Recent extreme weather events have had severe impacts on UK crop yields, and so there is concern that a greater frequency of extremes could affect crop production in a changing climate. Here we investigate the impacts of future climate change on wheat, the most widely grown cereal crop globally, in a temperate country with currently favourable wheat-growing conditions. Historically, following the plateau of UK wheat yields since the 1990s, we find there has been a recent significant increase in wheat yield volatility, which is only partially explained by seasonal metrics of temperature and precipitation across key wheat growth stages (foundation, construction and production). We find climate impacts on wheat yields are strongest in years with compound weather extremes across multiple growth stages (e.g. frost and heavy rainfall). To assess how these conditions might evolve in the future, we analyse the latest 2.2 km UK Climate Projections (UKCP Local): on average, the foundation growth stage (broadly 1 October to 9 April) is likely to become warmer and wetter, while the construction (10 April to 10 June) and production (11 June to 26 July) stages are likely to become warmer and slightly drier. Statistical wheat yield projections, obtained by driving the regression model with UKCP Local simulations of precipitation and temperature for the UK's three main wheat-growing regions, indicate continued growth of crop yields in the coming decades. Significantly warmer projected winter night temperatures offset the negative impacts of increasing rainfall during the foundation stage, while warmer day temperatures and drier conditions are generally beneficial to yields in the production stage. This work suggests that on average, at the regional scale, climate change is likely to have more positive impacts on UK wheat yields than previously considered. Against this background of positive change, however, our work illustrates that wheat farming in the UK is likely to move outside of the climatic envelope that it has previously experienced, increasing the risk of unseen weather conditions such as intense local thunderstorms or prolonged droughts, which are beyond the scope of this paper.
Globally, wheat is the most widely grown cereal crop by area, with more than
UK national and regional wheat yields.
Observed, direct impacts of climate change on crop yields are emerging globally (Brisson et al., 2010; Grassini et al., 2013; Hochman et al., 2017; Rigden et al., 2020), slowing the growth in global agricultural productivity (Ortiz-Bobea et al., 2021), and altering patterns of global food production (Ray et al., 2019). Rising temperatures under anthropogenic climate change are often detrimental to agricultural productivity (Ortiz-Bobea et al., 2021) and compound heat-drought impacts may directly affect crop growth: for instance, maize and soil yields are historically worse in places with strong associations between low rainfall and high temperature (Lesk et al., 2021). Cool and wet growth phases have also been linked to poor yields because it is hard to warm the surface when soils are wet and hard to dry wet soils during cooler periods. Thus it remains to be seen how warming and precipitation interact and whether future warming may help offset the increased precipitation by drying out waterlogged soils. This interaction depends on how the link between precipitation and soil moisture may evolve in the future (a topic drawing increasing attention in both climate and crop science, enabled by the rise in satellite-derived soil moisture observations). Combined with the nutrition demands of a rapidly growing global population, there is an urgent requirement to estimate these effects on future crop yields. Breeding and evaluating new wheat varieties tolerant of hotter, drier summers may take decades (Zheng et al., 2012) and it is unclear whether advances in agronomy are occurring fast enough to mitigate the impacts of any accelerating frequency of extreme climatic events (Chen et al., 2021). Changing climatic conditions may also affect yields indirectly by constraining the ability of farmers to undertake key management actions of tillage, sowing, and harvest or by causing damage to natural capital, such as soil erosion. These new constraints on yields may overtake any gains from physiological and phenological advances obtained through plant breeding.
In order to assess this risk to future food production, there is a critical
need to understand how climate extremes are likely to evolve during the
seasonal growth phases that are most relevant to the farming industry.
Observational evidence has revealed changes in the intensity, frequency,
duration, and extent of weather extremes, such as heavy rainfall events and
hot days, across certain regions and continents
(Rahmstorf and Coumou, 2011; Slater et al.,
2021). There has been much research relating weather indices to potential
crop variability or projected damage
(Harkness
et al., 2020; Iizumi and Ramankutty, 2016; Rosenzweig et al., 2001; Trnka et
al., 2014), but most work has described weather extremes by using seasonal
or annual metrics rather than focussing on the periods most relevant to crop
growth (Frich et
al., 2002; Zhang et al., 2011). There is also increasing research focus on
compound weather extremes (Zscheischler et al.,
2020) occurring simultaneously or in close succession, such as very warm
temperatures in the late autumn followed by abnormally wet conditions in
spring (Ben-Ari et al., 2018) and their impacts on crop
yields. Of the total annual crop losses in world agriculture, many are due
to direct weather and climatic effects such as drought, flash floods, heavy
rainfall in otherwise dry periods, frost, hail, and storms
(Ray et al.,
2019; Sultan et al., 2019). High temperatures and heat stress lead to
stomatal closure and therefore reduced photosynthesis due to restricted
CO
This work thus investigates (1) whether statistically significant associations exist between observed temperature/precipitation metrics and historical wheat yields during the three crop growth stages, in the three main wheat-growing regions of the UK, and (2) the extent to which projections of compound temperature and precipitation extremes under a high-emission scenario may impact future crop yields. To assess future changes in precipitation and temperature extremes, we employ state-of-the-art UK Climate Projections Local (UKCP 2.2 km) convection-permitting simulations, which constitute a step change in resolving small-scale processes in the atmosphere. These climate projections are considered the most reliable simulations presently available in terms of their ability to project future changes in meteorological extremes over the UK.
Geographically, we focus on the three main wheat-growing regions outlined using the EU “NUTS” classification (European Commission, 2010). These three regions are (i) north-eastern Scotland, eastern Scotland, and the north-east English region (SNE); (ii) the East Midlands, Yorkshire, and the Humber region (EMYH); and (iii) the south-east and eastern region (SEE; Fig. 1d). These three regions account for over 80 % of total UK wheat production by tonnage (DEFRA, 2020) and correspond with the yield reporting boundaries of available data. The regional wheat yield data were obtained from the UK Department for Environment, Food and Rural Affairs (Defra; DEFRA, 2020). The data are drawn from the England Cereals and Oilseeds Production Survey and Scotland Cereal Production and Disposal Survey, part of an annual survey of the UK agricultural industry. For full details of the survey methodology, see DEFRA (2018b). The data were summarized by Defra to average yield at the national (1885–2020) and regional (1990–2020) levels, resulting in 136 and 31 years of data, respectively.
The dates for the foundation, construction, and production growth stages are taken from benchmarks in the UK's “Wheat growth guide”, in Table 1 (AHDB, 2018). Prior knowledge on the effects of climate in different growth stages guides our choice of climate variables in the study (Table 1). Absolute anomalies of wheat yields were computed by fitting a locally weighted scatterplot smoothing curve (LOESS) to obtain the running mean (red lines shown in Fig. 1a and b), and subtracting this running mean from each annual value (resulting anomalies shown in Fig. 1c). We perform this calculation to remove the trend and thereby isolate annual anomalies, which we expect to be related to inter-annual climate variability rather than other factors such as long-term technological improvements, increasing atmospheric carbon dioxide, or climate warming.
Three standardized wheat growth stages, modified by 1 d to avoid overlap across stages (AHDB, 2018).
For historical climate data we employ the HadUK gridded 5 km observational data from the National Climate Information Centre (NCIC; Hollis et al., 2019). Provisional HadUK data were employed for the year 2020, produced as per previous years (Hollis et al., 2019); provisional data may have very small differences at regional scales compared with the final published dataset, available later in the year. Observed precipitation and temperature data were checked for completeness: any incomplete climate data during each of the crop growth stages (i.e. a foundation phase with less than 187 d of data; a construction phase with less than 60 d, or a production phase with less than 46 d) were removed, to ensure consistency and comparability across years.
To investigate the association with crop yields, we computed climate metrics
within each geographical region and wheat growth stage (Table 2)
using region-averaged values of temperature (
Association between observed climate metrics and wheat yields in
each crop growth stage and region. Table indicates Pearson's correlation
coefficients and their
For precipitation, we computed metrics representing the total
region-averaged daily precipitation within a growth stage
(
The UKCP Local simulations have a spatial resolution of just 2.2 km – providing exceptional detail in local rainfall changes. Importantly, such high resolution allows the climate model to explicitly represent convective precipitation events on the model grid (see Kendon et al., 2019, 2020, for details), thus providing credible projections of future changes in short-duration precipitation extremes and in particular for summer months. The UKCP Local simulations were initially released in September 2019 (Kendon et al., 2019) but were then updated in July 2021 after correction of an error in the representation of graupel (soft ice pellets; Kendon et al., 2021). Here we use the new updated Local 2.2 km projections. The local 2.2 km model (HadREM3-RA11M) spans the UK and is nested within the 12 km regional model (HadREM3-GA705), which is in turn driven by the 60 km global model (HadGEM3-GC3.05; Andrews et al., 2019; Williams et al., 2018). The 2.2 km projections are available for three 20-year periods of 1981–2000, 2021–2040, and 2061–2080. Known atmospheric greenhouse gas (GHG) concentrations are prescribed as forcings to the historical 20-year period. For the second and third periods, the projections employed follow the RCP8.5 scenario, which assumes substantial on-going human burning of fossil fuels. The 2.2 km projections consist of an ensemble of 12 members (Table 3), each of which can be regarded as a plausible realization of the climatic response to rising GHG levels. The local members are driven by different members of the global coupled model ensemble and corresponding regional model ensemble, created by perturbing uncertain parameters in the model physics within their bounds of uncertainty. Thus, the range of the 2.2 km projections provides an estimate of the uncertainty in future changes due to natural variability while additionally accounting for uncertainty in the physics of the driving global climate model. We computed regionally averaged UKCP temperature and precipitation projections for each of the three regions shown in Fig. 1d and for each of the crop growth stages indicated in Table 1. For a detailed discussion of modelling assumptions and limitations see Sect. 2.6.
Bias correction factors for region-averaged total daily precipitation and minimum/mean/maximum daily temperature for each of the three regions (columns) and each of the 12 UKCP ensemble members (rows) relative to HadUK observed data. These are the complete data (ensembles 02, 03, and 14 do not exist in the UKCP Local dataset). Bias correction is performed using daily data over the common historical period 1 December 1980 to 30 November 2000. The bias correction factors are multiplicative for precipitation and additive for temperature.
Given that the driving parent model of each UKCP Local simulation comes from a
perturbed physics ensemble, each ensemble member is typically regarded as a
different model and therefore is independently bias-corrected. UKCP Local
simulations of area-averaged precipitation and temperature were
bias-corrected against the 5 km area-averaged observed daily HadUK data
(Hollis et al., 2019) for each geographical region,
using the entire the historical period of December 1980 to November 2000
(Table 3). The bias correction scaling
factors were identified and applied with the “hyfo” (Xu,
2020) package written in the software R. This bias correction approach is a
simple scaling method which is additive for temperature and multiplicative
for precipitation (one correction factor per ensemble, per region), so it
preserves an absolute or relative trend, respectively. The UKCP data have 30 d in each month; therefore, to perform the bias correction we added
calendar days for each of the three 20-year periods (e.g. from 1 December 1980 to
30 November 2000 with only 30 d in each month) and merged the historical period
with observed data, removing any non-matched days (e.g. dropping the
31st of the month from the observed data, or dropping
29–30 February from the projections). This produced two overlapping
time series of equal length over the period of December 1980 to November 2000 to
perform the bias correction. We make the assumption these present-day biases
are likely to extend into the future periods, a key caveat of any bias
correction method. The bias correction factors are relatively small, which
suggests the simulations are well-aligned with the historical observations:
Bias correction of each UKCP 2.2 km ensemble member, for
Bias correction of each UKCP 2.2 km ensemble member, for
We first assess the association between climate metrics and crop yield by
using pairwise two-variable Pearson correlations (expressed as annual crop
yield versus each individual seasonal climate variable). The magnitudes of
the correlation coefficients and their
Second, to assess the additive or offsetting effects of different climate
conditions across crop growth stages, we develop a multiple linear
regression model between regional crop yields and climate (Eq. 1). We develop one model per region, with different observed temperature
and precipitation variables for each region. Using Table 2, we
purposely select just two continuous variables per growth stage to develop
the model (one temperature-based metric and one precipitation-based metric),
thereby avoiding correlated metrics. The equation used to fit the observed
data for a given region is formulated as
Statistics of the multiple linear regression model (Eq. 1) for
each region and nationally (historical observed data, 1990–2020). The low
Third, to assess future changes in crop yields, we drive the same multiple regression model with the bias-corrected projections of the same variables, computed from the 12 members of the UKCP Local simulations (i.e. we employ a hybrid approach; see Slater et al., 2022). This approach allows us to fuse together the data-driven regression model with the meteorological simulations for higher greenhouse gas emissions. We use the model results to understand how multivariate climate change could lead to compensating or compounding impacts on future crop yields.
One of the advantages of the empirical data-driven approach herein is that
there are fewer assumptions than in a process-based model approach. However,
such an approach makes some key assumptions nonetheless, listed here.
To assess the impact of extreme weather on crop yields, we assume that
the crop yields are affected by weather within the pre-defined crop growth
stages described in Table 1. We employ fixed-in-time growth stages
for practicality, but in reality these growth stages may be weather
dependent from year to year, as plant vulnerabilities to extreme
temperatures or precipitation may differ, e.g. from one July to another
July. We did not use the 99 detailed physiological growth stages
(AHDB, 2022) but rather the high-level growth stages which are
defined over long time periods to split the year into key stages of wheat
growth. A major assumption in our regression-based approach is that wheat
responses to climatic variables in the past are a reliable predictor of
responses in the future. One important uncertainty that we do not consider
is how wheat growth and water use might respond to increases in atmospheric
CO Spatially, we average the climate metrics over the three regions. This
aggregation to regional scales may mask variation in the weather conditions
occurring in individual grid cells (or farms) – for instance the regional
average could show little change, but this could hide large local changes
(such as less frequent but more intense bursts of rain), or contrasting
directions of change within the region. Other spatial metrics, such as
extracting the highest rainfall event within each region, may be worth
testing in future work. The multiple regression model describes the impact of compound climate
effects in different growth stages on wheat yields but not that of
antecedent conditions (memory effects). Compound effects are captured well
by our model, e.g. frost conditions during the foundation phase and heavy
waterlogging during the production phase might combine to produce poor
conditions across the whole year. However, the model cannot assess whether
the climatic impacts during the production phase are the same irrespective
of “memory” impacts from the climatic conditions in the earlier plant
development stages (for example, the antecedent effects of a warm winter and
wet spring in leading to a crop failure, e.g. Ben-Ari et al., 2018). For the future projections of altered meteorological conditions, the
UKCP18 HadGEM3 climate model simulations (in which the UKCP Local 2.2 km
simulations are nested) were only performed for the RCP8.5 pathway for
atmospheric greenhouse gas concentrations, and we do not address emissions
uncertainty from other scenarios. While the likelihood of such high on-going
emissions is now considered low
(Chen et al., 2021; Hausfather and Peters, 2020), the RCP8.5 scenario is
commonly used to facilitate detection of climate signals in future
projections above natural variations in the climate (due to the large
changes projected) and was deliberately chosen as the configuration for
UKCP Local simulations to maximize the signal to noise. Using a high-emission scenario also has the advantage that one can make estimates of
climate changes for lower-emission scenarios using scaling approaches. Our analysis employs one single model, the UKCP Local (2.2 km) climate
projections. As described in Sect. 2.3, the UKCP Local simulations are
driven by a perturbed physics ensemble (PPE) of a single forcing Earth
system model (ESM), i.e. the parameters within the physics of the driving
ESM are perturbed within their bounds of uncertainty. Thus, the 12 members
of the high-resolution ensemble describe both internal climate variability
and the climate modelling uncertainty in the driving model (i.e. they have
wider uncertainty than is typically represented in one single climate
model). The trends of the UKCP Local simulations therefore at least
partially cover the range of uncertainty and trends that would occur in the
ESMs developed by other climate research centres. However, the climate
modelling range of uncertainty is likely to be underestimated since the UKCP
Local ensemble lacks information from other international climate models. In
winter, the UKCP Local simulations show some higher-precipitation responses
compared to the full CMIP5 ensemble due to the improved representation of
winter-time convective showers in the Local 2.2 km model (Kendon et al., 2020).
UKCP Local projections also project relatively high temperature changes
compared with other climate models (see, e.g., The UKCP Local projections provide high spatial-resolution (2.2 km)
downscaling of global climate model projections specifically for the UK.
Such high-resolution simulations are able to at least partially resolve
convective storms and do not require a parameterization scheme to provide a
representation of convection, which is a simplification of the real world
and a known source of model deficiencies. These simulations are therefore
considered more reliable for projecting future changes in rainfall
characteristics. However, there is still uncertainty in the parameterization
of UKCP Local, and so it can be expected that as future research groups also
build convective-permitting models, differences will emerge that we are
presently unable to account for.
Since the late 1800s, and especially since the 1950s, there has been exceptional growth in UK wheat yields due to rapid advances in crop breeding, increasing farm mechanization, and the availability of agrochemical inputs, such as fertilizers (Fig. 1a). Sustained increases throughout the 1980s–90s reflect the development of farming technologies, varieties, improved nutrient use efficiency and effective pesticides and growth regulators. Available time series of crop yields are much shorter when disaggregated to the regional level (Fig. 1b) than at the national level (Fig. 1a). Of particular note, though, is that the EMYH and SNE regions exhibit some levelling of wheat yields since 1990, mirroring the national trend, while the southernmost region, SEE, has seen some continued increases (Fig. 1b).
In addition to increases in mean yields, the national yield time series
exhibits a visible increase in the variance of yields in the last few
decades (Fig. 1c). This increase
in volatility is not solely driven by increases in the mean of the time
series. A comparison of the variance of crop yields between the periods
1885–1989 (105 years) and 1990–2020 (31 years) using both Levene's test
(
We assess the association between seasonal climate and crop yields by using
precipitation and temperature metrics during the three crop growth stages.
We expect the association between climate anomalies and wheat yields to
differ regionally due to a range of factors, including the resilience of the
wheat plant, husbandry practices of farmers and agronomists, biophysical
conditions (e.g. soils, day length), and climatic differences (e.g. rainfall
tends to be more frontal in the north, with orographic rainfall over high
ground, and more convective in the south-eastern UK). Although only some of
the associations between the seasonal climate metrics and annual crop yields
are statistically significant, we show all the associations and their
relative strength for full transparency (Table 2). In
Figs. 4 and 5, we focus on
Association between observed wheat yields and climate during the
three wheat-growing phases. Anomalies of observed UK wheat yields are shown
for total area-averaged precipitation (
Association between wheat yields and climate during the three
wheat-growing phases and in each of the three UK wheat-growing regions.
Anomalies of observed UK wheat yields are shown for total area-averaged
precipitation (
From a crop physiology perspective, in the foundation phase (October to
early April; Table 1), prolonged waterlogging of the soil may
suppress wheat yields by restricting root development and plant growth
(AHDB, 2018). We find a significant negative association between
crop yields and the number of heavy rainfall days in the EMYH region
(Table 2,
While crops are growing rapidly during the construction phase (April to early June), both late frosts and dry weather can reduce crop growth (Table 1). For this period in each year, we find no significant associations between climate characteristics and crop yields (Table 2). This is not necessarily a contradiction, as reduced growth does not always carry through to reduced yield. Both low yields (e.g. years 1976, 2001, 2020; Fig. 4b) and some high yields (1962, 1984) have occurred during drier-than-average construction phases. Overall, wheat yields seem to be more sensitive to climate conditions during the foundation or production phases.
The clearest association between climate extremes and crop yields seems to
be in the production phase, which is the time from post-flowering to harvest
(summer: June and July). It is during this phase that yields may be
susceptible to both drought and water logging (Table 1). We find a
consistently negative association between heavy rainfall (both
It can be challenging to systematically identify the weather conditions to which wheat yields are most vulnerable within individual growth stages. Most of the correlations in the historical data are not statistically significant (Table 2). The often relatively weak association between climate anomalies and wheat yields at the level of individual growth stages can be explained partly by the shortness of observational records, the combined resilience of the wheat plant (i.e. physiological reproductive mechanisms), and the husbandry skills of farmers and agronomists in mitigating these impacts by adjusting to climatic extremes. There is thus a role for agronomic management in mitigating apparent relationships with climate: this role might not be as direct as irrigating in response to drought, but farmers can dampen the effects of climatic variation through crop management, for example, by changing fungicide regimes to respond to increased fungal disease brought about by wetter conditions, changing the timing or amount of inputs of nutrients, pesticides, and growth regulators (Knight et al., 2012). The relatively intensive nature of UK wheat production (Hillocks, 2012; Wesseler et al., 2015) may thus be sufficient to dampen crop responses to climatic variation (Gagic et al., 2017). Farmers can also change many other aspects of management, including wheat variety, tillage, sowing date, sowing rate, or harvest date, in response to forecast or current conditions. Wheat cultivars are bred with a measure of resistance to certain climatic variables, so a farmer can select a cultivar appropriate to local climatic conditions (Kahiluoto et al., 2019).
Low correlations between climate and yield anomalies over seasonal wheat growth stages may also reflect compensatory effects between growing phases. For instance, a less than ideal foundation phase might be offset by a favourable production phase or vice versa. It is equally important to note that growing phases in real plants are determined by their growth rather than calendar days. Thus a phase may last longer, resulting in delayed crop growth but maintaining the expected yield. Our calendar-fixed phases are a simplification of this process.
Conversely, cumulative detrimental impacts of climate across stages (e.g. accumulated rainfall and subsequent waterlogging) may be one of the most damaging factors affecting overall annual crop yields. In other words, the flexibility and techniques farmers have at their disposal to adapt to climate variability are bounded. For instance, low yields in the year 2018 were due to very dry conditions in the foundation stage, followed by very hot and dry conditions in the construction and production stage (DEFRA, 2018a). In contrast, very low yields in the years 2001 and 2007 were caused by a combination of high rainfall in the foundation and production stages (Fig. 4). The exceptionally wet winter of 2019 (affecting the 2020 harvest in Fig. 4) also imposed severe constraints on farming operations and resulted in a reduction in the areas of autumn-sown crops. These examples illustrate why a full understanding of projected changes to temperature and precipitation across wheat growth stages is required.
To try to assess the offsetting or additive effects across growth stages, we
develop a simple multiple regression model relating the observed wheat
yields in each region to just two metrics reflecting temperature and
precipitation conditions in the most important stages based on the outcomes
of Table 2: foundation
Temporal trends in wheat yields (t ha
At the annual scale, projections of future maximum hourly temperature are available for the periods 2021–2040 and 2061–2080 from the UKCP Local simulations. The interquartile range of projected temperature for 2021–2040 lies well above the median of historical extremes (Fig. 2a–c). Future high-temperature conditions generally fall beyond the bounds of annual variability experienced in the contemporary period for all three wheat-growing regions (Fig. 2c). As expected, changes are largest for the later modelled period 2061–2080, corresponding to higher atmospheric greenhouse gas concentrations. This exceedance of historical thresholds by temperature projections is true for all 12 UKCP Local ensemble members, independent of uncertainty in changes in the large-scale conditions sampled by perturbing parameters in the Hadley Centre global climate model. However, it is important to note that the 12 climate model members (Table 3) do not sample the full range of uncertainty, evident in differences between all available global climate models (Kendon et al., 2021); see Sect. 2.6 for a discussion.
For total annual precipitation (Fig. 2d), the projections do not
indicate a very obvious increase or decrease in any of the three regions
relative to the historical period, although SNE may seem very slightly
wetter and SEE very slightly drier on average (comparing medians) in the
later period (2061–2080). This lack of trend in yearly data may be explained
by the opposing changes in the different seasons: in general the winter
season is projected to become wetter and the summer drier
(Kendon et al., 2021). Importantly, there
are also changes in the underlying intensity and frequency of precipitation
(e.g. significant increases in
Trends in key climate metrics for the three growth stages (columns) and three regions (rows). Metrics are selected from (and defined
in) Table 2:
When considering UKCP Local projections by wheat growth stages (instead of at the annual scale), clearer patterns become apparent (Fig. 3). We expect to find spatial differences in the climate projections, as they exhibit north–south gradients in changes across the UK. Even in a single ensemble, there are gradients in the future changes in rainfall which differ from present-day climatology and relate to regional differences in increases in moisture availability as well as changes in circulation patterns. The question of compound climate change – i.e. the joint impacts of temperature and rainfall or moisture availability – is important for future crop yields.
Contrary to global expectations of declining yields under climate change, the multiple regression model indicates that projections of future temperature and precipitation change are likely to contribute to a continued growth of future wheat yields in the UK (Fig. 6). These projections rely on broad estimates of changing night/day temperature extremes as well as total rainfall in the foundation and production stages. It is possible that more data may provide greater information on changing water availability, atmospheric vapour demand, and plant stress; however with the existing observations our data-driven approach highlights that a changing climate may not be entirely negative for wheat yields. This can be explained as follows.
For the foundation phase (October to early April), all regions can expect to
see progressively warmer, wetter conditions in the coming decades according
to the UKCP simulations. Significant projected increases in
Projections for the construction phase (mid-April to mid-June) are not
included in the multiple regression model, due to the lack of significant
associations between climate and wheat yields (Table 2). During
this phase, the projections indicate significant decreases in
In the production phase (mid-June to end of July), UKCP simulations project
both much warmer and drier conditions (Figs. 3 and 7). The drying signal is relatively similar across the
three regions and becomes more apparent in the later simulations towards the
end of the century. It is important to note that the UKCP Local projects
stronger drying than CMIP5-6 models. Projected trends also indicate
significant, strong increases in
Overall, projections of future temperature and precipitation conditions
suggest a continued increase in future wheat yields when relying on
Lastly, the impact of rising atmospheric CO
Mean UK crop yields saw a rapid growth in the 1950s followed by a plateau in the 1990s and then substantial increases in the inter-annual variability in yields. This acceleration has been challenging for UK wheat farmers, since crop yields over the past 2 decades (2000–2020) have been significantly more volatile than over the previous century (Fig. 1).
A first question is thus our ability to explain such changes, and assess whether statistically significant associations exist between observed temperature/precipitation metrics and historical wheat yields during the three crop growth stages, in the three main wheat-growing regions of the UK. While the plateau in yields can be explained by a variety of technological and agronomic factors (Knight et al., 2012), we find some evidence that yields over the last 30 years can be partially explained by climate metrics such as warm night temperatures and heavy rainfall days in the foundation phase (principally in the EMYH region) or maximum daily temperatures, daily temperature variability, and total precipitation in the production phase (Table 2; with correlation strength and significance varying regionally). Significant statistical associations are found principally in the foundation and production phases and for regions EMYH and NAT. Yields are more fully explained when considering a multiple regression model characterizing additive and offsetting impacts of climate across growth phases (e.g. detrimental impact of very cold temperatures in foundation phase followed by very high precipitation in the production phase). However, it is unclear whether the added explanatory power of the regression model is from inter-stage compensation or compensation between variables within a single growth stage. This would be an area for further research. The data-driven regression could additionally be refined by including various thresholds (e.g. considering the beneficial impacts of a warm and dry production phase only up to certain limits relevant to plant stress). We find the association between historical climate and crop yields is most evident in years which saw compound extremes (Zscheischler et al., 2020), i.e. climate anomalies across multiple growth stages (e.g. 2007, 2012, 2020, Figs. 4 and 5), “escaping” the ability of farmers to adapt through agronomic means. Outside these combined extremes, the data indicate a strong inter-annual resilience of wheat production, implying that at present farmers can, and do, successfully utilize crop husbandry to maintain yield levels.
Our second question seeks to understand how projections of compound
temperature and precipitation extremes might impact future crop yields under
a high-emission climate scenario. Overall, the data provide a surprisingly
favourable outlook of climate conditions for future crop yields. During the
foundation phase, high seasonal values of night temperatures
(
In summary, this work provides evidence that wheat yields over the last 30 years are associated with combined temperature and precipitation extremes, especially across the crop foundation and production phases, in the EMYH region and nationally (Table 2). Although the climate projections provide a generally positive outlook for future yields across the UK, it is important to note that the relationships between past wheat yields and historic climatic conditions may not be adequate guides to the risks associated with projected future conditions, as future temperature extremes and rainfall lie outside the range of conditions that UK agriculture has so far experienced. Further, this work studies climate extremes at the regional scale but not local changes in rainfall intensity and variability, which are beyond the scope of the paper (e.g. drier average regional conditions may hide less frequent but more intense local thunderstorms). Out of caution, therefore, a priority is to continue developing agricultural systems resilient to emerging climate patterns, as the global demand for wheat and other crops has been projected to double from 2005 to 2050 (Tilman et al., 2011). As higher-resolution crop yield data become available, further research into robust process-based or AI-informed crop models, alongside improved collaboration across spatial, governance, and supply-chain scales (Holman et al., 2021), will be required to help farmers adapt to evolving climate conditions and maintain the security of wheat production.
All data employed in the paper are publicly available as described in
the “Methods” section. (1) HadUK gridded 5 km observational temperature and
precipitation data were obtained from the National Climate Information
Centre (
LJS led the coding, analysis and visualization. CH extracted the regional UKCP Local projections. All authors (LJS, CH, RFP, JWR, EJK) contributed to designing the experiments and writing the paper.
The contact author has declared that none of the authors has any competing interests.
Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We gratefully thank the editor and reviewers (Corey Lesk and two other anonymous reviewers) for their helpful comments, which greatly improved the paper. Louise J. Slater gratefully acknowledges funding from the UK Research and Innovation FLF scheme (grant no. MR/V022008/1). Chris Huntingford, Richard F. Pywell, and John W. Redhead gratefully acknowledge the Agland project. Chris Huntingford also acknowledges the NERC CEH National Capability Fund. Richard F. Pywell, Chris Huntingford, and John W. Redhead were supported by research programme LTS-M ASSIST – Achieving Sustainable Agricultural Systems (grant no. NE/N018125/1), funded by NERC and BBSRC. Elizabeth J. Kendon gratefully acknowledges funding from the Joint UK BEIS/Defra Met Office Hadley Centre Climate Programme (grant no. GA01101).
This research has been supported by the Natural Environment Research Council (grant no. NE/N018125/1), the Department for Business, Energy and Industrial Strategy, UK Government (grant no. GA01101), and UK Research and Innovation (grant no. MR/V022008/1).
This paper was edited by Gabriele Messori and reviewed by Corey Lesk and two anonymous referees.