Optimizing sowing window and cultivar choice can boost China's maize yield under 1.5 °C and 2 °C global warming

China's Maize Belt was divided into six maize planting regions (figure 1) based on geographic location and different cropping systems (table 1). We focused on 163 weather stations with observed climate data in the belt for rainfed maize plantings. Historically observed daily climate data of 163 weather stations from 1980 to 2016 were available from the China Meteorological Administration to drive the maize model and calculate the thermal time of different growing seasons and maize cultivars, including maximum temperature (°C), minimum temperature (°C), precipitation (mm) and sunshine hours (h). Daily solar radiation (MJ m⁻²) was estimated using the Angstrom equation with sunshine hours (Wang et al 2015a). Observed phenological data during 1980–2011 by 100 agro-meteorological observational sites across China's Maize Belt were used to set the actual maize sowing date and derive the potential sowing window. Observed data on maize cultivar, phenology (e.g. sowing, flowering, and maturity dates), yield and field management practices from 12 experimental sites were used to calibrate and validate the maize model. These field experiments were conducted across China's Maize Belt to investigate the responses of maize growth and development to sowing date. Detailed information on soil data can be found in supplementary S1 and table S1 is available online at stacks.iop.org/ERL/15/024015/mmedia.

Zoom In Zoom Out Reset image size

Future time series of daily temperature, precipitation and solar radiation data for the study area were statistically downscaled from monthly gridded climate data (with the resolution of 2°× 2.5°) obtained from the GISS-E2-H-CC (GE2) global circulation model (GCM), using NWAI-WG downscaling model (Liu and Zuo 2012). The spatial inverse distance-weighted (IDW) interpolation method was used to downscale gridded monthly data to sites, followed by bias-correction using qq-plotting approach by comparing observed and the GCM projected data for the period of 1960–1999. Then, the modified WGEN stochastic weather generator was used to downscale the bias-corrected monthly data to daily data. These climate outputs based on this downscaling approach have been extensively applied in climate change impact studies (Liu et al 2014, Anwar et al 2015, Wang et al 2016, 2017, 2018, Feng et al 2019, Yao et al 2020). In this study, we selected the GE2 model because it is superior in its ability to capture historical climate for China compared to the ability of other GCMs (Yang et al 2019).

To generate warming scenarios of 1.5 °C and 2 °C, 1986–2005 was set as the baseline period given that 1986–2005 is 0.61 °C warmer than the preindustrial period (IPCC 2013, UNFCCC 2015). The 1.5 °C and 2 °C warming scenarios were derived using 20 year time slice periods following the methods of previous studies (Gosling et al 2017, Leng 2018). Using the 20 year moving mean temperature method, 1.5 °C and 2 °C warming scenarios, additional warming of 0.89 °C and 1.39 °C above the baseline were anticipated during 2018–2037 and 2044–2063, respectively, for the GE2 model under representative concentration pathway (RCP) 4.5 (Schleussner et al 2016), which represents a moderate greenhouse gas emission scenario which is more closely to meet 1.5 °C and 2.0 °C target under current socio-economic conditions. Supplementary S2 and table S2 showed projected climate change during the maize growing season.

The Agricultural Production Systems sIMulator (APSIM) maize model (version 7.7) was used to investigate the future climate change adaptation potential of maize production by adjusting the sowing date and shifting the cultivar. APSIM simulates maize growth, development and yield formation in response to solar radiation, temperature, photoperiod, soil water and nitrogen (Keating et al 2003, Holzworth et al 2014). Maize phenology is determined by temperature together with photoperiod sensitivity of a given cultivar. Daily aboveground biomass accumulation is calculated by daily solar radiation interception and radiation use efficiency, reduced by soil water and nitrogen stress. Grain yield of maize is determined by grain number, daily grain-filling rate and assimilate re-translocation. Precipitation affects crop biomass accumulation and yields significantly by impacting soil water content in the APSIM. APSIM has been well evaluated and applied in China's Maize Belt and performed well in capturing the impacts of changes in climate and agronomic management practices on crop growth, development and yield (Liu et al 2013, Wang et al 2014, Bu et al 2015, Dai et al 2016). Here, we further evaluated the performance of the APSIM-Maize model in simulating the phenology and yield of maize in terms of different cultivars, sowing dates and regions across China's Maize Belt. The statistical metrics for model evaluation were shown in supplementary S3.

Three representative cultivars at different cultivar maturities (early-, middle- and late-maturing) were selected for each maize planting region to minimize the possible spatial heterogeneity in local cultivars (table 2). For regions I–III, the same representative cultivars were used due to similar cropping system and management options. Genetic parameters for four cultivars used in the APSIM were referenced directly from the literature, and the genetic parameters of the other eight cultivars were derived by model calibration and validation based on experimental field data with 'trial-and-error' method. Genetic parameters for maize phenology were determined by comparing observed and simulated flowering and maturity dates while genetic parameters for maize yields were determined by comparing observed and simulated maize yields. We also referenced the traits of maize cultivars given by breeders to ensure the physiological means of parameters. Experimental field data from the literature were extracted using WebPlotDigitizer software (https://automeris.io/WebPlotDigitizer/) from the figures in the literature. To calibrate and validate the APSIM-Maize model, field experimental data were separated into independent calibration and validation data by different sowing dates and years.

To evaluate the adaptation potential of maize production to climate change, long-term simulations driven by projected daily climate data including solar radiation, maximum and minimum temperatures and precipitation, for the baseline scenario and 1.5 °C and 2 °C warming scenarios were conducted without any nitrogen stress during the maize growth period under rainfed conditions. Nitrogen fertilizer was applied automatically within 50 cm depth in the soil to keep mineral nitrogen no less than 300 kg ha⁻¹ to avoid of any nitrogen stress. All simulations were conducted under rainfed conditions. For single-cropping maize planting regions (I–IV), the simulation of the first 10 years was discarded as a 'spin up' period to minimize the effect of the initial condition (Teixeira et al 2015, Tang et al 2019). For the double-cropping system (V–VI), the initial soil water content at the maximal root depth at sowing was reset at 30% and 81% of the plant available water holding capacity for regions V and VI, respectively, according to the observed historical averaged initial soil water content at sowing from agro-meteorological experimental sites. For all maize planting regions, the maize planting density was set as 67 500 plant ha⁻¹. The planting depth and row spacing were set as 5 cm and 60 cm, respectively. Our study also considered the impact of an increase in CO₂ concentration on maize (supplementary S4). For C₄ crop, the elevated CO₂ concentration mainly affects plant transpiration. Pervious study has showed that transpiration efficiency was increased linearly by 37% with CO₂ concentration increasing from 350 to 700 ppm (Lobell et al 2015), and therefore we nested this function into APSIM-Maize model.

In the case of without adaptation options, actual sowing date at each weather station used the average sowing date from historically observed values at each nearest agro-meteorological sites and actual maize cultivar was selected from local typical cultivars in table 2 based on the heat condition at the planting region (supplementary S5 and table S3).

When considering adaptation strategies, the optimal combination of sowing date and maize cultivar was selected from three typical cultivars in table 2 and the sowing date within the potential sowing window for each planting region. The potential sowing window for each region was defined as the period between the earliest and latest planting dates for the early-maturing cultivar at each planting region. The earliest planting date was set as the first day when the five-day moving average of daily average temperature was >8 °C, which is the baseline temperature for maize used in the APSIM-Maize model and is based on a literature review (Sánchez et al 2014) in regions I–IV. In region V, with a double-cropping system, the earliest planting date constrained by the harvesting date of the previous crop was set 3 d after harvesting the previous crop recorded in agro-meteorological sites. In region VI, with a mixture of a single-cropping system and a double-cropping system, the earliest planting date was set as the recorded averaged maize planting date from agro-meteorological sites. The latest planting dates in regions I–IV were set as the last day maize could mature before the first frost day calculated as the daily minimum temperature below 0 °C. In regions V–VI, the latest planting date was set as the day maize could mature at 7 d before sowing the next crop. Maize was harvested when one of the following three conditions was achieved: (1) at the physiological maturity date, (2) before the first frost day and (3) 3 d before sowing next season crop.

The optimal sowing window was prescribed as the sowing dates that achieve over 80% of the highest yield within the potential sowing window. Cultivar harvested the highest yield was chosen as the optimal cultivar for a given region.

There were four types of simulated yield changes (%) in our study compared with that of the baseline scenario: (1) ΔY_1.5/2.0 was the yield change under the 1.5 °C or 2.0 °C warming scenarios compared with baseline, (2) ΔY_1.5/2.0,s was the yield change caused by climate change from baseline to 1.5 °C or 2.0 °C warming scenarios and the adjusting sowing date, (3) ΔY_1.5/2.0,c was the yield change caused by climate change from baseline to 1.5 °C or 2.0 °C warming scenarios and the shifting cultivar, and (4) ΔY_1.5/2.0,s&c was the yield change when considering climate change from baseline to 1.5 °C or 2.0 °C warming scenarios, changing the sowing date and shifting the cultivar.

$$\begin{eqnarray}&&{\rm{\Delta }}{{Y}}_{1.5/2.0}\left( \% \right)=\left({{Y}}_{1.5/2.0}-{{Y}}_{{\rm{bl}}}\right)/{{Y}}_{{\rm{bl}}}\times 100\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{s}}}\left( \% \right)=\left({{Y}}_{1.5/2.0,{\rm{s}}}-{{Y}}_{{\rm{bl}}}\right)/{{Y}}_{{\rm{bl}}}\times 100\end{eqnarray} \tag{ 2 }$$

$$\begin{eqnarray}&&{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{c}}}\left( \% \right)=\left({{Y}}_{1.5/2.0,{\rm{c}}}-{{Y}}_{{\rm{bl}}}\right)/{{Y}}_{{\rm{bl}}}\times 100\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}( \% )=({{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}-{{Y}}_{{\rm{bl}}})/{{Y}}_{{\rm{bl}}}\times 100,\end{eqnarray} \tag{ 4 }$$

where subscripts s and c represent adjusting sowing date and shifting cultivar, respectively. Y_bl/1.5/2.0 was the simulated average yield for baseline or under warming of 1.5 °C or 2 °C without adaptation, Y_1.5/2.0,s was the simulated yield with the adjusted sowing time alone under warming of 1.5 °C or 2 °C, Y_1.5/2.0,c was the simulated yield with the shifted cultivar alone under warming of 1.5 °C or 2 °C, and Y_1.5/2.0,s&c was the simulated yield with the adjusted sowing date and shifted cultivar under warming of 1.5 °C or 2 °C.

ΔY_s, ΔY_c and ΔY_s&c (intermediate variables) were the yield changes that occurred after adjusting the sowing date, shifting the cultivar and combining the adjusted sowing date and shifted cultivar under the 1.5 °C or 2.0 °C warming scenarios compared with the yield under the baseline scenario. The relative contribution rate of climate change (ΔR_1.5/2.0), change in sowing date (ΔR_s), change in cultivar (ΔR_c), and combined change in sowing date and cultivar (ΔR_s&c) on yield change was calculated as follows:

$$\begin{eqnarray}&&{\rm{\Delta }}{{Y}}_{{\rm{s}}}\left( \% \right)=\left({{Y}}_{1.5/2.0,{\rm{s}}}-{{Y}}_{1.5/2.0}\right)/{{Y}}_{{\rm{bl}}}\times 100\end{eqnarray} \tag{ 5 }$$

$$\begin{eqnarray}&&{\rm{\Delta }}{{Y}}_{{\rm{c}}}\left( \% \right)=\left({{Y}}_{1.5/2.0,{\rm{c}}}-{{Y}}_{1.5/2.0}\right)/{{Y}}_{{\rm{bl}}}\times 100\end{eqnarray} \tag{ 6 }$$

$$\begin{eqnarray}{\rm{\Delta }}{{Y}}_{{\rm{s}}\&{\rm{c}}}( \% )=({{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}-{{Y}}_{1.5/2.0})/{{Y}}_{{\rm{bl}}}\times 100\end{eqnarray} \tag{ 7 }$$

$$\begin{eqnarray}{\rm{\Delta }}{{R}}_{1.5/2.0}( \% )={\rm{\Delta }}{{Y}}_{1.5/2.0}/{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}\times 100\end{eqnarray} \tag{ 8 }$$

$$\begin{eqnarray}{\rm{\Delta }}{{R}}_{{\rm{s}}}( \% )={\rm{\Delta }}{{Y}}_{{\rm{s}}}/{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}\times 100\end{eqnarray} \tag{ 9 }$$

$$\begin{eqnarray}{\rm{\Delta }}{{R}}_{{\rm{c}}}( \% )={\rm{\Delta }}{{Y}}_{{\rm{c}}}/{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}\times 100\end{eqnarray} \tag{ 10 }$$

$$\begin{eqnarray}{\rm{\Delta }}{{R}}_{{\rm{s}}\&{\rm{c}}}( \% )={\rm{\Delta }}{{Y}}_{{\rm{s}}\&{\rm{c}}}/{\rm{\Delta }}{{Y}}_{1.5/2.0,{\rm{s}}\&{\rm{c}}}\times 100.\end{eqnarray} \tag{ 11 }$$

Figure 2 shows the comparison of the simulated and observed phenology and yield of maize for eight typical local cultivars across China's Maize Belt. APSIM-Maize simulated the duration from sowing to flowering and from sowing to maturity well with root-mean-square error (RMSE) values of 2.1 and 4.1 d for the calibration period and 2.9 and 3.9 d for the validation period, respectively. The simulated maize yield was in good agreement with the observed maize yield with an R² of 0.89 for the calibration and 0.79 for the validation period. For the calibration period, the RMSE was 0.97 t ha⁻¹, the NRMSE was 9.9%; for the validation period, the RMSE was 1.38 t ha⁻¹, and the NRMSE was 17.7%. All cultivar parameters were shown in table S4.

Zoom In Zoom Out Reset image size

The simulated average rainfed yield without adaptation was 5.8–10.0 t ha⁻¹ across China's Maize Belt under the baseline climate scenario, with the highest yield in region III and the lowest yield in region IV (figure 3). In comparison with the baseline scenario, without adaptation, warming of 1.5 °C and 2 °C would increase maize yields by 0.9%–4.9% and 2.6%–7.9% in regions I–IV, respectively, and would reduce maize yields by 6.3%–7.6% and 0.1%–12.5% in regions V–VI, respectively. The factors resulting in yield change varied with region. An increasing CO₂ concentration would lead to maize yield increases in regions I, II and IV (figure 4 and table S5). For region III, the combined increase in growing period precipitation and CO₂ concentration would increase maize yield. However, a decrease in growing period precipitation would decrease maize yield in region V. A decrease in growing period solar radiation and an increase in growing period temperature would decrease maize yield in region VI.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Considering adaptation to climate change through the optimization of the sowing date and cultivar, the simulated average maize yield would increase under all scenarios (figure 3 and table S6). Across China's Maize Belt, simulated average maize yield would increase by 11.1%–53.9%, with the highest increase in regions II and IV and the lowest increase in region VI. Compared with the baseline scenario, under the warming scenarios, only adjusting the sowing date could increase the yield by −0.5% to 40.7%, while only shifting the cultivar could increase the yield by −1.8% to 24.9%.

Contribution rate of 44.5%–96.7% on yield improvement by adjusting the sowing date was higher in comparison with that of shifting the cultivar (0%–50.8%) and climate change (−53.1% to 23.0%) in most maize planting regions, except in regions III and V under warming of 2 °C and in region VI under both warming scenarios (table 3). Optimal cultivar would not change in region I implying rising heat resource under future scenarios of warming 1.5 °C and 2 °C could still not meet the thermal time required by middle-late maturing cultivars. The contribution rate of shifting the cultivar to yield improvement would increase with climate change in all maize planting regions. The combined effect of sowing date and cultivar shifting was not equal to the sum of their individual impacts in regions II–VI, suggesting an interaction effect between changing the sowing date and shifting the cultivar on maize yield change.

Figure 5 shows the potential sowing window, the optimal sowing window and the optimal maize cultivar maturity level across China's Maize Belt under the baseline scenario and the 1.5 °C and 2 °C warming scenarios. The earliest potential sowing date started in mid-April in regions III and IV, while the latest potential sowing date started in early June in region V under the baseline scenario. The longest potential sowing window was in region VI, with an average of 68 d, while the shortest potential sowing window was only 6 d in region I under the baseline scenario. With climate warming, the earliest potential sowing date advanced for all the maize planting regions except those in regions V and VI, where sowing dates were constrained by the harvest of the previous crop. The latest potential sowing date was delayed for all the maize planting regions. Therefore, the potential sowing window increased by 2–17 d under warming of 1.5 °C and 4–26 d under warming of 2 °C across the maize belt with a higher extension rate in higher latitude regions (e.g. regions I–III). The change in the optimal sowing window under climate change was consistent with the change in the potential sowing window. Significant advances in the start date of the optimal sowing window occurred in region I by 8 d and in region II by 10 d, while the end date of the optimal sowing window was delayed in all the maize plating regions, with a more significant delay in regions I–IV under the warming scenarios compared with under the baseline scenario. Therefore, climate warming would significantly lengthen the optimal sowing window, especially in regions I–IV, by an average of 10 d under warming of 1.5 °C and by 12 d under warming of 2 °C relative to under the baseline scenario. The actual sowing windows in regions I–II and VI occurred slightly earlier than the optimal sowing window, while they were close to the optimal sowing window in regions III–V. The optimal cultivar maturity level would change from an early-maturing cultivar under the baseline scenario and warming of 1.5 °C to a late-maturing cultivar in region II under warming of 2 °C but would not change with climate change for the early-maturing cultivar in region I and the late-maturing cultivar in other planting regions.

Zoom In Zoom Out Reset image size