GEPIC-V-R model: A GIS-based tool for regional crop drought risk assessment

Agricultural Water Management 144 (2014) 107–119

Contents lists available at ScienceDirect

Agricultural Water Management journal homepage: www.elsevier.com/locate/agwat

GEPIC-V-R model: A GIS-based tool for regional crop drought risk assessment Yuanyuan Yin a,b , Xingming Zhang a,c , Degen Lin c , Han Yu c , Jing’ai Wang a,c,d , Peijun Shi d,e,f,∗ a

The Key Laboratory of Regional Geography, Beijing Normal University, No. 19, Xinjiekouwai Avenue, Haidian District, Beijing 100875, China The Institute of Geographic Sciences and Natural Resources Research (IGSNRR), Chinese Academy of Sciences (CAS), 11A Datun Road, Anwai, Beijing 100101, China c The School of Geography, Beijing Normal University, No. 19, Xinjiekouwai Avenue, Haidian District, Beijing 100875, China d The State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, No. 19, Xinjiekouwai Avenue, Haidian District, Beijing 100875, China e Key Laboratory of Environmental Change and Natural Disaster, Ministry of Education, Beijing Normal University, No. 19, Xinjiekouwai Avenue, Haidian District, Beijing 100875, China f Academy of Disaster Reduction and Emergency Management, No. 19, Xinjiekouwai Avenue, Haidian District, Beijing 100875, China b

a r t i c l e

i n f o

Article history: Received 24 December 2013 Accepted 28 May 2014 Keywords: Large scale risk assessment model (GEPIC-VR model) Global Vulnerability curves Drought risk Maize

a b s t r a c t In recent years, food losses caused by drought accounted for approximately 60% of the total world food loss, seriously threatening the world’s food security and sustainable development. Against the background of frequent extreme climate events and “local warming and drying”, frequency and potential risks of global drought have tended to increase. As the scientific basis for disaster prevention and mitigation, disaster risk assessment has drawn widespread attention in the scientific community. Using the commonly used EPIC crop model, this study constructed a crop drought risk assessment model – GEPICV-R model – suitable for large regional scale, with functions to fit vulnerability curves and calculate risk. Additionally, global maize drought risk was assessed. From a global perspective, South Africa, Chile, Western and Central Europe, Russia and southeastern regions have elevated risks of maize drought; Chinese maize drought risk distribution is characterized by low risk in southern regions and high risk in northern regions. For once in 10- and 30-years, Pearson values between converted maize loss rate (CMLR) or Harikishan Jayanthi’s loss rate and loss rate are greater than 0.7, with a S.D. of 0.01. Rank correlation analyses of 28 provinces in China and seven countries in Africa generated Pearson, Kendall and Spearman values greater than 0.48, with a S.D. of 0.05. There was a close correlation between the results and statistical predictions or existing results. Therefore, the simulation results supply the theoretical support for acting based on local conditions to manage drought and drought risk. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Drought is one of the types of natural disaster with the widest effects, and it causes the greatest loss in worldwide agriculture, with nearly half the world’s countries suffering from severe drought (UNDP (United Nations Development Programme), 2004; IPCC (Intergovernmental Panel on Climate Change), 2007). With impacts of climate change and local warming and drying, droughts in

∗ Corresponding author at: Beijing Normal University, The State Key Laboratory of Earth Surface Processes and Resource Ecology, No. 19, Xinjiekouwai Avenue, Haidian District, Beijing 100875, China. Tel.: +86 10 58808179; fax: +86 10 58802158. E-mail address: [email protected] (P. Shi). http://dx.doi.org/10.1016/j.agwat.2014.05.017 0378-3774/© 2014 Elsevier B.V. All rights reserved.

the world have been occurring frequently, and their impacts are being aggravated. In the past 50 years, the global very dry areas have more than doubled (Dai et al., 2004). Especially in North America and Mexico, central and southern Africa and parts of South America, as well as northern China, drought is very serious (IPCC (Intergovernmental Panel on Climate Change), 2007). Based on drought disaster database provided by Center for Research on the Epidemiology of Disaster (http://www.emdat.be/), drought frequency increase in the global scale increased form less than once a year, between 1900 and 1959, to about 10 times a year, between 1960 and 2009. Between 2000 and 2009, in particular, drought occurred about 190 times, with 8.3 million affected people and 282 billion U.S. dollars loss per year on average caused by drought.

108

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

With the increase in frequency of extreme events, the management of extreme weather and climate events (i.e., droughts) based on risk assessment is an academic research hotspot (IPCC (Intergovernmental Panel on Climate Change), 2012). Disaster risk refers to the potential disaster losses of life, health status, livelihoods, assets and service systems. In the literature, different scholars and institutions give similar definitions (Crichton, 1999; ADRC (Asian Disaster Reduction Center), 2005), emphasizing the two basic aspects: possibility of disaster and possibility of expected loss. Therefore, agricultural drought risk, which is defined as the degree of harm to agriculture hazard-effected bodies caused by drought, can be considered as the probability of the occurrence of agricultural drought and the negative impact on agricultural production. According to data availability and model accuracy, crop drought risk assessment models can be divided into three types: non-quantitative risk assessment model, semi-quantitative risk assessment model, and quantitative risk assessment model (Shi, 2012). The above three types of risk are noted as risk level, risk grade, and risk. Risk level is calculated based on the index system. The core of the non-quantitative risk assessment model is the evaluation index system. The methods include weighted summation, determination matrix and disaster index (UNDP (United Nations Development Programme), 2004; IDEA and IDB, 2005; World Bank and Columbia University, 2005; Li et al., 2009). For example, Li et al. (2009) considered crop yield in drought risk as constituted by the drought frequency (DDF), drought intensity (DX), output level (PDL) (i.e., exposure level) and drought resilience (AV) (i.e., vulnerability) of the product. Risk grade is the ranking of disaster loss expectation through quantitative risk assessment and is computed based on the probability distribution of hazard and vulnerability matrix. Risk is defined as loss expectation calculated based on hazard intensity-probability distribution, vulnerability curve or matrix, and exposure. And this kind of risk assessment model pays more attention to the mechanism of crop vulnerability, and how building vulnerability curves. According to the elements contained in the assessment model, crop drought risk assessment models include the following three types: one-dimensional (crop hazard) crop drought risk assessment model, just considering hazard, chooses an appropriate crop drought index for evaluation and assesses the probability of occurrence, strength and impact area of drought (Mpelasoka et al., 2008; Qian et al., 2009; Dai, 2011). Alternative crop drought indexes include precipitation anomaly percentage, Bhalme–Mooley index, standardized precipitation index (SPI) (Thomas et al., 1993), available soil moisture storage capacity, palmer drought severity index (PDSI) (Palmer, 1965), crop moisture index (CMI) (Palmer, 1968), palmer moisture anomaly index (Z index) (Palmer, 1965), crop water stress index, water stress index and so on. However, these indicators just include precipitation, temperature and soil moisture, and can’t express formation mechanism of crop drought well. Two-dimensional crop drought risk assessment model defines risk as “the probability of harmful consequences or expected loss resulting from interactions between drought and vulnerable conditions”, and it is expressed by formula: risk = hazard × vulnerability (UN/ISDR (Unite Nations International Strategy for Disaster Reduction), 2004). At first most of two-dimensional crop drought risk assessment is to calculate risk level based on the index system of hazard, and vulnerability and exposure (Zhang, 2004; Li et al., 2009). With the crop model development and widespread application, it was used to calculate the loss risk of crop yield. And some scholars have constructed drought vulnerability curve of crop (i.e., rice, wheat and maize) at basins and national scales (He, 2010; Jia et al., 2012; Wang et al., 2013). Three-dimensional crop drought risk assessment is considered that crop drought risk intensity is co-determined by environmental instability, hazard intensity, and

vulnerability intensity and exposure degree of crop. Given the complexity of the relationship between the three elements, so far the existing studies are mostly focus on crop drought risk in the mountainous areas, i.e., southwest China and west of Henan Province in China (Zhou et al., 2012; Hu et al., 2012). And there have been no studies of crop drought risk with three-dimensional risk assessment model at national and global scale. Crop drought risk is related to three elements: environment, hazard and hazard-effected bodies (exposure and vulnerability). Water shortage is the direct cause of crop drought, which is one kind of stress phenomenon resulting when water demand exceeds supply. The main sources of water supply are atmospheric precipitation and irrigation water (surface water and groundwater). Water demand includes transpiration and evaporation, soil evaporation, and water for keeping the plant survive. Among those elements, atmospheric precipitation, irrigation, and soil evaporation are determined by the agri-environment. Water for keeping the plant survives is determined by crop itself. Plant transpiration and evaporation are affected by environment and crop’s biological physiological characteristics. Therefore, risk evaluation based on environment and crop drought vulnerability can provide theoretical base for the reduction of drought risk. Based on the environmental division, this article built a large regional-scale crop Drought Risk Assessment Model (GEPIC-Vulnerability-Risk model, GEPIC-V-R model). Based on this model, the method of constructing vulnerability curves through setting irrigation scenarios is explored. Finally, global maize drought risk was evaluated. The assessment result could be a scientific basis for maize drought risk management at global scale.

2. Development of GEPIC-V-R model 2.1. EPIC model and GEPIC model Building a quantitative drought risk assessment model for crops on regional scale is based on crop modeling. To meet the objectives of this study, a crop growth model needs to possess the following features: (1) flexibility for the simulation of different crops under a variety of climatic conditions, (2) ability to simulate yield and water stress, (3) minimum data requirements and (4) in-depth research in spatialization. The following five models were considered and compared: DSSAT, CropWat, WOFOST, CropSyst and EPIC. DSSAT combines a series of crop models for specific crops (IBSNAT (International Benchmark Sites Network for Agrotechnology Transfer), 1989). CropWat is one type of manmachine interactive operation interface of the system, and its code is not freely available. WOFOST describes crop physiology well but requires detailed input data (Monteith, 1996). CropSyst is not suitable for simulating rice because its rice parameters are not well calibrated (Confalonieri and Bocchi, 2005). Compared to the above four crop models, EPIC model, short for Erosion-Productivity Impact Calculator model, which was developed in the early 1980’s to assess the effect of erosion on productivity (Williams et al., 1984), was selected as the basic tool in this study for many reasons as following: (1) EPIC uses a unified approach to simulate more than 100 types of crops (Williams, 1995). (2) The model has continuously evolved since the 1985 RCA application, and the newest up to date version is EPIC 0810. Its code can be downloaded for free from the website of Texas A&M Agrilife Research (2012). (3) Plant growth is one of the most important physically based components in EPIC. Water stress and yield, which are used in constructing drought vulnerability curves and calculating drought hazard (caused by water shortage), can be simulated for all considered crops with one crop growth model using unique parameter values for each crop. (4) EPIC operates on a continuous

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

109

Fig. 1. The framework of GRPIV-V-R model.

basis using a daily time step and can perform long-term simulations for hundreds of years (Sharpley and Williams, 1990). (5) Because of its good performance in simulation, it has been widely applied in the United States and in other regions of the world across a broad spectrum of environmental conditions (Tan and Shibasaki, 2003; Gassman et al., 2005; Wriedt et al., 2009; Velde et al., 2010) to simulate soil erosion losses from wind erosion, soil carbon sequestration, climate change impacts on crop yield and erosion and so on. In recent years, EIPC has been used to simulate the impact of disaster on crop, crop-water relations and drought risk (Liu et al., 2007; Velde et al., 2012; Jia et al., 2012). Velde et al. (2010, 2012) investigated impacts of drought and heat stress on maize and wheat and estimated irrigation use and effects on maize yield during heat wave in France using EPIC. Jia et al. (2012) evaluated maize drought disaster quantitative risk in China based on EPIC. However, drought disaster quantitative risk of crop hasn’t been assessed at global scale. However, EPIC model can’t simulate spatial variability of crop yield, plant growth, water use and so on, because the drainage area considered by it is generally a field sized area (up to 100 ha) with a hypothesis that weather, soils, landscape, rotation, and management system are assumed to be spatially homogeneous (Sharpley and Williams, 1990). In order to use site-specific crop model at large geographic scales, Thorton (1991) discussed the possibility of linking GIS with crop models on regional level. Then, spatialization research of EPIC is in-depth through integrating with GIS techniques, i.e., “Spatial-EPIC” (Priya and Shibasaki, 2001), “EPIC-View” (Rao et al., 2000), “GEPIC” (Liu et al., 2007; Liu, 2009). GEPIC model (an extension of original EPIC), was developed linking ArcGIS (a Geographic Information System) with EPIC between 2004 and 2005 by the Swiss Federal Institute of Aquatic Science and Technology (Eawag) to address spatial variability of crop yield, plant growth, water use and so on from one place/region to other (Liu et al., 2007; Liu, 2009). In the GEPIC model, ArcGIS is used as not only an application framework and map displayer but an input editor and invoker. GEPIC, which treats each grid cell as a site,

simulates the crop-related processes (i.e., crop growth, water use) for each predefined grid cell at regional (i.e., regional, national, and global) scale by invoking spatially distributed inputs provided to the model in terms of GIS raster maps as well as text files. The necessary input include maps and data of many environmental elements: elevation and slope maps, irrigation maps, fertilizer maps, daily weather data and monthly weather statistics, soil parameters, etc. 2.2. GEPIC-V-R model GEPIC-V-R model, short for GEPIC-Vulnerability-Risk model, is a large scale (i.e., regional, national, continental, and global) crop risk assessment model. The formulas to calculate hazard, vulnerability and risk were written in VBA and MATLAB and were embedded into GEPIC (Fig. 1). Section 2.2.2 Vulnerability curve fitting and risk calculation detailed how to calculate hazard, vulnerability and risk based on simulated long term water stress and crop yield. GEPIC-V-R model includes four core modules: crop yield simulation, determination of crop parameters for each zone, construction of drought vulnerability curve of crop and risk calculation. Six types of input data are used for the GEPIC-V-R model: (1) climate data, (2) DEM and slope, (3) soil physical parameters, (4) land use data, (5) phenological data and (6) management data, such as irrigation and fertilizer application. 2.2.1. Methods for estimating yield and water stress 2.2.1.1. Yield. Crop yield is calculated by harvest index (HI), biomass-energy ratio (WA), leaf area index (LAI) and growth period length (Williams, 1995), as follows: YLD = HI ×

n  WA × RAi × [1 − exp (−0.65 × LAIi )] i=1

5000

,

(1)

where YLD is crop yield in kg/ha, HI is harvest index, WA is the biomass-energy ratio in (kg/ha)(MJ/m2 ), RA is solar radiation in

110

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

MJ/m2 /d, LAI is leaf area index, and n is the growth period length in days. In non-stressed conditions, harvest index increases nonlinearly from 0 at planting to HI at maturity according to the following equations (Williams, 1995):



HIAi = HIj

i 



(2)

HUIi HUIi + exp (11.1 − 0.1HUIi )

(3)

where, HIAi is the harvest index on day i, HIj is the harvest index for crop j, HUFH is the heat unit factor that affects harvest index, HUI is the heat unit index. For stressed conditions, harvest index is a function of heat and water use rate according to the following equation (Williams, 1995): HIA∗ = (HIA − HIA0 )

WUR + HIA0 WUR + exp (6.13 − 0.0883WUR)

(4)

where, HIA* is the final estimate of harvest index applied to above-ground biomass to estimate yield, HIA is the simulated potential harvest index, HIA0 is the minimum harvest index. The value of HIA is simulated daily using the following equation: HIAi = HIi



100HUIi 100HUIi + exp (11.1 − 10HUIi )



(5)

where, HIAi is the harvest index on day i, HIi is the potential harvest index on day i, and HUIi is the heat unit index on day i.



WUR = 100

K U i=1 i K E i=1 pi





(6)

2.2.1.2. Water stress (WS). The water stress (WS) factor is simulated as a function of supply and demand of water during the period of plant growth in the following equations (Williams, 1995):

M

EP =

u i=1 i,l Epi

Eo LAI 3

EP = Eo

0 ≤ LAI ≤ 3.0

LAI > 3.0

i=1

WSji



(10)

max (DHIk )

where, DHIjk is the drought hazard index of k grid in j year, WSji is the water stress of i day (water stress is the greatest of all stress) of k grid, n is the number of days affected by water stress during growing season, and max (DHIk ) is the maximum cumulative stress during growing season for all scenarios of k grid. Crop loss rate for certain hazard intensity under the influence of environment is calculated as the following formula: LRij =





max Lj − Lij max Lj

,

(11)

where LRij is the loss rate of a crop induced by drought of j grid in i year, Lij is the crop yield under different irrigation scenarios of j grid in i year, max(Lj ) is the yield under the optimal scenario of j grid. 2.2.2.2. Risk (R). The model provides three types of unit drought risk calculation: 0.5◦ × 0.5◦ grid, country unit and approximately equal area unit. According to regional disaster system theory (Shi, 2002), without considering the various drought mitigation capabilities, when exposure is one, drought risk is the function of hazard and vulnerability under influence of environment (Eq. (12)).



R = f (E, H, V ) = H < P, hE > × V

where, WUR is the simulated water use ratio, U is actual plant water use rate for day i, Epi is the potential water use rate for day i.

WSi =

n

DHIjk =

HUFHk

k=1

HUFHi =

Drought hazard index under the influence of the environment is the normalized cumulative water stress during the crop’s growing season (Eq. (10)). Drought hazard index ranges from 0 to 1, with larger values representing greater intensities of drought hazard.

(7) (8) (9)

where, WSi is the water stress factor for day i, ui, l is the water use rate in soil in layer l for day i, Ep is the potential water use rate for day i. Ep is computed from Eo and LAI. The change in LAI is split into two phases: from emergence to the start of leaf decline and from the start of leaf decline to the end of the growing season. 2.2.2. Vulnerability curve fitting and risk calculation 2.2.2.1. Vulnerability curve. Drought vulnerability curves were built using the method of simulated drought stress by a controlling management file (OPS file). The model controls the water stress by increasing or decreasing the amount of irrigation. The yield and water stress change along with decreases in irrigation. In this study, the optimal scenario, in which growing season water stress index is 0, produces the maximum potential yield, and the killing crop scenario, in which WS is 1, makes the crop fail. According to the given irrigation scenarios, crop yields and water stress under different scenarios are simulated with a daily time step. Then, drought hazard index and crop loss rate under the influence of the environment are calculated. This allows the fitting of crop drought vulnerability curves based on logistical functions.



hE , lE

(12)

where, E is the sensitivity of disaster environment; H is hazard risk; V is vulnerability; P is the probability of occurrence; hE is hazard index under influence of environment; and lE is loss rate under   influence of environment. H{ P, hE } is the hazard intensity under the certain probability. V {hE , lE } is the function of hE and lE . In the research, water stress of each day during the growth stage and yield for different irrigation scenarios could be simulated with the GEPIC-V-R model. Based water stress of each day, DHI was calculated with Eq. (10), and loss rate relative to yield under the sufficient irrigation conditions was calculated with Eq. (11). Therefore, the function of DHI and LR (vulnerability curve) was constructed with GEPIC-V-R model. Meanwhile, we simulated the water stress of each day during the growth stage from 1975 to 2004 by inputting meteorological data. 30 years of DHI data was computed with Eq. (10). The probability density function and the cumulative probability density function were built with Kernel density estimation method (The method has been embedded into GEPIC-V-R model in BVA and Matlab.), and the hazard intensity under the certain probability was calculated. With Eq. (12), the loss rate caused by hazard under the certain probability (risk) was calculated. For the formula of risk index of country unit and approximately equal area unit, see the following equation.

n

Rij =

R k=1 ijk ni

(13)

where, Rij is drought risk index of i unit for year j and ni is the number of grid cells in i unit. 3. Risk assessment of maize with GEPIC-V-R model 3.1. Data The sources and the spatial and temporal resolutions of the data used in this study are listed in Table 1. Climate data include two components: daily weather data and monthly weather statistics

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

111

Table 1 Dataset used in this paper. Datasets

Content

Spatial and temporal resolution and source

Foundational geographic information

World Administrative Region, river, lake, main city etc. ARC5 The digital soil map of the World Global soil profile data

ESRI company, national administration of surveying, mapping and geo-information, CRU TS 2.1 and DIVA-GIS 1971–2004, 0.5◦ × 0.5◦ 2007, FAO-UNESCO

Climate data Soil data

Land-use data DEM data Slope data Fertilizer data Irrigation data

Global land-use data Global DEM data Global slope data Global fertilizer data Global irrigation data Global maize yield Chinese maize yield

Crop yield data

Phenological data

American yield Australian maize yield Indian maize yield Global maize sown date

data of mean temperature, minimum temperature, maximum temperature, solar radiation, precipitation, wind velocity and relative humidity. The daily weather data for the period 1971–2007 was obtained from ARC5 from IPCC. The resolution is 0.5◦ . WXPARM, which is EPIC’s own software, can calculate the monthly weather statistics data for simulating a long time series yield. The digital elevation model (DEM) was obtained from the 0.0833-degree digital elevation model GTOPO30 of the US Geological survey (USGS). Terrain slopes were derived from FAO-GAEZ (global agro-ecological zones) slope map at an original scale of 0.0833 degrees. Raster data (0.5 degree resolution) were gained by resampling with ArcGIS. The soil data include general parameters and soil profile characteristics to describe the physical and chemical properties. They are derived from the FAO-UNESCO Soil Map of the World (SMW) at an original scale of 1:5 million, which is one of the most comprehensive soil maps with global coverage (Nachtergaele, 1996). Soil parameters are provided by the Digital Soil Map of the World (DSMW) (FAO (Food and Agriculture Organization of the United Nations), 2003). These data sets include parameters such as depth, percentage of sand, slit and clay, soil bulk, PH, organic carbon content, calcium carbonate content. Other parameters are included with the default setting. The maize distribution with 0.5-degree precision was obtained by overlaying rain-fed and irrigated field raster on maize yield raster. The maize yield data was provided by FAO, with 0.5-degree resolution. The rain-fed and irrigated field data was extracted from land use data, which were downloaded from the United States Geological Survey (USGS) divided into 26 categories, with 0.00833degree resolution. To be consistent with other data in spatial precision, we gained land use data with 0.5-degree through resampling with ArcGIS. Phenological information of crops used in the GEPIC-V-R model includes sowing date and growing season length. A database of maize phenology with 0.5-degree precision was constructed based on the map of the maize planting date and length of the growing period (Cui, 1994). The build process is as follows: (1) scan the map of maize plating date and length of growing period in world; (2) register, vectorize and import Attribute Information with ArcGIS, then obtain the Shape format fertility of data; (3) obtain 0.5-degree Raster data through the equivalent line interpolation method; (4) read sowing date and length of growing period of each grid through GEPIC-VR model read operation; (5) build a global database of maize phenological information. The database has four fields: longitude, latitude, sowing date and length of growing period. The sowing date uses a 365-day calculation system.

International Soil Reference and Information Centre—World Inventory of Soil Emission Potentials, 2000 0.00833◦ × 0.00833◦ , USGS 0.0833◦ × 0.0833◦ , USGS 0.0833◦ × 0.0833◦ , USGS 1961–2002, country average, FAO 2000, country average, Kassel university in Germany 1961–2010, country average, FAO 1975–2008, province average, Chinese Farming Information Network 1992–2010, state average, United States Department Agriculture 1995–2004, state average, Australian Bureau of Statistics 2001–2002, state average, Agricultural Statistics at a Glance World Agriculture Climate and Crops Climate (Cui, 1994)

Because detailed fertilizer and irrigation amounts, among other details, are difficult to obtain, this study used default values for various parameters. We assume that the irrigation water use and fertilizer use are equally distributed in each country. AQUASTAT provides the data for irrigation and fertilizer amount in individual countries. Raster data (0.5 degree) of fertilizer and irrigation in 2000 were constructed with ArcGIS. 3.2. Model calibration This paper determines the crop parameters of each maize suitability zone (Table A1 and Fig. A1 show the zones of maize suitability in detail.), and the comparative analysis method was applied to validate the model. To reflect the real production level, we added the layer of planting zone, and each grid OPS file to determine the parameters. The operation is shown in Fig. 2: (1) Building the crop parameters library. The EPIC model contains more than fifty crop parameters, including energy conversion rate (WA), harvest index (HI), the lowest harvest index (WSYF), the optimal growth temperature (TOP) and the lowest growth temperature (TBS). Based on literature research (Cabelguenne et al., 1999; Huang et al., 2006; Jonghan et al., 2009; Gaiser et al., 2010), the paper selected three sensitive parameters – WA, HI and WSYF – to define the crop parameters of each zone. The parameter adjustment scheme is in accordance with the original parameters (Table 2) to generate the 0.7, 0.8, 0.9, 1.0, 1.1, 1.2 and 1.3 times parameterization scheme libraries. To ensure the relationship between the lowest harvest index and harvest index, the present study suggests that the harvest index = lowest harvest index +0.1. Thus, total parameter sets are 49. (2) Yield simulations of each parameterization schemes. With other input parameters unchanged, the yield of each parameterization scheme was simulated. (3) Verify simulation and build the parameter for each zone. Comparing simulation yield and statistical yield, we selected the parameterization scheme with the smallest RMSE (root-meansquare error) as the optimal parameter programs to create partition parameter library (Table 2). According to the obtained data, yields in 2000–2002 on the national or provincial scale were used to determinate the crop parameters, and yields in 2002–2004 were used for validation (Fig. 3). The RMSE is 1.1103. From the spatial distribution of

112

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

Fig. 2. Flow chart to determine crops parameters zone by zone. RMSE: root-mean-square error.

Table 2 Maize parameters for each zone.

3.3. Drought vulnerability curve of maize

Zone

WA

HI

WSYF

Zones

WA

HI

WSYF

Prime parameters 101 102 103 104 105 106 107 108 109 110 111 201 202 203 204 205 206

40 28 24 24 24 24 24 24 24 28 28 40 44 24 24 32 24 28

0.5 0.35 0.3 0.6 0.3 0.35 0.3 0.35 0.3 0.3 0.3 0.3 0.7 0.65 0.45 0.65 0.4 0.3

0.4 0.25 0.2 0.5 0.2 0.25 0.2 0.25 0.2 0.2 0.2 0.2 0.6 0.55 0.35 0.55 0.3 0.2

301 302 303 304 305 401 402 403 404 501 502 503 504 505 601 603 604

36 28 24 56 40 24 24 28 44 24 24 24 24 24 36 36 28

0.65 0.65 0.6 0.35 0.3 0.3 0.3 0.3 0.65 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.7

0.55 0.55 0.5 0.25 0.2 0.2 0.2 0.2 0.55 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.6

Note: HI: harvest index; WA: biomass-energy ratio; lowest harvest index (WSYF).

Fig. 3. Comparison of statistical yield and simulation yield (2002–2004).

simulation and statistical yield in 2000, the high value areas were consistent, mainly in the United States, China, Chile, Australia, Western Europe and West Asia. We consider that the simulation accuracy can achieve a satisfactory level, and the parameter can be carried out a long time series of model simulations. However, due to errors in the data itself, data entry errors, crop parameters and the gap between field management practices and data with local, there are deviations between the simulation and statistical yields. Increasing the accuracy of these data will greatly improve the simulation accuracy of the GEPIC-V-R risk model.

According to the determined parameters and the generated OPS file library, the GEPIC-V-R model was used to simulate waster stress each day and yield of maize for different growing areas. Eq. (12) that considers drought hazard index and yield loss rate was used to calculate risk with the results of the GEPIC-V-R simulation. Finally, through the vulnerability-curve fitting platform of the GEPIC-V-R model, maize drought vulnerability equation of each region was regressed (Fig. 4).

3.4. Analysis of results Two maps of global risk of drought probability were produced (Figs. 5 and 6). Catastrophe risk with low probability is mainly distributed in 40◦ N–50◦ N and close to 20◦ S. Disaster risk with higher probability is mainly distributed in the lower latitudes, near the equator. The distribution area of the world’s largest maize disaster risk index is mainly in the range from 0 to 0.1, which are all micro maize drought disaster risk impact areas. For the 10-year and 30-year disaster return periods, areas affected by extremely light drought were 57.72% and 52.88% of the world’s maize-planting area. Areas affected by extremely severe and severe drought were 24.13% and 28.26% of the global maize-planting area. With all return periods for the risk, the highest index of maize drought, reaching more than 40, is present in South Africa, Chile, Western and Central Europe, Russia and Southeast Asia. There is a high risk in West Asia, northern China and northwestern India, of 24.13% in the 10-year return period, increasing to 26.12% over a 30year return period. In the Americas and Australia, maize drought disaster risk level is low. There are larger areas of high risk in South Africa, Asia and especially Europe, where the risk level is highest. China’s maize drought disaster risk is lower, showing a north-to-south spatial pattern. Most areas are micro-degrees or mildly affected maize drought disaster areas where the disaster risk index is less than 15. Moderate and severe drought disasteraffected zones for maize are distributed in China’s Northeast and North China regions, which resembles a northeast - southwest strip, distributed roughly parallel to the 400 mm isohyet.

3.5. Validation Based on the accessibility of data, we chose China, Kenya, Malawi and Mozambique as verification regions in this paper. Our results show that the evaluation results accurately reflect the spatial pattern of maize drought risk. Based on areas affected, inundated and demolished by drought in data released by China’s Ministry of Civil Affairs and on the crop sown area published by the National Bureau of Statistics of China, in provincial units, 17 provinces were selected, with rates of maize sown area and crop sown area above 0.1. This paper converted those data to maize loss rate to compare to the assessment result for China (Xu and Zhang, 2011; Xu et al., 2011; Wang and Zhang, 2013), using the following

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

113

Fig. 4. Maize drought vulnerability curve for each zone. Figure (a) and (b) is for Asia; figure (c) and (d) is for Europe; figure (e)–(h) is for North America, South America, Africa and Australasia, respectively. 101: Northern Central Asia temperate irrigated zone; 102: Southern Central Asia temperate zone; 103: West Asia temperate zone; 104: South Asia tropical rain-fed zone; 105: Southern Northeast Asia temperate rain-fed zone; 106: China’s Huang-Huai-Hai zone; 107: Central China sub-tropical-warm-temperate rain-fed zone; 108: Northwestern Southeast Asia sub-tropical rain-fed zone; 109: South China sub-tropical rain-fed zone; 110: Southeast Asia tropical rain-fed zone; 111: Northeast Asia-Japan warm temperate rain-fed zone; 201: Western European Plain temperate rain-fed zone; 202: Central European Plain temperate rain-fed zone; 203: Alps-Karbala mountain and East European Plain temperate rain-fed zone; 204: Iberian-Apennine temperate rain-fed zone; 205: Balkan temperate rain-fed zone; 206: VolgaUrals temperate rain-fed zone; 301: North America temperate irrigated zone; 302: North America temperate rain-fed zone; 303: North America warm temperate rain-fed zone; 304: North America sub-tropical irrigated zone; 305: North America sub-tropical rain-fed zone; 401: Latin America tropical rain-fed zone; 402: South America tropical rain-fed zone; 403: South America sub-tropical rain-fed zone; 404: South America temperate zone; 501: Africa tropical irrigated zone; 502: Africa tropical rain-fed zone; 503: Africa sub-tropical rain-fed zone;504: Africa sub-tropical irrigated zone; 505: Madagascar zone; 601: Oceania tropical rain-fed zone; 603: Oceania sub-tropical irrigated zone; 604: Oceania temperate zone; 605: Oceania sub-tropical rain-fed zone.



equation.(14)

Li = =















Ajsi × Yi × Ai /Si × Ljsi + Aczi − Ajsi × Yi × Ai /Si × Lczi + (Aszi − Aczi ) × Yi × Ai /Si × Lszi

Yi × Ai

 Ajsi × Ljsi + Aczi − Ajsi × Lczi + (Aszi − Aczi ) × Lszi

Si Li is converted maize loss rate of province i; Ajsi is the crop area demolished by drought of province i; Aczi is the crop area inundated by drought of province i; Aszi is the crop area affected by drought of province i; Yi is the maize yield of province i; Ai is maize sown area of province i; Si is crop sown area of province i; Ljsi is coefficient of yield reduction of demolished crop area of province i; Lczi is coefficient of yield reduction of inundated crop area of province i; and Lszi is coefficient of yield reduction of affected crop area of province i.

here,

The natural-disaster statistical system defines sown areas suffering more than 10%, 30% and 80% reduction from drought as the affected crop area, the inundated crop area and the demolished crop area, respectively. Therefore, in this study, Ljs , Lcz and Lsz are 20%, 55% and 90%, medium values for 10–30%, 30–80% and 80–100%, respectively. The correlation analysis between converted maize loss rate (CMLR) and loss rate indicated the loss rate of maize, once every 10 years and 30 years, respectively, had significant and positive correlations with CMLR (Fig. 7). The correlation coefficients are

114

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

Fig. 5. Maize drought risk (once every 10 years).

0.777 and 0.744, respectively, and all were statistically significant at the confidence level of 0.01. Kenya, Malawi, Mozambique and Niger are selected typical areas in Africa. Harikishan Jayanthi & Greg Husak have evaluated maize drought risk in these regions, and they calculated maize

loss rate in different return periods for different countries and provinces (Jayanthi and Husak, 2013). The paper analyzed correlations between the calculated loss rate and the previously calculated loss rate in Kenya, Malawi and nine provinces in Mozambique. There is positive correlation for each return period (Fig. 8). The

Fig. 6. Maize drought risk (once every 30 years).

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

115

Fig. 8. Comparison between simulated loss rate and Harikishan Jayanthi’s loss rate in Kenya, Malawi and nine provinces in Mozambique.

Fig. 7. Comparison between relative affected area rate and loss rate in 17 provinces in China.

correlation coefficients are 0.745 and 0.765 for once every 10 years and 30 years, respectively, with all probabilities significant at a confidence level of 0.05. 4. Conclusions and discussion 4.1. Discussion Our model was built based on GEPIC which was built by Junguo Liu (Liu et al., 2007; Liu, 2009), and was developed under the guidance of the regional disaster system theory (Shi, 1991, 2002). That model, which is a useful tool for regional crop simulation and drought risk assessment, has functions to fit vulnerability curve and calculate risk. Comparing the simulated yield with statistical yield from 2002 to 2004 for various countries and then evaluating existing studies, our results indicate that our model behaves well in simulating yield and assessing risk. Adjusting crop parameters zone by zone, high spatial resolution crop phenological information and validation data increase the resolution of the simulation results. In recent years, many scholars have researched simulation yields with crop models (Williams et al., 1989; Tan and Shibasaki, 2003; Liu et al., 2007) and have evaluated crop drought risk at global and national scales (Velde et al., 2012; Jia et al., 2012; Jayanthi and Husak, 2013). Although these results reach the given confidence level, all of the studies use the same set of parameters within the

study area. For the global level and for large countries such as China, this hypothesis cannot represent regional differences in varieties of crops. For crop sowing date and length of growing period settings, some authors presumed that the sowing date and length period of growing period are the same across regions (Jia et al., 2012). Others identified the month which has the largest yield over 12 months as the sown date (Liu, 2009). No matter what method is used, this hypothesis is unable to show the gap between different regions. Meanwhile, on the global scale, statistical yields in national units were used to validate the model, but it is unreasonable for large country. In our study, we obtain a good precision of simulation by automatically adding determination of the partition parameter module, using sowing date and growth stage length data with 0.5 degree and validating with provincial yields. In addition because the EPIC model includes maize, wheat, rice, cotton and other dozens of food and cash crops germplasm information, the GEPIC-V-R model can be used for drought risk assessment of wheat, rice, cotton and other crops at any regional scale. In the quantitative study of crop drought vulnerability, although some scholars have constructed drought vulnerability curves for major crops (i.e., maize and wheat) with rainfed and irrigated scenarios (Jia et al., 2012; Wang et al., 2013), existing models do not remove the effect of temperature stress on crop drought. Crop growth is simultaneously affected by moisture, temperature, nutrients and other common elements, and the relationship between yields with stress is non-linear. Therefore, this study addressed the relationship between water stress and yield loss by setting the most suitable temperature, nutrients and ventilation to meet the stress scenario. Maize is one of the most widely distributed crops in

116

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

Table 3 Correlation between ranks of converted maize loss rate (CMLR) and loss rate in China.

Pearson

Kendall’s tau b Spearman’s rho

Pearson Correlation Sig. (2-tailed) Pearson Correlation Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed)

CMLR

CMLR

y10

1

0.666** 0.000

CMLR

1

y30

0.666** 0.000 1.000 . 0.487** 0.000 1.000 . 0.666** 0.000

y10 CMLR y10 CMLR y10

0.487** 0.000 1.000 . 0.666** 0.000 1.000 .

CMLR

Y30

1

0.720** 0.000

0.720** 0.000 1.000 . .513** 0.000 1.000 . 0.720** 0.000

CMLR y30 CMLR y30

1 0.513** 0.000 1.000 . 0.720** 0.000 1.000 .

Note: Sample size is 28. ** Correlation is significant at the 0.01 level (2-tailed).

Table 4 Correlation between ranks of GDP maize and loss rate in Africa. y10

Pearson

Kendall’s tau b

Spearman’s rho

Pearson Correlation Sig. (2-tailed) Pearson Correlation Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed) Correlation Coefficient Sig. (2-tailed)

GDP maize

y10

GDP maize

1

0.844* 0.017

*

y10 GDP maize y10 GDP maize y10 GDP maize

1 0.791* 0.034 1.000 . 0.905** 0.004 1.000 . 0.964** 0.000

0.791 0.034

y30

1

GDP maize

0.905** 0.004 1.000 . 0.964** 0.000 1.000 .

y30 GDP maize y30 GDP maize

0.844* 0.017 1.000 . 0.905** 0.004 1.000 . 0.964** 0.000

1 0.905** 0.004 1.000 . 0.964** 0.000 1.000 .

Note: Sample size is 8. * Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).

the world. The extreme differences in the geographic environment of maize agriculture result in differences in maize varieties and drought resistance. We added a regional crop parameters determination module to GEPIC-V-R to get maize parameters of each environmental zone. Compared with GEPIC which uses only one specific set of parameters, GEPIC-V-R model can show regional differences in maize. Based on these, the study fitted the maize vulnerability curve to drought through controlling for the irrigation volume, and it completes the quantitative drought risk assessment of maize. Over all, the vulnerability curves for each continent show smaller differences in Europe, Asia and North America, while it shows bigger differences in Africa. Correlation analysis of maize loss rate of China and Kenya, Malawi and Mozambique, chosen based on data limitations, showed that the GEPIC-V-R model worked well in zones 105, 106, 107, 109, 501 and 503. To ensure the validity and accuracy of the research, three correlation coefficients (Pearson, Kendall and Spearman) between the new results and the previous results or statistics were calculated to validate the model results. Until now, global drought risk assessment has mainly concentrated on populations and economies (UNDP (United Nations Development Programme), 2004; World Bank and Columbia University, 2005). Li et al. (2009) evaluated the drought risk of major crops with a weighted summation method, determining the drought risk relative rank of each nation. We therefore selected China and Africa for validating the assessment results. For China, 28 provinces were chosen for correlation analysis between ranks of converted maize loss rate (CMLR) and loss rate. The results showed significant correlations (Table 3). For once every 10 years, Pearson, Kendall and Spearman values are 0.666, 0.487 and 0.666, respectively, with a S.D. of 0.01. For once every 30 years, Pearson, Kendall and Spearman

values are 0.720, 0.513 and 0.720, respectively, with a S.D. of 0.01. Raster data for economical exposition to drought events from 1980 to 2001 provided by UNDP was used in Africa (UNEP (United Nations Environment Programme) and UNISDR (United Nations International Strategy for Disaster Reduction), 2013). GDP, cereal harvest area and production, price and harvest area from 1991 to 2000 are downloadable from the website of Food and Agriculture Organization of the United Nations (FAO) (FAO (Food and Agriculture Organization of the United Nations), 1991–2000). Total value of maize exposed to drought events (GDP maize) was derived with Eq. (15). Of the 26 countries evaluated, seven countries were selected. These seven were in the top 15 ranked by loss rate, and their maize harvest area ratios (divide maize harvest area by cereal harvest area) were greater than 50%. Correlation analysis between ranks of (GDP maize) and loss rate showed that there were significant correlations (Table 4). For once every 10 years, Pearson, Kendall and Spearman values are 0.791, 0.905 and 0.964, respectively, with a S.D. of 0.01. For once every 30 years, Pearson, Kendall and Spearman values are 0.844, 0.905 and 0.964, respectively, with a S.D. of 0.01. GDP maize =

price × production × GDPexp GDP

(15)

where, GDP maize is the total value of maize exposed to drought events; price is maize price; production is maize production; GDP is the gross domestic product of the country; GDPexp is economical exposition to droughts events 1980–2001. With the deepening study of disaster risk, the trend of crop drought risk assessment considers three elements (environment, hazard and hazard-affected body) in a quantitative assessment.

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

The drought risk assessment with GEPIC-V-R takes these three elements into account. Due to the complex relationships among the three factors, the weighted sum method is usually used to calculate drought risk with an index system (Li et al., 2009). The GEPIC-V-R model takes DEM, slope and soil as input data and uses daily water stress and yield to calculate risk. Therefore, risk calculation based on GEPIC-V-R achieves a comprehensive consideration of these factors and avoids probing the direct quantitative relationship between environment and risk. The accuracy of the GEPIC-V-R output depends largely on the quality of the input data. Given comprehensive and detailed data for soil, irrigation and fertilizers, the simulation accuracy of GEPICV-R model can be enhanced. Use of only three crop parameters (HI, WSYF and WA) cannot exactly reflect the crop characteristics. Although EPIC has a generic pest component for simulating insect and disease damage, as well as a weed competition component, this function is difficult to apply on a global scale because the necessary information is difficult to acquire. More detailed datasets, including a global map of crop planting patterns, would help improve the simulation accuracy of the GEPIC-V-R model. The GEPIC-V-R model is a GIS-based crop growth model integrating a bio-physical EPIC model with ArcGIS to simulate the spatial and temporal dynamics of the major processes of crops. However, the minimum grid cell is still considered to be a homogeneous unit in a “pseudo-threedimensional” grid, and water and nutrient substances horizontal flow process are not reflected (Luo and Guo, 2008; Wang and Feng, 2012). In the future, specialization with three-dimensional units should be strengthened. 4.2. Conclusions Under the guidance of the environment region division, the GEPIC-V-R model was built through application of the drought assessment theory of crops to GEPIC, which was developed by the Swiss Federal Institute of Aquatic Science and Technology (Eawag). GEPIC-V-R provides a practical and systematic tool for simulating crop yield, vulnerability curve and drought risk with a relatively high precision on a global scale. According to maize parameters of each environment zone provided by the regional crop parameters determination module, GEPIC-V-R can fit the maize vulnerability curve to drought by controlling for irrigation volume. The validation of maize loss rate in China and Kenya, Malawi and Mozambique demonstrates that the model works well in simulating vulnerability curves. The results of worldwide maize drought risk assessment shows that high drought risk for maize is concentrated in South Africa,

117

Chile, Western Europe, Central Europe, Southwest Russia, and North and Northeastern China. The rank correlation analyses of 28 provinces in China and seven countries in Africa showed that there was a close correlation between the GEPIC-V-R result and statistics or existing results. Therefore, the simulation results supply the theoretical support for taking measures according to local conditions to manage drought and drought risk. Because of data limitations, the validation of this research was performed only in typical regions. To validate from the physical mechanism and in more regions, we should strengthen vulnerability curve research through field experiments and water-control experiments, as well as gathering agricultural drought data from many more regions. Furthermore, different crop varieties and different growth stages have different sensitivity to water stress and thus have different degrees of yield in response to water stress. In the future, much more attention should be paid to drought risk assessment of different crop varieties and different growth stages, including water stress index and loss rate. Acknowledgments This study was jointly supported by the State Key Scientific Program of China (973 Project): Global Change, Environmental Risk and Its Adaptation Paradigms (No. 2012CB955403) and the National Natural Science Foundation of China: Research of Regional Agricultural Drought Adaptability Evaluation Model and Risk Prevention Paradigms (No. 41171402), and the Foundation for Innovative Research Groups of the National Natural Science Foundation of China (No.41321001). We are grateful to Deyong Yu at Beijing Normal University of China for his valuable comments on the manuscript. We thank Jiayi Fang at the Beijing Normal University and Lu Gao at Fujian Normal University for their assistance during the revision. In particular, our heartfelt should be given to two anonymous reviewers for their helpful comments that greatly helped to improve the quality of this article. Appendix A. Appendix: Global maize suitability zones Consulting the maize planting division of China, this paper promotes the maize division of world with a top-down and bottom-up method, considering the similarities and differences among many elements and combinations, such as weather conditions (temperature, water, etc.) and agricultural techniques. Principles of the maize planting division are shown as follows: combined actions of environment elements, leading ecology elements, and hierarchy of ecology elements. Global maize division includes the following

Table A1 Codes and names of maize planting divisions. Code

Name

Code

Name

101 102 103 104 105 106 107 108 109 110 111 201 202 203 204 205 206 301

Northern Central Asia temperate irrigated zone Southern Central Asia temperate zone West Asia temperate zone South Asia tropical rain-fed zone Southern Northeast Asia temperate rain-fed zone China’s Huang-Huai-Hai zone Central China sub-tropical-warm-temperate rain-fed zone Northwestern Southeast Asia sub-tropical rain-fed zone South China sub-tropical rain-fed zone Southeast Asia tropical rain-fed zone Northeast Asia-Japan warm temperate rain-fed zone Western European Plain temperate rain-fed zone Central European Plain temperate rain-fed zone Alps-Karbala mountain and East European Plain temperate rain-fed zone Iberian-Apennine temperate rain-fed zone Balkan temperate rain-fed zone Volga-Urals temperate rain-fed zone North America temperate irrigated zone

302 303 304 305 401 402 403 404 501 502 503 504 505 601 602 603 604 605

North America temperate rain-fed zone North America warm temperate rain-fed zone North America sub-tropical irrigated zone North America sub-tropical rain-fed zone Latin America tropical rain-fed zone South America tropical rain-fed zone South America sub-tropical rain-fed zone South America temperate zone Africa tropical irrigated zone Africa tropical rain-fed zone Africa sub-tropical rain-fed zone Africa sub-tropical irrigated zone Madagascar zone Oceania tropical rain-fed zone Oceania tropical irrigated zone Oceania sub-tropical irrigated zone Oceania temperate zone Oceania sub-tropical rain-fed zone

118

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119

Fig. A1. Global maize planting division map. What is the code mean is showed in Table A1.

four parts: select division indexes, divide planting regions, name the division regions, and draw the division map. Division indexes include three hierarchies: annual accumulated temperature of ≥10 ◦ C, annual precipitation, and tendency yield. This global maize planting division includes 36 regions (see Table A1 and Fig. A1).

References ADRC (Asian Disaster Reduction Center), 2005. Total Disaster Risk Management— Good Practices 2005. Asian Disaster Reduction Center, Kobe, Japan. Cabelguenne, M., Debaeke, P., Bouniols, A., 1999. EPIC phase, a version of the EPIC model simulating the effects of water and nitrogen stress on biomass and yield, taking account of developmental stages: validation on maize, sunflower, sorghum, soybean, and winter wheat. Agric. Syst. 60, 175–196. Confalonieri, R., Bocchi, S., 2005. Evaluation of CropSyst for simulating the yield of flooded rice in northern Italy. Eur. J. Agron. 23, 315–326. Crichton, D., 1999. The risk triangle. In: Ingleton, J. (Ed.), Natural Disaster Management, A Presentation to Commemorate the International Decade for Natural Disaster Reduction (IDNDR). Tudor Rose, London, United Kingdom, pp. 102–103. Cui, D.C., 1994. Agricultural Climate and Crop Climate of the World. Press of Zhejiang Science and Technology, Hangzhou. Dai, A.G., 2011. Drought under global warming: a review. WIREs Clim. Change 2, 45–65. Dai, A.G., Trenberth, K.E., Qian, T., 2004. A global dataset of Palmer Drought Severity Index for 1870–2002: relationship with soil moisture and effects of surface warming. J. Hydrometeorol. 5, 1117–1130. FAO (Food and Agriculture Organization of the United Nations), 1991–2000. FAO, http://faostat3.fao.org/faostat-gateway. FAO (Food and Agriculture Organization of the United Nations), 2003. The digitized soil map of the world including derived soil properties (version 3.6). FAO, Rome, Italy. Gaiser, T., Barros, I., Sereke, F., et al., 2010. Validation and reliability of the EPIC model to simulate maize production in small-holder farming systems in tropical sub-humid West Africa and semi-arid Brazil. Agric. Ecosyst. Environ. 135, 318–327. Gassman, P.W., Williams, J.R., Benson, V.W., et al., 2005. Historical Development and Applications of the EPIC and APEX Models. Center for Agricultural and Rural Development Iowa State University, Ames, IA. He, F., 2010. Regional Agricultural Drought Disaster System Research—Case Study of Drought Disaster of Rice. Beijing Normal University, Beijing. Hu, R.H., Li, X.E., Wei, H., et al., 2012. Drought risk assessment model of mountainous area based on DEM: a case of southwest China. J. Nat. Disasters 21 (2), 157–162. Huang, M., Gallichand, J., Dang, T., et al., 2006. An evaluation of EPIC soil water and yield components in the gully regions of Loess Plateau, China. J. Agric. Sci. 144 (4), 339–348. IDEA, IDB, 2005. System of Indicators for Disaster Risk Management: Program for Latin American and the Caribbean. Inter-American Development Bank,

Universidad Nacional de Colombia—Sede Manizales, Instituto de Estudios Ambientales, Manizales, Colombia. IPCC (Intergovernmental Panel on Climate Change), 2007. IPCC Expert Meeting Report: Towards New Scenarios for Analysis of Emissions, Climate Change, Impacts, and Response Strategies. Intergovernmental Panel on Climate Change, Noordwijikerhout, The Netherlands. IPCC (Intergovernmental Panel on Climate Change), 2012. Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation: Special Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, New York, NY, USA. IBSNAT (International Benchmark Sites Network for Agrotechnology Transfer), 1989. Decision Support System for Agrotechnology Transfer Version2. 1 (DSSAT V2. 1). Dept. of Agronomy and Soil Science. College of Tropical Agriculture and Human Resources: University of Hawaii, Honolulu. Jayanthi, H., Husak, G., 2013. A probabilistic approach to assess agricultural drought risk. In: Background Paper Prepared for the Global Assessment Report on Disaster Risk Reduction 2013, Geneva, Switzerland. Jia, H.C., Wang, J.A., Cao, C.X., et al., 2012. Maize drought disaster risk assessment of China based on EPIC model. Int. J. Digital Earth 5 (6), 488–515. Jonghan, K., Giovanni, P., Evelyn, S., 2009. Using EPIC model to manage irrigated cotton and maize. Agric. Water Manage. 96, 1323–1331. Li, Y.P., Ye, W., Wang, M., et al., 2009. Climate change and drought: a risk assessment of crop-yield impacts. Clim. Res. 39, 31–46. Liu, J.G., 2009. A GIS-based tool for modeling large-scale crop-water relations. Environ. Modell. Softw. 24, 411–422. Liu, J.G., Williams, J.R., Zehnder, A.J., et al., 2007. GEPIC—modelling wheat yield and crop water productivity with high resolution on a global scale. Agric. Syst. 94 (2), 478–493. Luo, Y., Guo, W., 2008. Development and problems of crop models. Trans. CSAE 24 (5), 307–312. Monteith, J.L., 1996. The quest for balance in crop modeling. Agron. J. 88, 695–697. Mpelasoka, F., Hennessy, K., Jones, R., et al., 2008. Comparison of suitable drought indices for climate change impacts assessment over Australia towards resource management. Int. J. Climatol. 28, 1283–1292. Nachtergaele, F.O., 1996. From the Soil Map of the World to the Global Soil and Terrain Database: 1960–2002. Working group meeting Soils and GIS, Athens, Greece. Palmer, W.C., 1965. Meteorological Drought. U.S. Weather Bureau Research, Washington, DC, pp. 58. Palmer, W.C., 1968. Keeping track of crop moisture conditions, nationwide: the new crop moisture index. Weatherwise 21, 156–161. Priya, S., Shibasaki, R., 2001. National spatial crop yield simulation using GIS-based crop production model. Ecol. Model. 136, 113–129. Qian, B.D., Jong, R.D., Gameda, S., 2009. Multivariate analysis of water-related agroclimatic factors limiting spring wheat yields on the Canadian prairies. Eur. J. Agron. 30 (2), 140–150. Rao, M.N., Waits, D.A., Neilsen, M.L., 2000. A GIS-based modeling approach for implementation of sustainable farm management practices. Environ. Modell. Softw. 15, 745–753.

Y. Yin et al. / Agricultural Water Management 144 (2014) 107–119 Sharpley, A.N., Williams, J.R., 1990. EPIC—Erosion/Productivity Impact Calculator: 1. Model Document. U.S. Department of Agricultural Technical Bulletin. Shi, P.J., 1991. One the theory of disaster research and its practice. J. Nanjing Univ. 11, 37–42. Shi, P.J., 2002. Theory and practice on disaster system research in a third time. J. Nat. Disasters 14 (6), 1–9. Shi, P.J., 2012. Atlas of Natural Disaster Risk of China. Science Press, Beijing. Tan, G., Shibasaki, R., 2003. Global estimation of crop productivity and the impacts of global warming by GIS and EPIC integration. Ecol. Model. 168 (3), 357–370. Texas A&M Agri-life Research, 2012. EPIC & APEC Models, http://epicapex.tamu. edu/downloads/. Thomas, B.M., Doesken, N.J., Kleist, J.,1993. The relationship of drought frequency and duration to time scale. In: Proceedings of the Proc. Committee on Applied Climatology, Eight Conference on Applied Climatology. American Meteorological Society, Anaheim, California, pp. 179–184. UN/ISDR (Unite Nations International Strategy for Disaster Reduction), 2004. Living With Risk: A Global Review of Disaster Reduction Initiatives. United Nations, New York and Geneva. UNDP (United Nations Development Programme), 2004. Reducing Disaster Risk: A Challenge for Development. Hureau for Crisis Prevention and Recovery, New York, NY. UNEP (United Nations Environment Programme), UNISDR (United Nations International Strategy for Disaster Reduction), 2013. Global Risk Data Platform. UNEP, lang;http://preview.grid.unep.ch. Velde, V., Tubiello, F.N., Vrieling, A., et al., 2012. Impacts of extreme weather on wheat and maize in France: evaluating regional crop simulations against observed data. Clim. Change 113 (3–4), 751–765. Velde, V., Wriedt, G., Bouraoui, F., 2010. Estimating irrigation use and effects on maize yield during the 2003 heat wave in France. Agric. Ecosyst. Environ. 135, 90–97.

119

Wang, K., Zhang, Q., 2013. A new approach to assessment crop yield risk based on mixed source of data. Sci. Agric. Sin. 46 (5), 1054–1060. Wang, W.J., Feng, H., 2012. The progress and problems in the development of foreign crop models. Water Saving Irrig. 8, 62–68. Wang, Z.Q., He, F., Fang, W.H., et al., 2013. Assessment of physical vulnerability to agricultural drought in China. Nat. Hazard. 67, 645–657. Williams, J.R., 1995. The EPIC model. In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, Littleton, CO, pp. 909–1000. Williams, J.R., Jones, C.A., Dyke, P.T., 1984. A modeling approach to determining the relationship between erosion and soil productivity. Trans. ASAE 27 (1), 129–144. Williams, J.R., Jones, C.A., Kiniry, J.R., et al., 1989. The EPIC crop growth model. Trans. ASAE 32 (2), 497–511. World Bank, Columbia University, 2005. Natural Disaster Hotspots: A Global Risk Analysis. The World Bank Hazard Management Unit, Washington, D.C. Wriedt, G., Velde, M.V., Aloe, A., et al., 2009. Estimating irrigation requirements in Europe. J. Hydrol. 373 (3–4), 527–544. Xu, L., Zhang, Q., 2011. Assessment approach for agricultural catastrophic risk in Chin. Sci. Agric. Sin. 44 (9), 1945–1952. Xu, X.C., Ge, Q.S., Zheng, J.Y., et al., 2011. Drought risk assessment on regional agriculture: a case in Southwest China. Prog. Geogr. 30 (7), 883–890. Zhang, J.Q., 2004. Risk assessment of drought disaster in the maize-growing region of Songliao Plian, China. Agric. Ecosyst. Environ. 201 (2), 133–153. Zhou, Y., Wang, J.A., Yue, Y.J., et al.,2012. Drought risk assessment model of mountainous area based on DEM-A case study of henan in China. In: Association of American Geographers Annual Conference 2012. Association of American Geographers, New York, NY.