Summary

The submitted manuscript describes a study of the historic link between extreme weather conditions and wheat yields in the United Kingdom (UK), extended with an analysis of relevant future weather changes. It is shown that mean yields have systematically increased over time, which can be explained by technological advances, but interestingly year-to-year variability of yield has also increased in recent decades (Fig 1c). The authors then set out an analysis in which correlations between various (extreme) weather indices for temperature and precipitation, across three important wheat growing stages, and end-of-season yield are investigated. They develop a simple scoring metric that describes some of the cumulative effect of weather conditions on yield. This analysis is extended by very high resolution climate model-based projections of future temperature and precipitation conditions under a high-emission scenario, and describes future weather conditions for wheat growing.

General remarks

Whilst I appreciate that the topic is of large societal relevance, I struggle to see where the manuscript answers its research questions or if promises delivered. The link between weather and agricultural impacts is highly complex and non-linear. The authors acknowledge this in their introduction, and set out on logical path of addressing the topic and ultimately provide specific wheat sector-relevant climate projections. I fear however that the relationships are of such highly complex nature, that the present analysis does not provide satisfactory answers to the questions posed.

The first research aim (finding statistical associations) results in Table 2. Only one of three regions shows any stat.significant relations (at p=0.05 level) between the weather indices and yield. The accompanying text leans very heavily on anecdotical evidence, which I fear may lean towards overinterpretation of single events. To account for temporally compounding effects a simple scoring metric is developed (Fig 6), failing to take into account developments in agricultural science. Why haven’t the authors followed their own advise (last sentence) and employed process-based crop models or AI methods to find robust relationships between weather and yield?

Then, the future climate projections are based on a single climate model. Though high-resolution modelling without doubt adds value, a multi-model perspective is needed to provide ‘reliable’ projections. I advise the authors to add a comparison of projections in CMIP6 or CORDEX to UKCP, such that readers may get a feeling of where these projections lie within the larger model-related uncertainty. 

But more importantly, rather than analysing how crops respond to future weather conditions (which is what the title of this manuscript implies), changing weather conditions in three growing stages are discussed. Given the limited relationships between weather and yield that were found, does this really provide useful information for the UK agricultural sector?

Major points

Section 2.4 - In table 3 I see you have bias-corrected each ensemble member separately. Though I haven’t worked with perturbed physics large ensembles, the normal procedure for bias correcting would be to do a single bias correction for all ensemble members, as differences are due to internal variability not indicative of different mean climates. Are the imposed physics changes so large that this is different in this case?

Section 3.2 - The relationships noted in this section are very anecdotical. Table 2 provides the quantitative correlation coefficients, which are only in a few cases statistically significant. Please add more quantitative evidence of the suggested relations, or note in the text that despite anecdotical evidence there is no statistical link. I think it is easy to overfit/overanalyse seemingly simple relations (e.g. wet conditions lead to low yield), when in reality the interactions between plant and weather is very complex and highly non-linear. 

Line 295, line 318 - I hadn’t noted evidence in the results section supporting the conclusion that increased inter annual yield variability is linked to “one-to-one correlations with temperature or precipitation extremes” or “the recent increase in yield volatility is associated with combined climate metrics”. Please either emphasise this more in the results section, or remove from the conclusions.

I miss a discussion of the assumptions that went into this work, and how these assumptions might influence the results. One item would be the use of fixed-in-time growth stages, in reality these are weather dependent, and plant vulnerabilities to extreme temperatures or precipitation can thus be different from one July to another July (for example).

Minor points

Fig 1 caption - I very much dislike the bracket-way of scientific writing. There is no word limit in ESD, I strongly encourage the authors to rewrite their statement in two sentences. “Green (brown) labels indicate examples of years with anomalously high 485 (low) yields.”

Line 50 - Indeed there is a lot of climate research into weather extremes, but there is a vast quantity of climate impact research as well in how these will influence for example crops. A slight rearrangement of this sentence is asked for. And maybe a few more recent examples from the literature (e.g. Ben-Ari et al 2018). 

Line 85 - Please clarify “any incomplete crop growth stages”, does this relate to gaps in the PR/TAS data, or too fast progression through the stage? If the latter, would that not be caused by climate extremes?

Line 104 - A model realistically simulating present-day climate is a requisite for making a seemingly reliable projection, but it is not guaranteed of course. The response to GHG forcing can still be very wrong. And since only RCP8.5 is used as forcing, further doubt on  the ‘credibility’ of the projections is added (see e.g. Hausfather and Peters et al 2020). This does not discredit your analysis, but does put some limitations on the ‘credibility of projections’ made. Please add a sentence noting these issues (single model, single GHG scenario), maybe use the first paragraph of section 3.4 and remove it there.

Fig 2 - I originally thought the panels a-c showed the UK as a whole, only noting later that maybe the small region labels on the right count for all panels. Maybe add these inside the plot, or explicitly state this in the caption.

Line 154 - I don’t understand why you would think those are related, on the one hand growth due to technology and on the other hand increased variance? I’d say for the first you have a very good argument, and the second is an interesting question indeed, with the link to increasing weather extremes as a good hypothesis. 

Line 169, Fig 4/5 - These figures show a lot of data in a small panel. As you don’t discuss any correlation between PR and TAS, why would you show them in this way? Wouldn’t a ‘simple’ scatter plot between PR and yield better show this conclusion? Furthermore, yes- low yield years show an extreme PR in one of the seasons, but how many normal/high yield years do the same? E.g. from Table2 only EMYH in the production phase shows a statistically significant correlation between PR and yield. Then, what do the grey crosses add to the figure?

Table 2 - the horizontal line separating TAS and PR measures is one row too low.

Line 205 - Add here that the growing phases in real plants are determined by their growth, rather than calendar days. So a phase can last longer, to have the desired number of growing degree days for example, delaying the crop, but resulting in the expected yield. The calendar-fixed phases are a simplification of this proces.

Fig 6 - A few remarks: (i) Please separate the projections from the observed data, maybe in a second row of figures below the first one. (ii) I don’t understand where the future yield data are coming from? The relationship of black circles and triangles and grey shading is surprisingly (doubtfully?) linear, and fully captures the eye of the reader. The dots show very much variability, by eye alone I doubt one would have been able to draw the correct regression line through them. (iii) please add fitted regression lines using observed data (I assume the statistics plotted are those lines), and maybe for the national subplot also show data from before 1990.

Section 3.4 - The section title is misleading, general forced climatic changes are discussed, not crop-specific climatic changes.

Line 245 - I imagine the UKCP lie in the upper/lower-percentiles of the full CMIP5/6 ensemble, but not fully outside? “UKCP simulations tend to sample greater future warming and drying in summer compared to the full…”

Fig 7 - I’m not sure this is the best way of showing the data. 10 lines on top of each other, plus the ensemble mean, and then for each the regression line. I had to zoom in to 500% to read the data. Maybe consider only showing with shading the min-P25-P50-P75-max across the ensemble, and the ensemble mean regression line? This would show less data, but I think more information.

This paper seeks to understand historical links between climate and variability in UK wheat yields, and examine the implications of future climate change as projected using a convection-resolving climate model. It considers differential yield responses across growth stages, and tries to then aggregate these stages to assess compensating or amplifying impacts. This latter aspect is the main novelty of this research, which I think is useful and timely. I also appreciate the effort to consider compensating effects between growth stages and between heat and water via this aggregate climate scoring approach, and in Figures 4-5. There is increasing attention to these joint affects, so this paper has the potential to add some clarity here as well.

I have two main critiques that should be addressed. First, the statistical analysis is not adequately described, and based on what I can surmise from the sparse detail, it is probably not the strongest approach. Second, the assessment of future impacts is only driven by data on the climate side, and the crop impacts are only qualitatively discussed. This sells the historical climate-yield relationships short: why not use your historical results for a data-driven estimate of future impacts? Further, your results and other research show how multivariate climate variation/change could lead to compensating or compounding impacts on crops, the potential for which could be more robustly and objectively assessed through a more quantitative approach.

On statistical analysis: The methods is missing any description of the statistical analysis, justification for model specification, etc. This makes it fairly hard to assess the reliability of the results, and what they mean. I gather that the analysis is pairwise two-variable Pearson correlations (yield vs. each climate variable). The authors then use these results to develop a scoring system to combine variables/growth stages, which is not necessarily a bad approach, and the results in Figure 6 seem pretty strong. But this is not a widely used approach, and given lack of detailed methods, it is hard to assess. Rather, multivariate regression (i.e., temperature and precipitation variables for each growth stage all included in one yield model) is what is typical. There are both benefits and pitfalls to it, but it would improve confidence to try this more widely-vetted method and see if results are consistent, and would enable a more self-consistent way to assess compensations. Further, this multivariate regression approach is more suited to then actually projecting yield based on multivariate projections from climate models. You may also consider non-linear yield responses. Finally, only p-values are mentioned in the text, which only provide limited information. I see Pearson coefficients in a table, but their relative magnitudes are not discussed. And the effect size (i.e. slope coefficient) underlying these correlations also provide useful information (steepness of yield response to climate variable), so may be helpful to discuss.

Another methodological issue is reliance on interpreting specific years relative to statistical results, which often lead the paragraphs in the results. I actually really like this for its concreteness, but it is not a super robust method and seems prone to cherry-picking years that fit the narrative. I think this can be remedied by trying to frame these claims more as discussion points and reducing their prominence in the results. Alternatively, you could formalize your method for selecting key years, and describe it in the text.

Another important limitation of this research is its use of only one climate model under only one climate forcing scenario. This leaves important uncertainties in emissions trajectories and climate responses unquantified. The RCP8.5 scenario also is falling out of favor in some circles, as it assumes implausibly high emissions – the authors acknowledge this late in the paper, but don’t strongly justify why we should nevertheless be focusing on an unlikely future. It would probably be useful to include RCP2.6 or 4.5, or at very least acknowledge that the paper doesn’t address emissions uncertainty. The implications of using one climate model should also be justified – is the HadGEM3/HadREM3 nested model particularly useful for the region? The use of a 12 member ensemble helps, but I notice that some years (often with important yield impacts) in Figures 4-5 fall outside the whiskers of the historical model data, raising questions of whether this model can reproduce these conditions (historically or in the future). We know models have such deficiencies –using more than one can help at least partly constrain uncertainty.

Small comment: impacts of rising CO2 on crop water use will be important in the future, as you mention in the intro. It’s a huge uncertainty and hard to model, but should probably discuss its relevance for your projections.

Finally, I think you could consider in a bit more depth the interactions between temperature and precipitation both in the climate and for crops. For instance, very hot conditions in the UK can often only be reached with a dry land surface (visible as apparent negative temp-precip correlations during production, Fig’s 4-5). Miralles et al. 2019 is useful reference on these processes. Cool and wet conditions could also be linked physically, with implications for crop impacts. This raises questions about the independence of heat and moisture impacts, which is a problem here since they are only assessed one-at-a-time using Pearson’s correlations (multivariate regression could help capture the interaction).  Further, joint impacts of changes in temp and precip in the future could be discussed more, see line comments.

Thanks for the nice paper! I think it will be a useful publication once some issues are addressed.

Line comments:

Line 36: Could cite more recent papers on this: Ray et al. 2019, Ortiz-Bobea et al. 2021

Line 55: Ainsworth and Long 2021 would be a useful reference here

Line 56: Soil moisture, precipitation intensity/distribution ref?

Line 80: Could be helpful to motivate this step. Presumably, you do this to remove long-term yield trends (due to technology, climate, co2) and isolate annual anomalies relative to this.

Line 100: This threshold for heavy rainfall should be justified and/or its influence should be tested. For instance, Lesk et al. 2020 found extreme rainfall impacts only at high intensities >50mm/hr for US maize and soy (how this maps to daily scale is unclear, but a 10mm/hr threshold would preclude these damaging intensities). Others have used more holistic distributional measures like the daily rainfall GINI coefficient (Shortridge 2019). I’m not aware of equivalent studies for wheat, but these could be good references to add to Zampieri et al. 2017 in line 56 to bring in studies in sub-seasonal rainfall distribution.

Line 137: I think “1989-1960+1” was not intended to be included in text

Lines 159-161: I think the connection between temperature and precipitation is an issue worth discussing. The wet years with poor yields also tend to be relatively cool (especially during foundation). The dry years tend to be hot.

Line 191: I don’t see 1976 on the figure, and 2013 and 2018 don’t seem particularly extreme.

Line 200: This somewhat undercuts your preceding results. You do find climate-yield relationships so I don’t see strong basis for claiming they are masked by inputs. Further, it is not clear which inputs these would be. I do not know of any short-term adaptive solutions to excess moisture (farmers can improve drainage and soil texture over time, but not within a single season). Further, the usual adaptive management for heat or drought is irrigation, which is not widespread in the UK. Instead, what might be more important/interesting is analyzing (or at least speculating on) the role of inputs in raising mean yields (over decades), and how that may influence yield variability (which you are trying to attribute differentially to climate).

Line 205: This claim is interesting and usefully motivates the next section, but needs work, and here’s one place using multivariate regression may be useful. In this more standard method, multiple climate variables together usually explains less than half of yield variation (full-model adjusted r² < 0.5). Using individual pairwise correlations is less common, and so it’s unclear what would be high or low correlation. If the correlations are indeed low in a more robust assessment, it could be because of the myriad other environmental or social factors contributing to yield (climate explains less than half of yield variability).

Line 257-259: Here’s a place you could mention multivariate change. Cool and wet foundation phases have been linked to poor yields, and these are connected because it is hard to warm up the surface when soils are wet, and hard to dry out wet soils when it is cool. The projected warmer and wetter conditions are orthogonal to this connection, and some of that warming may help dry out waterlogged soils. Question is whether the warming will suffice to offset the increased precipitation, and this is the kind of question that a multivariate regression model could help answer. See for instance Rigden et al. 2020, Lesk et al. 2021, Ortiz-Bobea et al. 2019.

Line 265: Precipitation may not change much, but there is still warming, which will increase atmospheric vapour demand (all else being equal). So here’s a place where some acknowledgement or analysis of multivariate change would probably lead to more robust conclusions about the future. Zampieri et al. 2017 touches on some of this multivariate influence. See also Lobell et al. 2013 with detail on the evaporative role of temperature (it’s for U.S. maize, but relevant to interpreting future warming).

Line 277-279: Yes, especially since temperature could have non-linear impacts, see Barlow et al. 2015, a useful reference for frost effects too.

Line 285-288: Great, this offsetting is coming to light as an important mechanism/uncertainty, I just think it could be discussed in more depth.

Line 300-301: Consider using term ‘compound extremes’ here and in the intro to link to emerging literature on this topic. E.g. Zscheischler et al. 2020

Figures 4-5: I like that this shows the bivariate temperature-precipitation distributions. It is hard to differentiate the grey circles from diamonds, however. It may be easier to see if the 95% confidence ellipses are removed – I’m not sure what they add and could be replaced by simple dots showing point-estimates of mean yield. Otherwise, perhaps the climate model data should be presented on separate axes.

Figure 6: this is a pretty convincing figure notwithstanding my concerns above, but it’s hard to understand why the black data are showing y-axis values and an increasing trend, as I don’t see yield projection results or methods anywhere in the paper. I assume the points are different years, and aggregate climate scores evolve over time. If so, this data should probably be separate time axes. The black data also seem visually like trendlines on the yield/climate score scatters, but I don’t think they are so this may mislead readers.

References:

Ainsworth, E. A., & Long, S. P. (2021). 30 years of free‐air carbon dioxide enrichment (FACE): What have we learned about future crop productivity and its potential for adaptation?. Global Change Biology, 27(1), 27-49.

Barlow, K. M., Christy, B. P., O’leary, G. J., Riffkin, P. A., & Nuttall, J. G. (2015). Simulating the impact of extreme heat and frost events on wheat crop production: A review. Field Crops Research, 171, 109-119.

Lesk, C., Coffel, E., & Horton, R. (2020). Net benefits to US soy and maize yields from intensifying hourly rainfall. Nature Climate Change, 10(9), 819-822.

Lesk, C., Coffel, E., Winter, J., Ray, D., Zscheischler, J., Seneviratne, S. I., & Horton, R. (2021). Stronger temperature–moisture couplings exacerbate the impact of climate warming on global crop yields. Nature food, 2(9), 683-691.

Lobell, D. B., Hammer, G. L., McLean, G., Messina, C., Roberts, M. J., & Schlenker, W. (2013). The critical role of extreme heat for maize production in the United States. Nature climate change, 3(5), 497-501.

Miralles, D. G., Gentine, P., Seneviratne, S. I., & Teuling, A. J. (2019). Land–atmospheric feedbacks during droughts and heatwaves: state of the science and current challenges. Annals of the New York Academy of Sciences, 1436(1), 19.

Ortiz-Bobea, A., Wang, H., Carrillo, C. M., & Ault, T. R. (2019). Unpacking the climatic drivers of US agricultural yields. Environmental Research Letters, 14(6), 064003.

Ortiz-Bobea, A., Ault, T. R., Carrillo, C. M., Chambers, R. G., & Lobell, D. B. (2021). Anthropogenic climate change has slowed global agricultural productivity growth. Nature Climate Change, 11(4), 306-312.

Ray, D. K., West, P. C., Clark, M., Gerber, J. S., Prishchepov, A. V., & Chatterjee, S. (2019). Climate change has likely already affected global food production. PloS one, 14(5), e0217148.

Rigden, A. J., Mueller, N. D., Holbrook, N. M., Pillai, N., & Huybers, P. (2020). Combined influence of soil moisture and atmospheric evaporative demand is important for accurately predicting US maize yields. Nature Food, 1(2), 127-133.

Zscheischler, J., Martius, O., Westra, S., Bevacqua, E., Raymond, C., Horton, R. M., ... & Vignotto, E. (2020). A typology of compound weather and climate events. Nature reviews earth & environment, 1(7), 333-347.

The authors present an interesting analysis of UK’s wheat yield variability. They first explore the influence of different climatic conditions on wheat yields to then construct a scoring system for combined climate effects. In their analysis, they separate major plant development stages. Finally, they use climate model projections to estimate potential yields in a warmer climate.

Despite the confined regional focus, I would expect that the findings and the presented approach would be of interest to a wide readership. The manuscript is well written.

Major concerns:

The paper is based on statistical analysis and this analysis should be described in more detail including a description of underlying assumptions. Especially the part about the scoring system should be better introduced and potentially justified.

The analysis of climate effects during the plant development phases delivers interesting results. The authors argue that with their scoring system they can assess the combined effect of climatic conditions throughout the plant development. Here the question arises whether the climatic impacts during the production phase are the same irrespective of the climatic conditions throughout the earlier plant development stages. For example, Ben-Ari et al. 2018 describes a compound event where the combination of warm winter and wet spring lead to a crop failure. As I understand the analysis, it wouldn’t be able to capture such compound events if it is not generally bad for wheat to have warm winters and wet springs. This is just an example, but it might help to understand a limitation that comes from splitting up events. I would find it interested to read the authors view on this concern. These reflections could also be included in the discussion.

The use of only one climate model appears problematic to me. Furthermore, for this type of analysis I don’t see the benefit of high spatial resolution if in the end regional averages are used. I would find it more convincing to see a CMIP6 ensemble instead of one high-resolution model. On the other hand, the climate model projections are not the main part of the analysis. Therefore one could also think of comparing this climate model to the CMIP6 ensemble and discussing the differences and potential biases.

Minor comments:

The abstract could be improved. At the moment it reads a bit like a summary of different results and ideas. The aim of the study should be clarified more precisely and not all results have to be included in the abstract.

L9-10: “future impacts of climate projections on wheat”. I think this should be formulated differently.

L30-31: Is this due to climatic conditions only? Or does technology play a role here?

L97-101: Did you consider a different spatial aggregation methods for precipitation? While for temperature it seems reasonable to average over the regions, for precipitation there could be other meaningful choices. As an example, what would you think about area affected by extreme precipitation instead of regionally averaged precipitation?

L111: I think you should mention here, that the scientific community is not considering this scenario as a plausible future. I have seen, that you do so later on. Maybe still worth mentioning earlier.

L102: Although the UKCP Local simulations are surely great, there remains a large uncertainty with respect to forced changes in precipitation. The accurate representation of small features in these simulations does not necessarily reduce the uncertainty concerning the regional trend in precipitation. Therefore it would be good to compare the precipitation tendency from this model with climate models from other institutes. I have seen that you do so later in the manuscript.

L120: Could you add one or two sentences on the bias correction method? Is it a trend-preserving bias correction?

L173-175: Are these two sentences contradicting each other?

Section 3.2 and Table 2: How would you explain that the effects of climate conditions are different between the regions? I wouldn’t have expected different effects for the different regions. If there is a reason for that it would be good to mention it. You explain this in L194-206, right?

L220: What is the advantage of using this “score”. Couldn’t you also work directly with the correlations of table 2?

L246: Is “sample” the correct word here? I would have written “project”. But I’m not a native speaker.

L246: Is this statement true for the UK in particular? And how did you get there? I think it would be good to spend a few more sentences on this aspect to provide a good overview of potential biases over UK.

Figure 7: I think this figure could be improved a bit. What do you think about displaying the ensemble spread by a shaded area and the ensemble median by a line?

L292: “since crop yields” instead of “since inter-annual crop yields”?