Synergistic contributions of climate and management intensifications to maize yield trends from 1961 to 2017

Iowa is around the middle of the U.S. Corn Belt, which has an area of about 14.5 million ha, of which 86% is cropland. It had commonly ranked first in the US corn productions over the last decades, accounting for above 15% of the annual totals (USDA 2020). As in many other states of the Corn Belt, maize in Iowa is usually planted in April or May and harvested between October and November, mostly under rainfed conditions. We used the decision support system for agrotechnology transfer CERES-Maize model (Jones et al 2003) to simulate annual maize yields for each county of Iowa over six decades, from 1961 to 2017. The model simulations explicitly account for changes in management, cultivars, climate, and CO2 concentration. The climate and agronomic data used in this study are introduced in the supplementary information.

The process-based crop model was calibrated at each cluster including multiple counties (see SI Supporting text), and the simulations were later aggregated to obtain state-level simulations. Specifically, we were interested in the crop model simulated yields $\hat y = \eta \left( {{\boldsymbol{\mathit{\xi }}}{\text{|}}{\boldsymbol{\mathit{\varphi }}}} \right)$, with the model denoted by $\eta$, based on a vector of inputs ${\boldsymbol{{\xi }}}$ and a vector of, in principle unknown, parameters ${\boldsymbol{{\varphi }}}$. The calibration was to find one or multiple sets of unknown genetic and soil model parameters that minimize the differences between the simulated and the observed county yields during some of the years. In addition, we designed a parameterization strategy that accounts for the explicit representation of the changes in management, both in space and time, within a Bayesian framework for parameter optimization. We assume that the genotype parameters vary in time and space, while the soil parameters vary in space but not in time. The generic genotype coefficients for calibration at any county are considered as changing linearly with time t. The calibration used 22 out of the 57 years' data, most of which were randomly chosen, with the other 35 years being used for validation. Details of the model calibration and validation are provided in the supplementary Information. The calibrated and validated model was then used to simulate and compare state-level yield assessments under actual and defined conditions.

Using the calibrated model, we assessed the contributions of the different factors, including climate, temperature, solar radiation, precipitation, and management, to decadal and multi-decadal yield trends at the state level. We assess the trends and contributions during the entire period from 1961 to 2017 as a measure of the long-term impacts, during 1982–1998 and 1984–2013 to compare with the previous studies of Lobell and Asner (2003) and Tollenaar at al. (2017), respectively, and over each decade and multi-decades between 1961 and 2017. We evaluated the contribution of a factor by comparing the actual simulated yields (i.e. the simulated yields using the calibrated model considering the actual climate and management conditions every year) with the alternative simulated yields when the actual values of that factor were replaced with ones from other years. For example, to evaluate the climate contribution, we run the calibrated model for 1961 using the weather dataset from 1962, 1963, ... until 2017 at every county. Then we repeated the same run for 1962 using the weather data from the other years, and so on until the last year. After aggregating the county-level yields to the state level, we obtained a matrix of 57 × 57 yields, with the i, j element of the matrix being the yield under actual conditions in year $i$, except the climate factor, which is based on the weather records at year $j$. The differences between columns of this matrix were induced by the factor of interest, the climate in this case, while the differences between rows were induced by non-climate factors, e.g. management. To evaluate the contribution of the management intensification to the yield trend, we simulated the yield when replacing both the daily records of the weather series and the annual record of CO2 concentration from the current year with the records from the other years and analyzed the differences between the rows of the resulting matrix. The supplementary information provides detailed information for assessing the contribution of the individual climatic variable and contribution of the climate and management factors to yield changes.

To understand the synergistic effects of warming and management intensifications, we assessed the changes in yields under different management conditions considering warming temperature following a systematic increase of 0.0343 and 0.0604 °C per year, which are equivalent to a rise of about 2.8 and 4.8 °C in 80 years, comparable to lower and higher warming levels in the US by the end of the century (Easterling et al 2017). To minimize the influence of historical patterns of inter-decadal variability (Henson et al 2017) and retain important patterns of inter-annual variability, we first permuted continuous decades of weather records randomly, starting from a typical randomly chosen year (say, the third year of every decade), and then imposed the increments in temperature to the rearranged series and simulated the annual yields using the incremented temperature series. We evaluated the changes in yields associated with the systematic increases in temperature considering five of these permutations. To assess the synergistic effects of warming and management, we evaluated the trends induced by temperature or climate change of the yield differences between the imposed warming and the baseline from 1961 to 2017: $T_i^{{{\,\text{f}}_{p,q}}} = s\left( {y_i^{{{{\tilde{\mathrm{f}}}}_p}} - y_i^{{{\,\text{f}}_p}},\,y_i^{{{{\tilde{\mathrm{f}}}}_{p + 1}}} - y_i^{{{\,\text{f}}_{p + 1}}}, \ldots ,y_i^{{{{\tilde{\mathrm{f}}}}_q}} - y_i^{{{\,\text{f}}_q}}} \right)$ for each of the replicas, by fixing the other climate factors or management according to the conditions of each of those year. The accent in $\tilde f$ indicates the value of the factor considering warming levels.

Figure 1 shows the performance of the county-level and state-level calibrated model for the years included in the calibration and those excluded. The performance was good at both levels, but that at the state level was better than at the county-level for both calibration and validation, with the root mean squared error being about 0.2 Mg ha⁻¹ lower and R² being about 0.14 larger than the county level. This is likely because the county-level errors tend to be random, canceled out after aggregation to the state-level. The calibrated county-level yields underestimated the largest yields observed in recent years, while slightly overestimated the lowest yields. The estimations were particularly worse in 1993 than in the other years, with the estimated yields at the state-level being more than 2 Mg ha⁻¹ larger than the observed yield (corresponding to the red dot far from the diagonal in figure 1(c)). Yields in 1993 were about 30% lower than in previous years because of the severe floods that occurred in the summer of that year.

Zoom In Zoom Out Reset image size

Figure 2 shows the changes in maize yields induced by climate and management factors as well as the contributions of these factors to yield trends from 1961 to 2017 as well as two subperiods: 1982–1998 and 1984–2013, corresponding with the study period of Lobell and Asner (2003) and Tollenaar et al (2017), respectively. The result shows that management intensification contributes to approximately 90%, 74%, and 69% of the yield trends from 1961 to 2017, 1984–2013, and 1982–1998, respectively, climate combined contributes to approximately 10%, 26%, and 31% during the corresponding periods. The contributions of climate factors to yield trends are significantly varied among the three periods, with the contribution during the entire period (1961–2017) being much smaller compared to the periods selected in previous prominent studies (1984–2013 or 1982–1998), suggesting the importance of decadal climate variability in determining yield trend.

Zoom In Zoom Out Reset image size

We found that temperatures contributed to 27% of the yield trend from 1982 to 1998, coinciding with a period of decreasing temperature. It confirms the study of Lobell and Asner (2003), which suggested climate factors, especially temperature, had a significant contribution (25%) to the yield trend during this period. We also found that radiation contributed to approximately 20% of the yield trend from 1984 to 2013, coinciding with systematic increases in solar brightening. It corroborated the finding of Tollenaar et al (2017) showing that solar radiation has the most significant contribution (approximately 27%) among the climatic factors during this period. The result also showed that precipitation had a minor impact on yield trend compared to temperature and solar radiation, despite the significant increases in rainfall during the last three decades (Pryor et al 2009, Ford 2014, Mueller et al 2016).

Figure 3 shows decadal yield trends driven by the climate and management factors over different subperiods from 1961 to 2021. The change in climate conditions during the first half of the study period mostly had negative impacts on yields associated with adverse changes in temperature-driven trend (figure 3(b)). In contrast, the change over the second half generally had positive impacts (figure 3(a)) associated with favorable changes in solar radiation driven trend (figure 3(d)). The opposite impacts of the climate during the first decades and the last decades resulted in only a slight contribution of climate to the long-term yield trend over the entire period (figure 3(a)), suggesting nontrivial effects due to the choices of study periods. It is worth noting that the yield trends induced by temperature and solar radiation changes were, in size, comparable and considerably more significant than those attributable to precipitation (figure 3(c)) and lower than those attributable to management (figure 3(e)). However, they often differed in direction, meaning that they frequently caused counteractive effects on yields. On average, the increase of 1 °C (1 W m²) of the summer temperature (radiation) during the different decadal and multidecadal periods (figure 4) led to a decrease (increase) in yields of 0.35 (0.033) Mg ha⁻¹ (figure 3). Similarly, the increase of 100 mm in the summer precipitations (figure 4) led to an increase in yields of 0.05 Mg ha⁻¹ (figure 3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The trends driven by solar radiation showed lower interdecadal variability than those driven by temperature. The radiation contribution includes a prolonged period of large and favorable impacts after the 1980s following the small or negative effects in previous decades due to the multidecadal climate variations (figure 4), consistent with the patterns of global dimming and brightening (Wild 2009). The maximum decadal yield trend (in absolute terms) due to changes in temperature and solar radiation was about ±0.13 and ±0.11 Mg ha⁻¹ yr⁻¹, respectively, comparable to the maximum yield trend associated with management intensifications (approximately +0.14 Mg ha⁻¹ yr⁻¹), while considerably larger than the maximum trend due to changes in precipitation (approximately ±0.05 Mg ha⁻¹ yr⁻¹). This result suggests that temperature and radiation can play at least an equally important role as management in determining yield trends over a short period.

Figures 5(a) and (b) show the yield trends driven by combined or individual climate factors from 1961 to 1991 and 1961–2017, respectively, considering constant management and other climate factors in different years. From 1961 to 1991, the temperature had a positive trend while the radiation and the precipitation experienced little change (figure 4). The yield trends driven by different climate factors were always negative and substantially decreased with management intensification throughout the years (figure 5(a)). For example, the reduction in yields considering management conditions in 1991 was about six times greater than those in 1961 for the same increase in temperature. Unlike 1961–1991, during 1961–2017, the temperature showed almost no trend while precipitation and solar radiation increased (figure 4). The climate-driven yield trends were all positive or close to zero. They increased with management intensifications over time, with the precipitation-driven trend showing a strong positive effect from management intensification while the radiation-driven trend showed little effect. These results suggest that management intensifications can induce more yield gains with increased precipitation, but greater losses of yields with increased temperature, and radiation changes have little effect on yield gains from management intensification.

Zoom In Zoom Out Reset image size

It is also worth noting that yield gains due to management intensifications were considerably diminished under drought conditions (figure 5(c)). For example, in 1988, a year with severe drought, the management-induced yield gains were the lowest (+0.035 Mg ha⁻¹ yr⁻¹) from 1961 to 2017. The yield under the 1988 extreme drought (the most severe drought) and 1961 management (the lowest intensification) conditions were comparable and even better than the yields under management conditions over other years while increasing management intensification from 1961 to 2017 exacerbated the drought impacts (figure 5(d)). Similarly, under the climate conditions of 1983, 1976, and 2012, the years with severe droughts, the yield gains due to management intensifications were also low (between +0.062 and +0.064 Mg ha⁻¹ yr⁻¹).

Projections show that the Midwest summer temperatures will increase more than in many other regions of the U.S (US Global Change Research Program 2017). Figure 6 shows the trends of yield differences between baseline and imposed warming at two different levels along with changing temperature or combined climate factors from 1961 to 2017. All the trends are negative while considerably decreased with the intensification of management from 1961 to 2017, with those under the higher warming levels decreasing more than those under the lower warming levels. The between-replication variability of the trends were similar under different scenarios and management intensification levels. However, the variability of the temperature-induced trends was greater than that of the climate-induced trends because temperature-induced trends were calculated using constant precipitation and solar radiation records over years for each replication, while these records showing great variations between replications.

Zoom In Zoom Out Reset image size

Under the management condition of recent years, the average trend at the higher warming level was about twice lower than that at the lower warming level, and the sensitivity of yield to warming temperature increased with management intensifications from 1961 to 2017. It is worth noting that while the yield gains from intensification decreased at imposed warm levels (figure 6), management intensification provided increased yields under any climate conditions (figures 2–5). In other words, yields in warmer conditions increased with management intensification, but to a lesser extent. Under the imposed lower and higher warming levels, the trends along with management intensifications from 1961 to 2017 under the constant 1988 climate condition (the extreme drought year) were +0.034 and +0.032 Mg ha⁻¹ yr⁻¹, respectively, lower than the original condition without warming. The simulated losses from this study due to warming are comparable to previous studies (Brown and Rosenberg 1999, Mearns et al 2001) based on process-based crop simulation models, while lower compared to previous empirical model-based studies (Schlenker and Roberts 2009, Urban et al 2012) regardless of the warming levels.