Model Development and Evaluation in Northeast China

PUBLICATIONS

Journal of Advances in Modeling Earth Systems

RESEARCH ARTICLE Modeling irrigation-based climate change adaptation in 10.1002/2014MS000402 agriculture: Model development and evaluation in Northeast

Key Points: China This study developed a coupled crop production and river circulation Masashi Okada1, Toshichika Iizumi1, Gen Sakurai1, Naota Hanasaki2, Toru Sakai1, Katsuo Okamoto1, model and Masayuki Yokozawa3 The model can accurately capture the major features of hydrology and 1National Institute for Agro-Environmental Sciences, Tsukuba, Japan, 2National Institute for Environmental Studies, crop yield 3 The model is useful in assessing Tsukuba, Japan, Graduate School of Engineering, Shizuoka University, Hamamatsu, Japan climate change adaptation based on irrigation Abstract Replacing a rainfed cropping system with an irrigated one is widely assumed to be an effective Supporting Information: measure for climate change adaptation. However, many agricultural impact studies have not necessarily Supporting Information S1 accounted for the space-time variations in the water availability under changing climate and land use. Figure S1 Moreover, many hydrologic and agricultural assessments of climate change impacts are not fully integrated. Figure S2 Figure S3 To overcome this shortcoming, a tool that can simultaneously simulate the dynamic interactions between crop production and water resources in a watershed is essential. Here we propose the regional production Correspondence to: and circulation coupled model (CROVER) by embedding the PRYSBI-2 (Process-based Regional Yield Simula- M. Yokozawa, tor with Bayesian Inference version 2) large-area crop model into the global water resources model (called [email protected] H08), and apply this model to the Songhua River watershed in Northeast China. The evaluation reveals that the model’s performance in capturing the major characteristics of historical change in surface soil moisture, Citation: river discharge, actual crop evapotranspiration, and soybean yield relative to the reference data during the Okada, M., T. Iizumi, G. Sakurai, N. Hanasaki, T. Sakai, K. Okamoto, and interval 1979–2010 is satisfactory accurate. The simulation experiments using the model demonstrated that M. Yokozawa (2015), Modeling subregional irrigation management, such as designating the area to which irrigation is primarily applied, irrigation-based climate change has measurable inﬂuences on the regional crop production in a drought year. This ﬁnding suggests that adaptation in agriculture: Model development and evaluation in reassessing climate change risk in agriculture using this type of modeling is crucial not to overestimate Northeast China, J. Adv. Model. Earth potential of irrigation-based adaptation. Syst., 7, 1409–1424, doi:10.1002/ 2014MS000402.

Received 6 NOV 2014 Accepted 28 AUG 2015 1. Introduction Accepted article online 1 SEP 2015 Published online 24 SEP 2015 Many studies have widely deemed that replacing a rainfed cropping system with an irrigated one can be an effective measure for climate change adaptation [Porter et al., 2014; Challinor et al., 2014]. This is true as long as this replacement does not lead to unsustainable water extraction and/or as long as there is still sufﬁ- cient irrigation water available [Elliott et al., 2014]. However, many agricultural impact assessments have used time-constant future irrigation scenarios derived from statistical data or other independent studies (some exceptional integrated studies are seen in Deryng et al. [2011], Biemans et al. [2013], and Kummu et al. [2014]). General hydrologic models or land-surface models have been used to assess the hydrologic impacts due to climate change [Doll€ and Schmied, 2012; Hagemann et al., 2013; Hanasaki et al., 2013]. These hydrologic models have been intensively tested for their ability in capturing the major components of terrestrial water cycling, including river discharge, actual evapotranspiration, and soil moisture [Alcamo et al., 2003; Hanasaki et al., 2008; Biemans et al., 2009], but have not necessarily been evaluated for their ability in simulating crop growth and yield (a few exceptions are seen in Bondeau et al. [2007] and Biemans et al. [2013]). VC 2015. The Authors. Although a recent study combines the outputs of the hydrologic models and crop models to evaluate the This is an open access article under the potential impacts of water availability on future crop productivity [Elliott et al., 2014], the hydrologic models terms of the Creative Commons and crop models are separately simulated and thus still has not yet fully integrated. These facts necessitate Attribution-NonCommercial-NoDerivs License, which permits use and an improved assessment of potential crop production under changing climate, varying water availability, distribution in any medium, provided and expanding irrigated cropland areas, using a tool that can simultaneously simulate the dynamic interac- the original work is properly cited, the tions between crop production and water resources (e.g., river water) in a watershed. use is non-commercial and no modiﬁcations or adaptations are China is the leading crop-producing country worldwide and accounted for 20% of global cereal production made. in 2011 [Food and Agriculture Organization of the United Nations, 2014]. The Northeast China Plain is a major

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1409 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

crop-producing region and important for maintaining China’s food balance. However, the recent warming and drying trends in Northeast China are evident [Tao et al., 2003; Piao et al., 2010; Yu et al., 2014]. These changes in climate have likely led to an increase in the extent of drought-damage cropland area and a decrease in the area suitable for cultivation of some crops, such as spring wheat [Jiao et al., 2007, 2008]. A decreasing trend in water resources, including the river discharge of the Songhua and Liao Rivers, which are major river channels in this region, has accompanied these changes, increasing concerns about food supply in the coming decades [Piao et al., 2010]. Hence, it is valuable to revisit the following questions that are posed in a series of previous studies [Jimenez Cisneros et al., 2014; Porter et al., 2014], using North- east China as the example: Is the available water resources sufﬁcient for crop production under changing climate and expanding irrigated area? Given the pressure to maintain the increasing food demand in China, how much can subregional irrigation management increase the total crop production in a given watershed? To address these issues, this study (1) developed the regional production and circulation coupled model (CROVER) that can simultaneously simulate crop growth and yield, river discharge and their dynamic interactions by embedding the PRYSBI-2 large-area crop model [Sakurai et al., 2014] into the H08 global water resources model [Hanasaki et al., 2008]; (2) applied this coupled model to the Songhua River watershed in Northeast China and evaluated the model’s performance by comparing the historical model simulation, including soybean yield, river discharge, surface soil moisture, actual crop evapotranspiration, with the reference data; (3) performed simulation experiments using the coupled model to quantify the possible inﬂu- ence of subregional irrigation management in a drought year on regional crop production.

2. Study Area Our study area includes the Northeast China Plain, the main portion of which is located in the cool- temperate and partially semiarid zones and is surrounded by three mountain ranges: the Daxing’anling, Xiaoxing’anling, and Chanbai Mountains (Figure 1a). The main watersheds consist of the Songhua and Liao Rivers. The Songhua River flows across the Northeast China Plain from west to east after its tribu- taries join at its upper stream (Figure 1b). The riverhead of the tributary flowing from the north side (called the Nen River) is located in the Daxing’anling and Xiaoxing’anling Mountains, whereas that of the tributary flowing from the south side (called the Second Songhua River) is located in the Chanbai Mountains. The areal extent of the Songhua River watershed reaches 545,000 km2 (supporting information Figure S1a), and the total river length is 1927 km. The river flow is mainly sourced from precipitation in summer (which accounts for 60–80% of the annual runoff [Asian Development Bank, 2005]) and snowmelt from March to May. The mean quantity of water resources in the Songhua River watershed and its surrounding small watersheds in 2002–2011 is approximately 130 3 109 m3 [Ministry of Water Resources of the People’s Republic of China (MWR), 2015]. Twenty-four percent of the water resources (31 3 109 m3) are consumed by agriculture [MWR, 2015]. Two reservoirs with the storage capacity of >109 m3, the Baishan and Fengman reservoirs, are located in the upper stream of the watershed (Fig- ure 1b). The Northeast China Plain is the largest crop-producing region in China. Among other crops, soybean pro- duced in this region accounted for approximately 40% of the national production in 2011 [National Bureau of Statistics of China (NBSC), 2014]. Therefore, soybean was selected for this study. Soybean is mainly culti- vated in the northern part of the watershed (Figure 1a). The Liao River and its watershed were thus excluded from the analysis.

3. Data and Model 3.1. General Description of the CROVER Model As schematically illustrated in Figure 2, the PRYSBI-2 large-area crop model [Sakurai et al., 2014] was embedded into the H08 global hydrologic model [Hanasaki et al., 2008]. The coupled model has ﬁve major components: land surface, crop growth, river routing, reservoir operation, and anthropogenic water withdrawal. The land-surface process that was used in the coupled model is based on that of the PRYSBI-2 model, because the process that was used in the H08 model, the soil water balance in particular, is based on the simple leaky bucket model of Manabe [1969] (supporting information Table S1). As the land-surface

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1410 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

Figure 1. Geographical distributions of (a) elevation, soybean harvested area, mountain ranges, and major river channels around Northeast China. A blue line indicates a major river channel. A solid black line indicates a national border. A gray shaded area indicates the elevation. A green shaded area indicates the major soybean harvested area (>5% of a 18 grid cell). A triangle indicates the location of the river discharge observatories at Harbin and Jilin. (b) The river ﬂow directions and the locations of major reservoirs along with the Songhua River. An arrow indicates the direction of river ﬂow in the Songhua River watershed, and the different colors indicate the different (upper, middle, and lower) parts of the watershed. A star indicates the major reservoirs.

component that was used in the PRYSBI-2 model is based on that of the SWAT (Soil and Water Assessment Tool) [Neitsch et al., 2011], the coupled model can account for more detailed atmosphere-crop-soil interactions than the H08 model. The crop growth component was replaced by the PRYSBI-2 model instead of the SWIM (Soil and Water Integrated Model) [Krysanova et al., 2000],which was originally used in the H08 model. Therefore, the coupled model is expected to allow more sophisticated simulation of crop growth and water resources compared to either of the PRYSBI-2 and H08 because of the coupling (supporting information Table S1). The other processes (river routing, reservoir operation, and anthropogenic water withdrawal) used in the coupled model were based on those of the H08 model. The crop parameters values of the coupled model were derived by applying the Markov Chain Monte Carlo (MCMC) method to the PRYSBI-2 model from one grid cell to another (see supporting information Text S1 for more details). The grid-cell yield data in the odd years during the interval 1982–2006 obtained from the global data set of historical yields (GDHY) [Iizumi et al., 2014b] were used for the calibration subset. The single parameter set that showed the highest likelihood relative to the calibration subset was used in this study (supporting information Table S2). The identical parameter set was used for both rainfed and irrigated conditions. In contrast, the hydrologic parameters values of the coupled model were calibrated for the watershed (but not for each grid cell) using the trial-and-error method (supporting information Text S1). The surface soil moisture, actual evapotranspiration, and river discharge references in the earlier 3 years which were not severe drought years were used for the calibration subset. Importantly, both crop and hydrologic parameter sets were calibrated using the calibration subset, and the calibrated model was evaluated using the independent validation subset.

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1411 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

Figure 2. Schematic illustration of various water ﬂows that were considered in the coupled model proposed in this study.

3.2. Modeled Interactions Between River Discharge and Irrigation As already noted in section 1, demonstrating the potential benefits of coupling river water dynamics and crop growth dynamics in a watershed is the purpose of this study. For this reason, we used an idealized land use map, i.e., only two land use categories, rainfed soybean and irrigated soybean in the watershed. For a given grid cell, the coupled model first separately simulates the land-surface process and crop growth process in rainfed and irrigated conditions (Figure 2). In rainfed condition, no river water intake is considered, whereas in irrigated condition, the model accounts for the river water intake for irrigation when calcu- lating crop growth, actual evapotranspiration, infiltration, and surface and subsurface runoffs. Then the mean simulated quantity of interest (e.g., soil moisture) for a given grid cell across rainfed and irrigated conditions is calculated using the grid-cell extents of rainfed and irrigated areas derived from the MIRCA2000 (Monthly Irrigated and Rainfed Crop Area around the year 2000) data set [Portmann et al., 2010] as the weights. The irrigated area used for this study was not only for soybean, but was extended to the sum of four major crops (maize, soybean, rice, and wheat) (supporting information Figure S1b). The irrigated area in four major crops was accountable for much of that in all crops (approximately 80% over the watershed). The use of this idealized irrigation map was to avoid the potential underestimation of agricultural water intake and associated overestimation in river discharge. Finally, the lateral flow of river water along with river channels was computed and updates river discharge for a given grid cell. Therefore, the water intake in upper stream affects river water availability downstream. Importantly, in the coupled model, the river water availability for irrigation in a given grid cell depends on the location of the grid cell in the watershed and on the river water intake happened in the upstream. More detailed relationships between the crop water demand and river water intake are described as below. 21 In irrigated conditions, the crop water demand on a given day i (mm d ), Agrdem; i, was computed: Xn Agrdem; i5 FCl2SMl;i when ETi=PETi < thrs; (1) l51

Agrdem; i50 when ETi=PETi thrs; (2)

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1412 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

where FCl is the field capacity of the soil layer l (mm), SMl;i is the soil moisture of the layer l on the day i (mm), n is the number of the deepest soil layer where crop’s roots exist, PETi is the potential crop evapotranspiration on the day i calculated by the Penman-Monteith method [Monteith, 1965; Neitsch 21 et al., 2011] (mm d ), ETi is the actual crop transpiration on the day i calculated from the PETi and the available soil water in the root zone (mm d21), and thrs is the triggering aridity level at which irrigation operation starts (dimensionless). If sufficient grid-cell river discharge is available to maintain the crop water demand for the day i, water is taken from grid-cell river discharge and applied to the first (sur- 21 face) soil layer of the grid cell as the irrigation on the day i (mm d ), Irri, until the soil moisture of the whole soil layer approaches its field capacity:

Irri5Agrdem; i when Rivi Agrdem; i; (3)

where Riviis the amount of river discharge on the day i ðmm). If the grid-cell river discharge is insufﬁcient to fully maintain the crop water demand on the day, only the available river discharge is used for irrigation:

Irri5Rivi when Rivi < Agrdem; i: (4)

A thrs value of 0.9 was commonly used for all irrigated grid cells in the historical simulation (see section 4.3) in accordance with Neitsch et al. [2011]. Varying thrs values depending on the different irrigation management options were used for the simulation experiments.

3.3. Model Inputs The daily maximum and minimum 2 m air temperature, precipitation, solar radiation, 2 m relative humidity, and 10 m wind speed for the period 1979–2010, obtained from the 1.1258 daily GRASP (Global Risk Assess- ment toward Stable Production of food project) meteorological forcing data set [Iizumi et al., 2014a], were used as the weather inputs (supporting information Table S3). The weather data regridded to the grid size of 18 by using the inverse distance weighted averaging method were used to be consistent with the 18 river channel network map [Hanasaki et al., 2006]. Owing to the speciﬁc reservoir-river-channel combination, it was relatively harder to change the grid size of this map than that of other inputs. Hence, the model operated at the 18 grid size. The textural and hydrologic characteristics of the soil, including the percentage content of clay, silt, sand and organic carbon, moist soil albedo, bulk density, available water capacity, and maximum rooting depth for two soil layers, were derived from the ISLSCP-II (International Satellite Land-Surface Climatology Project Initiative II) soil data set [Scholes and Brown de Colstoun, 2011] (supporting information Table S3). The bot- tom depth of each soil layer corresponds to 300 and 1500 mm, respectively.

For other inputs, the following data were used: the annual atmospheric carbon dioxide ([CO2]) concentration reported by Keeling et al. [2009]; the soybean planting date derived from the global crop calendar circa 2000 [Sacks et al., 2010]; the geo-referenced major reservoir capacity [Hanasaki et al., 2006], for which the original information was based on the World Register of Dams 1998 [International Commission on Large Dams, 1998]; the geo-referenced domestic and industrial water intakes [Hanasaki et al., 2008], which con- verted the national values obtained from the FAO AQUASTAT database using the geographical population distribution [Center for International Earth Science Information Network at Columbia University and Centro Internacional de Agricultura Tropical, 2005] (supporting information Table S3).

3.4. References for Model Evaluation Supporting information Table S4 summarizes the references used for the evaluation of the coupled model’s historical simulation: the 0.258 daily surface 10 cm soil moisture reference for 1979–2010 derived from the Essential Climate Variables Soil Moisture (ECV-SM) product [Liu et al., 2012; Wagner et al., 2012], the 0.00838 (or 1 km) 8 day composite actual evapotranspiration reference for 2000–2010 derived from the MODIS (Moderate Resolution Imaging Spectroradiometer) Global Evapotranspiration Project (MOD16) product [Mu et al., 2011], the observed monthly river discharge data from 1979 to 1987 at the Harbin observatory (Figure 1a) obtained from the Global Runoff Data Centre (GRDC) [The Global Runoff Data Centre, 2014], and the 1.1258 yearly soybean yield references from 1982 to 2006 obtained from the GDHY product [Iizumi et al., 2014b]. Note that the GDHY product includes two differently sourced yield values for the study area: the reported district yield statistics and the yield estimates by merging satellite-derived net primary production with the FAO country yield statistics [Iizumi et al., 2014b].

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1413 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

As the grid size of the ECV-SM and MOD16 products was ﬁner than that of the coupled model (18), the values for soil moisture and actual evapotranspiration references were spatially aggregated to achieve a consistent comparison. When aggregating the soil moisture values, the 0.0838 (or 10 km) soybean harvested area map [Monfreda et al., 2008] was aggregated to have a grid size of 0.258, and the average value over the Songhua River watershed was calculated day by day using the grid-cell harvested area as the weights. The actual evapotranspiration values were ﬁrst aggregated to have the same grid size as the 0.0838 harvested area map, and the average value (weighted by the harvested area) over the watershed was computed for every 8 days. After the spatial aggregation, the values of the daily soil moisture and 8 day actual evapotranspiration references were aggregated month by month. A missing value was assigned for a given month if the number of effective data values was fewer than half of the number of days in a month. To provide additional information whether this aggregation method is acceptable to compare with the model outputs, we also provided the aggregated soil moisture and actual evapotranspiration references without areal weighting by the soybean harvested area. The GRDC monthly river discharge data were compared with the simulated monthly values (computed from the simulated daily values) in a grid cell where the Harbin observatory was located (Figure 1a). The GDHY soybean yield reference was compared with the historical simulation after averaging the simulated yields under rainfed and irrigated conditions using the extents of rainfed and irrigated areas of the MIRCA2000 data set as the weights.

4. Methods 4.1. Regional Water Balance To describe the major characteristics of the annual water balance in the study area, we analyzed the GRASP precipitation, MOD16 actual evapotranspiration, GRDC river discharge, and ECV-SM surface soil moisture references on a 12 monthly climatology basis. The area-mean monthly values for these data (weighted by the grid-cell soybean harvested area) were computed by averaging the grid-cell submonthly reference values over the study area (more precisely, the Songhua River watershed above Harbin observatory, Figure 1a). A missing value was assigned when the number of effective data values was fewer than half of the number of days in a month.

4.2. Meteorological, Hydrologic, and Agricultural Droughts To depict an overview of the temporal variations in meteorological, hydrologic, and agricultural droughts in the study area, we analyzed the soybean-growing season (May–September) GRASP precipitation, ECV-SM surface soil moisture, MOD16 actual evapotranspiration, GRDC river discharge for the Songhua River (at two sites, Harbin and Jilin, Figure 1a), GDHY soybean yield, and the extent of drought-damage cropland area in Northeast China (Heilongjiang, Jilin, and Liaoning provinces) obtained from the China Statistical Yearbook [NBSC, 2014]. The drought-damaged levels used in the statistical yearbook have three different classes: ‘‘slight,’’ ‘‘moderate,’’ and ‘‘severe’’ which was quantiﬁed based on yield reduction compared to normal yield. The monthly data for precipitation, soil moisture, and yield were aggregated over the study area (weighted by the grid-cell soybean harvested area) in the manner described in section 3.4 and used for analysis.

4.3. Historical Simulation The historical simulation was performed to evaluate the model’s performance in reproducing the major characteristics of temporal variations (and trends if any) in the surface soil moisture, actual crop evapotranspiration, river discharge, and soybean yield from 1979 to 2010 over the soybean harvested area in the watershed. The model operated with the calculation domain (418N–538N; 1198E–1338E, Figure 1a) covering the entire Songhua River watershed with a grid size of 18. The spin-up was conducted until the vertically integrated soil moisture in most (>95%) of the grid cells that were located within the calculation domain showed the convergence by iteratively using the weather data in 1979. The model then began the historical simulation for the 32 year period.

4.4. Simulation Experiments To quantitatively examine the response of regional soybean production to different subregional irrigation management options, we performed simulation experiments by applying various hypothetical irrigation

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1414 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

management conditions to the model. In this study, we focused on the severe drought year of 1982 as the test case (see section 5.1). We separated the watershed into three areas (the upper, middle, and lower) so that the soybean- production in each area was the almost same, instead of dividing the watershed according to the elevation. This setup provides an easier comparison of the production under different irrigation management conditions. More practically, all of the grid cells were first numbered in reverse order of river sequence; for example, the most downstream grid cell was numbered ‘‘1,’’ and the adjacent upstream grid cell was numbered ‘‘2.’’ Second, the grid-cell soybean production was calculated by adding the grid cells in numerical order until the area reached one third of the total production in the watershed, and these grid cells were assigned to the ‘‘lower’’ area. The ‘‘middle’’ and ‘‘upper’’ areas were determined in a similar manner. The grid cells that were categorized into the middle area were further divided into two types: middle area with and without an adjacent upper area. Lower areas with and without an adjacent middle area were specified in a similar manner. Finally, we had five areal categories: upper, middle with and without adjacent upper area, and lower with and without adjacent middle area. This categorization was important because the irrigation management options used in upper (middle) areas do not influence crop production in middle (lower) areas without adjacent upper (middle) area. The irrigation management options were expressed by a combination of five restriction levels of water intake (very weak, weaker, intermediate, stronger, and very strong restrictions) in the upper and middle areas. The very weak, weaker, intermediate, stronger, and very strong restriction levels correspond to the triggering aridity level (thrs) of 1.0, 0.9, 0.6, 0.3, and 0.2, respectively. We thus compared the impacts of 25 different irrigation management options (five restriction levels in the upper area 3 five restriction levels in the middle area) on regional soybean production. We did not examine the influences of irrigation management in the lower area because this management does not affect the river water availability in the upper and middle areas. The irrigation management in the lower area is set as the weaker restriction level 0.9 each irrigation management options to allow higher yields in the lower area. Here we defined that effective irrigation management is to maximize the total regional production, not water productivity (yield gain per additional irrigation water) at grid-cell level.

5. Results 5.1. Observed Hydrologic and Agricultural Impacts of Droughts The mean annual precipitation in the study area in 1979–2010 ranged from 300 to 750 mm yr21, with the monthly peak in July (Figure 3). Annually, approximately 97 6 19% of the precipitation evaporated, and the remaining 23 6 7% was distributed into the river discharge and surface soil moisture (the errors indicated by the standard deviation shows the year-to-year variability). Note that they did not add up to 100% or completely closed water balance because of the different time intervals across the data used and the lack of some terms in the water balance equation, such as the deeper-layer soil moisture. On a monthly basis, the amount of the actual evapotranspiration in spring and autumn was almost comparable to the precipitation, suggesting the importance of precipitation and river discharge in these seasons to maintain the crop water demand. The below-normal precipitation in the soybean-growing season (May–September) is often accompanied by hydrologic droughts (indicated by lower river discharge) and agricultural droughts (indicated by either or all of lower soil moisture, lower yield, and larger extent of drought-damage area). For instance, the observed soybean-growing season precipitation in 1982 was 308 mm season21, whereas the normal precipitation was 414 mm season21 (Figure 4a). The lower precipitation led to a substantial decrease in the river discharge at two locations of the Songhua River (Figure 4b). The lower precipitation and lower river discharge were further accompanied by lower soil moisture, lower soybean yield than in the subsequent years (Figure 4c), and a larger extent of drought-damaged cropland area (Figure 4d). Other observed major agricultural droughts indicated by the extent of drought-damage cropland area >5 3 106 ha included 1989, 1997, 2000, 2001, 2003, and 2004 (Figure 4d). The incidence of meteorological, hydrologic, and/or agricultural droughts in these years were corroborated by a governmental agricultural disasters report [Wen and Sun, 2007]. This report also notes that the drought in 1982 was the most severe drought of the past 300 years and resulted in unprecedented agricultural damage [Wen and Sun, 2007].

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1415 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

] 5.2. Evaluation of the Model’s

−1 150 Performance Prec The modeled year-to-year variations in 120 ET the surface soil moisture, as averaged Riv over the soybean-growing season and 90 Δ the soybean harvested area in the SM watershed from 1987 to 2010 were comparable to the ECM-SM soil mois- 60 ture reference, with a Pearson’s correlation coefficient (r) of 0.73 (n 5 21, 30 p < 0.001) and a root-mean-square error (RMSE) of 2.9 mm or 13% of the 0 multiyear mean value of the reference (Figure 5a). The model captured the −30 specific drought events, such as those in 2000 and 2001 (Figure 5a), as well as Amount of water [mm month 123456789101112 the drying trend of soil moisture in this Month area in recent decades. However, the modeled annual rate of the decrease in 21 Figure 3. Multiyear mean monthly water balance in the Songhua River water- soil moisture of 20.20 mm yr was shed. The mean values in 2000–2010 are presented for the precipitation (Prec), overestimated by 67% compared to surface soil moisture (DSM), and actual evapotranspiration (ET) references, 21 whereas the mean values in 1979–1987 are shown for the observed river dis- the observed rate of 20.12 mm yr . charge at the Harbin observatory (Riv). The overall results mentioned above were almost consistent even when we used the reference aggregated without areal weighting by the soybean harvested area (supporting information Figure S2a). While the limited length of the MOD16 actual evapotranspiration reference prevented us from drawing a conclusive statement, the model seemed well capturing year-to-year variations in the reference actual evapotranspiration over the soybean harvested area except year 2007, which was a drought year according to the observed soil moisture (Figure 5a). With the data in 2007, the goodness-of-fit statistics were low (r 5 0.05, n 5 8, p 5 0.90, RMSE 5 0.17 mm d21 or 9%, Figure 5b). However, these statistics improved to some degree if the data in 2007 were eliminated (r 5 0.23, n 5 7, p 5 0.62, RMSE 5 0.14 mm d21 or 7%). Interestingly, the agreement between the aggregated modeled and reference values became worse compared to those presented in the above when the reference aggregated without areal weighting by the soybean harvested area was compared (supporting information Figure S2b). The modeled river discharge in a grid cell where the Harbin observatory is located (Figure 1), as averaged over the soybean-growing season, accurately matched the GRDC data with an RMSE of 283 m3 s21 or 20% (Figure 5c). The model captured the year-to-year variations in the river discharge fairly well (r 5 0.80, n 5 6, p 5 0.06), although the references were limited for the first 9 years of the analyzed period (Figure 5c). The lower modeled river discharge in 1982 presented a good match with the reference. The modeled soybean yields, as averaged over the harvested area, from 1984 to 2006 were comparable to the GDHY yield reference, with an r value of 0.72 (n 5 11, p < 0.01) and an RMSE value of 0.2 metric ton ha21 or 12% (Figure 5d). The model was able to capture some yield loss events due to droughts, such as those in 2000 and 2003 (see section 5.1). However, both overestimation (e.g., in 1988, 1989, 1995, and 2005) and underestimation (e.g., in 1987, 1999, and 2001) of the yield were found. No clear tendency of the model’s overestimation and underestimation emerged because the model showed both types of error in drought years (1989 and 2001) and nondrought years (1987, 1988, 1995, 1999, and 2005). Importantly, the performance in simulating the yield was largely dependent on the reference that was used for the evaluation. The r value and RMSE values decreased to 0.44 (n 5 11, p 5 0.18) and 0.4 t ha21 or 18%, respectively, when the crop statistics were compared (black dashed and red lines, Figure 5d). Furthermore, some larger discrepancies between the modeled and reference yields, such as the yield values in 1995 and 2005, were found irrespective of the source of the reference.

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1416 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

800 28

] (a) Precipitation Soil moisture −1 600 24

400 20

200 16 Precipitation [mm 5months

0 12 Soil moisture [mm]

] 1980 1985 1990 1995 2000 2005 2010

−1 4000 s

3 (b) 3000

2000 Songhua River (Harbin) 1000 Songhua River (Jilin)

River discharge [m 1980 1985 1990 1995 2000 2005 2010 3.0 (c)

] 2.5 −1 2.0 1.5 1.0 Crop statistics Yield [t ha 0.5 GDHY 0.0 1980 1985 1990 1995 2000 2005 2010 (d) Slightly damaged Severely damaged 3000 Moderately damaged

2000

1000

Damaged area [1000ha] 1980 1985 1990 1995 2000 2005 2010 Year

Figure 4. (a) Reference soybean-growing season (May–September) precipitation and surface soil moisture, (b) river discharge at two locations along the Songhua River, (c) soybean yield from two different sources (the district crop statistics and the GDHY satellite-statistics- merged yield estimates), and (d) extent of area in which droughts-damaged crop growth. (a and c) Data averaged over the Songhua River watershed (weighted by the soybean harvested area) are presented. (d) The data indicate the sum across the Liaoning, Jilin, and Heilong- jiang Provinces.

5.3. Impacts of Subregional Irrigation Management on Crop Yield and Production In the severe drought year 1982, the higher modeled yields appeared around the upper area of the watershed (red arrow, Figure 6a) and in part of the lower area (blue arrow) compared to those in the middle area (green arrow). If the river water intake for irrigation in the upper area was restricted compared with the baseline (i.e., a triggering aridity of 0.3 was applied to the upper area and of 0.9 to the middle and lower areas in the simulation experiment), the modeled yields in the upper area decreased by 1–26% (red shaded area, Figure 6b), relative to the baseline (Figure 6a), whereas the modeled yields in the middle and lower areas increased by 1–5% (green shaded area, Figure 6b). As a result, this alternative water management option decreased total production in the watershed by 1.8% because the increased production in the middle and lower areas was not able to compensate for the decrease in production in the upper area (Figure 6d). In contrast, the modeled yield in the middle area decreased (red shaded area, Figure 6c) and the modeled yield in the lower area increased (green shaded area, Figure 6c) if the river water intake in the middle area was restricted compared to the baseline. In this case, no change in the yield or production there was simulated, as the river water intake in the upper area was unchanged (Figures 6c and 6e). However, the production in the middle area decreased by 16.5% (only the middle areas with adjacent upper area were considered), while that in the lower area increased by 4.4% (only the lower areas with adjacent middle area were used), resulting in a 1.5% decrease in the regional soybean production (Figure 6e).

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1417 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

r = 0.73 (p < 0.001) RMSE[%] = 13 n = 21 30 (a) y=−0.12x+265.7 25

[mm] 20 Reference

Soil moisture 15 Model y=−0.20x+413.4 1980 1985 1990 1995 2000 2005 2010 r = 0.05 (p = 0.90) RMSE[%] = 9 n = 8

] 2.4 (b) −1 2.1 Reference 1.8 Model

[mm day 1.5

Evapotranspiration 1980 1985 1990 1995 2000 2005 2010 r = 0.80 (p = 0.06) RMSE[%] = 20 n = 6 4000 (c) Observation ] 3000 −1 Model s 3 2000

[m 1000

River discharge 0 1980 1985 1990 1995 2000 2005 2010

Ref. (estimate): r = 0.72 (p < 0.01) RMSE[%] = 12 n = 11 Ref. (statistics): r = 0.44 (p = 0.18) RMSE[%] = 18 n = 11

] 4 (d) Reference (estimate) Reference (statistics) −1 Model 3

2 Yield [t ha 1 1980 1985 1990 1995 2000 2005 2010 Year

Figure 5. Modeled and reference yearly time series of (a) surface soil moisture, (b) actual crop evapotranspiration, (c) river discharge, and (d) soybean yield. (a, b, and d) Data averaged over the Songhua River watershed above the Harbin observatory (weighted by the soybean harvested area) and over the soybean-growing season (May–September) are presented. (c) Data at the Harbin observatory are presented. The red open (closed) circles show the calibration (validation) subset. The correlation coefﬁcient (r), root-mean-square error (RMSE, percentage of the multiyear mean value of the reference where both modeled and reference data are available), p value (in parentheses), and sample size (n) during the validation period are shown in each plot. The dashed line in Figure 5a shows the best ﬁt liner regression line for each of the modeled and reference data.

To identify the management option that would provide a greater regional crop production, we examined 25 different management options, consisting of ﬁve levels of restriction on the river water intake (triggering aridity level of 0.2, 0.3, 0.6, 0.9, and 1.0 correspond to ‘‘very strong,’’ ‘‘stronger,’’ ‘‘moderate,’’ ‘‘weaker,’’ and ‘‘very weak’’ restrictions, respectively) for the upper and middle areas (supporting information Figure S3). Any restrictions in the lower area do not affect the water availability in the upper and middle areas. As expected, the very weak restriction for the upper and middle areas (i.e., the upper and middle areas could take water as much as available for these areas, and only the remaining river water was available for the lower area, supporting information Figure S3) led to the highest regional production of 2.13 Mt across the different irrigation management options (Figure 7). The lowest regional production of 2.04 Mt was simulated when the very strong restriction for the upper and middle areas was used (i.e., the water intake was limited for the upper and middle areas, and the river water was maintained for the lower area). The difference in regional production associated with this irrigation management was 0.09 Mt or 4.2% of the highest production. Interestingly, the response of regional production to subregional irrigation management was weakly nonlinear (Figure 7).

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1418 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

(a) (b) (c)

012 −4 −2 0 2 4 Yield [t ha−1] Yield change [%] River flow direction (d) 5 (e) Upper 0 Middle (with) −5 Middle (without) −10 Production change [%] Lower (with) −15 Lower (without) (with) (with) (with) (with) Upper Upper Whole Whole (without) (without) (without) (without) Lower Lower Lower Lower Middle Middle Middle Middle

Figure 6. (a) Simulated soybean yields over the Songhua River watershed in the severe drought year 1982 under the assumption of a weaker restriction on river water intake for the entire watershed (baseline). (b) Changes in the simulated yields relative to the baseline (presented in Figure 6a) when a stronger restriction on river water intake is applied to the upper area and (d) the corresponding changes in soybean production in each area. (c) Changes in the simulated yields relative to the baseline when a stronger restriction on river water intake is applied to the middle area and (e) the corresponding changes in production in each area. The ‘‘with’’ (‘‘without’’) in parentheses means the parts of each area with (without) any upper river segments.

6. Discussion 6.1. Advantages and Disadvantages of the Coupled Model A strong beneﬁt for the PRYSBI-2 crop model achieved by this coupling is that the simulated information on the river water availability for irrigation at daily time steps is incorporated into crop growth simulation. For the H08 water resources model, the signiﬁcant improvements include increased representation of soil water dynamics as a result of the replacement of the leaky bucket model with the more sophisticated soil water balance model; increased representation of crop growth and yield formation dynamics achieved by the replacement of the SWIM crop model with the PRYSBI-2 model; replacement of a hypothetical planting date that gives the highest yield of a crop under a given weather condition [Hanasaki et al., 2008] with the more realistic crop planting date of Sacks et al. [2010]. A known disadvantage of the current version of the coupled model is that the model can simulate fewer crops. Whereas the H08 model considered 18 crops using the embedded SWIM crop model, the version of the PRYSBI-2 crop model used for this study simulated only soybean (this is in part ameliorated by using the latest version of the PRYSBI-2 model which deals with maize, soybean, rice and wheat). More broadly, the consideration of natural vegetation in the coupled model is lacking. This is a shortcoming compared to

land-surface models. Also the impacts of seasonal variations in atmospheric [CO2] on the water-use

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1419 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

efﬁciency of plant are less considered 2.12 in the coupled model (because of the use of annual [CO2] input) compared to terrestrial natural vegetation models 2.10 [Ito and Inatomi, 2012], although the

annual [CO2] concentration input is fre- 2.08 quently used in other crop models [e.g., Tao et al., 2003, 2009]. Restriction 2.06 in the middle area 6.2. Model Evaluation and Intermediate

Production [Mt] Uncertainties Weaker Very strong The results show a signiﬁcant drying 2.04 Very weak Strong trend in the soybean-growing season soil moisture in the Songhua River watershed in the last decades, whereas there was no clear trend in precipita- Weaker

Stronger tion during the same season (Figure

Very weak 4a). These results are consistent with Very strong Intermediate the findings of Yu et al. [2014], who reported a significant drying trend over Restriction in the upper area North China using the standardized precipitation evapotranspiration index Figure 7. Responses of the Songhua River watershed soybean production to 25 different irrigation management options, consisting of different levels of restric- and an insignificant decreasing precipi- tion on river water intake in the upper and middle areas of the watershed. tation trends in that region. These contrasting tendencies between precipitation and the drought index can be explained by the increased temperature and associated increase in the evapotranspiration rate. Despite decreased soil moisture, which could lead to lower yields because of more frequent, severer, and longer agricultural droughts, the soybean yield in the study area kept increasing, with some variations in the annual rate of increase in the yield across the data sources (Figure 4c). The increase in the yield was mainly accompanied with technological improvements stimulated by the agricultural policies of the Chi- nese national and local provincial governments [Liu, 2011]. However, the extent of the drought-damage cropland area and the severity of droughts in the study area seem to increase with time (Figure 4d). These changes in agricultural drought are qualitatively consistent with the decreased soil moisture. However, a caution is required when attributing these changes in agricultural drought to decreased soil moisture alone, as the cropland in this area has expanded over time to less-fertile areas or to areas with poor irrigation facili- ties [Jiao et al., 2007; Wang et al., 2011]. There are some limitations in the evaluation of the model’s performance. First, for all of the variables examined here (surface soil moisture, actual crop evapotranspiration, river discharge, and yield), the references were not available for the entire period. The actual evapotranspiration reference was available only for the later years (2000–2010), and river discharge data were available only for the earlier years (1979–1987). There was no overlapping period between these two variables (Figures 5b and 5c). This limited data availability prohibited us from evaluating the model’s performance in capturing the reference long-term trends in the variables (except soil moisture) and their covariations in drier and wetter years. Second, for all of the variables except river discharge, the references are modeled data rather than observations. The ECV-SM soil moisture is the product of merging the satellite data and land-surface model output [Liu et al., 2012; Wagner et al., 2012]. The MOD16 actual evapotranspiration product is also modeled data using the satellite-derived climates and leaf area index as the inputs [Mu et al., 2011], and part of the GDHY yield is the product of merging the satellite-derived net primary production and FAO yield statistics [Iizumi et al., 2014b]. Despite these limitations, we conclude that the overall performance of the model in capturing the major characteristics of soil moisture, actual crop evapotranspiration, river discharge, and crop yield across the soybean harvested area in the last decades is satisfactory accurate. Importantly, the model follows the reference drying trend in soil moisture, as well as in the year-to-year variation. In addition, the goodness-of-fit

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1420 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

statistics for the modeled and reference crop yields for the coupled model are similar to or better than those for other state-of-the-art large-area crop models [e.g., Tao et al., 2009]. The modeled soil moisture tends to be lower than that of the reference; therefore, the threshold aridity level to triggering irrigation (thrs) might vary by locations depending on local irrigation management. Also further analysis using in situ soil moisture data may be necessary to be more conclusive if this discrepancy between the modeled and reference soil moisture is due to a deficiency in the model. The fact that many of the references used here are modeled data may help explain the discrepancy between the modeled and reference data. Note, however, that the reliability of these products is intensively evaluated by comparison with in situ observations [Mu et al., 2011; Liu et al., 2012; Wagner et al., 2012]. Also these products are sufficiently reliable as the references when evaluating model simulations on a regional level, as discussed in previous studies [Loew et al., 2013]. If the reference and modeled data are simulated under similar conceptual models, it probably con- tributes to the better agreement between them. For crop yield, subnational crop statistics are in general expected to be closer to the actual yields, but the quality of yield statistics in the developing world is often questionable because of the frequent reporting of unrealistic yield values, which are suspected of be reporting errors [Monfreda et al., 2008; Ray et al., 2012; Iizumi et al., 2014b]. Therefore, it is not necessarily reasona- ble to consider the district crop statistics yield values used here as exclusively reliable compared to the GDHY yield estimates values. Different methods for spatial aggregation of the reference and modeled data (with and without crop harvested area) can lead to slightly different conclusion on the model performance for actual crop evapotranspiration, but not for soil moisture (Figure 5 and supporting information Figure S2). The difference in the original resolution between the references (0.258 for soil moisture and 0.00838 for actual evapotranspiration) and associated difference in the representativeness of cropland may explain the varying results, suggesting that the aggregation method with weighting may be more suitable to compare to large-area crop model simulation when a finer-resolution reference is available.

6.3. Potential of Subregional Irrigation Management to Increase Regional Crop Production The results show that the regional soybean production approaches the highest value when a very weak restriction on water intake is applied to the upper and middle areas of the watershed (Figure 7 and supporting information Figure S3). All of the restrictions in the upper and middle areas that were considered here decrease the regional crop production. This is not surprising because the relatively productive and exten- sive soybean harvested area is located in the upper area (Figure 6a). Given the geographical distribution of soybean productivity (including the difference between irrigated and rainfed conditions) and harvested area, in the current drought year, maintaining the yield in the upper area (by allowing as much water intake in the upper area as the crop demands) is more efficient than increasing yields in the middle and lower areas (by restricting the water intake in the upper area to make river water available for the middle and lower areas). The simulated influence on regional production of the different irrigation management approaches up to 4.2% of the highest value (Figure 7). A caution is needed as the value of 4.2% is derived from a single severe drought year, and a multiyear mean value (including wet years) may result in a smaller value, but this result suggests the potential of subregional irrigation management for improving regional crop production. As already stated, the response of regional production to subregional irrigation management was shown weakly nonlinear (Figure 7). The response curve can be explained via the response of soil water deficit to the irrigation management. As the irrigation restriction on the river water intake for the upper area was eased, the soil water deficit in the watershed decreased (Figure 8a). This was mainly caused by the increase of irrigated water in the upper area (Figure 8b). However, the excessive ease on the irrigation restriction (weaker and very weak) for the upper area brought reduction of the decrease rate in soil water deficit in the watershed (Figure 8a). The reasons for the reduction of the decrease rate in soil water deficit include; (1) the soil water deficits in the middle and lower areas increased at a bit accelerated rate because the river water available for the crop water demand in the middle and lower areas remarkably decreased by allowing as much water intake in the upper area (Figure 8b); (2) the frequency of depleted river storage in the upper and middle areas increased, which means the increase of situations that the irrigation water could not be applied to the soil surface due to no water in the river and then the crop could not be utilized irrigation water at an appropriate timing such as a drought condition. Therefore, the regional crop production

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1421 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

(a) 0.14

0.12

0.10 Soil water deficit

60 (b)

20 Supply / Demand [%] 0 55 (c) 50

5 Frequency of

depleted river storage 0 Weaker Stronger Very weak Very strong Intermediate Restriction in the upper area Upper Lower (with) Middle (with) Whole

Figure 8. Responses of (a) soybean soil water deficit, (b) soybean annual water supply and demand balance, and (c) annual frequency of occurrence of depleted river storage to five different irrigation restriction options in the upper area over the Songhua River watershed in the severe drought year 1982 under an intermediate irrigation restriction in the middle area. The soil water deficit data averaged over each area (weighted by the soybean production) are presented. The other data are presented as simple area-averaged value (non- weighted). The soil water deficit corresponds to the difference calculated by subtracting the crop water stress factor Wstress (supporting information Text S1.1) from one. The depleted river storage is represented by that the grid-cell river storage for the day approaches zero.

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1422 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

response curve suggests that an increase in regional crop production due to subregional irrigation management would be saturated at a certain level, i.e., the adaptation limit which has not yet widely quantiﬁed in the agricultural impacts assessment. Results presented here were based on a number of assumptions. Some assumptions, the idealized land use and the simple irrigation scheme, may be rather unrealistic for some parts of the study area. The use of these assumptions does not affect the value of this study because the main purpose of this study was to describe the beneﬁts of the coupled model through the simulation experiments. However, more realistic assumptions may be necessary when the model is applied to, for instance, the detection and attribution of climate change impacts in the real world.

7. Conclusions The CROVER model proposed in this study is designed to simultaneously simulate crop growth and yield, river discharge, their dynamic interactions, and the influences of subregional irrigation management on regional crop production. This model is a powerful tool to examine the potential and limitation of irrigation as the measure for climate change adaptation in a given region. The model accurately captures the major characteristics of hydrology and crop yield in the Songhua River watershed, Northeast China. The simulation experiments demonstrate that managing the locations at which irrigation is primarily applied has measurable influences on the quantity of regional crop production. Our finding reinforces the importance of simulating the river water availability and crop production in a single model so that one does not overestimate the climate change adaptation potential of crop production induced by irrigation.

Acknowledgments References The simulation data used in this study Alcamo, J., P. Doll,€ T. Henrichs, F. Kaspar, B. Lehner, T. Rosch,€ and S. Siebert (2003), Development and testing of the WaterGAP 2 global are available upon request to the model of water use and availability, Hydrol. Sci. J., 48(3), 317–337, doi:10.1623/hysj.48.3.317.45290. corresponding author (yokozawa@sys. Asian Development Bank (2005), People’s Republic of China: Songhua River Basin water quality and pollution control management: Situa- eng.shizuoka.ac.jp). This research was tional analysis, consultants’ report, Manila. supported by the Environment Biemans, H., R. Hutjes, P. Kabat, B. Strengers, D. Gerten, and S. Rost (2009), Effects of precipitation uncertainty on discharge calculations for Research and Technology main river basins, J. Hydrometeorol., 10(4), 1011–1025, doi:10.1175/2008JHM1067.1. Development Fund (S-10-2) of the Biemans, H., L. Speelman, F. Ludwig, E. Moors, A. Wiltshire, P. Kumar, D. Gerten, and P. Kabat (2013), Future water resources for food pro- Ministry of the Environment, Japan and the JSPS KAKENHI grant duction in five South Asian river basins and potential for adaptation—A modeling study, Sci. Total Environ., 468-469, S117–S131, doi: 26310305. We thank J. Jiao, X. Xu, and 10.1016/j.scitotenv.2013.05.092. H. Tan at the Heilongjiang Academy of Bondeau, A., et al. (2007), Modelling the role of agriculture for the 20th century global terrestrial carbon balance, Global Change Biol.,13(3), Agricultural Sciences for their 679–706, doi:10.1111/j.1365-2486.2006.01305.x. comments, Y. Ishigooka and T. Center for International Earth Science Information Network at Columbia University and Centro Internacional de Agricultura Tropical (2005), Kuwagawa at the National Institute for Gridded Population of the World Version 3 (GPWv3): Population Grids. Socioeconomic Data and Applications Center (SEDAC),ColumbiaUniv.,N.Y. Agro-Environmental Sciences for Challinor, A., J. Watson, D. Lobell, S. Howden, D. Smith, and N. Chhetri (2014), A meta-analysis of crop yield under climate change and providing the data for the drought- adaptation, Nat. Clim. Change, 4(4), 287–291, doi:10.1038/nclimate2153. damage cropland area, and S. Ohji at Deryng, D., W. Sacks, C. Barford, and N. Ramankutty (2011), Simulating the effects of climate and agricultural management practices on University of Tsukuba for her support global crop yield, Global Biogeochem. Cycles, 25, GB2006, doi:10.1029/2009GB003765. in satellite data handling. Doll,€ P., and H. Schmied (2012), How is the impact of climate change on river flow regimes related to the impact on mean annual runoff? A Acknowledgements are extended to global-scale analysis, Environ. Res. Lett., 7(1), 014037, doi:10.1088/1748-9326/7/1/014037. the reviewers for their valuable Elliott, J., et al. (2014), Constraints and potentials of future irrigation water availability on agricultural production under climate change, suggestions to improve the original Proc. Natl. Acad. Sci. U. S. A., 111(9), 3239–3244, doi:10.1073/pnas.1222474110. submission. Food and Agriculture Organization of the United Nations (2014), FAOSTAT, The Stat. Div. of FAO, Rome. [Available at http://faostat.fao.org/.] Hagemann, S., et al. (2013), Climate change impact on available water resources obtained using multiple global climate and hydrology models, Earth Syst. Dyn., 4(1), 129–144, doi:10.5194/esd-4-129-2013. Hanasaki, N., S. Kanae, and T. Oki (2006), A reservoir operation scheme for global river routing models, J. Hydrol., 327(1-2), 22–41, doi:10.1016/j.jhydrol.2005.11.011. Hanasaki, N., S. Kanae, T. Oki, K. Masuda, K. Motoya, N. Shirakawa, Y. Shen, and K. Tanaka (2008), An integrated model for the assessment of global water resources—Part 1: Model description and input meteorological forcing, Hydrol. Earth Syst. Sci., 12(4), 1007–1025, doi:10.5194/hess-12-1007-2008. Hanasaki, N., et al. (2013), A global water scarcity assessment under Shared Socio-economic Pathways—Part 2: Water availability and scarcity, Hydrol. Earth Syst. Sci., 17(7), 2393–2413, doi:10.5194/hess-17-2393-2013. Iizumi, T., M. Okada, and M. Yokozawa (2014a), A meteorological forcing data set for global crop modeling: Development, evaluation, and intercomparison, J. Geophys. Res. Atmos., 119, 363–384, doi:10.1002/2013JD020130. Iizumi, T., M. Yokozawa, G. Sakurai, M. Travasso, V. Romanenkov, P. Oettli, T. Newby, Y. Ishigooka, and J. Furuya (2014b), Historical changes in global yields: Major cereal and legume crops from 1982 to 2006, Global Ecol. Biogeogr., 23, 346–357, doi:10.1111/geb.12120. International Commission on Large Dams (1998), World Register of Dams, Paris. Ito, A., and M. Inatomi (2012), Water-use efficiency of the terrestrial biosphere: A model analysis focusing on interactions between the global carbon and water cycles, J. Hydrometeorol., 13, 681–694, doi:10.1175/JHM-D-10-05034.1. Jiao, J., X. Xu, Y. Hayashi, and F. Tao (2007), Overview of the impacts of climate warming on paddy rice production in Heilongjiang Province, the north most of China, Tsukuba Geoenviron. Sci., 3, 23–31.

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1423 Journal of Advances in Modeling Earth Systems 10.1002/2014MS000402

Jiao, J., et al. (Eds.) (2008), Development of Rural Economy and Increase in the Farmers’ Proﬁtability [in Chinese], 204 pp., China Agric. Press, Beijing. Jimenez Cisneros, B. E., T. Oki, N. W. Arnell, G. Benito, J. G. Cogley, P. Doll,€ T. Jiang, and S. S. Mwakalila (2014), Freshwater resources, in Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assess- ment Report of the Intergovernmental Panel on Climate Change, edited by C. B. Field et al., 76 pp., Cambridge Univ. Press, Cambridge, U. K.

Keeling, R. F., S. C. Piper, A. F., Bollenbacher, and J. S. Walker (2009), Atmospheric CO2 records from sites in the SIO air sampling network, in Trends: A Compendium of Data on Global Change, Carbon Dioxide Inf. Anal. Cent., Oak Ridge Natl. Lab., U.S. Dep. of Energy, Oak Ridge, Tenn. Krysanova, V., F. Wechsung, J. Arnold, R. Srinivasan, and J. Williams (2000), SWIM (Soil and Water Integrated Model) User Manual, 239 pp., Potsdam Inst. for Clim. Impact Res., Potsdam, Germany. Kummu, M., D. Gerten, J. Heinke, M. Konzmann, and O. Varis (2014), Climate-driven interannual variability of water scarcity in food production potential: A global analysis, Hydrol. Earth Syst. Sci., 18, 447–461, doi:10.5194/hess-18-447-2014. Liu, Y. Y., W. A. Dorigo, R. M. Parinussa, R. A. M. Jeu, W. Wagner, M. F. McCabe, J. P. Evans, and A. I. J. M. Dijk (2012), Trend-preserving blend- ing of passive and active microwave soil moisture retrievals, Remote Sens. Environ., 123, 280–297, doi:10.1016/j.rse.2012.03.014. Liu, Z. (2011), Analysis and suggestion of soybean production situation in Heilongjiang Province, in Studies on Farming System in Cold Region of China [in Chinese], edited by G. Han, pp. 31–35, China Agric. Press, Beijing. Loew, A., T. Stacke, W. Dorigo, R. Jeu, and S. Hagemann (2013), Potential and limitations of multidecadal satellite soil moisture observations for selected climate model evaluation studies, Hydrol. Earth Syst. Sci., 17(9), 3523–3542, doi:10.5194/hess-17-3523-2013. Manabe, S. (1969), Climate and the ocean circulation I. The atmospheric circulation and the hydrology of the Earth’s surface, Mon. Weather Rev., 97(11), 739–774. Ministry of Water Resources of the People’s Republic of China (MWR) (2015), China Water Resources Bulletin [in Chinese], Beijing, China. Monfreda, C., N. Ramankutty, and J. Foley (2008), Farming the planet: 2. Geographic distribution of crop areas, yields, physiological types, and net primary production in the year 2000, Global Biogeochem. Cycles, 22, GB1022, doi:10.1029/2007GB002947. Monteith, J. L. (1965), Evaporation and environment, Symp. Soc. Exp. Biol., 19, 205–223. Mu, Q., M. Zhao, and S. W. Running (2011), Improvements to a MODIS global terrestrial evapotranspiration algorithm, Remote Sens. Environ., 115, 1781–1800, doi:10.1016/j.rse.2011.02.019. National Bureau of Statistics of China (NBSC) (2014), China Statistical Yearbook, China Stat. Press, Beijing. [Available at http://www.stats.gov. cn/english/Statisticaldata/AnnualData/.] Neitsch, S. L., J. G. Arnold, J. R. Kiniry, and J. R. Williams (2011), Soil and water assessment tool theoretical documentation version 2009, Tex. Water Resour. Inst. Tech. Rep. 406, 618 pp., Tex. A & M Univ. Syst., College Station. Piao, S., et al. (2010), The impacts of climate change on water resources and agriculture in China, Nature, 467(7311), 43–51, doi:10.1038/ nature09364. Porter, J. R., L. Xie, A. Challinor, K. Cochrane, M. Howden, M. M. Iqbal, D. Lobell, and M. I. Travasso (2014), Food security and food production systems, in Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, edited by C. B. Field et al.,82 pp., Cambridge Univ. Press, Cambridge, U. K. Portmann, F., S. Siebert, and P. Doll€ (2010), MIRCA2000-Global monthly irrigated and rainfed crop areas around the year 2000: A new high-resolution data set for agricultural and hydrological modeling, Global Biogeochem. Cycles, 24, GB1011, doi:10.1029/ 2008GB003435. Ray, D. K., N. Ramankutty, N. D. Mueller, P. C. West, and J. A. Foley (2012), Recent patterns of crop yield growth and stagnation, Nat. Com- mun., 3, 1293, doi:10.1038/ncomms2296. Sacks, W., D. Deryng, J. Foley, and N. Ramankutty (2010), Crop planting dates: An analysis of global patterns, Global Ecol. Biogeogr., 19, 607– 620, doi:10.1111/j.1466-8238.2010.00551.x.

Sakurai, G., T. Iizumi, M. Nishimori, and M. Yokozawa (2014), How much has the increase in atmospheric CO2 directly affected past soybean production?, Sci. Rep., 4, 4978, doi:10.1038/srep04978. Scholes, R. J., and E. Brown de Colstoun (2011), ISLSCP II global gridded soil characteristics, in ISLSCP Initiative II Collection, edited by G. Forr- est et al., Oak Ridge Natl. Lab. Distributed Active Archive Cent., Oak Ridge, Tenn. Tao, F., M. Yokozawa, Y. Hayashi, and E. Lin (2003), Changes in agricultural water demands and soil moisture in China over the last half- century and their effects on agricultural production, Agric. For. Meteorol., 118(3-4), 251–261, doi:10.1016/S0168-1923(03)00107-2. Tao, F., M. Yokozawa, and Z. Zhang (2009), Modelling the impacts of weather and climate variability on crop productivity over a large area: A new process-based model development, optimization, and uncertainties analysis, Agric. For. Meteorol., 149(5), 831–850. The Global Runoff Data Centre (2014), The Global Runoff Database, Bundesanst. fur€ Gew€asserkunde, Koblenz, Germany. [Available at http:// www.bafg.de/GRDC/EN/Home/homepage_node.html.] Wagner, W., W. Dorigo, R. de Jeu, D. Fernandez, J. Benveniste, E. Haas, and M. Ertl (2012), Fusion of active and passive microwave observations to create an essential climate variable data record on soil moisture, paper presented at The XXII ISPRS Congress, Int. Soc. for Pho- togramm. and Remote Sens., Melbourne, Australia, 27 Aug. Wang, A., D. Lettenmaier, and J. Shefﬁeld (2011), Soil Moisture Drought in China, 1950–2006, J. Clim., 24(13), 3257–3271, doi:10.1175/2011JCLI3733.1. Wen, K., and Y. Sun (Eds.) (2007), The Meteorology Disaster Record Over China—Heilongjian Volume [in Chinese], 329 pp., China Meteorol. Press, Beijing. Yu, M., Q. Li, M. Hayes, M. Svoboda, and R. Heim (2014), Are droughts becoming more frequent or severe in China based on the Standar- dized Precipitation Evapotranspiration Index: 1951-2010?, Int. J. Climatol., 34(3), 545–558, doi:10.1002/joc.3701.

OKADA ET AL. MODELING IRRIGATION-BASED ADAPTATION 1424