1. Introduction

Canada is the fifth largest agricultural exporter in the world [1]. Over approximately the last five decades, farm market receipts have grown by 5.8% per year on average, with a variety of grains being among the largest crops [2]. Within Canada, the Prairie (i.e. western) and Maritime (i.e. eastern) provinces produce the majority of grain crops. These regions grow a wide variety of crops, but among them, wheat, barley, and oats are three of the largest crops in terms of hectares or in terms of seeded acres [3].

Technological developments in both harvesting machinery and crop genetics have boosted these crop yields [3]. Changes in domestic and foreign economic conditions, trading agreements, and agricultural policy have also been important influences [4]. In addition, climatological conditions can influence crop yield [5–7], which is our focus here.

A major mode of climate variability that can influence climatological conditions globally, including in Canada, is the El-Niño Southern Oscillation (ENSO) [8]. ENSO’s variation in sea surface temperatures (SST) over the equatorial Pacific typically cycles every ~2–7 years, thus its presence and effects are recognized on interannual, rather than multidecadal, timescales [6, 7, 9, 10]. ENSO influences global crop yields, although that influence can be positive or negative depending on the region and on the type of crop [6]. ENSO’s influence on crop yield has been studied in the Canadian context [1, 6–8, 11].

The mechanisms of influence are the change in surface air temperatures (SAT) and precipitation during the summer growing season, as it has been established that water stress due to dry heat is the most important factor leading to reduced yields [12]. Spring wheat and canola yields in the Canadian prairies have been found to be most significantly influenced by rainfall, which is patterned by La Niña (less rainfall) and El Niño (more rainfall), corresponding to lower and higher yields respectively [13] over a 55-year period. Maize yields in Midwestern American states over 1961 to 1991 were significantly higher during El Niño events and significantly lower during La Niña events [14], again with rainfall as a significant mechanism, and with differing results for soybean crops, which have a different tolerance for drought.

More poorly understood in the Canadian context is the relationship between the Atlantic Multidecadal Oscillation (AMO) [15] and crop yields. Rather than on interannual scales, the AMO experiences SST variability over many decades, as a full period from negative (cool) to positive (warm) phase occurs over roughly 50–80 years [10, 15]. The negative phases of AMO have been found to be correlated with increased risk of maize and winter wheat crop failures between 1960 and 2016 in rainfed regions of the eastern United States [16], and AMO was identified as the most important predictor for maize and wheat in many of the climate divisions of the same region over the same time frame [17].

Like ENSO, AMO influences SAT and precipitation on land in Eastern North America [15]. As we can expect ENSO’s influence on crop yield to manifest in the Canadian West, we might expect AMO to influence crop yields in the Canadian East, specifically in the Maritimes.

We leverage more than 110 years of data on annual Canadian wheat, barley, and oat yields, to our knowledge the longest analysis of crop yields, to understand their association with ENSO and AMO, in conjunction with SAT and precipitation as possible risk factors. Because AMO’s periodicity is multidecadal, any analysis of its influence on crop yields requires a dataset that covers many decades, enough to span multiple periods of the AMO. The AMO has been identified for several decades [18], but research on the influence of the AMO on crop yields has likely been suppressed by the scarce availability of data on a time period of sufficient length to identify a relationship. In addition to the longest-term analysis of AMO on crop yields, this study provides an analysis of the risk of yield during ENSO, AMO, and various categories of SAT and precipitation events, and within regions–the Canadian Prairies and Maritimes–that are less studied in terms of crop yields. Moreover, because different crops experience different risks associated with large-scale climate oscillations, our analysis of wheat, barley and oat yields stands to significantly contribute to our understanding of the relationship between large-scale climate oscillations and crop yields. This risk analysis is an advancement over previous studies that have examined correlation coefficients between yields and climatological conditions [16].

2. Methods

Time series of wheat, barley, and oat yield for Prairies (Alberta, Saskatchewan, and Manitoba) and Maritime (Prince Edward Island, New Brunswick, and Nova Scotia) regions in Canada from 1908–2017 were obtained from Statistics Canada’s CANSIM database, which can be accessed through: https://www150.statcan.gc.ca/t1/tbl1/en/cv.action?pid=3210035901#timeframe. Data for crop production (bushels) and seed harvested area (acres) were obtained.

To a large extent, variability in Pacific and Atlantic SST strongly dictate climatological conditions on land, including SAT and precipitation [9]. The ENSO climate pattern is defined by an index which reflects deviations in SST over the central equatorial Pacific region, referred to as the Niño 3.4 region (5°N-5°S, 170°W-120°W) from the base period of 1981–2010 [9]. Similarly, the AMO index is defined by deviations in SST over the North Atlantic basin (0°-70°N and 75°W-7°W) [15]. Monthly mean ENSO and AMO indices were obtained from the National Oceanic and Atmospheric Administration (NOAA), throughout the period of 1908–2017: https://www.esrl.noaa.gov/psd/data/climateindices/list/.

ENSO experiences variability on short time scales of ~2–7 years, in contrast to the AMO which experiences fluctuations in SST across the entire North Atlantic Ocean with a period of ~50–80 years [15]. We classified the summer ENSO based on its standard definition of negative index values ≤-0.5°C (La Niña), index values >-0.5°C and <0.5°C (neutral), positive index values ≥0.5°C (El Niño) [9]. Similarly, summer AMO into two categories for each year based on its index value; negative (cool) for values <0°C and positive (warm) for values >0°C [19]. The ‘multidecadal variability’ of ENSO and AMO time series were calculated using the Singular Spectral Analysis (SSA) technique, which was used to reconstruct the low frequency variability [10]. As wheat, barley, and oat growing seasons are primarily in the summer, we studied the influence of ENSO on Prairie yields and AMO on Maritime yields during the summer period, from May-September [11].

To our knowledge the only monthly land SAT and precipitation rate dataset that spans the period of 1908–2017 is the University of Delaware’s reanalysis version 5.01 with a spatial resolution of 0.5° by 0.5° [20]. The mean summer (May-September) land SAT and precipitation rate anomalies (i.e. deviations based on 1981–2010 summers [19]) were computed from 49–60°N and 240–265°E corresponding to the Prairie region as well as from 43.4–47.5°N and 292.2–299.2°E corresponding to the Maritime region. To better understand the shape of associations, rather than using summer SAT as a continuous variable in our models, we grouped SAT based on its common magnitude thresholds: ≤-1°C (extreme cold); >-1°C and ≤0°C (cold); >0°C and <1°C (warm); ≥1°C (extreme heat). The distribution of precipitation is smaller than SAT, hence summer precipitation rate anomalies were categorized into three categories: ≤-1mm/day (dry); >-1mm/day and <1mm/day (neutral); ≥1 mm/day (wet).

As a study spanning a relatively long period of human agricultural activity, we realize that aside from climate factors, crop yields are influenced by changes in technology, population growth, and global crises related to politics or health [21]. Regarding population growth, although Prairie and/or Maritime specific populations were not available in historical archives, we obtained data through Statistics Canada for estimated population records (https://www150.statcan.gc.ca/n1/pub/98-187-x/4151287-eng.htm). With respect to agricultural technology, although there is no single technological variable to rely on as an indicator of technological change, we know from historical records that the period following World War II (WWII) was the onset of the Green Revolution [21], with major developments for yield-enhancing technologies occurring and being implemented in Canada over a 30–40 year time frame. For this reason, a ‘Technology Era’ variable was created grouped into three windows approximately every 36 years: 1908–1944 (early technology era, before WWII); 1945–1981 (mid technology era, after WWII); and 1982–2017 (late technology era, recent period). Furthermore, Canada participated in WWI (1914–1918) and WWII (1939–1945), which could also impact crop production, along with global infectious disease pandemics [22]. We created a variable ‘Crisis’ with two values: No crisis & Yes crisis. Years of crisis in our data included WWI, WWII, and past pandemics (i.e. Spanish Flu H1N1 virus 1918–1920, H2N2 virus 1957–1958, H3N2 virus 1968–1969, swine flu 2009–2010). These years were inserted as “yes” in our data for a crisis that was taking place during that year, with other years noted as “no”.

Prairie and Maritime wheat, barley, oat yields were found to have a temporal autocorrelation from year to year, leading to a red noise spectra that is characterized by a reduction in the number of degrees of freedom [10, 23]. The variance in crop production and yields was found to be much greater than the mean, leading to over dispersion in the data. Thus, four separate models were created–two unadjusted (one for Prairie and one for Maritime) and two adjusted models (one for Prairie and one for Maritime), each model and region stratified by crop type, barley, oat, and wheat.

The unadjusted model shows a simple relationship between ENSO/AMO and each crop by the region, using negative binomial regression with autocorrelated residuals of order one [19]. This allowed other climate factors to be bracketed to allow us to understand raw relationships.

Although some past work elects to detrend yield data to minimize longitudinal effects such as technological and socioeconomic factors that are not included in the models, we instead opted to include these major influences in our models as covariates. The equation for unadjusted models is shown below. Since crop yields are a ratio of production and seeded area, we used crop production as our outcome, while using seeded area as the offset to our models [24]. where

The adjusted models assessed the association between ENSO on Prairie crops, with covariates (i.e. model 3), as well as between AMO on Maritime crops, with covariates (i.e. model 4), using negative binomial regression with autocorrelated residuals of order one [19], adjusted for SAT, precipitation rate, technology era, crisis periods, secular trend (i.e. year), and Canadian national population estimates, because Prairie and Maritime-specific estimates were not available prior to WWII. Such generalized linear models (which included negative binomial regression) are standard methodology in public health and other applied disciplines [25]. Henceforth, the estimated coefficients of the models are interpreted as a rate ratio (RR). If RR>1 then it implies a positive association, meaning each 1-unit or 1-level rise in the predictor, increases the risk of crop yield. Similarly if RR<1, then it implies a negative association, meaning each 1-unit or 1-level rise in the predictor, decreases the risk of crop yield. The reference category was set to RR = 1. A two-sided t-test was used and P<0.05 was deemed significant. The adjusted model with all the covariates is described in the equation below. All figures and analyses were conducted in a combination of MATLAB R2019b and SAS OnDemand for Academics. where

3. Results

Annual yields for wheat, barley, and oats in Prairie and Maritime Canada were significantly increasing from 1908–2017 (Fig 1A and 1B). Moreover, it seems that in both regions (Prairie and Maritime) of Canada, oats have not only the highest yield amongst the other two crops within the region, but also a steep rise in yield (see Table 1 for more statistics).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Time series of crop yield from 1908–2017.

a, Annual yield of wheat, barley, and oats in Prairie Canada. b, Annual yield of wheat, barley, and oats in Maritime Canada.

https://doi.org/10.1371/journal.pclm.0000259.g001

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Summary statistics for each crop type and climatological variable by region from 1908–2017.

https://doi.org/10.1371/journal.pclm.0000259.t001

From 1908–2017, summer ENSO experiences a strong interannual variability but weak multidecadal variability (Fig 2A). On the other hand, the summer AMO experiences a strong multidecadal variability on ~50 year time scales (Fig 2B). Relationships between summer ENSO and AMO with crops in Prairies and Maritime, respectively have been presented in Fig 2C and 2D. Summer SAT and precipitation rate anomalies in Prairie Canada (Fig 3A and 3C) is not as strong with less extreme values in comparison to SAT and precipitation rate anomalies in Maritime Canada (Fig 3B and 3D). Hence summer SAT and precipitation possibly serve as key climatological mechanisms linking these climate patterns to Prairie and Maritime crop yields.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2.

a, Interannual (grey) and multidecadal (green) variability in mean summer (May-September) ENSO index. b, Interannual (grey) and multidecadal (orange) variability in mean summer (May-September) AMO index. c, scatterplot of summer ENSO index versus crop yield in Prairies from 1908–2017. d, scatterplot of summer AMO index versus crop yield in Maritime from 1908–2017.

https://doi.org/10.1371/journal.pclm.0000259.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3.

a, b, Mean summer (May-September) SAT anomalies, along with their categories for the Prairie and Maritime. c, d, Mean summer (May-September) precipitation rate anomalies, along with their categories for the Prairie and Maritime.

https://doi.org/10.1371/journal.pclm.0000259.g003

To understand effects of summer ENSO and AMO on either Prairie and/or Maritime crops, two separate unadjusted models were considered for each of the three crops. These models allow understanding of the raw effects of ENSO/AMO by geographic region, without any other climatic modification or mediating factors (Table 2). From Table 2, it is clear that the effects of either an El Niño or La Niña (in comparison to the neutral phase) are insignificant on crop yields in either the Prairie or Maritime regions. On the other hand, a negative phase of AMO (in comparison to positive AMO phase) shows a significant reduction in crop yield by ~7–17%, primarily for the Maritimes and less for the Prairies. These results demonstrate the varying effects of AMO, suggesting these teleconnections have regional influences on agriculture.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Unadjusted associations of annual crop yield to summer (May-September) ENSO and AMO categories in both Prairie and Maritime regions, 1908–2017.

https://doi.org/10.1371/journal.pclm.0000259.t002

Accordingly, we document the risk that summer ENSO and AMO variability introduce to crop yield in Prairie and Maritime independently, using a separate model for each region (Table 3). We see that, while controlling for other climatological factors, technology, global crises, secular trend (i.e. year), and population growth, summer La Niña and El Niño had insignificant associations with crop yield in Prairie Canada. At the same time, a positive phase of the AMO (in comparison to negative phase) reduces the risk of Maritime wheat, barley, and oat yield by 12%, 7%, and 2.8%, respectively over the period of 1908–2017. Furthermore, from Table 3, extreme hot summer SAT is insignificantly associated with any of the three crops in the Prairies, but precipitation rate seems to play an important role, where dry conditions (in comparison to neutral conditions) were significantly associated with a decreased risk in barley, oat, and wheat yield by ~12%, ~15%, ~16%, respectively. On the other hand, in the Maritime region, extreme heat summers were significantly associated with decreased yield for all three crop by ~22%, and extreme summer precipitation rate was significantly associated with a decreased risk in barley and oat by ~7% and 5%, respectively.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Associations of crop to summer (May-September) ENSO and AMO categories, along with controlling for covariates in Prairie and Maritime Canada, respectively from 1908–2017.

https://doi.org/10.1371/journal.pclm.0000259.t003

From Table 3 we also note that the middle era of technology (i.e. 1945–1981), in comparison to the early-era, produced the greatest amount of yield, regardless of crop type or region, which we discuss below. Interestingly, we also note from Table 3 that during global crises there is a general tendency to increase crop yield. As a sensitivity analysis, we found that El Niño’s effects on Prairie crops are stationary and insignificant throughout various time periods of technology era changes (S1 Table).

4. Discussion

Meeting the demands of a growing Canadian population, crop yields have increased considerably over the past century in both the Prairies and Maritimes [1], as noted by our findings. These increases observed in the Canadian context reflect similar increases observed in other regions with similar climates and seasonality [26, 27]. Long-term increases in crop yields are attributable to scientific advancements in crop productivity, referred to as the “Green Revolution” [28]. The Green Revolution was a diffuse strategy in the post-World War II era among governments, private firms, and non-governmental organizations to increase crop yields, both to enhance profitability and to meet skyrocketing demand for food in developing nations [28]. The primary tool for achieving enhanced crop productivity was genetic improvements [29], but advancements in fertilizer use and mechanical harvesting technologies were also influential.

We find that the effect of large-scale oceanic variability on crop yields can be discerned apart from the trend of an overall increase in these yields. Despite extreme phases of ENSO leading both drought and flooding in Western North America which are known to have a damaging effect on crop yield [8], we find a weak association with crop yields in Prairie Canada, consistent with prior research [8, 11]. Our analysis also finds that multidecadal variability in ENSO is not associated with any of the three crop yields, simply because ENSO does not have a strong multidecadal signal [6, 7, 9, 10].

In contrast, we find evidence of a more significant negative association of summer AMO interannual and multidecadal variability on the risk of wheat, barley, and oat yield in Maritime Canada. The summer AMO also has a strong influence on summer SAT and precipitation in Maritime Canada. Since a positive phase of the summer AMO results in above-normal SAT and lack of summer precipitation over Eastern Canada [15], we found that both these climatological covariates can possibly lead to reduced levels of soil moisture and crop yields. Considering AMO’s multidecadal variability we note along with others that from the mid-1930’s to the late-1950’s the AMO remained in a positive phase [15, 30]. Some have argued that this positive phase shift in the AMO led to the onset of the “dust bowl” event around the 1930’s, which reduced harvest and crop yield due to extreme drought summer conditions stretching from Western to Eastern Canada [30]. However, as our data show, below-normal crop yield persisted until the AMO switched to a negative phase around the year 1960, increasing the risk of wheat, barley, and oat yield through enhanced levels of summer precipitation and cooler SAT over Eastern Canada [15, 30].

Interestingly, we found that the covariates of summer SAT and precipitation rates had opposite effects on crop yield (i.e. positive in the Prairie versus negative in the Maritime). These mixed relationships between SAT/precipitation and various crop yields has not only been noted in previous Canadian studies [31], but also worldwide [32]. We believe soil conditions and topography also play an important role in explaining such differing associations. The Prairie region contends with drought and high temperatures relatively often. Soil conditions are drier in the Prairies and water stress is the critical limiting factor for crop production [1]. In contrast, Maritime soil is more sandy/clay in texture, which aids in water retention, and the region tends to experience relatively wet conditions at baseline. Given these differences, we can expect to see different outcomes across the two regions due to increases in precipitation. In the Maritime region increased precipitation risks introducing more moisture than is optimal for crop yield, consequently reducing yields.

There is also some evidence of a narrowing in AMO’s multidecadal variability over the past few decades due to the intensification of global SST warming in recent decades [33]. Currently the AMO is believed to be in its peak positive phase [10], along with rising summer SAT in Eastern Canada [1]. This could also suggest that current Canadian Maritime crop yields are maximally suppressed by the summer AMO. With recent North Atlantic Ocean cooling, studies have shown the likelihood of an emerging negative phase of the AMO [34]. Given our findings, this emergent negative phase may increase crop yields in the Canadian Maritime provinces to above-normal levels. However, such effects may be undermined due to the impact of global warming [33]. Some of the broader implications of this study include that agricultural policy provide for monitoring agriculture yield during extreme ENSO years.

There are a few limitations to our study. First, the data for crop yield exists on an annual basis, while our climatological exposures were based on the summer mean, raising the possibility that the risk that these climate patterns play on crop yield may be misrepresented. Second, in terms of the association between Maritime Canadian crop yield and summer AMO variability, we relied on 110 years of data, yet we were only able to capture at most 1.5 cycles of AMO, since its periodicity varies from ~50–80 years [10, 15]. Nevertheless, this is an improvement on past research. Third, the summer AMO variability varies in magnitude of its risk on wheat, barley, and oat yield in Eastern Canada differently, proposing that each crop may have dissimilar biological characteristics, which may be affected uniquely due to climatological, soil, and topographic conditions. Finally, the dynamics of the interactions between large-scale climate oscillations and temporal variability in their periodicity due to both natural climate variability and anthropogenic climate change could also be influencing these results and require further research [35, 36].

5. Conclusion

The interannual and multidecadal variability in summer AMO can be used as a tool for long-term forecasts related not only to crop yield in Eastern Canada, but perhaps also in other regions of the United States and Western Europe. Future work is needed to understand the effects that summer AMO variability can have on other crops in North America and Western Europe. Although past research focused much more frequently on ENSO than on AMO, our findings point to the need for research to move towards a more balanced consideration of the effects of AMO relative to ENSO in the study of crop yields.

Supporting information

Acknowledgments

We thank Statistics Canada and NOAA for their open access data.