Supplementary Material
Drought stress impacts of climate change on rainfed rice in South Asia
Tao Li*, Olivyn Angeles, Ando Radanielson,
Manuel Marcaida III, Emmali Manalo
(International Rice Research Institute, Los Baños, Philippines)
*: Corresponding author
Name: Tao Li, PhD, Scientist on crop modeling
Contact information:
International Rice Research Institute
Los Baños, Philippines
Tel: +63-2-580-5600, ext. 2732
Fax: +63-2-580-5699
Email:
Web:
This Supplementary Materials file includes:
Supplementary Text
Supplementary Figures S1 to S4
Supplementary Table S1
References
Supplementary Text
- The performance of ORYZA2000 for multiple rice genotypes and environments
Table S1 and Figure S2 illustrate the evaluation results on the performance of ORYZA2000 in estimating rice growth and grain yield. More than 90% of the simulations fell within the confidence levels of actual measurements. ORYZA2000 simulations on both total above-ground biomass (TAGB) and panicle biomass (as grain yield, GY) were highly correlated to field measurements with coefficient of determination (R2) higher than 0.9 while the regression slope and intercept were close to 1.0 and 0.0 (p<0.0001), respectively. The differences between simulated and measured TAGB and GY were statistically insignificant (Student T-test Pt>0.10). ORYZA2000 efficiently represented the field measurements coming from different soil-climatic conditions and agronomic practices. It was also reliable to estimate the irrigated and rainfed rice grain yield for evaluating the impacts of drought stress caused by climate change on rainfed rice.
It is worthy to note that the model has ability to simulate the rice yields within the range of uncertainties of the observed yield during its calibration and its validation. Variation in simulation from crop models and climate models was reduced by increasing the number of climate models and crop models in an ensemble (Challinor et al. 2009, Li et al., 2015). However, the reduction in uncertainty due to ensemble size could be from several sources which importance is not certain (Challinor et al. 2009, Smith and Stern 2011). It would be interesting to see whether the similar trends can be achieved if using an ensemble model of simulation. That will suggest that the simulation outputs will be generic and less sensitive to the crop model hypothesis of representation of the rice cropping systems. From an early study, the uncertainties of ORYZA2000 predictions on rice yield were consistent with other rice crop models and model ensemble across environments (Li et al., 2015). Considering that this study focused more on the relative changes of rice yields caused by drought stress associated with climate change, ORYZA2000 was further confirmed to provide reliable predictions of rice yields for the goals of the current study.
- The determination of best sowing dates by ORYZA2000
The best sowing dates were the sowing dates used to achieve the highest attainable grain yield in given weather and soil conditions for specific crop management and cultivar. In this study, ORYZA2000 was used to estimate the attainable grain yields under rainfed condition for 24 sowing dates per year starting on 1 January at 15-day interval. To evaluate the capability of ORYZA2000 on the determination of the best sowing seasons of rainfed rice, additional simulations were conducted using historical weather data from 1980 to 2008. The best sowing dates were determined based on the attainable yields and variations among years (Fig. S3, see the methods of determination of the best sowing dates in section 2.3.3 in main text). The best sowing dates determined by ORYZA2000 were consistent with the reported sowing dates forrainfed rice in eastern India.
3. Drought stress indicators as impact of climate change on rainfed rice
Using ORYZA2000, two sets of data were obtained to represent rice yields under potential and rainfed conditions. No consistent relationships existed between potential and rainfed yields across spatial and temporal scales due to varying soil-climatic conditions (Fig. S5).
The examples of randomly selected rice cells in Figure S5 illustratethat the gaps between potential and rainfed yield varied across years and sowing dates. Considering the complex annual and inter-annual (i.e. temporal) variations in potential and rainfed rice yield, the standard difference (SDpa) calculated by Equation 3 (main text) was suitable for a statistical indicator of a yield gap caused by drought stress across a temporal scale. It effectively represented the degree of drought stress when different sowing dates were considered in specific rice areas.
However, the SDpa indicating the absolute gaps between potential and rainfed yields were meaningless in representing the spatial variation on effects of drought stress on rice yield. The equal gaps might not imply the same drought impacts due to the different potential yield (Fig. S5). For instance, with the same yield gap, the drought impact is stronger under low potential yield than under high potential yield. The variability in potential yield was the result of the different climatic conditions such as radiation and air temperature, and interactions of the crop with these climatic conditions. Therefore, the ratio of absolute yield gap to potential yield was more relevant to represent the relative yield reduction caused by drought stress as calculated by equation 4 (main text). This ratio, named RDI(Eqs. 3 and 4 in main text), not only represented the effects of drought stress on rice yield across spatial scales but also imbedded the temporal effects over years (Fig. 1 in main text).
The probability of crop failure over year-round sowing dateswas represented by CFF (Fig. 2 in main text). Higher value meant greater risk for rainfed rice in arice cell. CFF was determined by the frequency of severe drought stress that caused crop failure in available rice-growing seasons over years (Figs. 2 in main textand S5). For areas with low CFF values, a slight shift of sowing date would create a chance for higher grain yield (Fig. S5c). CFF in this study referred to an agronomic crop failure, meaning no grain yield (i.e. grain yield is 0 t/ha), which differed from commercial crop failure (i.e. the market value of harvested grain could not compensate for the cost for production). Although the latter is a more realistic indicator, it was difficult to use in this study because of the varying costs of fertilizer, labor, fuel, pesticides, and other production inputs.
Unlike RDI and CFF, MRGY and CVGYwere two specific indicators ofdrought impacts in the best sowing seasons of rainfed rice over years. In term of MRGY, the simulated maximum rainfed yield was higher than 3t/ha in more than 70% of the riceareas in South Asia. The highest yield estimated was 9.4t/ha under climate change (Figs. 3in main text and S5). In comparison with earlier results simulated by CERES-RICE, the experimental and farmer-site measurements in the early 2000s (Aggarwal et al. 2008; Pathak et al. 2009) and the highest recorded experimental yield of 7 to 7.2 t/ha under upland conditions (De Datta and Beachell 1972; Kawano et al. 1972; Schmitter et al. 2011), the simulated MRGY was slightly higher but was consistent with field measurements due totimely and appropriate fertilizer application (Fig. 3 in main text). It could be argued that the values of MRGY did not represent annual rainfed rice yield because multiple rainfed rice seasons might be availablein a rice cell. The examples in Figure S4 demonstrated that available rainfed rice seasons per year were possibly more than one in some areas, although they were not the seasons with the most stable and/or highest yield. Therefore, annual rainfed yield might be higher than what was presented in Figure 3 (main text). This situation was true for some areas at the southern end and in coastal areas of South Asia. About 5% of the rice-growing areas could have a total yield from two rainfed rice seasons of more than 1.5 times that of the one best rainfed season. However, rainfed rice yield in these areas with multiple seasons,could suffer higher inter-annual variation, implying a higher risk of large yield loss. In general, the highest yield with the lowest yield variation was the best indicator of local rice productivity, which could be identified with the indicatorsMRGY and CVGY.
Among the four indicators, RDIwas the most independent of crop characteristics. It was a relative expression of water-limited production to potential production. RDIwas mainly subject to local climatic-soil conditions. The other three indicators (MRGY, CVGY, and CFF)were dependent on crop characteristics and strongly represent the interactions of the crop and its environment. Across the temporal scale in a rice cell, the RDI and CFF represented the impacts of drought stress on year-round rice production while CVGY and MRGY represented the impacts in the best sowing season of rainfed rice. On a spatial scale, the variations in these four indicators resulted from the spatial variation of drought stress caused by climate change. Because we used the same virtual rice cultivar for all years and all rice cells, spatial variations of these indicators mainly resulted from the variations in soil-climatic environments and also revealed the variations in interactions of the rice crop and soil-climatic environments under rainfed conditions.
- Changes of yield and the best sowing season of rainfed rice
In comparing the MRGY in the far to near future climate conditions, the large positive values of ΔMRGY occurred in most rice areas in South Asia indicated the significant increase of rainfed rice yield. Climate change could be beneficial to rainfed rice production in South Asia if the rainfed rice season would be adjusted accordingly (Figs. 3 and 4 in main text, and S4), implying that CO2 fertilization effects overcome the reduction caused by the increase in air temperature (Baker et al. 1992; Matsui et al. 1997; Ainsworth and Ort 2010). This result was inconsistent with early evaluations as that climate change may result in slight (~6%, Soora et al., 2013) or significant (15-17%, Aggarwal and Mall, 2002) decreases of rainfed rice in India, or 'no effect trends’ in rice yield (Knox et al., 2012).
The positive effect of climate change on rainfed rice, hypothetically, applied when the best sowing season of rainfed rice was utilized, which might change significantly in most areas in South Asia (Fig. 4). The assessment in this study was based on analysis of modeling results. The current farming practices consistentin the study area, and the suggested change of sowing seasonmight be different with the current sowing seasons of local farmers. In addition, assessment results would not be the sameif varieties with different drought tolerant traits or crop growth duration were applied in different regions (Li et al. 2013). This study would like to demonstrate that the adjustment on the sowing seasons of rainfed rice would be essential as the progress of climate change. Reviews in climate change effect in South Asia also recommended that the shiftinplanting dates would be the adaptive strategy at individual levels, and it would be more helpfulby accompaning with adoption of improved varieties and improved rice cultivation technologies (Knox et al., 2012). Simulations study applied for Soya crop has also suggested that shift in planting date wouldprovide adaptation but could be limited by low temperature as observed for the Gangetic plain (Mall et al., 2004). The variability of the best sowing season (BSS)would be mainly attributed to the interaction amongenvironmental factors.
Supplementary Figures:
Figure S1. Rainfall pattern of rice cultivated area in South Asia (colored cells)under SRES A1B (a andd) and A2 (b and c). Each grid cell represents the 5×5 geographic arc-minute ground area. Panels a and b present the average seasonal rainfall in the best rainfed rice season under these two SRES scenarios in the near future(2015- 2044), while panels d and e show the relative changes in seasonal rainfall in the best rainfed rice season from the near tofar future (2045-2074). Panel c is the histogram distribution of seasonal rainfalland panelf is the histogram distribution of the relative changes in rainfall for both SRES scenarios.
Figure S2. The performance of ORYZA2000 in estimating total above-ground biomass and grain yield (t/ha) using validation datasets from field experiments conducted at 32 locations in Asia for a total of 91 rice genotypes (Castañeda et al. 2002, Sudhir-Yadav et al. 2011, and Li et al. 2013). Stars on the map represent the locations of field experiments used in the analysis. The graphs show the measured and simulated data pairs for total above-ground biomass (a) and grain yield (b). Solid lines are the 1:1 relationship while the broken lines mark the range defined by the coefficient of variation of measurements.
Figure S3. The variation of simulated attainable (or maximum)rainfed grain yield (AGY) at different sowing dates and the best sowing dates determined by ORYZA2000 using historical weather data for the period from 1980 to 2008. The thick red arrows are the sowing dates reported in literatures,the black circles indicate the regions considered, and the numbers in red refers to references for this study(1. Mahmoodet al., 2004;2. Mandal and Shenoi, 1986; 3. Misraet al., 1986).
Figure S4: Spatial pattern of the coefficient of variation of maximum rainfed rice grain yield (CVGY) in near future (2015-2044) and the change in CVGY (blue triangles) from the near to far future (2045-2074) in SRES scenarios A1B (a) and A2 (b). Panel cis a histogram distributionof the difference in CVGY between two A1B and A2 duringthe period from 2015 to 2074.
Figure S5.Simulated grain yield (t/ha) under potential (blue) and water-limited (maximum rainfed yield, red) conditions using 24 different sowing dates over a 10-year period. The code of the sowing period represents the 24 sowing dates starting from 1 January at 15-day intervals.
Supplementary Table:
Table S1. Validation results of ORYZ2000 performance in simulating total above-ground biomass (TAGB; t/ha) and grain yield (as panicle biomass, GY; t/ha). Data pairs of measured and simulated values were composed from field experiments conducted in 32 locations of Asia for a total of 91 rice genotypes.
Item / Datapairs / Measurement / Simulation / Regression / Pt / RMSE / RMSEn / ME
Mean / sd / Mean / sd / b / a / R2
TAGB / 639 / 5.55 / 4.70 / 5.74 / 4.85 / 0.99 / 0.22 / 0.93 / 0.49 / 1.327 / 0.239 / 0.92
GY / 315 / 3.16 / 2.46 / 3.08 / 2.54 / 0.99 / -0.04 / 0.91 / 0.71 / 0.762 / 0.241 / 0.90
Note: sd is standard deviation (t/ha); a, b, and R2are the intercept, slope, and coefficient of determination; Pt is the probability of student T-test to accept that the simulation is the same as the measurement; RMSE (t/ha) is the root mean square error between simulated and measured values, while RMSEn is RMSE normalized by measured mean; and MEis the model efficiency in representing the measurements.
References
Aggarwal PK, Hebbar KB, Venugopalan MV, Rani S, Bala A, Biswal A, Wani SP (2008)Quantification of Yield Gaps in Rainfed Rice, Wheat, Cotton and Mustard in India. Global Theme on Agroecosystems Report No. 43. ICRISAT, Patancheru, Andhra Pradesh, India.
Aggarwal PK, Mall RK (2002) Climate change and rice yields in diverse agro-environments of India. II. Effect of uncertainties in scenarios and crop models on impact assessment. Climatic Change 52:331–343.
Ainsworth EA, Ort DR (2010) How do we improve crop production in a warming world? Plant Physiol 154:526–530.
Baker JT, Allen LH, Boote KJ (1992) Growth and yield responses of rice to carbon dioxide concentration. J AgricSci 115:313-320.
Castañeda AR, Bouman BAM, Peng S, Visperas RM (2002) The potential of aerobic rice to reduce water use in water-scarce irrigated lowlands in the tropics. In: Bouman BAM, Hengsdijk H, Hardy B, Bindraban PS, Tuong TP, Ladha JK (Eds.), Water-wise Rice Production. Proceedings of the International Workshop on Water-wise Rice Production, 8–11 April, 2002, Los Baños, Philippines. International Rice Research Institute, Los Baños, Philippines, pp. 165–176.
Challinor A, Osborne TM, Morse A,Shaffrey LC, Wheeler TR, Weller H,Vidale PL (2009) Methods and resources for climate impacts research: achieving synergy Bull.AmMeteorolSoc 90:836–48.
De Datta SK, Beachell HM (1972) Varietal response to some factors affecting production of upland rice. IRRI, Los Baños, Philippines.pp. 685–700.
Kawano K, Sanchez RA, Nurena MA, Velez IR (1972)Upland rice in the Peruvian jungle. In: Rice breeding. International Rice Research Institute, Los Baños, Laguna. pp. 637–643.
Knox J, Hess T, Daccache A, Wheeler T (2012) Climatechangeimpactsoncropproductivity in Africa and South Asia. Environ Res Lett 7:034032.
Li T, Hasegawa T, Yin X, Zhu Y, Boot K, et al. (2015) Uncertainties in predicting rice yield by current crop models under a wide range of climatic conditions. Global ChangeBiol 21:1328–1341.
Li T, Raman AK, Marcaida III M, Kumar A, Angeles O, Radanielson M (2013) Simulation of genotype performances across a larger number of environments for rice breeding using ORYZA2000. Field Crops Res 149:312-321.
Mall RK, Lal M,Bhatia VS,Rathore LS, Singh R (2004) Mitigating climate change impact on soybean productivity in India: a simulation study. AgricForMeteorol 121:113-125.
Mahmood R, Legates DR, Meo M (2004) The role of soil water availability in potential rainfed rice productivity in Banglandesh: application of the CERES-Rice Model. Appl Geog 24:139–159.
Mandal B, Shenoi PV (1986) Rainfed rice in West Bengal. In: Maclean J. (ed) Progress in rainfed lowland rice. International Rice Research Institute, Los Baños, Philippines. Pp. 53–56.
Matsui T, Namuco OS, Ziska LH, Horie T (1997) Effects of high temperature and CO2 concentration on spikelet sterility in Indica rice. Field Crops Res 51:213–221.
Misra B, Mukhopadhyay SK, Flinn JC (1986) Production constraints of rainfed lowland rice in eastern India. In: Maclean J. (ed) Progress in rainfed lowland rice. International Rice Research Institute, Los Baños, Philippines. pp. 25–52.
Pathak H, Ladha JK, Yadvinder-Singh, et al (2009) Resource-conserving technologies in the rice-wheat system of South Asia: field evaluation and simulation analysis.In: Integrated crop and resource management in the rice-wheat system of South Asia. Ladha JK, Yadvinder-Singh, Erenstein O, Hardy B.IRRI, Los Baños, Philippines.
Schmitter P, Dercon G, Hilger T, Hertel M, Treffner J, Lam N (2011) Linking spatio-temporal variation of crop response with sediment deposition along paddy rice terraces. Agric EcosystEnviron 140:34–45.
Smith L, Stern N (2011) Uncertainty in science and its role in climate policy. Phil Trans R Soc A 369:4818–41.
Soora NK, Aggarwal PK, SaxenaR, Rani S, Jain S,Chauhan N (2013) An assessment of regional vulnerability of rice to climate change in India. Climatic Change 118:683-699.
Sudhir-Yadav, Li T, Humphreys E, Gill G, Kukal SS (2011) Evaluation and application of ORYZA2000 for irrigation scheduling of puddled transplanted rice in north west India. Field Crops Res 122:104–117.
1