Given the broad scope of the research question, the study was designed as a semi-systematic review (Snyder, 2019), combining quantitative mapping of the literature with qualitative thematic synthesis.

The initial step consisted of a selection of scientific articles in the Web of Science (WoS) archive records, using the set of keywords provided in Table 1. A final search update was conducted on 17 February 2026, focusing on the period 2015–2025, possibly entailing that some articles with a publication date in late 2025 might not be included yet.

The aim was to choose keywords in such a way as to keep as many avenues open as possible, i.e., keeping the restrictions limited to not miss potentially relevant articles. Still, in order to keep a sufficient and necessary focus, we demanded that variants of “urban” appear as a part of the title. The only topical keyword specified was “climat*” (to cover “climate”, “climatic”, “climatological”, etc.), i.e., we refrained from pre-defining topical terms referring explicitly to hazards such as heat, flooding, drought – or to methods, such as modelling, remote sensing – in order to avoid creating a bias from the start. Conversely, we specified the geographic context as completely as possible, both at the continental level (“Africa*”) and at the national level by including all 54 African country names, following the United Nations Department for General Assembly and Conference Management, according to which 54 countries belong to the African Group and are Members of the United Nations.

No explicit hazard-related terms (e.g. heat, flooding or drought) or methodological terms (e.g. modelling or remote sensing) were prescribed in advance. This was deliberate, as the hazards and methodological approaches represented in the literature were intended to be outcomes of the analysis. Prescribing such terms at the search stage would have required defining these categories a priori and could have biased their observed representation in the resulting corpus. A consequence of requiring climat* as a search term is that studies addressing individual urban hazards without explicitly referring to climate may not be retrieved. The resulting corpus therefore represents literature that explicitly connects hazards affecting African cities to a climate context, rather than the complete literature on urban hazards.

The literature search was conducted using the Web of Science Core Collection (state of 17 February 2026) and was restricted to English-language records published between 2015 and 2025. No restrictions were applied by Web of Science Category, and all document types were considered except books, book chapters and book reviews. The search returned a list containing 1832 records, which was subjected to manual relevance screening. Because the broad search deliberately omitted hazard-specific terms, a large proportion of the retrieved records were clearly outside the scope of the review, and the screening consisted primarily of removing these records (Fig. 2).

All titles were screened first, and where relevance could not be established from the title alone, the abstract was consulted. Records were retained where a climate hazard affecting an African urban setting constituted a substantive focus of the study. Records were excluded where the African or urban context was incidental, or where climate change or climatic hazards primarily provided context for another research topic without substantive analysis of the underlying hazard. Examples include studies focusing on urban crime or conflict following climatic shocks, climate-induced migration, adaptation planning and governance, greenhouse-gas mitigation, urban agriculture, building energy performance, and environmental issues such as pollution or waste management.

Screening was conducted by a single author. We considered this appropriate for our semi-systematic review because the broad search retrieved a large proportion of clearly out-of-scope records, making most exclusion decisions straightforward. Where relevance could not be established from the title alone, the abstract was consulted, and where uncertainty remained after consulting the abstract, the record was retained, this conservative approach being intended to minimize the risk of excluding potentially relevant studies. This screening procedure resulted in a final corpus of 272 articles. The bibliographic details of the articles constituting the final review corpus are provided as Supplement Material S1. Full-text PDFs were obtained for all papers retained in the final corpus, including through institutional subscriptions where articles were not openly accessible.

Applying the WoS “Analyse” function to the selected body of articles revealed that authors with an African affiliation account for 48.1 % in the sample. The countries contributing most to the publication record were found to be non-African (UK and US). The highest-ranking African countries in the sample were, in order, South Africa, Nigeria, Ethiopia and Ghana.

The method to conduct the analysis of the 272 selected articles is outlined in Fig. 3. The upper row of the figure shows, from left to right, the progression from the WoS export list towards bibliographic (Zotero) and PDF files-archive for each of the article records.

The WoS article list was analysed quantitatively with respect to (left half of Fig. 3)

the representativeness of the sampled cities in terms of (1) Köppen-Geiger climate zone and (2) population size;

The WoS list of 272 selected articles was examined for city names in the Title and Abstract fields by extracting all capitalised words from these texts, enabling rapid identification of potential city names. This was implemented in MS Excel using a Visual Basic for Applications (VBA) function that identifies and extracts words starting with a capital letter and stores them in a separate column.

This column was then reviewed to identify city names, by comparing city names occurring in the GHS Urban Centre Database (GHS-UCDB) R2024A from the Copernicus Global Human Settlement Layer (GHSL), which includes population estimates for each city (Melchiorri, 2022). If an extracted name from a selected article matched a city in the list, it was retained. If the same city had already been identified earlier, its occurrence count was increased by one. Nouns that resembled city names but were not present in the GHS-UCDB list were further checked using Wikipedia and Google Maps to determine whether they referred to neighbourhoods or sub-units of larger cities. In such cases, the names were assigned to the corresponding host urban agglomeration.

The analysis focused on urban agglomerations as defined in the GHS-UCDB list rather than on smaller units such as sub-cities or neighbourhoods. This choice was made for reasons of cross-country consistency, as the administrative subdivision of cities differs between countries. For example, a large city in one country may be recorded as a single entity, while in another it may be subdivided into multiple smaller administrative units.

The cities thus identified are shown in Fig. 4 as solid dots, scaled to the number of mentions (references) in the literature sample. For each of the cities, the Köppen-Geiger climate zone (Beck et al., 2023) was determined. This enabled a comparison of the frequency of climate zones represented in the literature sample with the baseline distribution of African urban populations across climate zones.

The comparison (Fig. 4) shows that the climate zones represented in the sample of urban climate studies reflect the actual distribution of African agglomerations rather well. The tropical savanna (Aw) zone, for instance, dominates both the literature sample and the population baseline.

Some imbalances are nevertheless apparent. Cities in highland (Cwb) and Mediterranean (Csa, Csb) climates receive disproportionately high attention relative to their urban population share, likely reflecting the visibility of climate impacts (e.g., flooding, landslides) and/or the presence of strong local research infrastructures. Conversely, urban populations in monsoon (Am), tropical forest (Af), and hot desert (BWh) climates are somewhat underrepresented, despite their demographic significance. These patterns highlight that while the overall distribution across climate zones is well captured, there remains a tendency for research to concentrate in hazard-prone or research-intensive regions, leaving important populations in other climate zones relatively less studied.

From a geographical point of view, Fig. 4 shows a relatively higher coverage of referenced cities (1) near the Gulf of Guinea (West Africa), (2) along an axis extending from East to Southern Africa, and (3) in North Africa along the Mediterranean. Such a pattern is consistent with findings by other researchers (Leal Filho et al., 2023; Montfort et al., 2025). From Fig. 4 it also emerges that a considerable number of the frequently studied cities are located along the coast.

We compared the city size distribution (measured by number of inhabitants) of the city sample with the actual population distribution across different size categories:

small cities (<100000 inhabitants)

We also examined the distinction between capital and non-capital cities in the sample. Out of 112 cities, 28 (25 %) are capitals. However, when weighting by the number of times each city is referenced in the selected articles (as represented by dot size in Fig. 4), the share of capital cities rises to nearly 45 %. This reflects the tendency for capital cities to appear in multiple studies, while smaller cities are typically mentioned only once.

A keyword analysis was conducted using VOSviewer, a software tool for constructing and visualizing bibliometric networks (http://www.vosviewer.com, last access: 21 October 2026) (Van Eck and Waltman, 2017). Using VOSviewer, a network map (Fig. 6) was created from the bibliographic data contained in the WoS output file, based on a co-occurrence analysis of both author keywords and Keywords Plus.

The terms “climate (change)” and “urban/city” were excluded from the keywords, given they were the primary search keywords (hence should not re-appear on the network map). Our focus was on identifying additional keywords that emerged from the search. To reduce clutter and avoid less relevant terms, only keywords appearing at least nine times were retained, resulting in a final list of 47 keywords. This threshold of 47 was determined through trial and error, aiming for a balance between including as many relevant terms as possible while maintaining a clear visualization. A VOSviewer “thesaurus” file (see Van Eck and Waltman, 2017) was developed and applied to (1) filter out non-relevant keywords (e.g., city names); (2) merge spelling variants (e.g., “heat island” vs. “heat-island”); (3) unify conceptually similar terms (e.g., “earth observation” and “remote sensing”).

For visualization, the VOSviewer Scale parameter was set to 100 %, fully allowing circle sizes to reflect keyword frequency (larger circles representing more frequent keywords); the Size Variation parameter was set to 50 %, in order to avoid excessively small labels for less frequent keywords. Figure 6 shows the keyword co-occurrence network generated by VOSviewer. Node size reflects the frequency of keyword occurrence, while the thickness of the connecting lines indicates the strength of co-occurrence. Colours represent clusters, the algorithm positioning related terms closer together to reveal strong internal relationships. The occurrence and connectivity of keywords is also provided as a ranked table in Supplement Material S2, including keyword occurrence frequencies and Total Link Strength (TLS) values.

In Fig. 6, four clusters emerge. The red cluster is dense and cohesive around heat-related terms: UHI (urban heat island), LST (land surface temperature), and thermal comfort). Here, LST refers to the radiative temperature of the Earth's surface, as inferred from emitted thermal radiation, and is distinct from near-surface air temperature. The green cluster centres on flooding and social aspects (risk, vulnerability, poverty). Smaller clusters focus on hydrological extremes (blue) and water- and health-related issues (yellow).

Several nodes occupy central positions in the network, bridging across clusters: these include urbanisation, remote sensing, LULCC (land use/land cover change), and modelling. Unlike hazard-specific terms such as heat or flooding, these concepts represent methodological tools or overarching processes that naturally connect different research areas. For instance, remote sensing is strongly tied to land surface temperature (LST), urban heat island (UHI), and urbanisation in the red cluster but also links to flooding and land-use studies. Modelling serves as another methodological bridge, connecting the flooding and vulnerability literature with research on precipitation and variability.

Peripheral nodes represent less frequent or more isolated keywords. Strikingly, many of these are impact-related, such as mortality, vector-borne disease, malaria, or the aggregated term soil impacts (covering landslides, soil erosion, and subsidence). Broader socio-economic dimensions like poverty, informality, and hazard also appear in the periphery. Their position suggests that explicit links to concrete societal impacts remain less developed and relatively fragmented. Emerging methodological terms, such as machine learning, are also visible at the margins, indicating directions that are beginning to enter the field but have not yet reached central importance.

Taken together, Fig. 6 demonstrates that hazard-specific research (e.g., heat, flooding) forms the structural backbone of the network, while methodological concepts such as remote sensing and modelling provide essential connections across themes. At the same time, the peripheral position of impact-related keywords highlights a gap in the literature: despite frequent references to hazards and methods, systematic integration of health, poverty, and other vulnerability dimensions into the research is still relatively limited. This suggests an important direction for future work, namely the stronger linkage of urban climate hazard research to the societal impacts most directly affecting populations.

Considering that Earth Observation (EO) constitutes a rapidly evolving field, the selected articles were analysed with respect to the trend of the use of EO data. The regression was done over the time span 2015–2024, i.e., excluding articles published in 2025 as that might not be complete (the sampling of the 2015–2025 articles having been finalised in February 2026), which could introduce a bias.

The total annual output of articles shows a clear upward trend over the period 2015–2024 (Fig. 7, left), increasing from only a few articles in 2015 to more than 40 in 2024. The number of studies using satellite remote sensing similarly increases over this period. In contrast, the relative share of remote-sensing studies varies substantially during the first years of the period but remains relatively stable at around 40 % in the later years (Fig. 7, right). This indicates that, while the absolute number of studies using remote sensing has increased alongside the overall growth of the literature, its proportional contribution to the field has not markedly increased in recent years.

The list of 272 articles was further analysed in terms of the climate hazards covered. This was done by experts in the respective hazard domains, screening the title and abstracts of each article, labelling them with a hazard indication, including windstorms, sea level rise, water stress, extreme heat, slow-onset events, land surface temperature (LST), and flooding. Here, slow-onset events refer to gradual changes in climatic conditions, such as increasing mean temperatures or changing precipitation patterns, consistent with the IPCC framing of slow-onset events (Calvin et al., 2023).

In cases where an article covered multiple hazards, the dominant hazard was identified, obviously inducing a certain level of subjectivity. In addition, each article was labelled for the occurrence of (or reference to) one or more conceptual or methodological aspects, including urbanisation (including such terms as urban growth, or urban dynamics), modelling, informality, data science, future climate, remote sensing.

The outcome of this exercise is shown in Fig. 8. A close look at the distribution of hazards reveals clear thematic imbalances. Research on flooding and LST dominate the field, together accounting for nearly half of all publications. Studies on slow-onset events and extreme heat are also well represented, underlining the importance of both gradual climate variability and acute heat extremes in African cities. It is worth noting that the two heat-related categories (extreme heat and LST) were separated in our classification. While this provides analytical clarity, it may also dilute the apparent weight of heat-related hazards in the dataset: taken together, these account for well over a hundred articles, confirming heat as one of the most prominently studied urban climate risks in Africa. The prominence of land surface temperature in particular reflects the reliance on remote sensing, which offers a consistent and widely accessible data source in data-sparse regions such as Africa. By contrast, water stress, sea level rise, and windstorms are covered only sparsely, despite their known relevance for urban vulnerability. This thematic skew suggests a concentration of research attention on hazards that are both visible in cities and relatively straightforward to monitor, while other critical but less studied risks remain underexplored.

In addition to classifying articles by their dominant climate hazard, we also identified several recurring methodological and contextual themes that characterise how climate risks in African cities are studied. These include urbanisation processes (urban growth and land-use change), deterministic modelling approaches (e.g., urban climate or hydrological models), Earth Observation, and data science techniques such as artificial intelligence and machine learning. We further assessed whether studies explicitly consider future climate projections and whether they address urban informality, a key structural feature influencing exposure and vulnerability in many African cities.

Analysing Fig. 8 in terms of these conceptual and methodological aspects reveals a host of information:

Remote sensing data is used in a minority (10 %–20 %) of studies, except within the land surface temperature (LST) topic, for obvious reasons. A deeper analysis showed that most literature cases made use of Landsat imagery, either to observe patterns of land surface temperature, or to assess historical urban growth.

To complement the dominant-hazard classification, papers were additionally screened for substantive consideration of other hazard categories. Titles and abstracts were examined, with the full text consulted when necessary. Merely incidental references to other hazards were not counted.

The recorded dominant and additional labels indicated consideration of more than one thematic category in 30 papers (11 %).

A more detailed analysis shows that the cross-hazard nature differs between the themes:

Slow-onset studies can encompass several distinct hazards within a broader assessment of climatic change and its impacts. For example, research in Pietermaritzburg examines how drought, heatwaves, floods and other climate stressors affect urban livelihoods and coping capacities (Hlahla and Hill, 2018).

Some papers explicitly examine distinct hazards within a shared urban context. For example, two studies in Accra and Tamale address flooding and extreme heat and their impacts on healthcare provision and on water and electricity services (Codjoe et al., 2020; Kayaga et al., 2021). Consideration of multiple hazards should be distinguished from analysis of compound or cascading processes. A clear example of compound-hazard analysis is provided by the Maputo study, which examines the combined influence of rainfall and tidal conditions on urban flooding (Matos et al., 2023).

A few recent review studies provide important context for understanding recent urban climate research in Africa. These include systematic mapping of adaptation responses (Hunter et al., 2020), bibliometric analyses of research trends (Mhedhbi et al., 2023), thematic assessments of urban thermal change (Li et al., 2022b), emerging multi-risk perspectives (Fekete and Subramanian, 2024), and broader scoping reviews of urban climate risk and governance (Ansah et al., 2024). Together, these review articles describe a rapidly expanding but unevenly developed knowledge base.

A consistent finding is the strong growth in publications since the mid-2010s (Mhedhbi et al., 2023). However, research remains geographically concentrated. A limited number of countries – particularly South Africa, Nigeria, Kenya, Egypt and Ghana (also see the findings of Sect. 2.1) – account for a disproportionate share of studies, while large parts of Central Africa and many secondary cities remain sparsely represented (Hunter et al., 2020; Mhedhbi et al., 2023). Capital cities and economically prominent agglomerations dominate the record.

Thematically, heat and flooding receive substantial attention within the urban climate literature, reflecting their relevance for African cities. Flooding is frequently examined through hydrological modelling, exposure assessments and spatial analyses (Hunter et al., 2020). Heat-related research often focuses on land surface temperature and urban heat island characteristics derived from satellite data (Li et al., 2022b). Other hazards, such as windstorms, coastal flooding and water stress, appear less consistently across the literature. Also, even though calls for integrated multi-hazard perspectives are increasing (Fekete and Subramanian, 2024), most studies still address hazards in isolation.

Methodologically, the literature combines Earth Observation, spatial analysis and a range of modelling approaches. While these methods present substantially improved spatial diagnostics in data-scarce environments, high-resolution urban climate modelling and physically detailed representations of urban processes remain limited (Fekete and Subramanian, 2024; Li et al., 2022b). Explicit coupling of hazard, exposure, vulnerability and adaptive capacity is comparatively uncommon (Ansah et al., 2024; Hunter et al., 2020). Although social dimensions (poverty, informality, governance capacity and inequality) are frequently acknowledged, they are less often incorporated into quantitative risk assessments.

Forward-looking assessments also remain limited. While some studies employ downscaled climate model outputs, very few integrate future climate scenarios with projected urban growth, demographic change or adaptation pathways (Fekete and Subramanian, 2024; Li et al., 2022b). As a result, the literature provides important diagnostics of present-day patterns but offers fewer insights into long-term urban transformation under climate change.

In contrast to the reviews discussed above, the present study takes a hazard-centred perspective and explicitly quantifies the representativeness of the literature. In particular, research coverage is compared with the distribution of African urban populations across climate zones and city-size classes, alongside an assessment of methodological characteristics, including the use of Earth Observation and modelling.

In the following subsections, the selected articles – organised by hazard type as occurring in Sect. 3.5 – were assessed by thematic experts. The length and level of detail of the subsections vary according to the scope and characteristics of the literature identified for each hazard. Rather than providing an exhaustive study-by-study account, the discussion focuses on synthesising the main insights emerging from the reviewed literature for each hazard. Section 4.3 brings these findings together in a synoptic table, providing a concise synthesis across the different hazard themes.

Building on the thematic assessment in Sect. 4.2, Table 2 provides a synoptic overview of the main findings emerging from the reviewed literature across the different hazard themes. The table complements the more detailed discussion above by highlighting the principal synthesis for each theme.