General comments

This a valuable contribution as it links drought entries in a recently constructed disaster database (GDIS), which is based on EM-DAT, with actual droughts identified based on a state-of-the-art reanalysis dataset. The work is important because typically disaster databases are not based on information derived from meteorological variables and establishing such links gives credence to the databases but also serves to identify hazard thresholds and differences in vulnerability. 

Generally the manuscript is very clear and easy to follow but it would benefit from a more in-depth discussion of a number of points that are only briefly mentioned. More detailed comments and suggestions follow below.

L10: “We found that ERA5-Land soil moisture accurately captured the socio-economic impacts of drought shown in GDIS.” I think I understand what the authors mean but as it is this statement is not correct. ERA5-Land cannot capture “impacts of droughts”, rather it provides information that droughts actually occurred when disasters labelled as drought-driven where recorded. Similar statements can be found in other places throughout the text. Please adjust the language to more correctly represent your findings and the relationship between drivers of disasters and the recorded impacts.

L12: “were robust”: better us “were less vulnerable”

L106: The authors specify some criteria for the analysed events but it seems the number of analysed events is not affected by this. In line 100 it says “The 282 drought events…were analysed”, which suggests that this is the number of events contained in GDIS. In Line 162 it is stated that there are 282 drought events. Please clarify how many events are available in GDIS and how many are finally analysed in this study.

L 163: “We recognized that the drought indices successfully capture the GDIS drought events if the two distributions are not statistically the same.” This is only true for the full distribution, not at the individual event level. There could still be many events where no drought is evident from a soil moisture perspective (there is also a strong overlap in the distributions). Furthermore, the test doesn’t tell you how the distributions differ. Theoretically it would be possible that all disaster events show less extreme drought conditions compared to the control and the KS test give significant results. Please adjust this statement accordingly and make it more nuanced. This should also be reflected in the abstract.

L 165: unclear what “socio-economic drought events in GDIS” means. I assume “socio-economic droughts” are the droughts events and impacts recodred in GDIS?  Better use something like "drought disasters" or similar

L170: Initially I wasn’t completely sure what the purpose of the drought clustering is. I assume the idea is to check whether spatially large droughts typically also lead to impacts (as recorded by EMDAT/GDIS). Given that the drought definition is percentile-based, every location experiences drought with the same frequency, and differences from a rather homogeneous distribution (as for instance visible in the US, Canada and Russia in Fig. 8) are driven by the distribution of continents and the choice of the spatial cutoff (100000km2). With this background, this could be a useful analysis but it would be good to motivate it better and discuss the results in more detail (e.g. it seems that some regions experience more contiguous/large-scale droughts than others, why?). The finding that drought disasters tend to occur in regions that are characterise by frequent large-scale droughts is then quite interesting and novel, especially because from a meteorological perspective and at a pixel level, the frequency of drought occurrence is the same everywhere (20% of the time in this study). So it seems that drought disasters occur when droughts occur over large areas. These findings could be described and discussed in more detail. 

Another point to discuss is that you’re using a relative percentile, which means that in generally wet regions, a drought defined in this way might not be that impactful because the absolute amount of available water is still quite high. This will to some extent also determine where drought disasters occur and might confound the vulnerability assessment. It makes sense to use a relative percentile given that ecosystems/societies are usually adapted to the water availability in their region. However, it is worth discussing this choice.

L206: You could mention some more details in the results. For instance, how large are the differences between the medians for the different soil moisture layers in Figs 3 and 4. I assume the SDI for the whole period is approx 1 by definition (Fig. 4)? I would also mention that for clarification.

L219: It maybe worth checking how these regions differ in their absolute SM values (averaged over time). Regions with lower water availability in general might experience drought disasters more frequently even though the relative deviation from normal conditions is small (see the comment higher up). Independent of the findings, the interpretation that the identified regions are more vulnerable to drought probably still holds, but I think it can slightly change the interpretation and consequences for resilience planning. If water is typically abundant, it’s much easer to be drought resistant.

L258: Note that the results of Fig. 5 also confirm results from Tschumi & Zscheischler (2020) who also found smaller climate anomalies in less developed countries during EMDAT disasters for different climate variables (their Fig. 9). 

L280: unclear what “reproducibility of ERA5-Land” means. Please clarify.

L295: again, unclear what “reproducibility” of reanalysis products means. It seams the authors mean that using different climate datasets, one obtains similar drought maps (which makes sense) whereas this does not hold for socioeconomic impacts. This of course depends on how these things are defined. If the same drought definition is used, the choice of (climate) dataset has only little effect on the analysis as this information is comparably well constrained by observations. For socio-economic impacts, no clear/objective definition of disasters etc. exists and uncertainties in impact estimates are usually high. So a comparison across datasets is difficult also because different datasets use different definitions.

L328: “GDIS only covers about 60% of droughts in EM-DAT” Can you comment on why this is the case? Was more detailed information on the remaining events not available?

In this manuscript, the authors use the new GDIS data and link it to hazards (i.e. soil moisture). The contribution is timely and innovative. I enjoyed reading the paper. Currently, we have little understanding of the relationships between impacts and hazard data. Hence, the proposed approach provides a way to understand these linkages. I particularly appreciate that instead of the traditional approach (e.g. hazard x vulnerability x exposure = risk), the authors go from the disaster itself, and from it try to identify the hazard drivers.

The manuscript is well written, yet, some issues need to be addressed:

• My main criticism is that the authors claim to have assessed the “regional vulnerability to drought” (see line 258). Figure 5 shows that the SDI alone cannot explain drought disasters. Indeed, in Sub-Saharan Africa and South Asia, the SDI is not as high as in Europe and North America, yet, many disasters are observed. One explanation for this is indeed, the socio-economic vulnerability in these regions is higher. However, it could also be that it is the population exposure. Since you have not accessed any of these, you cannot make claims regarding the vulnerability.
•  
• You mentioned that you disregarded drought events that lasted for less than 2 months. However, I assume most of the EM-DAT data only has a year, and few have a start and end date. In such cases, you mentioned that you considered the whole year as a drought year. That can be very problematic, as probably many of these events could actually be 2 months. It would be great if you could bring this limitation in the discussion section, where you already mentioned the limitation of the impact data.
• Why was landcover data from 2020 used (line 127) if the drought events go from 1964 to 2018? Likewise, why did you consider soil moisture data from 2019-2020 (line 132) when your impact data is from previous years?
• Item 3.3 and 4.2. Similar to the first comment, stratifying your SDI data by different regions is not equivalent to “understanding the vulnerability in each region”. What you did here was to analyze the SDI according to different regions. This is definitely an important analysis, but it is not a vulnerability analysis. You can hypothesise that the vulnerability causes these differences, but you cannot confirm this with the present data. The statements in line 255 are pretty strong and cannot be made without considering the socio-economic vulnerability. I suggest toning them down and writing that “the vulnerability could explain these differences”. However, for this, further analysis is required.
• Item 3.4. It is not clear to me what the advantage of clustering is. Why would the results be different with or without clustering? Could you please elaborate?
• In line 259, you say that your findings are consistent with previous ones that used GDP, however, you have not used GDP here. Please rephrase the sentence, or do the analysis
• In line 266, you write that the vulnerability is explained by the lack of structural measures. Sure, these are contributing factors, but socio-economic variables and exposure could be even more important. I understand the author’s point of view, being an engineer myself. However, “command and control” approaches in which drought risk is controlled solely with conventional engineering measures is an outdated view. The authors cannot ignore the vast literature on socio-economic vulnerability and risk perception. You do not need to cite these papers, but have a look at the GAR chapters on vulnerability and droughts:

https://gar.undrr.org/sites/default/files/chapter/2019-05/Chapter_3.pdf

https://gar.undrr.org/chapters/chapter-6-special-section-drought

• Line 282: Spain is not shown in Figure 5. Hence, you cannot make this statement based only in the Figure. It could well be that the other European countries are “robust” and push the mean up, whereas Spain pushes the mean down. I suggest removing the sentence or doing the analysis for Spain only.
• Line 301: I find it good that you can find good linear correlations. However, I was surprised as soil moisture and drought impacts usually do not have a linear relationship. Indeed, there are non-linear relationships (there is a lag between soil moisture and forestry impacts for instance). Furthermore, you have multi-year drought events, which add complexity to the interpretation of spatial and temporal differences in drought prevalence. Indeed, it could be that a drought is not so severe in biophysical terms. However, because they last so long, the impacts are high. I assume you got good correlations because EMDAT usually do not have long term impacts but rather focuses on immediate ones. Please add a discussion of these issues in your discussion section.

Some minor comments:

Line 9: What is meant by “drought information” here? Do you mean Drought impact information (i.e., the GDIS data?). Please be more specific.

Line 55-57: I do not understand the purpose of these 3 last sentences. When you write about vegetation growth, do you mean forest or crops? Regarding the last sentence. Why do we need to treat drought impacts as they are socially perceived? If you think the section is relevant, please expand it to make it clearer to the reader.

Line 95: This information is repeated in the introduction. I suggest removing it: “GDIS is the geocoded….”

Line 122: You do not “show the drought vulnerability”. Land cover is an exposure data, not vulnerability.

Line 207: “significantly higher”: could you please provide the test results that made you reach this conclusion.