|The authors have adequately addressed all my comments in the first round of revisions. The manuscript is now significantly improved, both in content and clarity. In general, I recommend publication after a couple of minor issues are addressed:|
1. With all the additions from the previous revision round, the paper is a little lengthy now and the chapters could be separated better. Especially the chapters 3.1 and 3.2 repeat content that is visible in the tables, and describe things that have not been used (e.g. l. 264 “Selecting SST regions based on the preseason state of the Niño 1+2 anomaly index instead of MEI did not materially change results at Piura”, l. 305–307 “A quantile mapping approach (…) did not substantially differ (…)”. Slightly cutting these unnecessary parts should be enough.
2. Please either merge “Results” and “Discussion” to “Results & Discussion” or separate more strictly. The Discussion chapter should give an honest evaluation of the overall results, and put them into context by citing relevant literature. In my opinion, chapter 4.2 is Methods rather than Results, while 4.1 contains Discussion parts (e.g. comparison to Bazo et al. in l. 354). The Discussion then presents 3 more figures. Some repetitions could be avoided from merging the sections.
3. I doubt that the term “principal component regression PCR” is adequate to describe your method. In my understanding, the term PCR suggests that a regression is applied in principal component space, by which I mean that all variables have been included in the PCA, and the resulting regression is then transformed back into the original feature space. You are using only 1 PC of selected variables and most other predictors are regular variables. I would rather write of linear regression and a PC predictor component. Also I find it a bit pointless to always stress the “multiple” linear regression, as most people doing linear regression use multiple predictors. It’s ok if you just use your abbreviation MLR. The formulation in l. 271 “coupled principal component analysis and multiple linear regression” is wordy.
4. Please introduce abbreviations at the first occurrence of the term, and then always use the abbreviation afterwards. “Multiple linear regression (MLR)”, “Threat Score (TS)”, and others.
5. Table 5 and Table 6 could be merged when arranging in rows rather than columns, similar to Table 3 (which looks good now). When doing so, it is much easier to visually compare the different models by all metrics. Consider to highlight the best score per metric in bold font.
6. l. 46-54 The thematic jump from exposure/vulnerability to high temperatures in London requires rephrasing. As the rest of the article is about floods, the sentence should start with something like “In the context of heatwaves in London, (…)”
7. l. 65 needs a comma. In addition, consider to split the long sentence after “protocols” (l.66)
8. l. 94 the assumption here is actually that the errors in individual models are uncorrelated. Correlated errors would not cancel out.
9. l. 95 full stop after “individual model”. Please rephrase the subsequent sentence.