Assessing the impacts of extreme agricultural droughts in China under climate and socioeconomic changes

Yu, Chaoqing et al. Article — Published Version Assessing the impacts of extreme agricultural droughts in China under climate and socioeconomic changes Earth's Future Provided in Cooperation with: Leibniz Institute of Agricultural Development in Transition Economies (IAMO), Halle (Saale) Suggested Citation: Yu, Chaoqing et al. (2018) : Assessing the impacts of extreme agricultural droughts in China under climate and socioeconomic changes, Earth's Future, ISSN 2328-4277, Wiley, Hoboken, NJ, Vol. 6 5, pp. 689-703, https://doi.org/10.1002/2017EF000768 , https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2017EF000768 This Version is available at: https://hdl.handle.net/10419/179965 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ Assessing the Impacts of Extreme Agricultural Droughts in China Under Climate and Socioeconomic Changes Chaoqing Yu 1 , Xiao Huang 1 , Han Chen 1 , Guorui Huang 1 , Shaoqiang Ni 1 , Jonathon S. Wright 1 , Jim Hall 2 , Philippe Ciais 3 , Jie Zhang 1 , Yuchen Xiao 1 , Zhanli Sun 4 , Xuhui Wang 3 , and Le Yu 1 1 Ministry of Education Key Laboratory for Earth System Modeling, Department of Earth System Science, Tsinghua University, Beijing, China, 2 Environmental Change Institute, Oxford University, Oxford, UK, 3 Laboratoire des Sciences du Climatet de l’Environnement, CEA-CNRS-UVSQ, Gif sur Yvette, France, 4 Leibniz Institute of Agricultural Development in Transition Economies (IAMO), Halle (Saale), Germany Abstract Agricultural food production in China is deeply vulnerable to extreme droughts. Although there are many studies to evaluate this issue from different aspects, comprehensive assessments with full consideration of climate change, crop rotations, irrigation effects, and socioeconomic factors in broad scales have not been well addressed. Considering both the probability of drought occurrence and the consequential yield losses, here we propose an integrated approach for assessing past and future agricultural drought risks that relies on multimodel ensemble simulations calibrated for rice, maize, and wheat (RMW) in China. Our results show that irrigation has reduced drought-related yield losses by 31 ± 2%; the largest reductions in food production were primarily attributable to socioeconomic factors rather than droughts during 1955–2014. Unsustainable water management, especially groundwater management, could potentially cause disastrous consequences in both food production and water supply in extreme events. Our simulations project a rise of 2.5~3.3% in average rice, maize, and wheat productivity before 2050 but decrease thereafter if climate warming continues. The frequency of extreme agricultural droughts in China is projected to increase under all examined Representative Concentration Pathway (RCP). A current 100-year drought is projected to occur once every 30 years under RCP 2.6, once every 13 years under RCP 4.5, and once every 5 years under RCP 8.5. This increased occurrence of severe droughts would double the rate of drought-induced yield losses in the largest warming scenario. Policies for future food security should prioritize sustainable intensification and conservation of groundwater, as well as geographically balanced water resource and food production. 1. Introduction Drought and extreme heat are the largest climate-related threats to global agricultural production (Lesk et al., 2015). These threats are particularly acute in China, where agricultural food production is deeply vulnerable to extreme droughts. At least 15 catastrophic drought events lasting longer than 3 years and affecting multiple provinces have occurred during the past 1000 years, along with numerous shorter-lived droughts lasting a few months at a time (Zhang, 2005). The most severe droughts have caused widespread famine and loss of life (Edgerton-Tarpley, 2008; Zhang, 2005; Zhang & Liang, 2010). Despite the recent increase in the severity of meteorological droughts (Piao et al., 2010), annual food production in China has successfully transitioned from near famine-level through basic-level to consistent high-level supply during the past 60 years (Figure 1). These changes occurred as the national economy progressed from food rationing (1955–1993) through a decade-long free market stage (1994–2003) to subsidized farming (2004 up to present). In the past six decades, the foremost adaptation policy in China for reducing drought-induced yield loss has been the development of irrigation infrastructure. Irrigated area has increased by more than 400% from 1950 (~15 Mha) to 2013 (~63.5 Mha) (NBS of China, 2010). Spatial imbalances between irrigation and economic development have resulted in a northward shift in the center of Chinese food production, particularly after the implementation of the Economic Reform and Opening-up Policy in 1978 (Figure S1a in the supporting information), which replaced the collective farming system with a household responsibility system. Planting area for grain foods in the south (Figure S1b) has declined by 22.4% between 1978 and 2014, owing primarily to the intensive labor inputs relative to economic opportunities associated with urbanization. The rapid expansion of irrigation area in northern China (Figure S1c) has also played a key role in the northward shift of crop production by improving productivity and reducing risk in drought-prone areas. Expansion of YU ET AL. 689 Earth's Future RESEARCH ARTICLE 10.1002/2017EF000768 Key Points: •The development of irrigation in China has alleviated negative drought impacts on food productivity, the presence of irrigation systems explaining 31 ± 2% of avoided yield loss •Frequency and intensity of extreme droughts are projected to increase significantly under future climate change, leading to double the drought-induced yield losses in the largest warming scenarios •The sharpest drops in grain production during the past 60 years have been driven primarily by socioeconomic disturbances rather than by large droughts Supporting Information: •Supporting Information S1 Correspondence to: C. Yu and P. Ciais, [email protected], [email protected]; [email protected]l.fr Citation: Yu,C.,Huang,X.,Chen,H.,Huang,G.,Ni,S., Wright,J.S.,etal.(2018).Assessingthe impacts of extreme agricultural droughts in China under climate and socioeconomic changes. Earth’s Future,6,689–703. https://doi.org/10.1002/2017EF000768 Received 21 NOV 2017 Accepted 11 APR 2018 Accepted article online 25 APR 2018 Published online 8 MAY 2018 ©2018. The Authors. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. irrigation infrastructure in water-scarce northern China has increasingly been accomplished via the consumption of nonrenewable groundwater resources (Yu et al., 2011). Irrigation expansion slowed in the south after the 1960s because the area of flat and irrigable lands is limited in this part of China. Government policies for exploiting groundwater in northern China commenced in the late 1960s (Qian, 2009). The number of irrigation wells has subsequently increased from 0.19 million in 1965 to 5.01 million in 2010 (Ministry of water Resources, China, 2011), with more than 90% of these wells located in the north. Groundwater has become the primary source of water for irrigation in many northern provinces (Figure S2). Available records indicate that total groundwater storage in the plains in northern China declined by 91 billion cubic meters between 1996 and 2011 (Figure S1c) (Ministry of water Resources, China, 2014). Groundwater depletion threatens future water availability and food security in the face of a changing climate and must be considered in quantifying agricultural drought impacts. Quantitative evaluation of meteorological drought impacts on past and future food supplies provides important context of policy making for agricultural sustainability. It requires a comprehensive understanding of the relations among the spatiotemporal variations of climate, the complex cropping systems, the irrigation effects, and the availability of water resources. There are multiple approaches being reported to address such relations. For example, survey methodology is helpful to collect first-hand information of how drought has affected crop yields and how farmers adapted to drought (Chen et al., 2014; Huang et al., 2015; Wang et al., 2009; Yan et al., 2016). Empirical models have been applied to analyze the statistical relations among meteorological factors and yields (Huang et al., 2013; Qin et al., 2014; Zhang & Huang, 2012; Zhang et al., 2017). More research efforts have focused on evaluating drought risks based on meteorological (Das et al., Figure 1. Sixty-year changes in food production and food prices in China. (a) Historical evolution of food production in China since 1950 under an assumption of national self-sufficiency. Data include annual food production per person (AFPP, kcal per person per day, left axis), total staple food production (10 8 tons, right axis), and population growth and mortality rates (%, right axis). (b) Available historical market prices for rice and the average prices for rice, maize, and wheat (RMW) in Chinese yuan (left axis), as well as profit to cost ratios for cereal crop farming (calculated as net profit divided by cost, right axis). The threshold of annual food production per person for famine-level supply is defined as 1,800 kcal per person per day because the mortality rate has historically increased sharply at values below this level. Basic-level supply is defined as 2,250 kcal per person per day. China did not achieve this level of supply until the institution of the Open Door Policy in 1978. High-level supply, defined as 3,000 kcal per person per day, was not achieved until recent years. The shaded regions along the upper edge of Figure 1b mark the food rationing (1955–1993) and free market (1993–2003) stages. Subsidized farming has been in effect from 2004 through the present, with RMW prices protected and subsidies provided to farming households. Market prices and government subsidies are both included in the net profit ratio shown in Figure 1b. Costs include material, machinery, and labor. 10.1002/2017EF000768 Earth's Future YU ET AL. 690 2016; Li et al., 2016; Naresh Kumar et al., 2009), hydrological (Abebe & Foerch, 2008; Shukla & Wood, 2008), or composite (Ellis et al., 2010; Huang et al., 2011; Palmer, 1965) drought indices (Zargar et al., 2011). Piao et al. (2010) used precipitation and runoff data together with the Palmer Drought Severity Index (PDSI) to show that north and northeast China have experienced significant increases in hydrometeorological drought since the 1960s (Piao et al., 2010). They also noted that it is difficult to produce a clear assessment of the impacts of drought on Chinese agriculture and food security under climate change. Such difficulty embodies the knowledge gap between the physical characteristics of extreme hydrometeorological events and their actual and potential impacts as recently emphasized by the World Meteorological Organization (2015). Realistic crop growth models (Di Paola et al., 2016; Kroes et al., 2000; Mearns et al., 1999; Morgan et al., 1980; Steduto et al., 2009; van Diepen et al., 1989) are arguably one of the best tools to quantify the influences of short- and long-term variations in weather conditions and human management on soil moisture availability, crop growth, and harvest yields (Jin et al., 2016; Niyogi et al., 2015; Singh et al., 2013) at regional (Yu et al., 2014) and global (Asseng et al., 2015) scales. In China, for example, the Environmental Policy Integrated Climate model was applied to evaluate the impacts of drought on yields of cereal crops in China (Jia et al., 2012; Li et al., 2017). The Crop Environment Resource Synthesis model was used to assess the effects of future climate change on crop yields (Tao & Zhang, 2011; Ye et al., 2013). The AquaCrop model was utilized to estimate the effects of irrigation on winter wheat yields on the Loess Plateau (Wang et al., 2013). And the Decision Support System for Agrotechnology Transfer model was used to simulate drought impacts on wheat and maize yields in North China Plain (Hu et al., 2014). In spite of such progress, the following issues need to be improved for enhancing our understanding of agricultural drought risks in China. First, a recent study by Martre et al. (2015) reveals that multimodel ensemble mean (or median) generates more accurate results than using any single crop model due to compensation of the limitations in individual models’structure and parameterization (Martre et al., 2015). Second, the complex cropping systems have not been well considered when modeling the nationwide food production. Third, the contribution of irrigation, especially groundwater irrigation, to the food productivity has not yet been well understood. Fourth, the potential consequences of extreme droughts in light of yield losses and water availability are not well quantified. Finally, none of the reported research has truly built the linkage between drought impacts and a clearly defined food security levels for the country (e.g., as shown in Figure 1). We aim to present a comprehensive analysis of agricultural drought impacts in China, integrating climate change, water resource, and socioeconomic factors using multiple models. The multimodel ensemble simulations performed in this study include complex cropping system and provide long historical series for probability calculation. Based on a supercomputer system, the research objectives are to provide quantitative evaluation of (1) the current agricultural drought risk of the major grain crops, (2) the contributions of irrigation to the reduction of agricultural droughts, (3) the potential drought risks under extreme droughts in the scenario of groundwater depletion, (4) the impacts of future climate change on agricultural drought risk, and (5) other social economic effects (e.g., market and policies) on major reductions of grain production, including rice, maize, and wheat. 2. Methods and Material 2.1. Agricultural Drought Risk Assessment Drought impact on crop yields represents the combined effects of reduced precipitation, increased temperature, increased solar radiation, decreased soil moisture, and human management in an event. It is also related to the frequency of drought events when evaluating the long-term impacts on food production in a specific place because the concept of risk consists of damage (D) and probability (p). Agricultural drought risk in a given place therefore is defined as the following equation: R¼∫1 0DpðÞdp(1) where Rrepresents risk. Drepresents drought-induced yield loss in a given drought event relative to a target yield. And prepresents probability of this drought event. Rcan be understood as integral of annual yield losses caused by all possible drought events in this place. Dis a combined result from the effects of climate, soil, technology, irrigation, fertilization, and other management on crop growth. Choosing the maximum yield or the average yield as the target yield does not change the probability distribution when evaluating 10.1002/2017EF000768 Earth's Future YU ET AL. 691 drought risk based on a consistent input data set. In the current research, we use the achievable maximum yield from history to define D. 2.2. Drought Frequency Analysis We use 60-year historical climate data to simulate daily crop growth and yield formation and derive time series of drought-induced yield losses relative to local maximum yields (Yu et al., 2014). We have tested a variety of distribution types for goodness of fit tests using the χ-square statistic, including the Weibull, Pearson Type III, Generalized Logistic, Gumbel, Generalized Additive Model, and Generalized Extreme Value (GEV) distributions, and find that the GEV is the best distribution for drought frequency analysis. The historical time series yield loss data are then used to generate the parameters of GEV distributions for each county, each province, and the country as a whole. Damage-probability curves are then generated for China and its subregions. An identical set of GEV parameters is also used for drought frequency analysis in each location for future climate change scenarios. 2.3. Crop Model Setting We evaluate drought impacts on major cereal crops in China based on yield loss and probability as simulated by multiple crop models at the county level. We build agricultural databases containing the parameters required by these models to simulate the daily growth of three main cereal crops (rice, maize, and wheat, or RMW; Figure S3) in the 2403 counties of China (Figure S4). The data include 1,510 counties with rice, 1,704 counties with maize, and 2,007 counties with wheat. The baseline for crop modeling is defined with the identical (2007–2011 mean) conditions for land use, soil, crop rotations, and fertilization. We use observed daily climate data from 1955 to 2014 to simulate county-level daily growth and assess historical drought impacts on yields. We then examine potential future changes in drought impacts using projected daily climate data for 2006–2100 from nine global climate models (GCMs). There are multiple cropping systems in China. Most of the northeastern, northwestern, and Tibetan Plateau regions only have single-cropping systems. Double cropping is dominating in other major agricultural regions. Some places in the south have triple cropping systems. Balancing the water and nutrient budgets in crop modeling can become challenging if only studying a single crop without considering the actual rotation systems. We apply 11 crop rotation systems commonly used in China: rice, maize, spring wheat, winter wheat, maize/winter wheat, rice/winter wheat, rice/rice, rice/vegetable, rice/rice/vegetable, vegetable/winter wheat, and vegetable/maize. The assumed distribution of these crop rotation systems is based on geographical information and county-level archives for all three crop models. 2.4. Irrigation Scenarios Irrigation is crucial for evaluating drought risk in China, but there is a lack of reliable information on when, how, and to what extent farmers irrigated their crops. Our baseline simulations assume that irrigation demands can always be fully met, although water resources may not be sufficient to meet irrigation demands during drought years. Such overestimation of irrigation water use can be reduced through optimizing the parameters to minimize the gap between the modeled and observed yields. Three core irrigation scenarios are considered: baseline irrigation (2007–2011 mean), 1958 irrigation (irrigation is reduced to levels recorded for 1958), and rainfed (no irrigation). The 1958 scenario is used to evaluate the climate and socioeconomic impacts on the reduction of grain production in the following 3 years (Figure 1) with the irrigation level in 1958. We also perform sensitivity studies to evaluate the potential impacts of sustainable groundwater use or total groundwater depletion in northern China on national cereal crop yields. The method is to reduce irrigation areas in each province by an amount consistent with the ratio of groundwater overuse to all groundwater use. Groundwater overuse is defined as groundwater consumption that exceeds annual recharge. 2.5. Model Calibration and Validation We conducted simulations of the three major crops in China on the Sunway Taihu Light Supercomputer hosted by the National Supercomputing Center in Wuxi City, China. Considering energy, water, and fertilization as the primary drivers of grain yield variations, we chose the following crop models in this research: the nitrogen-oriented DNDC (DeNitrification and DeComposition) (Li et al., 1994; Yu et al., 2014), the wateroriented AquaCrop (Steduto et al., 2009), and the radiation-oriented SWAP (soil-water-atmosphere-plant, using WOFOST (WOrld FOod STudies) for crop growth modeling, Kroes et al., 2000). Another reason of 10.1002/2017EF000768 Earth's Future YU ET AL. 692 choosing these models is that the source codes can be transferred to fit the special hardware structure of the supercomputer with relatively less efforts than using other models. We use observed yields from 1998 to 2007 to calibrate the parameters in all three models and observed yields from 2008 to 2010 to validate the resulting configurations. The sensitive parameters selected for optimization are listed in Table 1. For DNDC and AquaCrop, the prior ranges of these parameter are set as 100 ± 20% of the defaulted values (Li, 2016; Raes et al., 2017). For SWAP, we use the recommended parameter range in the user manual (Kroes et al., 2009). All the other crop-related parameters are set as the defaulted values. We perform Markov chain Monte Carlo simulation for parameter optimization for all three crop models in each of the 2,403 counties using DREAM (DiffeRential Evolution Adaptive Metropolis) algorithm (Vrugt et al., 2009). We run the models on the supercomputer 2000–3500 samples in each county. With the convergence achieved after the first 1,200–1,800 samples, we use the last 600 postconvergence samples for analysis. These postconvergence samples are then used to estimate uncertainty and determine the optimal parameter sets (defined here as the parameter sets that produce the maximum posterior probability). 2.6. Ensemble of Multiple Crop Models To reduce the predictive uncertainty of model structure, we use the Bayesian Model Averaging (BMA) method to obtain the multimodel ensemble by the linear combination of individual model predictions (Huang et al., 2017). The DREAM algorithm is also used to derive the weights and variance for the individual ensemble members. Figure S5 shows the results in the calibration and validation periods, including statistics that summarize model performance relative to recorded data from all 2,043 counties (r 2 , root-mean-square error, and biases). After model calibration, the observed or statistical data were applied to validate the modeling results from the site level, the provincial level to the national level. 2.7. Uncertainty Estimation Three sources of uncertainty are considered: model-specific uncertainties estimated using the DREAM algorithm, statistical uncertainties in time mean quantities, and potential measurement errors in national planting area. Uncertainties in model-generated yield loss rates and drought risks are estimated as a 95% confidence interval constructed from postconvergence Markov chain Monte Carlo samples. Statistical errors in time mean quantities are estimated as twice the standard error of the time mean with the number of degrees of freedom adjusted to account for temporal autocorrelations. Potential measurement errors in national Table 1 The Parameters Used for Model Calibration (GDD: Growing Degree Day) Crop model Parameter Definition Unit DNDC MaxB Maximum biomass kgC/ha T opt Optimal temperature °C WD water requirements Kg G-CN Carbon/nitrogen ratios of grain kg/kg AquaCrop WP 0 Water productivity normalized for evapotranspiration and CO 2 g/m 2 HI 0 Reference harvest index % CGC Canopy growth coefficient Fraction per GDD CDC Canopy decline coefficient Fraction per GDD T 1 Time from sowing to flowering °C/d T 2 Time from sowing to maturity °C/d K 1 Upper threshold of soil water depletion for canopy expansion - K 2 Upper threshold of soil water depletion for stomatal control - WP 0 Water productivity normalized for evapotranspiration and CO 2 g/m 2 SWAP TSUM 1 Time from emergence to anthesis °C/d TSUM 2 Time from anthesis to maturity °C/d AMAXTB 1 Maximum leaf CO 2 assimilation rate at the first development stage of crop maturity kg·ha 1 ·hr 1 EFFTB Initial light-use efficiency for CO 2 assimilation by single leaves as a function of daily temperature % PERDL Maximum relative death rate of leaves due to water stress % CFET Correction factor transpiration rate - RGRLAI Maximum relative increase in leaf area index d 1 Note. SWAP = soil-water-atmosphere-plant; DNDC = DeNitrification and DeComposition. 10.1002/2017EF000768 Earth's Future YU ET AL. 693 planting area are estimated as the time mean relative difference (5%) between the national planting area reported by the National Bureau of Statistics of China and the sum of the provincial planting areas reported by the Ministry of Agriculture during 1970–2006. Standard error propagation rules are used to combine and propagate the uncertainties, under the assumption that the three sources of uncertainty are mutually independent. 3. Results 3.1. Impacts of Irrigation on Agricultural Drought Risk To quantify the contributions of irrigation to reducing agricultural drought risks in China, we simulate daily growth of RMW under a baseline scenario (assuming irrigation fully meets the crop demand) and a rainfed scenario (zero irrigation). We also study other permutations to these limiting case scenarios, such as sensitivity simulations in which baseline irrigation rates are adjusted to match 1958 levels (the “1958 scenario”)orto exclude unsustainable groundwater consumption (the “GDW scenario”). Land use, soil conditions, crop rotations, and fertilizer use for all scenarios are based on average records from 2007 to 2011. Figure 2a shows variations in national RMW yields under the baseline irrigation and rainfed scenarios, which are summarized from daily simulated crop growth in all 2,403 counties between 1955 and 2014. Figures 2b–2d show yields for wheat, rice, and maize, along with the weights assigned to the DNDC, AquaCrop, and SWAP models in the BMA ensemble for each crop. In these simulations, crop parameters are held constant at their calibrated values for 1955–2014, neglecting changes in breeding and agricultural practice that have contributed to the observed yield increases shown in Figure 1. Simulated yield variability thus mainly reflects the national-scale impacts of climatological and meteorological droughts (Figure 2). Figure 2a shows that irrigation substantially reduces the risks of yield loss associated with meteorological droughts and therefore plays a central role in ensuring national food security. The maximum simulated Figure 2. Simulated Bayesian model average (BMA) grain yields in mainland China during 1955–2014 based on observed meteorological variations under different irrigation scenarios. (a) Integrated RMW yields under the baseline, rainfed, and 1958 scenarios during 1955–2014. Yields under the 1958 scenario prescribe irrigation areas at 1958 values. The GDW scenario shows expected national RMW yields if groundwater had been unavailable for irrigation in the northern plains during 1999–2014. The green dot represents the national maximum RMW yield, which occurred in 1990, and the red dots are used to highlight model results for 2003 and 1959–1961 (see text for details). BMA yields of (b) wheat, (c) rice, and (d) maize under the baseline and rainfed scenarios. The relative contributions of each crop model to the weighted BMA are also shown. RMW = rice, maize, and wheat; SWAP = soil-water-atmosphere-plant; GDW = groundwater. 10.1002/2017EF000768 Earth's Future YU ET AL. 694 national RMW yield for the past 60 years occurred in 1990 under both the rainfed and baseline scenarios (Figure 2a). Applying the simulated maximum yield, we find that the agricultural drought risk Rfor RMW (Figure 3a) is much larger under the rainfed scenario (average yield losses of 35 ± 2%) than under the baseline irrigation scenario (4 ± 2%). These results indicate that approximately one third of current RMW production in China (31 ± 2%) can be attributed to irrigation (Figures 2a and 3a). Drought risks and the contributions of irrigation to reducing these risks vary by crop. Estimated drought yield losses during 1955–2014 are 55 ± 2% for wheat, 27 ± 2% for maize, and 33 ± 2% for rice under the rainfed scenario compared to 7 ± 2% for wheat, 7 ± 2% for maize, and 2 ± 2% for rice under the baseline scenario (Figures 2b–2d and 3b). Figures 3a and 3b summarize agricultural drought risks for integrated RMW and rice, maize, and wheat individually under the baseline irrigation and rainfed scenarios as a function of agricultural drought severity. The probability curves for the yield losses of RMW (Figure 3a) indicate that China can maintain its current high-level food supply (3,000 kcal per person per day) even during extreme drought events so long as agricultural technology, water supplies for irrigation, and planting areas are maintained at the baseline levels. By contrast, without irrigation, basic level food supply (2,250 kcal per person per day) would be threatened approximately every other year (i.e., every year with agricultural drought exceeding the 50th percentile; Figure 3a). The northern and northeastern parts of China have grown drier over the past 60 years, with increasing frequencies of meteorological drought (Dai, 2013; Piao et al., 2010). Without irrigation, we estimate that Figure 3. Agricultural drought risks (R) in mainland China. (a) Damage-probability curves for RMW under the rainfed and baseline irrigation scenarios. Damage (D)is quantified as integrated yield loss rates of RMW relative to maximum yields (vertical axis) in a given probability (p, horizontal axis) of a drought event. p= 0.5 corresponds to a 2-year drought event, p= 0.01 to a 100-year event, and so on. The values of the famine-, basic-, and high-level food productions can be read in Figure 1a. (b) Damage-probability curves for rice, maize, and wheat individually under the rainfed (dotted) and baseline (solid) scenarios. (c) Spatial distribution of agricultural drought risk map (expressed as annual expected yield loss ratio in percent) for RMW under the rainfed scenario. (d) Agricultural risk map for RMW under the baseline irrigation scenario. The spatial resolution of the land use map is 10 km, with gridded values interpolated from model results in 2,403 counties (Figure S4). RMW = rice, maize, and wheat. 10.1002/2017EF000768 Earth's Future YU ET AL. 695 this signal of increased drought severity during 1995–2014 would have reduced average yields by ~5% for wheat and maize relative to 1955–1994 (Figures 2b and 2d). Irrigation has reduced these rises in droughtrelated losses after 1995 to less than 1.5% for maize and wheat. Rice yields were negligibly affected (<0.5%) since rice is mainly grown in southern China (Figures 2c and S3) where droughts were less severe. Therefore, irrigation has thus far prevented recent increases in the frequency and severity of meteorological droughts from significantly affecting agricultural productivities. The maximum simulated yields for different counties may occur in different years. County-level agricultural drought risks are derived from model results using maximum county-level yields, local planting areas for each crop, and local irrigation areas. Figures 3c and 3d show 10-km gridded spatial distributions of agricultural drought risk interpolated from county-level results. Spatial distributions of agricultural drought impacts on RMW yields under the baseline and rainfed scenarios confirm that drought risks and the effects of irrigation in reducing these risks are larger in northern China, particularly in northwestern China. Corresponding spatial distributions of drought risks for each crop (Figure S8) show that drought risks are larger for wheat than for maize or rice. Simulated drought risks for all three crops compare well with the results of site-level irrigation/nonirrigation experiments (Figure S7). The longest of these experiments, which has been conducted in Liuan County of Anhui Province, in the transitional zone between north and south China, indicates that irrigation has improved rice yields by 34% between 1959 and 2013 relative to control plots without irrigation (Shi, 2015). This observed improvement is consistent with our simulated drought risk for rice in this county (i.e., 33 ± 2%). 3.2. Water Resources and Extreme Droughts Our simulations indicate that the severest national agricultural drought between 1955 and 2014 (a 165-year drought event) occurred in 2000 under both the baseline and rainfed scenarios. This event caused simulated RMW yield losses of 44 ± 2% under rainfed conditions and 9 ± 2% under baseline irrigation. The magnitude of this difference indicates that irrigation was crucial for food production during this drought. The shift in the center of food production in China from the water-abundant south to the water-scarce north has been facilitated in large part by unsustainable consumption of groundwater resources by agriculture in northern China (Figure S1). Reducing this unsustainable groundwater consumption is one of the most pressing grand challenges to agricultural drought management in China (Yu et al., 2011). We treat groundwater resources as the sum of a long-term storage component and an annual recharge component and then define groundwater overuse as consumption that exceeds the annual recharge component. We construct a groundwater depletion scenario that assumes the elimination of long-term storage due to unstainable water management, such that water for domestic, industrial, and agricultural use can only be withdrawn from surface water sources (streams, lakes, and reservoirs) and the annual recharge component of ground water. We use detailed groundwater use statistics from provincial records (Ministry of water Resources, China) to estimate groundwater overuse during 1999–2014 (Figure S1c) and then reduce the prescribed irrigation rates by this amount in the crop model simulations (the groundwater depletion scenario, or GDW). Simulated yield reductions under this groundwater depletion scenario are shown in Figure 2a (pink line). Our simulations indicate that eliminating groundwater overuse in northern China would have reduced national RMW yields by 3 ± 2% on average during 1999–2014 and by as much as 7 ± 2% during drought years. The impacts of recent extreme droughts at the provincial scale illustrate the importance of sustainable groundwater management. The 2002 agricultural drought was relatively mild nationally (an ~8-year return period) but was much more severe in northern China with a 20-year return period in Hebei Province and a 122-year return period in Shandong Province. Our simulations indicate that this drought would have only reduced RMW productivities by 6 ± 2% in Hebei Province and by 10 ± 2% in Shandong Province relative to the nondrought year 1998 if irrigation demands were fully met. Recorded yield data from statistical yearbooks show that cereal crop yields in 2002 were 7% less than those in 1998 in Hebei Province and 9% less than those in 1998 in Shandong, in agreement with our estimates. These relatively mild losses stand in stark contrast to simulated RMW yield losses in these provinces (51 ± 2% in Hebei and 52 ± 2% in Shandong) under the GDW scenario, which assumes zero long-term groundwater storage at the beginning of the year 2002. Table 2 shows annual water resources (surface water plus groundwater recharge from precipitation during 1 year) as recorded by the Ministry of Water Resources. Annual water resources during 2002 were more than 60% less than normal even though annual precipitation in these provinces was only about 30% less than its 10.1002/2017EF000768 Earth's Future YU ET AL. 696 Tao, F., & Zhang, Z. (2011). Impacts of climate change as a function of global mean temperature: Maize productivity and water use in China. Climatic Change,105, 409–432. Taylor, K. E., Stouffer, R. J., & Meehl, G. A. (2012). An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological Society,93, 485–498. Tso, T. C. (2004). Agriculture of the future. Nature,428, 215–217. van Diepen, C. A., Wolf, J., van Keulen, H., & Rappoldt, C. (1989). WOFOST: A simulation model of crop production. Soil Use and Management,5, 16–24. Vrugt, J. A., ter Braak, C. J. F., Diks, C. G. H., Robinson, B. A., Hyman, J. M., & Higdon, D. (2009). Accelerating Markov chain Monte Carlo simulation by differential evolution with self-adaptive randomized subspace sampling. International Journal of Nonlinear Sciences and Numerical Simulation,3, 273–290. Wang, J., Huang, J., Zhang, L., & Huang, Q. (2014). Do incentives still matter for the reform of irrigation management in the Yellow River Basin in China? Journal of Hydrology,517, 584–594. Wang, J., Mendelsohn, R., Dinar, A., Huang, J., Rozelle, S., & Zhang, L. (2009). The impact of climate change on China’s agriculture. Agricultural Economics,40, 323–337. Wang, J., Yang, Y., Huang, J., & Chen, K. (2015). Information provision, policy support, and farmers’adaptive responses against drought: An empirical study in the North China Plain. Ecological Modeling,318, 275–282. Wang, X., Wang, Q., Fan, J., & Fu, Q. (2013). Evaluation of the AquaCrop model for simulating the impact of water deficits and different irrigation regimes on the biomass and yield of winter wheat grown on China’s Loess Plateau. Agricultural Water Management,129,95–104. World Meteorological Organization (2015). WMO guidelines on multi-hazard impact-based forcast and warning services, WMO-No.1150. Yan, H., Wang, S., Wang, J., Lu, H., Guo, A., Zhu, Z., et al. (2016). Assessing spatiotemporal variation of drought in China and its impact on agriculture during 1982–2011 by using PDSI indices and agriculture drought survey data. Journal of Geophysical Research: Atmospheres, 121, 2283–2298. https://doi.org/10.1002/2015JD024285 Yang, S., Ding, Z., Li, Y., Wang, X., Jiang, W., & Huang, X. (2015). Warming-induced northwestward migration of the East Asian monsoon rain belt from the Last Glacial Maximum to the mid-Holocene. Proceedings of the National Academy of Sciences of the United States of America, 112, 13,178–13,183. Ye, L., Xiong, W., Li, Z., Yang, P., Wu, W., Yang, G., et al. (2013). Climate change impact on China food security in 2050. Agronomy for Sustainable Development,33, 363–374. Yu, C., Gong, P., & Ying, Y. (2011). China’s water crisis needs more than words. Nature, 307. Yu, C., Li, C., Xin, Q., Chen, H., Zhang, J., Zhang, F., et al. (2014). Dynamic assessment of the impact of drought on agricultural yield and scaledependent return periods over large geographic regions. Environmental Modeling and Software,62, 454–464. Zargar, A., Sadiq, R., Naser, B., & Khan, F. I. (2011). A review of drought indices. Environmental Reviews,19, 333–349. Zhang, D. E. (2005). Severe drought events as revealed in the climate records of China over the last thousand years. Acta Meteorologica Sinica, 485–491. Zhang, D. E., & Liang, Y. (2010). A long lasting and extensive drought event over China in 1876–1878. Advances in Climate Change Research,1, 91–99. Zhang, Q., Zhang, J., & Wang, C. (2017). Risk assessment of drought disaster in typical area of corn cultivation in China. Theoretical and Applied Climatology,128, 533–540. Zhang, T., & Huang, Y. (2012). Impacts of climate change and inter-annual variability on cereal crops in China from 1980 to 2008. Journal of the Science of Food and Agriculture,92, 1643–1652. Zhang, X., Tang, Q., Pan, M., & Tang, Y. (2014). A long-term land surface hydrologic fluxes and states dataset for China. Journal of Hydrometeorology,15, 2067–2084. 10.1002/2017EF000768 Earth's Future YU ET AL. 703