1.The manuscript aims to synthesize “climate risks in African cities,” but this is an extremely wide topic covering hazards, vulnerability, governance, EO, AI, urban growth, health, infrastructure, and planning. The paper identifies many gaps, but many are already well known: limited data, geographic imbalance, lack of multi-hazard studies, weak integration of social vulnerability, and few future-looking studies. The authors need to make clearer what this review adds beyond existing review papers, rather than mainly confirming broad and expected gaps. The abstract itself already lists many generic findings, such as underrepresentation of Central Africa, secondary cities, multi-hazard studies, urban modelling, and social vulnerability integration.

2. The authors used “urban* OR city OR cities OR metropol*” in the title and “climat*” as the only topical keyword, without hazard-specific or risk-specific terms. This may miss many relevant studies on flooding, heat, drought, landslides, coastal hazards, vulnerability, exposure, and adaptation that do not use “climat*” prominently. At the same time, the broad strategy created many irrelevant records, later filtered manually. The authors need to justify this design more rigorously and provide a reproducible PRISMA-style screening process, including exclusion criteria, reviewer agreement, and examples of included/excluded papers.

3. The paper states that 1,832 records were reduced to 273 articles through title/abstract screening, with uncertain cases retained. However, it does not explain how many reviewers screened the papers, whether screening was independent, whether disagreements were resolved, or whether inter-reviewer consistency was checked. For a semi-systematic review, this is a major weakness. Without this information, the final corpus may reflect subjective judgement rather than a reproducible evidence base.

4. The authors used SciSpace outputs and then processed more than 50,000 words with GPT-5.2 to generate knowledge gaps. This is potentially useful, but the manuscript does not provide enough information about validation, hallucination control, prompt sensitivity, reproducibility, or whether outputs were checked against the original papers. This is a serious issue because AI-generated “research gaps” can easily reproduce generic language rather than extract evidence-based findings. The authors should either move this to supplementary material as an exploratory check or provide a much stricter validation protocol.

5. Figures on climate-zone representation, city-size representation, keyword networks, EO trends, and hazard frequency are useful, but the interpretation often remains surface-level. For example, the city-size comparison shows strong underrepresentation of small cities and overrepresentation of large cities, but the implications for bias in risk knowledge are not deeply analysed. The authors should connect these patterns more directly to what kinds of adaptation decisions may be distorted by this bias.

6. Each article was assigned a dominant hazard, while many urban climate risks are compound by nature. This can artificially separate flooding, sea-level rise, water stress, heat, and LST, even when they interact. The manuscript itself criticizes the literature for studying hazards in isolation, but its own classification partly repeats this limitation. The authors should add a multi-label hazard classification and report how many papers address compound or cascading hazards.

7.Some thematic sections read like mini-literature reviews with many citations but limited synthesis. The flood section, for example, contains many claims and examples, but the writing becomes dense and sometimes repetitive. Other hazards, such as windstorms, receive much shorter treatment. This imbalance weakens the review. The authors should reduce citation-heavy narrative and replace it with clearer synthesis tables: key methods, geographic focus, limitations, evidence strength, and research needs for each hazard.

8.Currently, many gaps are repeated in different forms across the abstract, results, thematic review, AI synthesis, and conclusion. The final message would be stronger if the authors organized the contribution around a clear framework: hazard evidence, exposure/vulnerability evidence, future scenario evidence, modelling/EO evidence, and decision-use evidence. This would make the review more useful for researchers and urban planners.

9. The conclusion states that research has grown but remains limited by data scarcity, high-resolution modelling gaps, geographic imbalance, and weak integration of social and forward-looking dimensions. This is reasonable, but not sufficiently novel. The authors should end with a more actionable research agenda, including specific datasets, minimum reporting standards, recommended cross-city comparison designs, and priority regions/city types.

Overall:

The manuscript by De Ridder et al. presents a synthesis of 273 articles on climate hazards in urban areas in Africa published from 2015 to 2025. The manuscript reports interesting aspects, such as the distribution of the reviewed articles in relation to major climate zones, cities population, and hazard types. It also shows a network of keywords co-occurrence, discusses methodological aspects in the selected articles, and identify scientific knowledge gaps.

In my view, the manuscript is well structured, presents research that is relevant and within the scope of NHESS, and has potential to be published. 

However, additional work is needed, as listed below.

Specific comments:

1. From the initial list of 1832 articles, 273 were selected, which represents a substantial reduction. In this process, there was a relevance screening that was conducted manually, as explained by the authors. In spite of a "conservative inclusion strategy" was applied, this process may be somewhat subjective and not easy to reproduce, so it is important to mention this in the text, or reconsider the methodology.

2. Fig. 3 needs a scale for the solid dots size, which reflects the number of mentions in the literature, as explained in the text.

3. Although Fig. 5 graphically display how nodes/keywords are more or less connected, it is important to have this information also summarized in a table displaying how nodes/keywords rank in relation to the number of connections, contributing to clearly show which ones are more important.

4. Data and analysis in Fig. 6 bring interesting aspects, informing that the number of total publications and the number of studies using satellite-based remote sensing had both a ~linear increase from 2015 to 2024, while the proportion of remote-sensing studies in the period varied substantially in the first half of this period, but it remained nearly constant and close to 40% in the following years. 

However, the time interval from 2014 to 2025 yielded a sample size of 10 points, which is small to discuss these patterns in terms of coefficients of correlation and p-values. Discussion on results from Fig. 6 is already interesting by qualitatively evaluating the trends, and it would be much more precise without mentioning correlation coefficients and their statistical significance.

5. It is important to include somewhere at the beginning of the text what the authors mean by Land Surface Temperature (LST) because it may the confounded with air temperature near the surface. A clear explanation comes only at section 4.2.5.

6. Section 4 needs a table to summarize the findings from the meta-review and the expert-based thematic reviews.

7. In my view, it is important to consider if the topic "Slow-onset climate trends" in section 4.2.6 may be renamed to something like "Urbanization-climate interactions" mentioned in Table 2, which in my view it is easier to understand.