The past, present, and future of multiple wheat and maize breadbasket shocks


 Simultaneous yield shocks in multiple breadbaskets pose a potential threat to global food security, yet the historical risks and causes of such shocks are poorly understood. Here, we compile a dataset of subnational maize and wheat yield anomalies in 25 countries dating back to 1900 to better characterize the past, present, and future risk of multiple breadbasket shocks. We find that years in which at least half of all maize or wheat breadbaskets fall 10% (5%) below expected yields has occurred in ~2-3% (~14-16%) of years over the last century. Importantly, multiple breadbasket shocks have been decreasing in frequency from 1930 to 2017. The El Niño Southern Oscillation (ENSO) most strongly affects the probability of multiple maize breadbasket shocks, while the North Atlantic Oscillation (NAO) most strongly affects the probability of multiple wheat breadbasket shocks, each influencing the probability by up to 40%. The effect of climate change on climate stress in maize and wheat breadbaskets is mixed; extreme heat will increase uniformly, agricultural soil moisture stress will remain constant or increase, but hydrological stress (as measured by runoff) will remain constant or decrease in breadbasket regions.

been established, the strength of the effect has not been studied. 54 Characterizing long-term climate-related risks to maize and wheat breadbasket regions, how-55 ever, is complicated by climate change, which is already affecting crop yields [16][17][18]   To date research on multiple climate-forced breadbasket shocks has been constrained by 68 the necessity of using short statistical records of 38-47 years 7, 9, 10, 23 to characterize the causes, 69 probability, and change in probability, of presumably rare events. With limited crop yield data 70 available, studies detailing the physical causes of multiple breadbasket shocks have focused on 71 case-studies 11,13,15 . Such case studies neither address the likelihood of such events occurring nor 72 do they identify whether the proposed climate risks are those that are most relevant to food security, 73 making it impossible to contextualize the importance of these scenario-based analyses. 74 Here, we construct a new century-long maize and wheat yield dataset and use a comprehen-75 sive set of climate stress variables to provide the most complete picture to date of how climate 76 has affected multiple breadbasket shocks in the past, present, and may affect them in the future. 77 We have compiled a dataset of subnational crop yield anomalies for maize and wheat consisting 78 of over 34,000 observations from 132 subnational units in 25 countries dating back to 1900 ( Fig.   79 1), which we use to robustly estimate the probability of multiple breadbasket shocks, changes to that probability over the last 100 years, and to identify what has caused such shocks in the histor-81 ical record. We furthermore use measures of growing season climate stress that account for both 82 moisture supply and demand, as well as extreme maximum and minimum temperatures, to assess 83 present and future climate risks in major wheat and maize breadbaskets. 84 History of multiple breadbasket shocks The relevance of multiple breadbasket shocks to food 85 production differs by crop. For maize, the total production is dominated by yields in the United 86 States and Northern China, which reflects the concentrated nature of maize production 24 . Joint 87 maize breadbasket shocks between Northern China and the United States, therefore, are particu-88 larly relevant to global food production stability. For wheat, however, production is more evenly 89 distributed among breadbaskets such that the net number of breadbasket regions with low as com-90 pared to high yield anomalies closely corresponds to the total breadbasket production anomaly 91 (Fig. 2). In this context, the net number of breadbasket shocks is highly relevant to the global mar-92 ket. Throughout our analysis we define yield shocks as being either modest or major disruptions 93 using a threshold of 5% or 10% below expected yields, respectively.

94
The historical incidence of widespread moderate disruptions among breadbaskets has not 95 been uncommon, but widespread major disruptions have been exceedingly rare. We find that 96 ∼3% (∼16%) of years from 1930 to 2017 had four of eight maize breadbaskets experience major 97 (moderate) disruptions (Fig. 2). For wheat, ∼2% (∼14%) of years had five out of nine breadbaskets 98 experience major (moderate) disruptions simultaneously. While 14% -16% is a significant fraction 99 of all years, a yield anomaly of 5% below expected is only a modest shortfall in production. translated into substantially more frequent maize or wheat breadbasket shocks (Fig. 2). In fact, 105 the number of simultaneous maize and wheat shocks has been decreasing at both the 5% and 106 10% yield threshold, regardless of whether the trend is measured beginning in 1900 or 1931 -the 107 earliest year for which all regions report data -and regardless of whether regions affected by World 108 War 1 and 2 are removed. The decreasing trends are all significant at the p<0.05 level with the 109 exception of the frequency of wheat yield shocks at the 5% yield threshold when the trend begins 110 in 1931 and regions affected by World War 1 and 2 are removed, which is only significant at the 111 p<0.1 threshold. The overall trend towards stability for both moderate and major maize and wheat 112 shocks is likely a result of the green revolution -the introduction of fertilizers, irrigation, and new 113 varieties in many regions -that substantially increased crop yields and so made yield anomalies 114 relatively more stable compared to total yields.

115
Despite the overall trend towards more stable crop yields, there are a number of decades and 116 regions in which breadbasket shocks were relatively more common. For maize yields, the late 117 1940s and early 1950s stand out as a time with many breadbasket shocks, with crops failing across 118 China, India, Europe, and Southeast South America (Fig. 2). When considering the full period,  The IOD has an effect similar to that of ENSO in many regions, although it has a statistically 160 significant effect on fewer pairs of regions (Fig. 4). The strongest influence of the IOD on joint 161 wheat yield shocks is between pairs of regions that include Australia, while for maize it most 162 strongly affects the probability of joint crop yield shocks in pairs of regions that include South 163 Africa or India.

164
The NAO does not significantly affect as many regions as does ENSO, but the direction of 165 its influence is uniform (Fig. 4). Positive NAO events reduce the probability of joint wheat yield 166 shocks in regions that include the Western Mediterranean, Northern China, and India by ∼0-25%, 167 while negative NAO events increase the probability of joint wheat yield shocks in these regions.  At the global scale, we find that the NAO and ENSO affect the probability of multiple bread-173 basket shocks by up to 40%, while the IOD has only limited statistically significant effects. The 174 negative phase of the NAO is most strongly associated with an increased probability of simulta-175 neous wheat breadbasket shocks despite ENSO affecting a larger geographic area and a greater 176 number of regions than does the NAO, which only affects the climate of the northern hemisphere.

177
That the NAO affects the probability of multiple wheat breadbasket shocks more strongly than 178 ENSO is a result of its unidirectional influence on the probability of joint wheat shocks, as de-179 scribed above.  We find that temperature stress in breadbasket regions will uniformly increase, but the change 198 in agricultural and hydrological moisture stress depends on crop and forcing scenario. For wheat, 199 the breadbasket areas experiencing temperature stress will increase by 100-200%, covering up to 200 80% of all breadbasket areas by the end of the century. The marginal increase in hydrological 201 moisture stress will remain smaller than differences due to internal variability due to the NAO, 202 but the much larger increase in agricultural moisture stress will considerably exceed the observed 203 envelope of natural variability in both forcing scenarios. For maize, temperature stress will increase 204 by three-fold even in the RCP 4.5 scenario, reaching near-complete coverage of all breadbasket 205 areas in the RCP 8.5 scenario. Note that empirically-derived damaging temperature thresholds 206 could not be established for Mexico based on our data, but it is likely that temperatures will be 207 damaging in Mexico as well in this scenario. Hydrological moisture stress in maize breadbasket 208 regions, however, will decrease by ∼30-50% in RCP 4.5 and 8.5 forcing scenarios. Agricultural 209 moisture stress will increase only in the high-forcing scenario, but then it will increase by over text and SI Fig. 3).

232
By the end of the century, climate change will compound the stress that ENSO, the NAO, and 233 the IOD place on breadbaskets. Consistent with past literature, we find that the risk of damaging 234 maximum temperatures increases for all breadbaskets, and does so more strongly in maize than in 235 wheat breadbaskets. Changes to hydroclimate stress, however, will depend on the crop, variable, and forcing scenario. We consider both soil moisture and runoff, which may be captured and used 237 for crop water management. For wheat, although damagingly dry soil moisture conditions are 238 projected to expand in both high and moderate forcing scenarios, runoff conditions will largely 239 remain constant. For maize, damagingly dry soil moisture conditions will only increase in the high 240 forcing scenario while runoff during the growing season will actually become more favorable.

241
These contrasting results imply that for both maize and wheat breadbaskets, water management 242 may become increasingly important as soils dry but, in some regions, runoff increases. These 243 results indicate that major breadbaskets will experience warmer growing seasons with drier soil 244 moisture but constant or increased water available from runoff.    climate stress, which we relate to crop yield anomalies. We first create dataframes at the regional 364 level, in which each observation consists of a z-score for climate stress in a particular subnational 365 unit during a particular year in a given region and the corresponding yield anomaly for that year. 366 We then relate the climate stress covariates to the yield anomalies using a generalized additive