The Largest Crop Production Shocks: Magnitude, Causes and Frequency

Jehn, Florian Ulrich; Mulhall, James; Blouin, Simon; Gajewski, Łukasz G.; Wunderling, Nico

doi:10.31223/X55B25

Preprints

https://doi.org/10.31223/X55B25

Preprints

17 Sep 2025

| 17 Sep 2025

The Largest Crop Production Shocks: Magnitude, Causes and Frequency

Florian Ulrich Jehn, James Mulhall, Simon Blouin, Łukasz G. Gajewski, and Nico Wunderling

Abstract. Food is the foundation of our society. We often take it for granted, but stocks are rarely available for longer than a year, and food production can be disrupted by catastrophic events, both locally and globally. To highlight such major risks to the food system, we analyzed FAO crop production data from 1961 to 2023 to find the largest crop production shock for every country and identify its causes. We show that large crop production shocks regularly happen in all countries. This is most often driven by climate (especially droughts), but disruptions by other causes like economic disruptions, environmental hazards (especially storms) and conflict also occur regularly. The global mean of largest country-level shocks averaged -29 %, with African countries experiencing the most extreme collapses (-80 % in Botswana), while Asian and Central European nations faced more moderate largest shocks (-5 to -15 %). While global shocks above 5 % are rare (occurring once in 63 years), continent-level shocks of this magnitude happen every 1.8 years on average. These results show that large disruptions to our food system frequently happen on a local to regional scale and can plausibly happen on a global scale as well. We therefore argue that more preparation and planning are needed to avoid such global disruptions to food production.

Received: 04 Sep 2025 – Discussion started: 17 Sep 2025

Download & links

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (0 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

03 Feb 2026

| Highlight paper

The largest crop production shocks: magnitude, causes and frequency

Florian Ulrich Jehn, James Mulhall, Simon Blouin, Łukasz G. Gajewski, and Nico Wunderling

Earth Syst. Dynam., 17, 151–166, https://doi.org/10.5194/esd-17-151-2026,https://doi.org/10.5194/esd-17-151-2026, 2026

Short summary Chief editor

Florian Ulrich Jehn, James Mulhall, Simon Blouin, Łukasz G. Gajewski, and Nico Wunderling

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-4350', Navin Ramankutty, 25 Oct 2025
General comments
This is an impressive and engaging analysis of global food production shocks and their causes. The study makes good use of FAO data to identify and interpret the largest national food production shocks (in total calorie terms) over 1961–2023. I found the paper clear, rich in narrative detail, and well organized. It provides a valuable dataset and synthesis that will be of interest to both researchers and policymakers. My comments below are intended to help the authors strengthen the framing, clarify assumptions, and ensure that key methodological choices are transparent and well justified.

Specific comments
This paper clearly builds on Cottrell et al. (2019). It would help readers if the introduction and discussion more explicitly distinguished this study’s new contributions—for example through expanded data coverage, new cause-attribution methods, or additional insights.

The assumption that all crops could be diverted to human consumption in a crisis is strong and may not hold in practice. Many of the listed crops (for example, seed cotton, maize, soy, and barley) are primarily non-food or feed crops. I encourage the authors to discuss this limitation in more detail or, if feasible, re-analyze using food-only fractions or “delivered calories” that account for feed conversion losses (see Cassidy et al., 2013). Even a short sensitivity check would substantially strengthen the robustness of the findings. Otherwise your findings would be weighted heavily toward feed crops and those countries that produce them (e.g., consider that only about a quarter of produced calories from maize ends up as human calories).

The Savitzky–Golay filter is well motivated, but it would be useful to show that results are not overly sensitive to this choice. I recommend testing one or two alternative detrending methods (e.g., Gaussian or LOESS) and examining whether the identified “largest shocks” or their magnitudes change materially. Similarly, reconsider the “must be below last year” rule or provide a robustness check without it.

In your synchrony analysis, how do you account for the differences in size and production of different countries? Larger producers will naturally drive global totals, so weighting by production share or caloric contribution could yield a more accurate view of which regions most influence global variability. Clarifying or adding this weighting step would help the “buffering” interpretation.

Some of the observed geographic patterns may reflect differences in crop composition rather than exposure or governance. The authors could discuss this possibility, or test whether patterns persist when comparing regions growing similar crop mixes.

A few recent studies might offer useful methodological or interpretive ideas (and very sorry for the self citations—these do not need to be cited if not directly relevant):
On synchrony (sections 2.4, 3.4), see Mehrabi and Ramankutty (2019) and Egli et al. (2021). Drawing on their framework for quantifying and decomposing synchrony could strengthen your analysis.

On the role of trade (section 4.3), see Bajaj et al. (2025).

On diversification of trade (section 4.4), see Hertel et al. (2021).

Technical corrections
Consider renaming section 3.2 to something like “Geographic patterns of shock types”, because the geographic patterns of shocks has already been seen in figure 2 of section 3.1.

Lines 221-222: The conclusion that “pests and diseases are not a major factor for the largest shocks” may be too strong given only one observed data point.”

Lines 237-238: The reference to “North America” appears to describe tropical storm impacts in the Caribbean. “Central America” might be more appropriate, and the map (with only one blue country) suggests that this pattern could be toned down in the text.

Lines 250-251: You discuss a decade of mismanagement under Idi Amin – it’s not clear how a decade-long effect would result in a single year shock?

Lines 274-275 & Figure 6: Because the 2020s decade includes only four years of data, the low count of shocks is expected. Normalizing by the number of years per decade (i.e., showing average shocks per year) would provide a fairer comparison.

Lines 338-341: Please add supporting citations for these statements.

Lines 342-347: I had a hard time understanding the argument in this paragraph. How does the Zhang et al. study (which only examined climate driven shocks, so did not compare it to other shocks) support the finding that climate is the main driver of shocks?

References
Bajaj, K., Z. Mehrabi, T. Kastner, J. Jägermeyr, C. Müller, F. Schwarzmüller, T. W. Hertel, and N. Ramankutty, Current food trade helps mitigate future climate change impacts in lower-income nations, PLOS ONE, 20(1), e0314722, 10.1371/journal.pone.0314722, 2025.

Cassidy, E. S., P. C. West, J. S. Gerber, and J. A. Foley, Redefining agricultural yields: from tonnes to people nourished per hectare, Environ. Res. Lett., 8(3), 034015, 2013.

Egli, L., Z. Mehrabi, and R. Seppelt, More farms, less specialized landscapes, and higher crop diversity stabilize food supplies, Environ. Res. Lett., 16(5), 055015, 10.1088/1748-9326/abf529, 2021.

Hertel, T., I. Elouafi, M. Tanticharoen, and F. Ewert, Diversification for enhanced food systems resilience, Nature Food, 2(11), 832-834, 10.1038/s43016-021-00403-9, 2021.

Mehrabi, Z., and N. Ramankutty, Synchronized failure of global crop production, Nature Ecology & Evolution, 3(5), 780-786, 10.1038/s41559-019-0862-x, 2019.
Citation: https://doi.org/10.5194/egusphere-2025-4350-RC1
- AC1: 'Reply on RC1', Florian Ulrich Jehn, 13 Nov 2025
  
  Dear Navin Ramankutty,
  please find our responses in the attached pdf.
  Kind regards,
  Florian Jehn, on behalf of all co-authors
  
  Citation: https://doi.org/10.5194/egusphere-2025-4350-AC1
RC2:
'Comment on egusphere-2025-4350', Anonymous Referee #2, 29 Oct 2025

Nicely done study on crop production shocks, however I'm not sure if it brings so much new information to the table.
The basic methodology of the study is analogous to Cottrell 2019 and Anderson 2023 which the authors already state. The added sophistication and differentiation in my view comes from the use of a LLM to identify possible drivers of crop production shocks, as well as a different filter in the methodology (does the employment of a Gaussian filter change results much? Would be a relatively easy sensitivity test I imagine). The authors use FAOSTAT which provides more countries than Anderson, although less temporal extent; This is more useful for looking at strong shocks in individual countries, while globally synchronous shocks can already be mainly covered by a small number of countries. They also neglect the marine aspect which is already included in Cottrell, which in this paper's case, with its focus on individual countries and large shocks, may be releavant as these are often island countries with low production so indeed marine sources of food could be interesting.
However, as the authors also note, the results are quite similar to what is already in the literature, that climatic factors, also ENSO are strong drivers of production shocks, along with geopolitical factors.
An added element here that could make the paper more interesting, also harnessing its integration of the LLM into the methodology, is to qualitatively trace the biophysical impacts back to human impacts - i.e. in years with production shocks were there reports of price inflation, shifts in global trade patterns, hunger indices, etc. This may be more possible now with the LLM doing the first screening.

Finally, they note that some country-level data appears erroneous or unreliable. Can these be given an initial screen, or some way to account for reliability, especially the earlier FAOSTAT data is often quite dodgy.
Thank you very much.

Citation: https://doi.org/10.5194/egusphere-2025-4350-RC2
- AC2: 'Reply on RC2', Florian Ulrich Jehn, 13 Nov 2025
  
  Dear Reviewer,
  please find our responses in the attached pdf.
  Kind regards,
  Florian Jehn, on behalf of all co-authors
  
  Citation: https://doi.org/10.5194/egusphere-2025-4350-AC2

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-4350', Navin Ramankutty, 25 Oct 2025
General comments
This is an impressive and engaging analysis of global food production shocks and their causes. The study makes good use of FAO data to identify and interpret the largest national food production shocks (in total calorie terms) over 1961–2023. I found the paper clear, rich in narrative detail, and well organized. It provides a valuable dataset and synthesis that will be of interest to both researchers and policymakers. My comments below are intended to help the authors strengthen the framing, clarify assumptions, and ensure that key methodological choices are transparent and well justified.

Specific comments
This paper clearly builds on Cottrell et al. (2019). It would help readers if the introduction and discussion more explicitly distinguished this study’s new contributions—for example through expanded data coverage, new cause-attribution methods, or additional insights.

The assumption that all crops could be diverted to human consumption in a crisis is strong and may not hold in practice. Many of the listed crops (for example, seed cotton, maize, soy, and barley) are primarily non-food or feed crops. I encourage the authors to discuss this limitation in more detail or, if feasible, re-analyze using food-only fractions or “delivered calories” that account for feed conversion losses (see Cassidy et al., 2013). Even a short sensitivity check would substantially strengthen the robustness of the findings. Otherwise your findings would be weighted heavily toward feed crops and those countries that produce them (e.g., consider that only about a quarter of produced calories from maize ends up as human calories).

The Savitzky–Golay filter is well motivated, but it would be useful to show that results are not overly sensitive to this choice. I recommend testing one or two alternative detrending methods (e.g., Gaussian or LOESS) and examining whether the identified “largest shocks” or their magnitudes change materially. Similarly, reconsider the “must be below last year” rule or provide a robustness check without it.

In your synchrony analysis, how do you account for the differences in size and production of different countries? Larger producers will naturally drive global totals, so weighting by production share or caloric contribution could yield a more accurate view of which regions most influence global variability. Clarifying or adding this weighting step would help the “buffering” interpretation.

Some of the observed geographic patterns may reflect differences in crop composition rather than exposure or governance. The authors could discuss this possibility, or test whether patterns persist when comparing regions growing similar crop mixes.

A few recent studies might offer useful methodological or interpretive ideas (and very sorry for the self citations—these do not need to be cited if not directly relevant):
On synchrony (sections 2.4, 3.4), see Mehrabi and Ramankutty (2019) and Egli et al. (2021). Drawing on their framework for quantifying and decomposing synchrony could strengthen your analysis.

On the role of trade (section 4.3), see Bajaj et al. (2025).

On diversification of trade (section 4.4), see Hertel et al. (2021).

Technical corrections
Consider renaming section 3.2 to something like “Geographic patterns of shock types”, because the geographic patterns of shocks has already been seen in figure 2 of section 3.1.

Lines 221-222: The conclusion that “pests and diseases are not a major factor for the largest shocks” may be too strong given only one observed data point.”

Lines 237-238: The reference to “North America” appears to describe tropical storm impacts in the Caribbean. “Central America” might be more appropriate, and the map (with only one blue country) suggests that this pattern could be toned down in the text.

Lines 250-251: You discuss a decade of mismanagement under Idi Amin – it’s not clear how a decade-long effect would result in a single year shock?

Lines 274-275 & Figure 6: Because the 2020s decade includes only four years of data, the low count of shocks is expected. Normalizing by the number of years per decade (i.e., showing average shocks per year) would provide a fairer comparison.

Lines 338-341: Please add supporting citations for these statements.

Lines 342-347: I had a hard time understanding the argument in this paragraph. How does the Zhang et al. study (which only examined climate driven shocks, so did not compare it to other shocks) support the finding that climate is the main driver of shocks?

References
Bajaj, K., Z. Mehrabi, T. Kastner, J. Jägermeyr, C. Müller, F. Schwarzmüller, T. W. Hertel, and N. Ramankutty, Current food trade helps mitigate future climate change impacts in lower-income nations, PLOS ONE, 20(1), e0314722, 10.1371/journal.pone.0314722, 2025.

Cassidy, E. S., P. C. West, J. S. Gerber, and J. A. Foley, Redefining agricultural yields: from tonnes to people nourished per hectare, Environ. Res. Lett., 8(3), 034015, 2013.

Egli, L., Z. Mehrabi, and R. Seppelt, More farms, less specialized landscapes, and higher crop diversity stabilize food supplies, Environ. Res. Lett., 16(5), 055015, 10.1088/1748-9326/abf529, 2021.

Hertel, T., I. Elouafi, M. Tanticharoen, and F. Ewert, Diversification for enhanced food systems resilience, Nature Food, 2(11), 832-834, 10.1038/s43016-021-00403-9, 2021.

Mehrabi, Z., and N. Ramankutty, Synchronized failure of global crop production, Nature Ecology & Evolution, 3(5), 780-786, 10.1038/s41559-019-0862-x, 2019.
Citation: https://doi.org/10.5194/egusphere-2025-4350-RC1
- AC1: 'Reply on RC1', Florian Ulrich Jehn, 13 Nov 2025
  
  Dear Navin Ramankutty,
  please find our responses in the attached pdf.
  Kind regards,
  Florian Jehn, on behalf of all co-authors
  
  Citation: https://doi.org/10.5194/egusphere-2025-4350-AC1
RC2:
'Comment on egusphere-2025-4350', Anonymous Referee #2, 29 Oct 2025

Nicely done study on crop production shocks, however I'm not sure if it brings so much new information to the table.
The basic methodology of the study is analogous to Cottrell 2019 and Anderson 2023 which the authors already state. The added sophistication and differentiation in my view comes from the use of a LLM to identify possible drivers of crop production shocks, as well as a different filter in the methodology (does the employment of a Gaussian filter change results much? Would be a relatively easy sensitivity test I imagine). The authors use FAOSTAT which provides more countries than Anderson, although less temporal extent; This is more useful for looking at strong shocks in individual countries, while globally synchronous shocks can already be mainly covered by a small number of countries. They also neglect the marine aspect which is already included in Cottrell, which in this paper's case, with its focus on individual countries and large shocks, may be releavant as these are often island countries with low production so indeed marine sources of food could be interesting.
However, as the authors also note, the results are quite similar to what is already in the literature, that climatic factors, also ENSO are strong drivers of production shocks, along with geopolitical factors.
An added element here that could make the paper more interesting, also harnessing its integration of the LLM into the methodology, is to qualitatively trace the biophysical impacts back to human impacts - i.e. in years with production shocks were there reports of price inflation, shifts in global trade patterns, hunger indices, etc. This may be more possible now with the LLM doing the first screening.

Finally, they note that some country-level data appears erroneous or unreliable. Can these be given an initial screen, or some way to account for reliability, especially the earlier FAOSTAT data is often quite dodgy.
Thank you very much.

Citation: https://doi.org/10.5194/egusphere-2025-4350-RC2
- AC2: 'Reply on RC2', Florian Ulrich Jehn, 13 Nov 2025
  
  Dear Reviewer,
  please find our responses in the attached pdf.
  Kind regards,
  Florian Jehn, on behalf of all co-authors
  
  Citation: https://doi.org/10.5194/egusphere-2025-4350-AC2

Peer review completion

AR – Author's response | RR – Referee report | ED – Editor decision | EF – Editorial file upload

ED: Reconsider after major revisions (02 Dec 2025) by Somnath Baidya Roy

AR by Florian Ulrich Jehn on behalf of the Authors (03 Dec 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (04 Dec 2025) by Somnath Baidya Roy

RR by Navin Ramankutty (10 Dec 2025)

Suggestions for revision or reasons for rejection

Thanks to the authors for their diligent revisions. I am satisfied with most of the revisions and/or responses to my comments from the first round. I have just a few more comments, mostly a conversational response, with some recommendations for minor revisions. Will be good to see your paper published.

Comments
1. I was originally still not entirely convinced by the argument that humans could eat crops intended for livestock feed, but granted I am not the expert on this. When I first made that comment, I was thinking of the #2 yellow corn predominantly grown in the US for livestock feed, which is a different variety from the sweet corn that people normally eat. But having done some more reading, it does seem like one could eat #2 corn in situations such as famine (e.g., https://tinyurl.com/47e85wk9; feel free to add this or other citations to make your argument stronger). I also tried to review the citations you added to support your argument that livestock feed has historically been diverted to human consumption during times of crises. Two of your citations (Collingham, Offer) are books I could not easily access. But I was able to quickly review the Meng et al. (2015) paper, which does say that peasants in rural China ate green crops (illegally) from the fields. So I accept this point. And besides, you have added a sensitivity analysis by removing cotton, soy and rapeseed.
2. I would suggest adding a short sentence (either to lines 213-215 or right after) saying that the spearman correlation also accounts for different production magnitudes of countries.
3. Regarding pests & diseases, I see your point. But there is a potentially important message here that you could touch upon in your discussion if you wish to. It is widely believed that pests and diseases are a major source of crop losses (e.g., see https://doi.org/10.1038/s41559-018-0793-y). This is in fact why farmers apply a lot of pesticides and why a lot of the GM crops are focussed on traits to reduce pest burden. But while the burden of pests and disease for crop production may be high, maybe it just remains at a high baseline without too much year-to-year fluctuation, and therefore does not create the types of episodic shocks that you are examining in this paper. In other words, some drivers are of high magnitude but without large fluctuations, and therefore we don’t see their effects in terms of shocks.

Navin Ramankutty

Hide

RR by Anonymous Referee #2 (22 Dec 2025)

Suggestions for revision or reasons for rejection

Thank you for the responses and the changes. I still find it hard difficult to distinguish novel fatures of this article as it stands.
Of the three objectives listed in the introduction (L83) my impression is that these already exist in the literature, so this paper needs to argumentatively and/or technically assess a bit more.

I) I understand that the paper aims to focus on "the worst production shock" L78 and therefore aggregates, however, certain crops are particularly important for certain regions, and temporal dimensions also crucial i.e. alternating cropping seasons across hemispheres, as well as particular trade routes of particular crops. Furthermore, despite the justifcation on 'worst' shocks, the paper then goes on to describe production/shock distributions and coutry profiles, making the analysis extremely similar to Cottrell.

Furthermore, compared to Cottrell 2019 for instance, this paper is not only more aggregated, but actually covers less data, as Cottrell disaggregate between livestock, crops but also marine sources of food, which especially for small island states which this paper highlights crop shocks in, can be quite improtant. The famine framing allows this paper to assume all diversion of feed to direct consumption, but caloric content and calories available for human consumption can end up quite different (although I would disagree with soybean being one of the more difficult ones), and livestock may still play some roles in such cases, in any case, it is a strong simplification that already existing studies actually did not make. Especially as global disaster events could imaginably onset suddenly, adaptations of processing sectors may not happen very fast to make feed crops consumable to the general population.

II) Regarding the synchrony analysis, how do the correlation coefficients change over time i.e. could analyse with a moving window, and see how the (a)synchrony that cannot be qualitatively ascribed to climate factors match with conflict or policy or other aspects? Or does synchrony change with different crops, northern hemisphere maize could be compensated by south american production for instance, but perhaps wheat is harder to do so?

III) Regarding the qualitative aspect, is there any indication that an AI-assisted search provides more systematic answers than a human-based one, and in which ways? Access to untranslated documents perhaps? but this needs to be validated, and/or mentioned, beyond the fact that it certainly saves time for the authours.
Finally, I don't necessarily think closing the loop from driver to shock to (qualitative)human impact is beyond the scope of this analysis, especially given the framing of this paper on 'worst', human disaster, catastrophe etc, and how it aims to 'ground' such framings in historical data. I think it's extremely pertinent to then see if, given the 'worst' shock a country has faced in quantified history, how human impacts appeared, i.e. via price shocks, distributional impacts (on producers vs urban poor for instance), shifts in consumption, etc? And if these did not appear, what mechanisms allowed countries to avoid such impacts, i.e. was storage or trade particularly important under certain policy constraints. This would also allow for providing specific understandings of possible recommendations beyond the generalized ones given in section 4.4, would strenghten the framing and novelty of the paper much more beyond repeating an existing analysis.

Hide

ED: Publish subject to minor revisions (review by editor) (02 Jan 2026) by Somnath Baidya Roy

AR by Florian Ulrich Jehn on behalf of the Authors (09 Jan 2026) Author's response Author's tracked changes Manuscript

ED: Publish as is (14 Jan 2026) by Somnath Baidya Roy

AR by Florian Ulrich Jehn on behalf of the Authors (17 Jan 2026) Manuscript

Journal article(s) based on this preprint

03 Feb 2026

| Highlight paper

The largest crop production shocks: magnitude, causes and frequency

Florian Ulrich Jehn, James Mulhall, Simon Blouin, Łukasz G. Gajewski, and Nico Wunderling

Earth Syst. Dynam., 17, 151–166, https://doi.org/10.5194/esd-17-151-2026,https://doi.org/10.5194/esd-17-151-2026, 2026

Short summary Chief editor

Florian Ulrich Jehn, James Mulhall, Simon Blouin, Łukasz G. Gajewski, and Nico Wunderling

Data sets

Code and Data Repository Florian Ulrich Jehn and James Mulhall https://github.com/allfed/Historical-Food-Shocks

Model code and software

Code and Data Repository Florian Ulrich Jehn and James Mulhall https://github.com/allfed/Historical-Food-Shocks

Florian Ulrich Jehn, James Mulhall, Simon Blouin, Łukasz G. Gajewski, and Nico Wunderling

Viewed

Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.

Total article views: 825 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
817	0	8	825	0	0

HTML: 817
PDF: 0
XML: 8
Total: 825
BibTeX: 0
EndNote: 0

Views and downloads (calculated since 17 Sep 2025)

Month	HTML	PDF	XML
Sep 2025	518	0	518
Oct 2025	103	2	105
Nov 2025	57	6	63
Dec 2025	43	0	43
Jan 2026	96	0	96
Feb 2026	0

Cumulative views and downloads (calculated since 17 Sep 2025)

Month	HTML	PDF	XML
Sep 2025	518	0	518
Oct 2025	103	2	105
Nov 2025	57	6	63
Dec 2025	43	0	43
Jan 2026	96	0	96
Feb 2026	0

Viewed (geographical distribution)

Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.

Total article views: 819 (including HTML, PDF, and XML) Thereof 819 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 04 Feb 2026

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint

Final revised paper

Short summary

Large crop failures happen regularly around the world, threatening food security. We analyzed sixty years of global crop production data and found that every country has experienced major crop losses. Climate events like droughts cause most severe disruptions, with some African nations losing up to eighty percent of production. While global crop shocks above five percent are rare, regional disruptions occur frequently. These findings show our food system faces regular large-scale threats.


Total:	0
HTML:	0
PDF:	0
XML:	0