Many thanks for taking the time to comment on our paper. All comments from Referee #1 are addressed below in bold, after the original comment.

This paper provides an assessment of drought reporting by collating newspaper articles related to droughts in England check their alignment with meteorological anomalies. The study suggests that media coverage tends to coincides with meteorological drought, but that drought only coincides with media coverage when additional conditions are “favorable” (e.g., seasonality, long-term precipitation shortage). Additionally it concludes that media often reports about water shortages, water use restrictions, and the consequences that follow.

First, many thanks for taking the time to read and review our paper, and doing this so expeditiously. Your comments are very helpful to us, as they make clear that:

1.  … some of the work involved in writing a paper like this is invisible, and thus makes the findings appear perhaps more trivial than they are. When revising the paper, we will clarify the methodological contribution we make here, which builds on previous work on text analysis generally, studies of the media related to a variety of environmental phenomena, and a very small subset of studies related to hydrology and the media.
2. … our decision to link droughts to temperature and precipitation anomalies which captured important elements of weather as perceived by society weakened the link to droughts in a hydrological sense. We made this choice because drought indices effectively by definition combine the variables we used into a unidimensional value, which then provides less scope to explore the relationships we were interested in. However, we fully accept the argument that linking to existing hydrological indices as well would provide additional insights, and in our revisions we will explore different forms of drought indices representing aggregated precipitation, streamflow and groundwater deficits (SPI, SSI & SGI as provided by the UK CEH: e.g. https://eip.ceh.ac.uk/hydrology/water-resources/about/#availableData)
3. … the link between our quantitative, descriptive statistics and qualitative work (i.e. when is drought discussed and what are the topics that arise) is weak. Topic modelling is a well-accepted and commonly used method in text analysis, which provides useful insights into debates, using computational methods.
4. … we need to better motivate our study and argue for its relevance to hydrology. It is worth noting that the idea from the study came from a hydrologist (JS) who was interested in the question of media (and thus societal) understandings of drought, and how their representation might differ from discourses in hydrology. The results, though appearing trivial are, at least we would argue, not obvious, since we clearly show that discussions about droughts are not simply driven by droughts (c.f. Figures 3 & 4 in the paper). In the revision of the paper, we will explore in more depth the underlying differences in the peaks, for example by directly linking the topic modelling to the temporal trends (essentially, every document can be assigned a probability of belonging to a topic, and therefore we can present the distribution of topics over time for peak events).

The first part of the analyses uses statistical tests (chi-squared test) to quantify that most report on droughts occur when precipitation has been below-average and temperatures have been higher than expected (Fig 1) and that these effects tend to be more reported in Spring and Summer compared to other seasons.

These findings are fine but given how trivial they are it also seems extremely marginal for a contribution to HESS (in my opinion).

As we explained above, we agree with your summary of the results of the first part of the paper,  and recognise that we need to better motivate our work, present the methodology and associated challenges, and discuss the results more holistically to make clear why our work is not trivial. Our analyses demonstrate that media seems to respond more to droughts when these are felt more at the time and not to whether there might be consequences further ahead. While this might not be surprising in today’s media landscape, we still find it important to state that media is reporting on socioeconomic rather than meteorological/hydrological droughts.

Before I can recommend this paper for publication in HESS (or any other significant hydrological outlet) I expect some more depth to this finding. I expect the following points to be addressed thoroughly:

• The paper shows the statistically significant more frequent reporting on drought when there are anomalously dry and hot conditions, and states this can be because “a potential confirmation bias”. Can the paper also explore the more logical explanation that people report on droughts when there are droughts? Also can anything be said more than there is a statistically significant relationship?

By incorporating drought-related indices (i.e. SPI, SSI, SGI) related to precipitation, streamflow and groundwater we can explore this relationship in more depth (i.e. on top of temperature and precipitation), for example performing a multiple regression, and potentially also partitioning the indices regionally (a hypothesised London-centric bias in media might lead to drought having more influence in some regions than others).

• Can it be be more robustly reported how likely it is that (according the scientific defintion) drought events are actually picked up significantly by the media, and how this differs between different types of drought, different types of drought intensities, and differs per season. Hereby can the statistics that express this go beyond a chi-squared test that only tests whether two variables are independent or not. The plots 3 and 4 try to give some more depth but are only qualitative suggestions and no formal tests or numbers that quantify this.

Our answer to the previous point makes some proposals as to how to address this. However, we would also emphasise that visualisations of data like these are also a powerful way of reporting our results, and that we would intend to use a combination of quantitative and qualitative approaches to presenting our results.

• Optional: can something be said if drought reporting has become more or less frequent over time (i.e. if the same drought event now gets different coverage than in the past?).

The time period over which we are working is probably too short (20+ years), but we can take a look at this aspect. There are in practice many issues, also related to normalising for overall online media, which make this non-trivial. But we can certainly comment on it in the paper.

• How do we know that news items actually are about a current drought and do not reflect on conditions in the longer past? How does this affect the analyses? Can this be screened for?

This is a good question. In practice, in text analysis, so-called distant reading is always based on assumptions, e.g. people are talking about current events here. We did create several filters to remove, e.g. articles talking about droughts in other locations or irrelevant articles to droughts, but in the revised version, we can test on a sub-sample of data through close reading (essentially, we will take a percentage of all articles, read and classify them) if this assumption is valid. We will then report values of precision (i.e., the proportion of articles that are relevant according to our assumptions).

• The paper interprets that more common reporting of summer droughts “could indicate a confirmation bias driving media coverage –drought is more likely to be reported in warmer conditions since it confirms a simplistic narrative relating drought to warm weather.” Is an alternative considered where maybe there is just a mismatch between the scientific definition of the drought used, and the drought society and reporters care about (when they actually feel impacts will be during summer drought and not really during “droughts” in for example winter, because then, while conditions are anomalous the impacts on society will be absent or minimal.. You already acknowledge this yourself in the paper and therefore also omit data from other regions of the UK.

We agree, as you note, we indeed acknowledge this aspect. We don’t claim that a confirmation bias is the only explanation - rather that it is one possible one. However, confirmation bias is a well known phenomena in social media and the media more generally, and we believe it is important to present it as one possible explanation. We will give more thought as to how to present a number of possible explanations of our results more systematically, and to potentially propose further work to explore these issues in more depth in the future.

• The paper is about drought and reporting on drought, but ends up using an index based on accumulated weather conditions (P and T) and not a regular drought index. This seems to add an additional layer of inconsistency in the analysis as right now it balances between droughts people likely care about (bc they are reported in the news), drought as hydrologists typically define (e.g, standardized indices), and droughts as defined in this paper (P and T anomalies). This additional step op not using a typical drought index makes it harder to connect the findings to the existing literature on drought

We addressed this point above, and will include a number of drought indices in our analysis to complement and extend our existing approach. However, please note that we explicitly want to distinguish between different types of drought conditions and to be able to distinguish between drought conditions during warm respective cold situations.       

• The part were drought are assessed by their specific keywords. Can the authors express what we truly hydrologically learn here and how this is a scientific contribution that fits within the scope of HESS? (“HESS encourages and supports fundamental and applied research that advances the understanding of hydrological systems, their role in providing water for ecosystems and society, and the role of the water cycle in the functioning of the Earth system.”) I personally struggle to see how the contributions of this section align with these objectives of HESS.

We submitted this paper to a special issue in HESS titled <Drought, society, and ecosystems>. In the call, the editors state that “we aim to showcase the diverse interdisciplinary research being done on the interactions between drought, ecosystems, and people". Furthermore, the topics listed include "drought risk management and communication". We believe that our work is interdisciplinary and directly addresses the topic of communication by and to non-experts, which is in turn very important in understanding and developing future policy. Mismatches between scientific discourses and popular understandings are central to challenges in changing behaviour more broadly. We also think that work like ours fits to HESS more broadly, in terms of the link to water and society, but we explicitly prepared and submitted the paper to this very relevant special issue.

• The statistical tests (chi-squared test) assumes all observations are independent. To what extent is this the case in the dataset of this paper?

The assumption of independence here relates to events, their precipitation and temperature anomalies, and the seasons in which they occur. We think this assumption is reasonable.   

Detailed comments

L12-14: this final sentence of the abstract is unclear in its meaning (to me). Please check.

We will reword this sentence to make clearer that we are describing commonly occurring topics with respect to drought in our analysis.

L21: Tjideman is misspelled and not listed in the reference list…

Thanks, we will correct this, and also read through the paper to find other similar cases.

L25: I am unsure this the universal case for what a drought is. In some cases the balance of a drought will also be influenced by e.g. how much discharge has drained the system. In addition, the sentence after already starts to contradict the statement in line 25.

The universal cause for droughts is a deficiency of precipitation and to quantify these deficiencies it is useful to relate these to evapotranspiration. These deficiencies can then translate into different types of droughts. We will reformulate this sentence to make this clearer and avoid misunderstanding..

RQ1. How does the media coverage relate to meteorological conditions and seasonality in England?” “How […] relates” is pretty vague and not reflects any rejectable hypothesis. Maybe consider phrasing this bit more precise.

We will rephrase this research question appropriately, and give some thought as to whether it ought to include a rejectable hypothesis. We can certainly do this, in terms of the statistical analysis, but we are in some ways more interested in the overall ways in which droughts are being described, which leads to a more explorative research question.

RQ2. What are typical topics in media content, and how do they differ for different major drought events? This question needs be rephrase for clarity (what media content are you talking about, can “typical topics” be defined more clearly” etc).

Again, we can rephrase the research question, and this makes sense given that we intend linking the two parts of the analysis more closely.

The authors of this study developed a database of drought-related newspaper and tabloid articles in England from 2000 to 2023. Using this database the authors then looked at how the number of articles in the database varied seasonally. They found that the largest number of articles occurred in typically warmer and dryer months. Next, the authors used an unsupervised natural language processing method called topic modelling to  cluster articles in the databases into 10 topic areas. The clusters that emerged covered topics that all intuitively related to drought. While the results are not all that surprising, I think the methods and the presentation (i.e., figures) are interesting and would contribute a unique perspective to HESS.

Thanks for this supportive and helpful comment. We are glad that you find the methods and presentation interesting, and see the unique perspective we give to HESS. We agree that the results are not per se surprising, but as you correctly imply, that is not what we claim as a contribution.

First, I recommend the introduction be restructured to improve flow. For example, the paragraph starting on line 44 starts with a topic sentence summarizing previous methods used to explore media coverage responses to extreme weather events; however, this paragraph ends with a discussion of the role of newspapers. I think this newspaper discussion could be moved to another part of the introduction (maybe earlier?) and the authors could more concisely review all the methods that have been used to do this type of work. Additionally, by the end of Line 69, I’m left wondering what are the missing (methodological) pieces that this study aims to fill? I think the authors need to be more direct here. What are they doing differently compared to the other studies they list? Last, I’m not seeing any description of topic modelling in the introduction, which seems like an important component of the study that the authors point out in the abstract, but don’t mention again until the methods section. See my line comments for more suggestions.

These comments on restructuring the introduction (and your line by line comments) are very helpful, and we will take them up to make our argument more direct. Mentioning topic modelling in the introduction is an excellent idea.

Second, I recognize that no drought index is necessarily perfect at capturing all aspects of drought, but I recommend the authors including a WMO standardized drought index such as SPI for precipitation as well as EWP in the Supplement. I think this would allow them to compare this study to other studies conducted nearby (e.g., O’Connor et al. 2022) as well as globally. It might also be of interest because of the experimental design of this study; SPI leverages precipitation anomalies. At a minimum, I recommend the authors include a discussion (potentially in the Supplement) of how EWP compares to SPI and how CET compares to other indices calculated across the UK in terms of their behavior, benefits, and challenges. I had some other questions about the methods, which are included in my line comments.

This suggestion was already made in an earlier review, and we intend to take it up (see AC1 “By incorporating drought-related indices (i.e. SPI, SSI, SGI) related to precipitation, streamflow and groundwater we can explore this relationship in more depth (i.e. on top of temperature and precipitation), for example performing a multiple regression, and potentially also partitioning the indices regionally (a hypothesised London-centric bias in media might lead to drought having more influence in some regions than others).”

Third, the figures were very creative, but some additional information is needed to help the reader understand them. For example, a scale color key is omitted from several images in the main and Supplemental text. Some of the abbreviated axes labels need to be explained further. See my line comments for more details.

Again, this is a very helpful comment, which mirrors the suggestions of RC3. We will improve the figures, and their explanations to make sure that readers can fully understand them.

Last, I wanted to say that I greatly appreciate the authors transparent and reproducible approach; specifically, their willingness to share their R code and R session information. I think that’s awesome and was glad to see it.

Thanks a lot - we appreciate this positive feedback a great deal.

Given these general comments as well as my line comments below, this paper needs a great deal of revision before it will be ready for publication. Therefore, I recommend this manuscript be accepted with major revisions and resubmitted to HESS.

Thanks to the authors for this interesting article and for the opportunity to review this manuscript.

Line Edits

We don’t reproduce all of the very helpful line by line comments, but in revised version of the manuscript we will work through these individually, and respond to all of them. Thanks for taking so much time to give us this detailed and useful feedback.

Many thanks for taking the time to comment on our paper. All comments are addressed below in bold, after the original comment.

1. The discussion about the results displayed in the figures is lacking, such that the reader may have a hard time understanding the take home or some of the data. I would recommend additional content within the manuscript to walk the reader through the results.

In the revision, we will carefully revisit the discussion of the results and expand the relevant sections to provide clearer explanations of the findings. We will ensure that the key take-home messages are more explicitly conveyed, by improving the clarity of the manuscript. 

2. Many of the figures are challenging to read and the text overlaps. Further, ensure that all figures satisfy any color-blind considerations. 

Thanks for this helpful feedback, which mirrors some comments from RC2. We will revise the figures and their annotation, and use palettes suitable for color-blind readers.

3. Some of the the paragraphs in the discussion have a really clear take home lessons/message within the paragraph and making them explicit (see the first in 4.1). However, in many of the paragraphs, the phrasing/format is a bit awkward. See Lines 321-325 as an example. Much of the content in the discussion seems disjointed and doesn't have the coherence needed to unpack this complex research. As such, I'm having a hard time following and understanding the primary points the authors are trying to make. 

Thanks for this hint - we will revise the discussion, also taking into account the detailed comments in RC2, to improve both phrasing and argument. We’ll also make sure that the thread of our argument is easier to follow.

4. I would recommend the authors to do a deeper dive in the existing literature to further contextualize their results to current knowledge beyond the work by Marsh and Dayrell in the Discussion. The big "so whats" seem to be missing. For example, studies have been done that have looked at changes in water consumption based on media coverage of extreme events: https://www.science.org/doi/10.1126/sciadv.1700784. See other examples, which may be more or less pertinent since the geographies are different (certainly no obligation to include, but added to encourage further analysis, reflection, and contextualization of the within this article): Molina et al. https://www.science.org/doi/10.1126/sciadv.1700784, Rutledge-Prior and Beggs https://onlinelibrary.wiley.com/doi/abs/10.1111/ajph.12759, Helm Smith et al. https://journals.ametsoc.org/view/journals/bams/101/10/bamsD190342.xml. As such, I encourage the authors to ensure that the novelty and enhancement to the field is obvious and that there is a clear, explicit link between the results detailed in Section 3 and the discussion of said results. There are limited call-backs and I think that diminishes the potential impact of this piece. 

Thanks for this very helpful set of literature and the general comment with respect to “so what”. We’ll use some of these suggestions to better contextualise and discuss our results and their broader implications.

5. I am still a little unclear on the link between the demand side and its effect on drought and how that's drawn out in this paper. As mentioned in the limitations section, linking water demand (population density, industrial needs) to drought coverage could strengthen the analysis and offer a more comprehensive understanding of media biases.

This is an interesting point. We interpret the link between demand and drought as being related to topics describing restrictions on water use. One possible route to exploring this question would be to use regional indices, and explore whether topics related to demand occur more in areas where urban demand is likely to be higher. We will look into whether this is feasible, and at a minimum add a more detailed discussion around this point.

6. The national aggregation of meteorological variables may obscure regional variations in drought severity and media coverage. Incorporating more granular data could provide further insights into regional biases or differences in public perception. If not included, a further discussion about the results may be warranted. 

As we mentioned in the limitations, analysis at finer granularities would be challenging. Nonetheless, we agree that this could form a more substantive part of the discussion - issues that are important here include how newspaper articles refer to drought in a regional sense, and the availability of indices which can be matched at finer scales. In the end, we think aggregating to the country level, which allows us to explore in more temporal detail, is the most appropriate choice.

7. Some topics (particularly the N/As) may need further detail in their relevance to the central theme of droughts. Clarifying could improve the interpretability of results. Further, it's the ordering of Table 1 is confusing when it's not numerical.

Topics which cannot be labelled are a typical output of the unsupervised process involved in topic modelling. Some authors choose to simply not report these topics; we think it is better to present them and not label them. We will revise this table, and the connected text to make these important methodological points, and their implications for our study, clearer.

8. The authors did a good job to ensure that they attribute and discuss the appropriateness of your methods through citations. 

Thank you!