In line  33 we read "One method to partially address spatial mismatch between administrative and other geographic units such as natural hazards". As a matter of fact administrative boundaries are a way to measures land and territoires so there is no mismatch. the Auhtor should rephrase the sentence

in line 90 and further we derive the agricultural GDP as multiplying derived quantity for wholesale price in FAOSTAT: however we don't know if ALL THIS PRODUCTION will be sold, so I would suggest terms as "possible or potential GDP" more than GDP.

The Estimating Local Agricultural GDP across theWorld paper s a very interesting article, and faces in an innovative way the issue of integrating offical economic statistics, often scarse, with geospatial data available and of high quality. 

However, is it suggested a re-wording in the title and in the text for the term "agricultural GDP". Techically speacking Agricultural Value Added (which is a percentage of GDP) is more precise. 

Moreover the Authors should better explain how of the total livestock, crops etc. that could teoretically contribute to the Agricultural value added are netted out of the quantity related to natural losses, self-consumption by farmers, or simply are unsold in the market. In all these cases we have a physical quantity that do not reach the buyer, and therefore can't contribute the the agricultural value added as meant by the SNA and economic statistics. If these aspects are not considered by authors, it is suggested to use the term " potential value added". 

with these changes and/or further explanations, the article is welcome to be published.

1 The manuscript:

1.1 Are the data and methods presented new?

The AgGDP dataset is unique. The methods used to generate the dataset are not new per se, but it is the combination of diverse input data and the use of multiple methods to generate AgGDP that is new.  I commend the authors for taking on the challenge of producing this important dataset

1.2 Is there any potential of the data being useful in the future?

Yes I am sure the layer will be used in many future global modelling exercises that require detailed agricultural economic activity/output information. See comments in the data quality section that would aid future users in their understanding and usage of the dataset.

1.3 Are methods and materials described in sufficient detail?

The methods and data section is comprehensive.

I recommend some discussion on the impact of choices/assumptions such as those made on line 144.  This is just one example, other assumptions should also be addressed in the discussion.

1.4 Are any references/citations to other data sets or articles missing or inappropriate?

I did not miss anything.

1.5 Is the article itself appropriate to support the publication of a data set?

Yes with modification following the recommendations below.

2 The data quality:

2.1 Is the data set accessible via the given identifier?

The data is accessible at the following location https://datacatalog.worldbank.org/search/dataset/0061507

2.2 Is the data set complete?

The data and metadata are incomplete

• The metadata is quite sparse (perhaps limited by the WB Data Catalog format)
• The units are not mentioned in the metadata.
• Why is the dataset floating point? This level of precision seems unjustified – is it float because integer formats cannot deal with the large range of values? Even so, reporting GDP in USD to decimal places seems unjustified.
• The metadata does not link back to the preprint.
• The first published date is after the last updated date – please check and corect.
• Sea areas where no AgGDP data is possible (because marine-based AgGDP is allocated to land) and territories where there is no data available (due to a lack of data for now) are treated the same – this is not very elegant. Consider using different pixels values to distinguish these two "no data" types.

2.3 Are error estimates and sources of errors given (and discussed in the article)?

No. There are no error estimates or validation data in the dataset, though they are discussed in the article. See below for comments on validation and sensitivitiy analysis.

2.4 Are the accuracy, calibration, processing, etc. state of the art?

The pre-processing steps to generate the AgGDP dataset are appropriate. The section on uncertainties in each input layer and how those uncertainties may be compounded when they are combined is rather brief.

Some quantification via a sensitivity analysis would be a welcome addition to the paper. The lack of a quantitative uncertainty assessment is a weakness of the paper and the dataset in the absence of a robust validation.

2.5 Are common standards used for comparison?

Th AgGDP dataset is correlated with night time lights (section 2.5). This somewhat contradicts the introduction that states that night timelights are not always a good indicator of agricultural economic activity. The choice is not well justified. I understand that the AgGDP dataset is unique and depends on many input datasets on production value (limiting the availability of possible datasets against which to validate), but I would like to see a much better choice of validation/comparison with a strong justification too. 

The statements on lines 263 onwards are not sufficient indicators of quality. This section can be substantiated by reference to national studies that have also spatially decomposed AgGDP or similar measures of agricultural economic output.

The correlation table (Table 2) is a very high-level aggregation, which is does not reflect the highly spatially disaggregated AgGDP and Night Lights data. It is not very convincing or useful.

The validation section seems to compare this cross-entropy model against spatial allocation model  based on rural population.  The description of the rural population based comparison dataset is not sufficient for a reader to fully understand what it is, how it was made and thus what is being compared against what. I assume that national and subnational Ag GDP is disaggregated based on head count giving every rural person an equal share of the AgGDP? But I am guessing. 

Either way, the validation is a case of comparing one model against another with the argument that the assumptions in one model are more valid than those in another.  This is not a very satisfactory validation and I amnot sure what message is intended by showing the two have different degrees of correlation in differnet parts of the world.

Why exclude areas from the analysis with values that are less than 200,000 (USD?) ?. No summary or regional statistics on the correlation are provided. This cprrelation map and the rural per capita GDP should be provided as spatial datasets with the AgGDP dataset.

2.6 Is the data set significant – unique, useful, and complete?

From my perspective the AgGDP dataset is unique.

I have given recommendations above to make it both useful and complete. In addition to that, I recommend that a table of the production values (priors) per country and the collated GDP data would be very valuable additions to the dataset. Where appropriate these layers should also be provided in spatial data formats. This would help users understand the spatial patterns and artifacts in the AgGDP (alloc.tif) dataset and help ensure appropriate use

3 Article and data set:

3.1 Are there any inconsistencies within these, implausible assertions or data, or noticeable problems which would suggest the data are erroneous (or worse). If possible, apply tests (e.g. statistics). Unusual formats or other circumstances which impede such tests in your discipline may raise suspicion.

I note the edge effects above which could give potential users pause for thought before using this data

The authors could subnational representations of the data, both spatial and tabular which would make it easier to assess whether there are any noticeable problems due to modelling or assumptions.

3.2 Is the data set itself of high quality?

The effort is impressive; detailed estimates of Ag GDP are valuable. 

The datasets value is detracted from by the lack of validation data, spatial artifacts and the presentation of the data (level of precision is not justified, file name choice, sparse metadata, lack of tabular summaries).

4 Presentation quality:

4.1 Is the data set usable in its current format and size?

The format is suitable for use in both open and proprietary GIS software and can be easily read in open source sofware such as R or Python for statistical analysis

There are edge effects in northern latitudes and on some country and subnational boundaries that are rather inelegant – can these be dealt with better?

Recommend to change the file name from alloc.tif to something more meaningful

Recommend the dataset is converted to integer not float – both to reduce size and to be more realistic about the precision og the GDP estimates. Estimates could even be rounded up to the nearest 1000 USD.

Recommend that the different types of no data are treated differently

4.2 Are the formal metadata appropriate? 

See previous comments on metadata completeness – this needs to be addressed.

5 The publication:

5.1 is the length of the article appropriate?

Length is fine, but more space can be given to (i) a more robust validation or (ii) a sensitivity analysis to understand the impact of choices in methodology and/or the contribution of the uncertainties in the input layers.

5.2 Is the overall structure of the article well structured and clear?

Structure is fine though the natural hazards component seems like an add-on that does not add much value to the paper and dataset, which is really about AgGDP.  The hazards part is one of many possible applications.  Is it essential to the paper to focus on one use case like this?   If a use case is a requirement of the journal then fair enough.

Starting the conclusions section with a paragraph on hazards is a curious choice given that this is not the core purpose of the paper. Again if this is a requirement of the journal then fair enough.

5.3 Is the language consistent and precise?

The language would benefit from professional English editing.  The text is largely understandable but many lines in the text jar due to non-standard English. This reduces the readability. I had to pause and re-read some lines several times, e.g., lines 17 and 18.  The dataset and documentation is an extremely valuable resource and I commend the author’s efforst for developing it; please bring the text up to the same level of value as the data.

Line 8

The paper estimates the exposure of areas with at least one extreme drought during 2000 to 2009 to agricultural GDP is an estimated US$432 billion of agricultural GDP circa 2010, where nearly 1.2 billion people live.

Alternative We estimate that US$432 billion of agricultural GDP (circa 2010) was exposed to at least one extreme drought during 2000-9.

If hazard exposure is important, consider adding it to the title.

Line 2 of the abstract is hard to parse

Line 6 – consistency needed – either small “a” on agricultural GDP throughout the paper or capital A.

Line 12 remove “the” , same in line 15, same in line 92 and many other instances of non-standard use of the definite article Check and correct throughout the text.

Line 15 – location variation in what?

Line 15 – the possible implications of the mismatch are not clear

Line 72 –AgGDP not agricultural GDP – check paper that this abbreviation is used henceforth.

Line 79 – what does efforts varied mean?

Line 93/94 is the repetition necessary? Aim to be concise.

Line 118 – clarify the pixel areas. Is this land area or simply the total area of each 5 min pixel? Depends on how the densities were computed, but this is not clear from the paper.

Line 125 – first sentence seems superfluous. Also was the start of civilisation really the first use of forest resources? What about hunter/gatherer societies?

Some statements are superfluous and can be removed.For example

Line 270 The correlation of AgGDP with night light varies across world regions as it requires areas to emit light (Table 2).

Line 279 The exposure to drought is not uniform across the world.

Lines 303-304 are more or less repeated in lines 317-318.

5.4 Are mathematical formulae, symbols, abbreviations, and units correctly defined and used?

Yes

5.5 Are figures and tables correct and of high quality?

Maps are clear.

Figure captions do not need to start with “This map…” Just state what the figure shows.

Recommend an Equal Area projection and remove the E and N coordinates and graticules– they do not add useful information.

Check capital letter usage in map legend title; production instead of Production

Table 1 is shown before it is referenced in the text. Check capital letter usage in column headings in tables. Table 1 caption is not self explanatory – conversion factor should be explained.

Rating

On a scale of 1 (excellent) to 4 (poor) I would give the datasets and paper a 2.5 at the moment with the potential to be closer to 1 than 2.  The dataset is unique, potentially significant and will be widely used. To reach it's full potential users need to fully grasp how it was made, what the inputs were, where the major uncertainties are and thus how to properly use the dataset in further research.  These are areas for improvement (completeness and data quality) in the manuscript and associated datasets that could be included with the AgGDP layer.  The presentation quality of the manuscript can be improved - see comments above to do justice to the impressive work conducted so far to produce this unique global spatial dataset.

This paper extends the authors’ previous Brazil-focused work on gridded agricultural GDP to the entire world. It uses cross-entropy optimization to disaggregate an impressively large number of national and subnational datasets to 5 arc-minute grid cells. The authors previously applied this method to the development of gridded global crop data (“MapSPAM”). Their development of gridded agricultural GDP in this paper is a valuable advance given that agriculture is essential to human survival, remains the dominant economic sector in rural areas of most countries (especially low-income countries), and is threatened by climate change. The authors necessarily make some strong assumptions to construct this new dataset, but I don’t view their assumptions as being any less tenable than those that underlie standard national accounts statistics.

My overall reaction to the paper is favorable, but I have two general suggestions for making it easier to understand and more convincing. First, it can be organized better. It flows naturally through section 2.3. After that point, I suggest reorganizing it as follows:

• Create a new section 3, titled “Results and Validation.” This section would begin with the presentation and interpretation of the new dataset on gridded agricultural GDP, which is displayed in Fig. 7. It would then compare the new dataset to the night-time lights (NTL) data, the point of which (as I understand it) is to demonstrate that NTL is not a good proxy for gridded agricultural GDP. Nor are gridded total GDP or gridded population (Table 2). Hence, the new dataset does indeed provide new information. The material in current section 2.5 would be integrated into this new section.
• Confidence in these findings depends on the validity of the new dataset, so new section 3 would next cover validation. This subsection would begin with the acknowledgment of limitations of the new dataset presented in current section 3.2 (including 3.2.1 and 3.2.2) and wrap up with the presentation of the validation findings in current section 3.1 (including Fig. 10).
• A new section 4 would follow and would be titled something like “Illustration of use: drought risk.” It would integrate information from sections 2.4 and various parts of section 3 and would include Fig. 5, 6, 8, and 9 and Table 2.
• The paper would finish with the existing concluding section.

Second, and more substantively, the authors need to address several issues with the construction of the wood production component in section 2.1.3:

• The Lebedys and Li (2014) estimates used by the authors are, to my knowledge, the best available estimates of forest sector GDP, but they focus on industrial roundwood (and products derived therefrom) and largely exclude fuelwood, which accounts for half of global wood harvests. As a result, even allowing for fuelwood’s unit value being much lower than industrial roundwood’s, the current wood production component underestimates the contribution of wood harvests to agricultural GDP. The easier option for the authors would be to stick with the current estimated component but acknowledge that it underestimates the wood harvest value. The harder option, but the one I encourage the authors to consider, is to figure out a way to add the value of fuelwood harvest to the component. Fuelwood harvests are usually correlated with the collection of nonwood forest products, so perhaps the authors can use information in Siikamaki et al. (2015) to impute gridded values for fuelwood harvests. I note that Siikamaki et al. refer to some of the studies they reviewed as having included information on fuelwood values. Annual data on national harvests of fuelwood from FAOSTAT-Forestry might also be useful in the imputation.
• The authors write, “The value of wood products per pixel is calculated based on forest loss from year 2010 to year 2011 ….” This statement requires qualification and, ideally, some additional analysis. The MODIS dataset the authors use to calculate “forest loss” measures tree cover, which includes perennial tree crops such as oil palm plantations, cocoa plantations, orchards, etc. in addition to wood-producing forests. This is a well-known deficiency of satellite-based “forest cover” datasets (Tropek et al. 2014; https://www.science.org/doi/10.1126/science.1248753). The authors’ estimate of “forest loss” thus includes the replanting of perennial tree crops that occurs when the trees have reached the end of their economic lifetime. The resulting upward bias in “forest loss” can be substantial. For example, oil palm is replanted every 20-30 years, which implies a 3-5%/year “deforestation rate” that is many multiples of the annual loss rate for true forests reported in standard sources (e.g., FAO’s Global Forest Resources Assessment). Fig. 3 in the paper illustrates this problem, as it shows wood production occurring in parts of Malaysia and Indonesia that are virtually 100% oil palm plantations. I know there are remote sensing products that show the locations of oil palm plantations, and perhaps there are ones for other non-forest tree crops too. I encourage the authors to use these products to estimate forest loss more accurately by masking out areas with tree cover that are not forests.
• Also requiring qualification is the statement, “forest loss due to fire should be removed because it does not result in wood products.” Land clearing often involves a first stage of wood harvests followed by burning to eliminate remaining vegetation and woody debris. The authors’ assumption that wood harvests do not occur in areas with fires thus results in underestimating the area harvested for wood products. I can’t think of a way to fix this problem, but the authors should acknowledge it.

Specific comments

1. Abstract: State the year of the new gridded dataset, i.e., 2010. Precede the penultimate sentence on the drought analysis with a phrase like, “To illustrate use of the new dataset, the paper ….” Such a phrase would clarify that the paper is not primarily about drought risk. The paper would need to be completely rewritten if that were the case.
2. Line 22: The authors could note that detailed agricultural data are also needed to evaluate forest restoration opportunities (e.g., P. Shyamsundar et al., “Scaling smallholder tree cover restoration across the tropics,” Global Environmental Change 76, 2020; https://doi.org/10.1016/j.gloenvcha.2022.102591), which have become a focus of “nature-based” climate solutions (B. Griscom et al., “Natural climate solutions,” Proc Natl Acad Sci USA 114, 2017; https://doi.org/10.1073/pnas.1710465114) since the launch of the UN Decade on Ecosystem Restoration (https://www.decadeonrestoration.org/). Agriculture is the main land-use competitor for forestry. The dataset developed in the paper will help researchers and policymakers better understand the opportunity cost of converting land from agriculture to forest.
3. Line 45: Given that the paper is about GDP, “income” would be better than “wealth.”
4. Line 50: The phrase “the uniform distribution of labor in agriculture is another key concern” is vague and should be clarified.
5. Line 65: The authors refer to two main contributions of the paper, with one being the drought analysis. I view that analysis as an illustration of the use and value of the new dataset, not as a main contribution. For the latter to be the case, the authors would need to provide more context for the drought analysis and evaluate it more directly against prior analyses. Constructing the new dataset is a sufficient contribution to justify publication of the paper in my view.
6. Line 92: State the year of the producer price data. 2010? Mean of 2009-2011? Relatedly, the authors need to explain somewhere whether their agricultural GDP estimates are purchasing power parity (PPP) estimates or market-price estimates. Information in footnote 9 is pertinent to this point. The authors should explain the implications for interpretation of the dataset if the prices that underlie it are not using consistently defined (i.e., some prices are in PPP terms while others are market prices).
7. Footnote 3: This point should be incorporated into the text. Prices for agricultural, forestry, and fishery products can vary greatly within countries and their subdivisions. The use of national prices is unavoidable given current data limitations, but it is a shortcoming of the new dataset that the authors should acknowledge in the text.
8. Line 116: The use of uniform livestock conversion factors across countries seems like an unnecessary simplification. Why not use country-specific FAOSTAT data on the value of products from each type of animal?
9. Lines 130-132: The authors’ use of forestry terms is unconventional. I recommend the following rephrasing: “The trees are harvested for fuelwood and industrial roundwood, which is processed into a variety of products including lumber, plywood, furniture, and paper products.” Mentioning fuelwood is necessary given that it accounts for half of global wood harvests.
10. Footnote 5: The MODIS land cover data used by the authors is quite coarse, ~500 m at the Equator. I doubt it reliably measures selective harvesting or forest degradation. I recommend rephrasing the footnote as follows: “The measurement is limited to detection of land cover change from satellites and might not fully account for selective harvesting or forest degradation.”
11. Lines 188-189: Mention the typical level of the subdivisions in the dataset here or earlier. Lines 230-231 imply they are mostly Level 1 subdivisions (i.e., states or provinces).
12. Lines 214-219: The authors state, “Theoretically, the sum of these components should be close to the official values obtained from the World Development Indicators.” This statement prompts two thoughts. First, as part of the validation of the new dataset, I recommend presenting information on the ratio of the sum of the components to the official values and interpreting any systematic discrepancies that are observed across regions, countries, or subdivisions. Second, I wonder whether the components the authors have constructed actually correspond to GDP components in all cases. GDP refers to value added, i.e., output value minus expenditure on intermediate inputs. I believe that some of the authors’ components refer to output value (e.g., the crop and livestock estimates) whereas others refer to value added (e.g., the forestry estimates). If I am correct, then there is a conceptual inconsistency across the components that the authors must acknowledge and whose implications they must discuss.
13. Line 241: The authors need to explain why they have chosen two drought indicators instead of one. Are two indicators necessary? If the purpose of the drought analysis is to illustrate the use and value of the new agricultural GDP dataset, then why not use only one? Moreover, given global concerns about climate change, why not illustrate use of the new dataset by using a forward-looking indicator of climate-change risks? The SPEI and WCI indicators are backward-looking, which makes them of dubious value given that climate change is altering drought risks.
14. Table 2: I suspect that the correlations are not significantly different within some of the regions. I recommend adding information on the significance of the differences between the following pairs of correlations within each region: AgGDP/NTL vs. GDP/NTL, and AgGDP/NTL vs. POP/NTL.
15. Figure 7: Given that grid cells become smaller at higher latitudes, shouldn’t the map show $/km2 instead of $?
16. Lines 306-307: The authors state, “One advantage of the cross-entropy is the volume preserving pycnophylactic property, which ensures the sum of the gridded data is the original value ….” Spatial regression presumably violates this property. Does the analysis of predictive accuracy in the Brazil study by Thomas et al. (2019) indicate how much spatial regression violates it? In the current paper, the authors’ comparison of the cross-entropy dataset to the naïve dataset based on rural population would be more compelling if Thomas et al. find that spatial regression violates the property a lot and thus is internally less consistent than the cross-entropy dataset.
17. Line 317: The authors state, “Since we cannot perform an evaluation of prediction accuracy for all countries ….” Why not? I’m not saying they should perform such an evaluation. I am just unclear as to why they cannot perform it. Can they perform it for a subset of countries?
18. Lines 320-324: Doesn’t the finding that the naïve and cross-entropy maps are not significantly different imply that one might as well use the (presumably) simpler and more transparent naïve approach instead of the cross-entropy approach? I.e., what is the advantage of the cross-entropy approach over the naïve approach if the two approaches yield statistically indistinguishable results? Preservation of the pycnophylactic property? If so, can the authors provide information on the degree to which the naïve approach violates that property?
19. Line 334: The authors need to define “MAUP.”
20. Lines 336-337: The authors write, “The data are most appropriate for applications at global, continental and regional scales (You and Wood, 2006).” Aren’t the data also appropriate for applications in countries that contribute data from a relatively large number of subdivisions to the cross-entropy optimization (e.g., Thailand)?
21. Line 381: Starting the “Conclusions” section with discussion of the drought analysis is odd given that the main contribution of the paper is the construction of the new gridded dataset. The current second paragraph in the section would work better as the starting paragraph.
22. The manuscript includes an Appendix B but no Appendix A. Is Appendix A missing, or is Appendix B mislabeled?