Rice yield benefits from historical climate warming to be negated by extreme heat in Northeast China

Rice is currently benefiting from climate warming in Northeast China, but whether such positive effect will continue in the future remains unknown. Here, we evaluate the impacts of individual and combined climate variables on rice yields in Northeast China during 1980–2015. Results show that there is 10% yield increase induced by climate change in Northeast China since 1980. At present, the reduced chilling results in 5.4% yield increase (approximately 28,000 tons) and the higher growing degree-day contributes to 4.6% yield increase (approximately 24,000 tons), while the high-temperature extreme reduced yield by 0.054% (approximately 280 tons). However, with continuous warming, the harmful impact of such high-temperature extreme will outweigh other positive climate effects when the temperature increases by 3.36 °C. Therefore, high-temperature extremes cannot be ignored despite their influence on rice yield being quite limited at present in Northeast China. Climate change mitigation and heat tolerance breeding are thus necessary for rice production in Northeast China.


Introduction
Rice is the main staple food in China, providing 28% of the protein and 20% of the calories for the 850 million people in the country (Peng et al. 2020). Northeast China is a key rice-growing area that contributes to approximately 17.7% of the annual national rice production (National Bureau of Statistics 2020), accounting for 60% of China's high-quality exported rice. Therefore, maintaining the stability of rice production in Northeast China is vital and significant for rice supply in China.
The average temperature of the earth's surface between 2011 and 2020 is 1.1 °C warmer than the average temperature at the end of the nineteenth century. A significant climate-warming trend has been observed in Northeast China. The annual mean temperature in this region has increased by 1.71 °C since 1950 (Li et al. 2004(Li et al. , 2010Ding et al. 2007); this change is approximately 0.51 °C higher than the average warming magnitude in China (Piao et al. 2010). Climate models have projected that future warming in Northeast China is likely to reach 2-4 °C by year 2100 under low-greenhouse-gas-emission scenarios, and 5-7 °C under high-emission scenarios (Wen et al. 2016;Niu et al. 2018;Xu et al. 2018). Increasing temperature results in higher-frequency heat extremes (daily maximum temperature > 35 °C); high-temperature days have been projected to increase by 10-30 days in the 2080s compared to 1980-2010 (Liu et al. 2018). Moreover, climate warming has been reported to reduce the occurrence of cold events Wen et al. 2016;Li et al. 2019;Chou et al. 2021). The extreme temperatures recorded in Northeast China from 1961China from to 2003 show that the number of days with extremely 1 3 low temperatures (daily minimum temperature < 0 °C) has tended to decrease (You et al. 2011). The trend of cold damage events in Heilongjiang Province has been significantly reduced (Zhu and Liu 2011). Over 2021-2100, climate models projected that the lowest temperature would be elevated by 1.1 °C and 2.6 °C under the Representative Concentration Pathways (RCPs) 4.5 (medium-greenhouse-gas-emission scenarios) and RCP 8.5 (high-emission scenarios), respectively Chou et al. 2021).
Climate change has diverse multiple influences on rice growth and production. A significant number of studies have reported a beneficial warming effect on rice yields in Northeast China (Sun and Huang 2011;Tao et al. 2012;Zhou et al. 2013;Zhang et al. 2014;Wang et al. 2014). Zhou et al. (2013) found a 3.6% yield increase for each degree of warming in the Heilongjiang Province of Northeast China. Likewise, Tao et al. (2012) estimated that rice yields increased by 4.5-14.6% per degree due to the increased minimum temperature from 1951 to 2002. Despite these reported positive effects of warming on rice, it remains unclear which climatic variables are responsible for this yield increase. The future warming rate has been projected to be accelerated (Zhou et al. 2013;Wen et al. 2016) and unprecedented  in Northeast China. Therefore, it is still unclear whether the beneficial effects described above will continue under further warming or whether these positive effects will become negative with such a magnitude of warming in the future.
Our study is based on the annual county-level crop datasets recorded by the Chinese Academy of Agricultural Sciences with 6125 data points in 175 sites over 1980-2015 in Northeast China. We analyze the panel data using a fixedeffect regression model ("Methods"). Here, we focused on the effect of extreme temperature (chilling damage and heat stress) and optimal growth temperature variations on rice yield and project possible tipping points for future rice yield changes. It should be noted that our dataset and method cannot be used to quantify the non-climate effects and potential changes in yield brought by new agricultural technologies.

Study region and data sources
We studied the response of rice yields to climate in Northeast China (Fig. 1a), including in the provinces of Heilongjiang, Jilin, and Liaoning. The temperature varies between 13.4 and 24.3 °C, with 369.9-669.5-mm precipitation over the rice-growing season. Since 1980, the rice-growing areas have expanded, with the most rapid increases in Heilongjiang Province. Rice-growing areas in Jilin and Liaoning increased from 385.7 to 544.9 ha and from 252.7 to 761.7 ha over the past three decades, respectively. The rice-growing areas increased from 210.4 to 3147.8 ha in Heilongjiang (Fig. 1b).
We obtained county-level rice yield statistics (from 175 counties) in Northeast China between 1980 and 2015 from the Chinese Academy of Agricultural Sciences. Climate data recorded at ground-based meteorological observation sites were downloaded from the China Meteorological Data Sharing Service System (http:// cdc. cma. gov. cn) and Fig. 1 Characterization of the study region. a Our study region and the distribution of rice-growing areas; and b temporal variations of ricegrowing areas in three provinces then interpolated to the above counties using the DAYMET algorithm (Thornton et al. 1997). These interpolated climate data have also been used in our previous studies .

Climate indices
We use the growing degree-day (GDD, °Cd), extreme heat degree-day (HDD, °Cd), and cooling degree-day (CDD, °Cd) to quantify the effects of the climate and associated temperature extremes on rice production. GDD are used to estimate the growth and development of plants, and the accumulation between the base temperature level required for the rice to start to develop and the optimum temperature level (above the temperature the rice can be damaged and not developed) is used in the calculation. An HDD denotes high-temperature effects when the temperature is beyond the optimum growth temperature threshold. A CDD represents chilling stress that occurs during the reproductive phase. In addition to the three temperature-related indices, we also calculated the growing-season mean radiation (RAD, MJ·m −2 ·day −1 ).
Following an earlier study (Lobell et al. 2012(Lobell et al. , 2020, we define the accumulated degree-days between T base and T opt over the rice-growing season as GDDs (Eq. 1) and the accumulated degree-days above T opt as HDDs (Eq. 2). These values are calculated using the hourly temperatures derived using a cosine interpolation between the daily minimum and maximum temperatures (Eq. 3): where T h is the hourly temperature (°C), T min andT max are the minimum and maximum temperatures, respectively, h is measured from 1 to 24 o'clock in a given day, N is the total number of hours in the rice-growing season (from 1 May to 30 September) (Gao and Li 1992;Sun and Huang 2011), and T base andT opt are the base growth temperature threshold (8 °C) and the optimum growth temperature threshold (30 °C) for rice growth, respectively (Bouman 2001).
To calculate CDD, we adopted the expression shown below (Bouman 2001;Shimono 2011) to quantify the chilling temperature injury at the flowering stage (Eq. 4): where T mean is the daily average temperature, and the summation of CDD is a cooling degree-day below 22 °C during the time from the booting stage to the flowering stage (Bouman 2001; Shimono 2011). The period from the booting to flowering stages is determined based on earlier literature (Gao and Li 1992;Sun and Huang 2011). We used the Angstrom equation (Eq. 5) to calculate the RAD based on sunshine hours (Angstrom 1924).
where RAD is the total solar radiation (MJ·m −2 ·d −1 ); c is the astronomical radiation (MJ·m −2 ·d −1 )); n is the actual sunshine hours (h); N is the number of shineable hours (h); and a and b are empirical coefficients, referring to the results of existing studies (Zhu et al. 2010). The calculation of c follows the (Allen et al. 1998).

Statistical analysis
Panel model regression analyses have been widely used to study climate change impacts (Lobell and Burke 2010;Zhang et al. 2016;Lobell et al. 2020). Rice yields and climate data at the 175 county-level stations collected in three provinces over 1980-2015 were combined into a panel dataset, unit root tests were conducted for each variable to ensure data stationarity (p < 0.05), and then conducted a panel regression model. Since the Hausman test for the fixed-effects model was significant (p < 0.05), a fixed-effect panel data model was established (Eq. 6): where Y i,t is the observed yield at time t at location i. We follow a common approach of expressing yields in log units, which assumes that a given change in climate variables will have the same percent impact on yields. f (Clim i,t ) is a function of the climate variables at location i and time t; is a parameter vector reflecting the sensitivities of climate variables, including GDD, HDD, CDD, RAD, and RAD 2 , to yield changes; i,t is a systematic random error term that represents the effect of factors other than weather on yield; i and 1i ⋅ t are location-specific and location-year fixed effects, respectively; and 2i ⋅ t 2 is the fixed effect of the year squared at a specific location. Location-specific intercepts account for spatial variations in crop management and soil quality, and location-specific time trends and location-specific time squared trends account for yield nonlinear growth due to technological progress. The squared term of radiation represents the non-linear effect of radiation on rice growth, as rice growth accelerates when radiation is suitable, but excessive radiation can have a negative impact on rice yield. By calculating the variance inflation factor, the potential presence of collinearity in the model is negligible (Supplementary  Table S1, Note S2). Figure 2 shows the trends of the climate indices in Northeast China from 1980 to 2015, weighted by the ricegrowing area in each county. The mean growing season temperature in Northeast China varies between 18 and 25 °C. In the past 35 years, there has been a significant warming trend, with the mean growing season temperature increasing by 0.31 °C/10 a. The minimum and maximum temperatures grow by 0.33 and 0.29 °C per decade, respectively. Fig. 3 demonstrates the spatial distributions of climate variables in Northeast China from 1980 to 2015. The average temperature trends in Northeast China ranged from − 0.4 to 0.6 °C/10 a. The central and northern areas of Northeast China demonstrated the fastest warming rates (0.36 °C/10 a). In contrast, the rates in the southern regions are much slower (0.26 °C/10 a). (Fig. 3c). The increase in minimum and maximum temperatures also shows an uneven spatial distribution, with the greatest increase in the central region ( Fig. 3a and b).

Temporal trend and spatial distribution of historical climate
Regarding the degree-day indices, there is an increase in GDDs, with a rate of 43.5°Cd/10 a, while the HDD trend is much slower, presenting an average increase of only 0.22°Cd per decade (Fig. 3f). For CDDs, a 4.55°Cd decrease per decade is observed. In terms of its spatial distribution, the GDD trend is lower in the east but higher in the west (Fig. 3f). The spatial distribution of HDDs also varies, with a decreasing trend in the northeast and an increasing trend in the southwest (Fig. 3e). The CDD trends are generally slower towards the north (Fig. 3d). It should be noted that HDDs do not trend significantly in most locations, which may result in spatial heterogeneity in HDDs.

Yield responses to climate
Regression analysis of historical data is used to relate past rice yields to climate. Table 1 includes climate variables (i.e., GDD, HDD, CDD, and RAD) as explanatory variables for f (Clim i,t ) to examine rice yield variabilities. According to the results (Table 1), significant positive effects were found on rice yields in GDD and RAD (p < 0.05), while HDD and CDD presented significant negative effects (p < 0.05). The effect of temperature on rice yield is quantified by the yield trend induced by the trend of temperature indices. During the rice-growing season, for each 10 degree-day increase in GDDs, rice yield increased by 0.28%. In contrast, CDDs decreased yields, with a 3.54% penalty for each 10 degreeday increase in CDDs. HDDs reduced yields by 3.03% for each 10 degree-day increase. In the results of the main text, we did not include the moisture index because the effect of moisture was insignificant in the regression model (Supplementary Table S3, Note S3) because 95% of rice in Northeast China is irrigated (Gan and Liu 2005;Wang et al. 2017).   Figure S1, Note S4).

Impact of historical climate trends on rice yields
Next, we projected yield changes based on the historical trends in individual and combined climate variables (Fig. 4). We set the reference period to the beginning of a 5-year period (e.g. , 1980-1984). There was no noticeable time trend in the yield changes induced by HDD (slope = − 0.015%/10a, p = 0.9). The effects of GDD and CDD were stronger than those of HDD. On average, the CDD trend resulted in a 5.4% yield increase (p < 0.01) over the past 35 years, while the GDD-induced yields increased by 4.6% (p < 0.01). HDD caused a few yield changes (− 0.054%) (Fig. 4). Consequently, with the historical climate trends in Northeast China, the combined trend in temperature extremes (HDD + CDD) was projected to improve yield by 5.4%; this increase was slightly greater than that imposed by GDD alone (3.9%). Regarding the spatial differences in the study area, the increasing trend of CDD in the northwest resulted in a higher increase in rice yield compared to the southeast, with an overall increasing trend. The trend in HDD resulted in a decrease in rice yield in the south and an increase in the north. The combined effect of the trend in HDD and CDD was consistent with the effect of the trend in CDD. In the entire study area, the increasing trend of GDD had a positive impact on rice yield, with a slightly higher increase in the north than in the south (Fig. 5).

Tipping points of warming for rice yields in Northeast China
To investigate the impacts on rice yields induced by continuous warming, we artificially increased the historical temperature by 1-6 °C and projected yield changes under each warming scenario. With higher temperatures, rice yields are projected to initially rise and then decline when the temperature is higher than a certain threshold (Fig. 6). We present the tipping point temperatures on the map based on the county-level data to facilitate comprehension of this spatial variation. In Fig. 6a, sensitivity to warming varies by region, and the tipping point temperature gradually increases from south to north and from west to the east (Fig. 6a). This threshold is the lowest in Liaoning (2.12 °C), while in Jilin and Heilongjiang, the values are 3.78 °C and 3.99 °C, respectively. The average tipping point temperature is approximately 3.36 °C in the study region (Fig. 6b). Figure 7 exhibits the yield impacts associated with individual climate variables under various warming scenarios. We found that the positive effect of GDD grows steadily with warming, while the effect of CDD becomes weaker. HDD shows an increasingly greater contribution to rice yields as warming increases. For example, in Liaoning, approximately 1.8% of the yield reduction is attributed to HDD under the 1 °C warming scenario, while the contribution of HDD increases to 31.4% under the 6 °C warming scenario. Such a greater harmful impact of HDD offsets the beneficial effects of GDD and CDD, and the contribution of HDD becomes dominant with higher warming scenarios.

Discussion
We provide an empirical assessment to benchmark the effects of diverse climatic variables on rice yield. Our results are based on long-term climate and rice yield observations Our results suggest that historical warming is beneficial for rice yields in Northeast China. Warming causes greater HDDs, but such harmful effects are quite limited. Comparably, the historical increase in GDDs and decrease in CDDs pose much stronger positive effects. Increase in average growing season temperature improved heat resources, increases the photosynthetic rate and biomass accumulation time of the crop, and provides the possibility to adopt late maturing varieties. As a result, a significant increase in GDD contributed to the increase in rice yield in Northeastern China. Besides, during the reproductive growth stage, temperature extremes can damage rice reproductive organs, leading to spikelet sterility, empty grains, and reduced yields (Hussain et al. 2019). Climate warming has significantly reduced the occurrence of cold damage, and the reduction of CDD is the main reason for the increase in rice production in Northeast China. During the flowering period, exposure to high temperatures for just 1 h can harm normal rice development. High temperatures during the flowering stage inhibit pollen grain expansion and rupture, causing poor anther dehiscence and low pollen production, leading to yield reduction (Matsui et al. 1997(Matsui et al. , 2001Satake & Yoshida 1978). Overall, the combined beneficial effects of GDD and CDD play a more dominant role than HDDs on yields in Northeast China during 1980-2015. However, such beneficial effects were projected to decrease with future warming in Northeast China. In contrast, the contribution of HDDs was projected to become more dominant with warming. When the warming degree was greater than the threshold values predicted by the model, the contribution of HDDs rapidly increased, and yields began to decline. This result suggests that the harmful impacts of high temperature should not be ignored as future warming. We compared our model results with previous studies to test the robustness of our results, finding that the estimated yield increases based on our county-level data fall within the range of earlier assessments derived from using process-based modeling approaches and other data sources (Sun and Huang 2011;Tao et al. 2013a, b;Wang et al. 2014;Yu et al. 2021). Secondly, when the above statistical analysis was repeated using climate data from different warming scenarios, the responses of rice yield to 1-6 °C climate warming derived here were close to the estimates based on crop model as well as climate model projection data (Fei et al. 2020;Luo et al. 2022;Saud et al. 2022;Yu et al. 2021;Zhang et al. 2021a, b). This reflects the robustness of our panel model configuration and output results. This suggests that our study approach can also be applied to other crop analyses by setting up various panel model configurations. Finally, we considered extreme climate indices at vegetative and reproductive stages (Supplementary Table S6), and found that yield changes due to 1 °C warming is quite similar derived from the model considering the different growth stages and the original model form in the main text (Supplementary Figure S3).
Our estimates differ from those reported in earlier studies (Liu et al. 2013;Tao et al. 2013a). For example, Liu et al. (2013) found that the yields associated with coldtemperature injuries fluctuated over time and did not result in any clear trend in yield impacts over time. This discrepancy could be interpreted by the uses of different temperature indices to reflect extremely low temperatures. Liu et al.'s study quantified the effects of extremely low temperature based on the number of days with temperatures below the lowest temperature threshold and did not account for the effects of cumulative temperature on rice chilling (Supplementary Note S1; Table S2). We reestimated the effects of low-temperature extremes using the same indices (Supplementary Table S4) and found that the index that was based on the number of days with temperature below the lowest temperature threshold also returned a negative regression coefficient (Supplementary Table S4), but the model projection did not show an obvious time trend over 1980-2015. This suggests that the cumulative temperature calculation used when considering extremely low temperature does have an effect and projects different results compared with those obtained by counting only low-temperature days, as performed in earlier studies (Liu et al. 2013).
Based on our results, two methods can be suggested to reduce the climate risks faced by rice. First, climate-warming trends should be mitigated to 3 °C in Northeast China. At such a warming rate, the general effects of climate change on rice would be beneficial. Climate models have projected that temperature would increase by 2 °C in Northeast China under the RCP2.6 scenario (Sun et al. 2019) and that increasing temperature will not exceed the threshold temperature by the mid-to late twenty-first century. However, under the RCP4.5 and RCP8.5 scenarios, climate models have projected that temperature will increase by 4 °C under RCP4.5 and 6 °C under RCP8.5 by approximately 2050 (Chong-Hai and Ying 2012;Zhang 2012;Chu et al. 2017;Sun et al. 2019;Zhang et al. 2021a, b), thus exceeding the threshold temperature inferred from our study. To control warming, we should constrain greenhouse gas emissions. This may be achieved by the development of clean energy or through the implementation of reasonable policies. The second method involves reducing the sensitivity of rice to heat, as the contribution of extreme heat temperature was low under historical climate conditions but will become more dominant under future warming scenarios. This may be achieved by breeding (Matsui et al. 2001;Lawas et al. 2019;Malumpong et al. 2020) to introduce more heat-resistant cultivars or by adjusting the sowing date (Pal et al. 2017;Bai et al. 2019) to reduce the exposure of rice to temperature extremes. Future investigations regarding the priority of these adaptation measures are needed to enable efficient adaptations to cope with future climate risks.
Several limitations and uncertainties of our study need to be addressed. First, similar to other statistical models, statistical models fix time windows and cannot consider changes in sowing time and rice phenology over time. However, an earlier study showed that the rice heading dates and the growth period length remained nearly constant over time due to the widespread late-maturity rice cultivars planted (Tao et al. 2013b). We conducted sensitivity analyses for different calendar period windows, and the results showed that shifting the time window had no significant impact on the output of the statistical model (Supplementary Table S5). Second, statistical models assume that the climate sensitivity (e.g., regression coefficients in the model) is derived based on the historical climate applied to warmer scenarios. It should be stressed that nonclimatic factors, e.g., the introduction of heat-tolerant cultivars (Lobell et al. 2012;Tayade et al. 2018) or changes in sowing densities (Lobell et al. 2020), can potentially alter the regression coefficient. Therefore, our results presented in this study should be viewed as a model projection based on historical yield responses and existing agronomic technologies; they indicate that the future will demand improvements in agronomic technologies to achieve sustainable agricultural development in Northeast China.

Conclusion
We employed a fixed-effect panel model to examine the effects of climate change on rice yields in Northeast China using recorded county-level rice yield statistics. Historically, climate change has been favorable for rice in the study region, mainly due to the positive effects of CDDs (i.e., low-temperature extremes) and GDDs outweighing the negative effects of HDDs (i.e., hightemperature extremes). The contribution of CDDs to yield changes was 1.15 times greater than that of GDDs and 100 times greater than that of HDDs under current climate conditions. Therefore, the decline in CDDs is currently the most significant climate variable driving rice yields over time in Northeast China. However, with future climate change, continuous warming would result in fewer CDDs and greater HDDs. The contribution of HDDs would thus become more dominant over the contributions of the other temperature indices. Such a shift in climate influences was used to determine tipping point temperatures; the yield trends would change from increasing to decreasing with 3.36 °C of warming in our study region. Our results suggest that despite the relatively weak impact of HDDs under historical climate conditions, the harmful impacts of HDDs will become more dominant as future warming progresses. Breeders and farmers will benefit from our research by adopting effective approaches to address the effects of climate change in rice production activities in northeastern China. Slowing down climate change rate and speeding up the variety adaptation to heat stress conditions are necessary.