The present paper sets out to compare different datasets and methods for estimating the economic value and usage types (exposure modelling) of buildings in Germany at the object level (building-based). Conventional risk models characteristically present data in aggregate form for extensive regions (e.g. neighbourhoods or cities). In contrast, the present article aims to determine the most transparent, sustainable and accurate building-based exposure model for Germany. The development of a hand-labelled benchmark dataset represents a significant contribution to the field. The study is suitable for publication, but several issues require further clarification and discussion. 

Introduction: The introduction offers a strong and comprehensive overview of exposure modelling, clearly motivating the need for object-level approaches in impact forecasting and local risk management. The discussion of asset value concepts such as replacement cost, depreciated cost, and net asset value is appropriate, but their relevance for different modelling applications could be stated more explicitly. Clarifying early on that the study intentionally compares models with different cost bases would help frame later results and avoid potential misinterpretation.

Data and Methods: The selection and description of datasets are thorough and well justified, and the modelling workflows are described with commendable transparency. The use of LoD1 cadastral data, Eurostat accounts, BEAM, and EHRE reflects realistic choices faced by exposure modellers. The benchmark dataset is a major strength of the study, but its limitations should be more clearly emphasised. In particular, the reliance on a single region with a predominantly residential building stock raises questions about representativeness, especially for industrial and service buildings. A clearer discussion of potential benchmark uncertainty and regional bias would strengthen the methodological credibility of the evaluation.

Results: The sector classification results are clearly presented and reveal important structural patterns. The strong performance of LoD1-based models for residential buildings contrasts sharply with their near-complete failure to identify industrial assets, highlighting a fundamental limitation of cadastral building function categories rather than a modelling error. This finding is important and should be more explicitly framed as a warning against uncritical use of authoritative building function data for economic sector classification.

The comparison of regional asset values shows substantial divergence between models, underlining that exposure-related uncertainty can be of similar magnitude to uncertainty in vulnerability modelling. While the manuscript correctly attributes much of this divergence to differences in cost basis, the interpretation would benefit from more clearly separating effects driven by accounting concepts from those driven by spatial or sectoral disaggregation. The per-asset comparison against BKI construction costs is informative and convincingly demonstrates systematic underestimation, but it should be stated more explicitly that this reflects conceptual alignment with stock-average or depreciated values rather than an inherent model deficiency.

Conclusions and Limitations: The conclusions are well aligned with the results and appropriately cautious. The emphasis on transparency, maintainability, and local validation is well supported and constitutes an important message for the community. However, the limits of generalising the findings beyond the Ahrweiler region should be stated more clearly. Strengthening this point would enhance the credibility of the study without diminishing its contribution.

I would like to thank the authors for sharing this interesting piece of work. Overall, the piece is well written and interesting. I do have (perhaps obviously as a reviewer) a couple of comments and remarks.

• I would consider rephrasing the title a little. The authors are indicate in the discussion that the results cannot directly be generalized to the rest of Germany, because it was such a specific use case. Of course, I do agree that certain results could be generalized across Germany, but it would require a couple of more case studies to validate that. As such, perhaps it would be better to emphasize the Ahrweiler region as well in the title.
• The introduction is well written and I do not have much to say about it. Besides that it reads very well.
• In the methods, I miss a brief section that explains how the different models/datasets are being compared. Given that you compare them based on precision and recall, it is generally common practice to already explain this in the methods section.
• For the OSM classifications, what is the ratio of CLC vs OSM land use categories. And for consistency, why not fully base this in CLC, instead of only filling the gaps. Especially given the OSM results (and that OSM is also used a lot, so many people are interested in how OSM performs), it would be nice to go into a bit more detail on this. I would be curious to see how OSM performs when fully using CLC land cover classes.
• To better understand the results, how do the different geometries compare of the buildings? That is not entirely clear yet. LoD1-based and OSM-landuse have all 844 objects, but EHRE has only 699 but it is based on OSM (and some additional 100x100 polygons added?). Line 311 to 329 also reflect a bit on this, but I find it a little bit hard to follow (especially the bit from 321 to 329).
• I also do not fully understand why OSM was omitted for the total asset value Many people use OSM, so it would actually be interesting to see how this compares. I mean, the other datasets also have a very skewed distribution of unit values, which are also slightly hard to follow sometimes on why that is precisely the case. I am not sure if the wrongly mapped service section would be a sole reason to not estimate total asset value. Or at least it would be interesting to see how that would be different. Especially also because industrial was actually very well mapped for OSM? And given that in many countries (outside of Europe) Lod1 type of data is typically not available, I would still proceed on including an OSM-based approach to the end, also to “warn” users on the potential pitfalls. Could be a good source for many people to cite 😊
• I propose to remove Table 6. Next to the subjectivity of the weights and scores, they also different so little (6.9 vs 6 vs 6.3) that it still does not really say much. I think lines 416 to 436 already explain in enough detail how the models/datasets vary.
• In line 444, one could also argue that risk models, especially for flooding, often do not include this rich detail in exposure characteristics. A more coarse but well-validated unit value through a “simple” sector-classification would already be a very good baseline to aim for.

The manuscript entitled „Building-level exposed asset value modelling for Germany “ presents the development of an exposure asset values dataset for Germany, having the district of Ahrweiler, a recently flooded area, as case study. The goal of the work is much important for disaster risk studies, given that the exposure is an essential component for impact and risk assessment, but is heterogeneously handled in such studies. Despite not being a novel research question, it is still a timely needed development, as new auxiliary datasets become available each year. The developed dataset mixes various object datasets and asset values datasets. Some of them are freely available, others are not. Moreover, features of each dataset vary, as some object datasets are at a land-use scale, others at a building scale. Values datasets also have different features, some based on the replacement cost, other on the depreciated cost. Making them compatible is perhaps the main challenge of this development.

That said, the manuscript requires major revisions to clarify each step of the development as well as how these incompatibilities were handled. Many steps are unclear. Comparing seemingly incompatible objects might be possible, usually by making certain assumptions and clarifying them beforehand. Some of these unclear points, depending on how they were handled, could undermine the validity and value of the final produced dataset. Below is a list of specific comments.

I would suggest a complete restructuring of the manuscript. Instead of focusing on candidate “models”, I suggest presenting how each aspect or component was tackled, that is: (1 – objects) how different building function, land use classes, etc. were harmonised; (2 – economic values) how the economic sectors were harmonised; (3) how the sectors and building functions were matched; and (4) how replacement costs and depreciated costs etc. can be compared (and if not, what were the assumptions to make the comparison in this manuscript). In the current form, these questions are presented per candidate model, but very unclear, and above all, when comparing them it is also unclear.

With that suggestion I bring another point: over reliance of supplementary material. This work definitely needs supporting material (the suggested restructuring would not avoid that). But in many places an average or range of values could be given in the main script, providing the reader a quick and valuable reference without constantly flipping to the supplementary tables.

When the candidate models are compared, mostly on the object-level, another very unclear part is about the mismatching building functions, what happened to them, how they were dismissed or counted separately (?).

Before any classes are reclassified or adjusted or partially dismissed, do the original datasets agree on the TOTAL assets within the pilot region?

I highly appreciate the work, as this is a very important component of risk assessment and in need of more studies. And I highly commend the effort to match seemingly incompatible datasets. However, there are still too many unclear steps that undermines the value of the final work as it is currently presented.

It is a pity that the final published data (the Sanitized data) includes the chosen economic sector, but not the original building function, which could make transparent the matching procedure adopted.

Specific comments:

Title: “exposure” instead of “exposed”

Line 48: “e.g. by federal state in 1946” seems to be a very German-specific information. Please clarify.

Line 63: specify better what is meant by “procedure for estimating exposed assets”.

Lines 83-85, 98-101, and maybe elsewhere: In the Introduction much is said about loss modelling, giving context and objective for this development. However, exposure and loss should be kept separate. Some references given (and some other points in the Introduction) seem to be only about loss modelling and not the exposure.

Lines 90-95: Would it rather be “per asset type” than “per-asset” or “per-building” replacement cost?

Fig 1.: Provide sources of rivers and administrative limits datasets.

Line 122: The gathered data are not all “freely available”, as in Tab. 1.

Tab. 1: could be separated per type of information, similar to Tab S2.

Line 152: BEAM was developed “for” the JRC, not “by”.

Ch 2.2.2 + Tab2: BEAM already explicitly separates “building” and “content” values. Was there a need to do that again?

Lines 171-174: If EHRE values are based on the Census 2011, are the values “referenced to 2020” or corrected to 2020?

What is meant by “Supplementary Data Table 1”? It appears many times at different points. Should they be one of the suppl tables?

Candidate 1 (2.3.1):

I understand that there are different pathways for Residential and Non-Residential assets. Fig 3, however, does not make that clear. The text also goes back and forth about it.

Tables S3-S6 could be better condensed (which is only 1 step before comparing to the other datasets in the other candidate models; reinforcing what I suggest above about restructuring the manuscript).

Would it be more correct to say this candidate is LoD1 + Eurostat + Insee?

Data 2.2.2. and Candidate 2.3.2:

BEAM values are NUTS2-level averages of €/ per “land” m², per land-use type; that is, not per m² of building, but of land use. That means, the residential building assets € per m² of e.g. dense urban fabric will be the same in all dense urban fabric within the NUTS2. It does not differentiate per building, but per land use class, regionally at the NUTS2 level. This is not wrongly explained in the manuscript, but also not clear.

Moreover, why did you sum the total value in Ahrweiler to redistribute per building instead of keeping them per polygon (land use type, differentiating for example low dense and dense urban areas)?

Candidate 2.3.3:

Line 255-258: Do I understand correctly, the EHRE-class “other” includes the OBM buildings for which there is no equivalence in ESRM20 asset values and are therefore dismissed. The EHRE-“mixed” do have asset values in ESRM20, and are matched to the benchmark-dataset “ambiguous”, is that correct? Fig 5 does not list the mixed-ambiguous class.

Candidate 2.3.4:

Why give preference at the land-use level to OSM over CORINE?

Benchmark:

Line 277: such a mixed used is an existing class in ALKIS/LoD1.

Line 283: where does this regional factor comes from?

Fig7: I understand that Agriculture is not considered in the Benchmark. But what happened to Eurostat “Production” and BEAM “Industrial” sectors?

Line 298: The “many industrial or ambiguous” are exactly 66 industrial and 68 ambiguous, correct (based on Fig 7)?

Lines 298-300: What is meant by “insufficient resolution”?

Line 311-312: Why? Chapter 2.3.3 does not mention LoD1 at all.

Line 316: EHRE-unmatched are still shown in Fig 7. What was discarded from other models?

Line 333: “over-estimation”?

Line 339-340 – As far as I know, building functions in LoD1 follow the ALKIS standard. If they do, there are “industrial” labels in it, as well as mixed uses (e.g. residence with commerce).

Line 355-356: confusing sentence, please rephrase.

Area BEAM-Industrial and Eurostat-Production equivalent?

Tab 3:

-I suggest presenting all values at Mi € or all at 10^9

-Market services with 23€/m seems too low.

-Are residential m² of footprint or floor area? (please, clarify this everywhere in the manuscript)

-What are “remainder” in the EHRE? How are there decimal-count of buildings?

Lines 361-362: Rewrite in a form not to mix LoD1+BEAM €/ building m³ with BEAM’s DC, which is per land area m².

Lines 367-368: Both have only 19 buildings.

Tab 4: I would expect Benchmark classification to be objective and follow Table S12.

All criteria here were given subjectively? By one author or as average of all authors? With small or large divergences?

Data availability: Not all data sources are freely available.

[optional] Lines 45-58 + Lines 376-378: In these paragraphs, the different types of values in loss modelling are discussed. I would suggest taking into consideration the paper of Molinari et al. 2020, which has loss models using different asset values as well.