories (for example, physical aspects and economic aspects). Second, they identified the top three opportunities (adapting the list in ref. 19), reflecting on the data gathered during the field campaign and their own knowledge of the site. Although we requested site leaders to identify three of each, some identified two to four in some sites, as they considered some to be equally important, or only one to be relevant. Note that even if not cited in one site, some constraints and opportunities might still apply, they were just not considered as the top three most important by the study site leaders. Third, we combined the information from the ten sites to identify general constraints and opportunities across mountain regions, those cited in most sites.
Transformational adaptation
Before the last webinar, we requested study site leaders (co-authors) to reflect on transformational adaptation at their study site, by applying the framework of ref. 15. This framework considers five aspects (change agents, learning with engagement, generalizability of pathways, impacts across scales and sectors and sustainability of change) to determine if change is incremental or transformational. During the last webinar, through a process of collective qualitative assessment, the case studies were allocated points along the incremental to transformational continuum. The process analysis throws light on ways that characterize change, reflecting on ongoing social dynamics and multiple dimensions to think about transformational change, rather than deciding whether a particular change is transformational or not15, as it is known that incremental changes may aggregate over time to become transformational. During this last webinar, we also reflected on these findings to identify key priorities for moving forward climate change adaptation in African mountain regions and beyond—summarized in Box 1.
Study limitations
We report a range of adaptation responses, which can help inspire adaptation options in other mountain regions. However, we did not investigate which are complementary or substitutions, nor their effectiveness or long-term sustainability, aspects which require further investigation, as highlighted by ref. 16. We focused on climate change impacts as the main challenge to farmers’ livelihoods, but population change, new technologies, globalization, agricultural policies and social change are all exerting increasing influence on rural smallholder farmers56, and should also be considered when designing future adaptation interventions. Also, because of financial constraints, we did not engage local actors to reflect on transformational adaptation processes; this step was carried out by co-authors only. To imagine, initiate and maintain transformational change, we recommended engaging with local actors in a deliberative process in the future. Engaging national actors in the deliberative process in the future is also recommended to address systemic issues that constrain adaptation43.
Ethics statement
The research was approved following an ethical review at the University of York. Informed consent was obtained from all research participants before entering the study. State permissions were obtained from the relevant authorities in each country: Tanzania—the Tanzania Commission for Science and Technology (COSTECH) (2019-68-NA-2018-205); Kenya—the National Commission for Science, Technology and Innovation (NACOSTI) (NACOSTI/P/21/11045); Rwanda—the National Council for Science and Technology (NCST) (no number given); Uganda—the Uganda National Council for Science and Technology (UNCST) (NS282ES); Ethiopia—the authorities of the Oromia regional state (no permit number given); Burundi—the Faculty of Sciences, University of Burundi (no permit number given); and Democratic Republic of the Congo—the Faculty of Sciences of the Université Officielle de Bukavu (001/FS/VDR/BZI/UOB/2021-2022). At the local level, traditional authorities (for example, village chiefs and paramount chiefs) were consulted before starting this research, explaining study objectives, methods and potential benefits of the findings. We followed the guidelines on ethical research of the British Sociological Association57 when conducting interviews.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.