Effects of temperature frequency trends on projected japonica rice (Oryza sativa L.) yield and dry matter distribution with elevated carbon dioxide

Study area

Japonica rice is typically grown in temperate regions (Chauhan, Jabran & Mahajan, 2017; Mackill & Lei, 1997). The major japonica rice cultivation areas are located in northeastern China, Korea, Japan, New South Wales of Australia, and European countries among the Mediterranean Sea (Fig. 1). We calculated the annual mean cumulative growing degree days (GDD) over 1901 through 2016 for these regions based on the temperature data from the Climate Research Unit-National Centers for Environmental Prediction (CRUNCEP; Viovy, 2011). The cumulative GDDs were mostly between 1,000–2,600  °C day in these regions, and the base temperature was set to 9 °C in our calculation based on Singh, Ritchie & Godwin (1993). For this study, we selected the Zhongwei (37.53°N, 105.18°E) and Yongning (38.25°N, 106.25°E) rice field stations, which are located about 125 km apart in Ningxia province, China (Fig. 1) and have a typical temperate continental climate. Due to the high production and excellent quality, japonica rice has been planted in this region for about 2,000 years (Wang et al., 2013) where the GDD is about 1,800 °C day based on the temperature data for the period of 1901–2016.

Input data for the CERES-Rice model

Historical weather data, field management information, soil properties, and rice cultivar parameters are needed to drive the CERES-Rice model. Historical weather data were derived from the China Meteorological Forcing Dataset (CMFD) (He et al., 2020; Yang et al., 2010), including the daily maximum temperature (T_max), daily minimum temperature (T_min), daily precipitation, and daily incoming solar radiation. These data cover China with a spatial resolution of 0.1° at 3-hour time steps for the period of 1979 to 2015. In this study, we averaged the weather data for each station over the nine surrounding grids to reduce the uncertainty. The field management information required by the CERES-Rice model was provided by the Ningxia Meteorological Bureau, China. This information included detailed phenology dates, yields, amounts and dates of fertilization and irrigation, etc. Soil data were collected from the China soil database (http://vdb3.soil.csdb.cn/) as shown in Table 1. We used the rice cultivar Ningjing No. 16, a common japonica rice cultivar in the study area. All data from the Zhongwei station were used to calibrate the CERES-Rice model, and data from the Yongning station were used for model validation.

The future climate data used to drive the CERES-Rice model were derived from the outputs of three GCMs (Table 2) with four Representative Concentration Pathways (RCPs) (2.6, 4.5, 6.0, and 8.5) from the Coupled Model Intercomparison Project Phase 5 (Taylor, Stouffer & Meehl, 2012). These three GCMs represent different warming intensities, where the temperature increase was the most aggressive in IPSL-CM5A-MR, moderately aggressive in MIROC5, and least aggressive in GFDL-ESM2G (Hsu et al., 2013). The variables we used from these GCMs were T_max, T_min, precipitation, and incoming solar radiation. We also used the projected future CO₂ concentrations from Prather et al. (2013) to consider the effects of CO₂ on rice growth. Future projections in this study were made using the cultivar Ningjing No. 16, and the transplanting date was set to May 12 each year, a time when rice transplanting is commonly done (Liu et al., 2017; Sang, Liu & Qiu, 2006). According to the local farmers’ experience, transplanting was set at a hill spacing of 25 ×10 cm with six seedlings per hill (Chen, 2012; Ma et al., 2003; Wang, Yin & An, 1996). A total of 290 kg/ha of nitrogen fertilizer was applied to the rice field each year, the maximum value in the records, to ensure that there was no nitrogen stress during the rice growth period. The other field management information, such as full irrigation, and soil data were the same as in the historical runs.

Methodology

Downscaling and trend calculation

For this study, statistical downscaling was performed to transform the coarse-resolution GCM data (1.3°–2.5°) into the station data from Zhongwei and Yongning to reduce the uncertainty in the GCM data. The CMFD data were used as observations for downscaling. We first used bilinear interpolation to downscale these 3-hourly GCM data, except precipitation, to the resolution of CMFD (0.1°). Second, the downscaled data were statistically bias-corrected by maintaining the probability distributions of historical values similar to those of CMFD through linear regression models (Dettinger et al., 2004). In addition, we used daily downscaled precipitation data (Abatzoglou, 2013) with the method described by Buotte et al. (2019) to create 3-hourly precipitation data for the selected GCMs, and then interpolated these new precipitation data to 0.1° using bilinear interpolation. Through downscaling and bias-corrections, the biases in these GCM outputs were reduced for the historical period, making the future projections more reliable. Meanwhile, the long-term trends of the downscaled climate and simulated rice growth and development variables were calculated for the period of 2006–2100. The significance of these trends was examined with the Mann–Kendall test (Hamed, 2008) using R programming language.

Parameter estimation and model evaluation

The crop genotype parameters in CERES-Rice were optimized with the Generalized Likelihood Uncertainty Estimation (GLUE) package (He et al., 2010) using seven-year available agricultural experiment data from the Zhongwei station. These optimized parameters were validated with data from the Yongning station (Table 4). In this study, we used the absolute relative error (ARE) and the least normalized root mean square error (nRMSE) to evaluate the performance of the CERES-Rice model. The largest AREs of the anthesis and maturity dates were less than 11% for the two stations. Except for the Yongning station in 1997, all the AREs of the simulated rice yield were less than 13% (Figs. 2 and 3). Ritchie et al. (1998) indicated that the quality of DSSAT simulations with AREs of less than 15% are considered acceptable. In addition, He et al. (2017) suggested that a crop model performance is considered reasonable when the nRMSE is less than 15%. Therefore, the GLUE-optimized genotype parameters for the CERES-Rice model can be employed to study the effects of future climate on rice growth and development.

Climate change projections

GCM data downscaling

We calculated the root mean square error (RMSE) differences between the downscaled and original GCM data for the period of 1979–2015 (Fig. 4). The observational data used for the calculations were the CMFD data. All the RMSEs were reduced with the downscaled data except for those with precipitation in the three red boxes simulated by IPSL-CM5A-MR (Fig. 4). These RMSE increases were very minor and negligible. Thus, statistical downscaling improved the quality of the historical GCM data and should be capable of generating more reliable future projections.

Trends of climate variable projections

The annual averaged T_max and T_min showed a significantly upward trend with most greenhouse gas emission scenarios over the rice growth period, while the solar radiation and precipitation trends were mostly insignificant over the same period (Fig. 5). In most of our study cases (92%), the trends of T_max and T_min passed the 95% significance level (p < 0.05). The significant trend for T_max ranged from 0.08 to 1.12 °C /decade, while it ranged from 0.09 to 0.86 °C /decade for T_min. The largest T_max and T_min trends occurred with IPSL-CM5A-MR for RCP 8.5. Thus, these significant trends may affect future rice growth and development projections, as analyzed in the following sections.

Impact of climate change on rice growth and development

Photosynthetic production

In this study, the accumulated photosynthetic production for the entire rice growth period showed different trends under different climate change scenarios (Fig. 8). The IPSL-CM5A-MR and MIROC5 models generated negative trends for all emission scenarios, while the GFDL-ESM2G model produced positive trends. The majority of the trends (in blue and red) passed the 95% significance level test. As discussed previously, when the temperature was out of the optimal range, this reduced the photosynthetic production (Eq. (1)). Our simulations showed that the temperature shifted from the low and optimal ranges toward the high and extreme ones under almost all study scenarios (Fig. 6), and such shifts tended to reduce the photosynthetic production. Thus, it is clear that the negative trends of accumulated photosynthetic production were controlled by rising temperatures.

Aboveground biomass, leaf, stem, panicle, and yield

The dry matter accumulation of rice organs was strongly affected by the photosynthetic intensity, whose long-term changes were very sensitive to the temperature frequency trends. Figure 9 shows that the trends of aboveground biomass at harvest were very similar to those of accumulated photosynthetic production, since the carbohydrates of the rice plant are produced by photosynthesis. The leaf and stem dry matter at harvest decreased significantly in the MIROC5 and IPSL-CM5A-MR models (p < 0.05; blue boxes in Fig. 10), while the trends of such dry matter mostly did not pass the 95% significance test in the GFDL-ESM2G model (white boxes in Fig. 10). The trends of panicle dry matter at harvest did not pass the 95% significance test for approximately 68.8% of the study cases in the MIROC5 and IPSL-CM5A-MR models (white boxes in Fig. 11). Two RCP 2.6 cases with MIROC5 and two RCP 8.5 cases with IPSL-CM5A-MR showed a significant downward trend of panicle dry matter, while an upward trend of this variable occurred for the MIROC5 with RCP 8.5 case with a p-value of less than 0.05, which was an exception. The panicle dry matter increased meaningfully for all study cases in the GFDL-ESM2G model (p < 0.05; red boxes in Fig. 11). Yield trends were very similar to those of the panicles (Fig. 12), because the panicle dry matter is mostly composed of grain weight (yield).

Trends of temperature frequency

Based on above results, we can see that HT and EMT become dominant under future climate change with the decreased LT and OT frequencies (Fig. 6). Such shifts not only reduced rice photosynthesis (Eq. (1); Fig. 8), but limited fertilization as well (Fig. 11). Noticeably, with the daily maximum temperature over 35 °C for even a few hours, rice fertilization was restricted, severely reducing the yield, as verified by field experiments (Madan et al., 2012; Jagadish, Craufurd & Wheeler, 2007; Jagadish, Craufurd & Wheeler, 2008; Yoshida, Satake & Mackill, 1981; Satake & Yoshida, 1978). In the CERES-Rice model, this process was also considered. However, the model was run at a daily time step, which could cause errors in simulating rice fertilization.

Photosynthetic production

In this study, the IPSL-CM5A-MR and MIROC5 models generated negative trends for all emission scenarios, while the GFDL-ESM2G model produced positive trends (Fig. 8), which need to be analyzed. As rice is a C₃ plant, the elevated CO₂ concentrations in the GCM projections enhance the activity of Rubisco, which accelerates the carboxylation of ribulose-1,5-bisphosphate (RuBP) and increases the net CO₂ fixation rate, strengthening the photosynthetic production (Busch et al., 2013; Sasaki et al., 2007; Makino, 1994). Thus, the positive trends in the GFDL-ESM2G model indicate that the CO₂ concentration can override the effects of rising temperature and increase photosynthetic production. Among the three GCMs, the GFDL-ESM2G model had the least temperature increase (Fig. 5), and the upward trends of the HT and EMT frequencies in this GCM were the slowest (Fig. 6). Thus, the CO₂ concentration exerted a stronger effect on photosynthetic production than temperature in the GFDL-ESM2G model. On the other hand, rising temperature played a dominant role in affecting photosynthetic production than CO₂ concentration in the other two GCMs. Different from the increased photosynthetic production in the field experiments (Wang et al., 2019a; Nam et al., 2013; Roy et al., 2012; Kim et al., 2011a; Cheng et al., 2010), the most scenarios in our studies showed a significant decrease in rice photosynthesis, which can be attributed to the random combinations of temperature and CO₂ concentration in the field experiments.

Dry matter of leaf, stem, panicle, and yield

The dry matter distribution among rice organs was affected by both photosynthesis and pollen fertilization. The rice photosynthesis in the field experiment mostly increased (Wang et al., 2019a; Nam et al., 2013; Roy et al., 2012; Kim et al., 2011a; Kim et al., 2011b; Cheng et al., 2010). These studies found that the dry matter of each organ increases with the increased CO₂ concentration and temperature (Roy et al., 2012). On the other hand, the elevated temperature reduces the dry matter allocation to panicle due to the decreased spikelet numbers, significantly increasing the leaf and stem dry matter (Nam et al., 2013; Kim et al., 2011a; Kim et al., 2011b; Cheng et al., 2010). However, these results are based on the limited and random combinations of temperature and CO₂ concentration. When photosynthesis is weakened, or when both photosynthesis and fertilization are restricted, the dry matter accumulation and distribution are uncertain.

Usually, leaf and stem dry matter is transferred to the panicles to form rice grains during the grain filling period (Stoy, 1980). When the leaf and stem dry matter decreased (Fig. 10), we did not see a significant decrease in the panicle dry matter except for two RCP 2.6 cases with MIROC5 and two RCP 8.5 cases with IPSL-CM5A-MR, where the panicle dry matter had a negative trend with a p-value of less than 0.05 (blue boxes). The trends of the projected fertilized spikelets (Fig. S1) were determined and were consistent with the changes in panicle dry matter in our CERES-Rice model projections. The insignificant changes in the fertilized spikelets and panicle dry matter imply that the demand and transfer of leaf and stem dry matter remained at a certain level, even if this dry matter decreased quite obviously due to the reduction in photosynthetic production. However, the panicle dry matter showed a distinct downward trend in the MIROC5 with RCP 2.6 cases and IPSL-CM5A-MR with RCP 8.5 cases. For the RCP 2.6 scenarios, the HT frequency in the MIROC5 model had the largest increase among the three models (Fig. 6), showing the largest photosynthetic production reduction (Fig. 8), and therefore decreasing the panicle dry matter much more than in the other two models. In these cases, the low carbon concentration with RCP 2.6 could not compensate for the large reduction in photosynthetic production and panicle dry matter caused by the increased HT frequency. In addition, the EMT frequency had the highest increase in the IPSL-CM5A-MR with the RCP 8.5 case among all the study cases, dramatically reducing the number of the fertilized spikelets and thus the panicle dry matter as shown in Fig. 11.

The rice panicle receives the photosynthetic product with a higher priority than the leaves and stem in the CERES-Rice model. The increase in panicle dry matter was clearly caused by the increased photosynthetic production in the GFDL-ESM2G model due to the more dominant effects of carbon over temperature (Figs. 10 and 11). This model generated an upward trend in the number of fertilized spikelets (Fig. S1) with the elevated photosynthetic production, indicating that the demand and transfer of the photosynthetic product to the panicles needed to hold a similar trend. This demand and transfer increase could reduce the percentage of photosynthetic product distribution to the leaves and stem and suppress their trends. To a very large extent, these processes led to insignificant changes in the leaves and stem (Fig. 10).

Yield trends were very similar to those of the panicles (Fig. 12), since changes in the panicles resulted directly in changes in the yield. Thus, the two variables shared similar trends behind their changes. The most dramatic yield reduction occurred in the IPSL-CM5A-MR with the RCP 8.5 case for the two study stations. As mentioned, the EMT frequency had the largest increase in this case. Previous studies indicated that EMT weakens the activity of pollen and inhibits anther dehiscence, resulting in poor pollen shed on the stigma (Prasad et al., 2006; Matsui, Omasa & Horie, 2000; Matsui, Omasa & Horie, 2001). The resulting processes severely weakened the rice fertilization and, thus, reduced the spikelet numbers and panicle dry matter. Finally, the highest EMT frequency led to the largest reduction in the rice yield.

The effects of high temperature and CO₂ on other rice type

The combined effects of increased temperature and CO₂ concentration on the dry matter distribution in the organs of other rice types such as the indica rice may be different from those for japonica rice. The primary reason can be attributed to the higher heat resistance of indica rice than that of japonica rice (Wang et al., 2019b). Wang (2016) indicated that the spikelet degeneration rate of japonica rice cultivars increases more quickly than that of indica rice cultivars under high temperatures, implying that the indica rice may have a better heat adaptability. Such adaptability can affect the dry matter accumulation and distribution among the rice organs. Therefore, the indica rice may have a different response to the increased temperature and CO₂ concentration, which needs to be further investigated.

Limitations

Several limitations still exist in this study. Only two sites were selected for our simulations, and additional simulations need to be done in broader areas to draw more general conclusions. The single rice cultivar and fixed planting date are also possible shortcomings of this study. Different cultivars may have different sensitivities to high temperature, and the vegetative and reproductive growth phases usually vary with different cultivars. All these can affect the dry matter accumulation and distribution among the rice organs. Meanwhile, the effect of EMT on rice fertilization is likely avoided by adjusting the planting date, which need to be further tested. In addition, our simulations were conducted with sufficient water and fertilizer, which cannot always mimic the reality. These limitations need to be overcome in the future studies.