Agricultural Systems 147 (2016) 76–86

Contents lists available at ScienceDirect

Agricultural Systems

j ourna l homepage: www.e lsev ie r .com/ locate /agsy

The value of seasonal forecasts for irrigated, supplementary irrigated, andrainfed wheat cropping systems in northwest Mexico

Melissa A. Ramírez-Rodrigues a,⁎, Phillip D. Alderman b,1, Lydia Stefanova c, C. Mariano Cossani b,2,Dagoberto Flores b, Senthold Asseng a

a University of Florida, Department of Agricultural and Biological Engineering, Gainesville, FL 32611, USAb International Maize and Wheat Improvement Center, Global Wheat Program, Texcoco, Estado de México, Mexicoc Florida State University, Center for Ocean-Atmospheric Prediction Studies, Tallahassee, FL 32310, USA

⁎ Corresponding author.E-mail address: meliram@ufl.edu (M.A. Ramírez-Rodr

1 Present affiliation: Oklahoma State University, DepartStillwater, OK 74078.

2 Present affiliation: South Australian Research and DSouth Australia, Australia, 5064.

http://dx.doi.org/10.1016/j.agsy.2016.05.0050308-521X/© 2016 Elsevier Ltd. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 25 July 2014Received in revised form 12 April 2016Accepted 3 May 2016Available online 7 June 2016

Half of global wheat production occurs in irrigated cropping regions that face increasingwater shortages. In theseregions, seasonal forecasts could provide information about in-season climate conditions that could improve re-sourcemanagement, helping to savewater and other inputs. However, seasonal forecasts have not been tested inirrigated systems. In this study, we show that seasonal forecasts have the potential to guide crop managementdecisions in fully irrigated systems (FIS), reduced irrigation systems (supplementary irrigation; SIS), and systemswithout irrigation (rainfed; RFS) in an arid environment. We found that farmers could gain an additional2 USDha−1 season−1 in net returns and save up to 26USDha−1 season−1 inN fertilizer costswith a hypotheticalalways-correct-season-type-forecast (ACF) in a fully irrigated system compared to simulated optimized N fertil-izer applications. In supplementary irrigated systems, an ACF had value when deciding on sowing a crop (plussupplementary irrigation) of up to 65 USD ha−1 season−1. In rainfed systems, this value was up to123 USD ha−1 when deciding whether or not to sow a crop. In supplementary irrigated and rainfed systems,such value depended on initial soil water conditions. Seasonal forecasts have the potential to assist farmers in ir-rigated, supplementary irrigated, and rainfed cropping systems to maximize crop profitability. However, fore-casts currently available based on Global Circulation Models (GCM) and the El Niño Southern Oscillation(ENSO) need higher forecast skill before such benefits can be fully realized.

© 2016 Elsevier Ltd. All rights reserved.

Keywords:Arid environmentIrrigationSupplementary irrigationWheatMexicoSeasonal forecast

1. Introduction

Wheat provides approximately 20% of calories consumed byhumans (Food and Agriculture Organization 2012), and wheat cropscover approximately 22% of the world's cultivated land (Licker et al.2010) across developed and developing countries. Irrigated wheat pro-duction accounts for almost half of global wheat production, and ap-proximately 90% of irrigated wheat production occurs in developingcountries (Shiferaw et al. 2013). Groundwater depletion (Balwinder etal. 2011; Chen et al. 2010; Liu et al. 2013; Lv et al. 2013; Zhao et al.2013), limited surface water resources, and increasing soil salinity(Seifert et al. 2011) create enormous challenges for regions that dependon these irrigated wheat systems. Demand for wheat will likely

igues).ment of Plant and Soil Sciences.

evelopment Institute, Urrbrae,

increase, given that global population is projected to exceed 9 billionby 2050 (BeVier 2012). Increases in water price (Shiferaw et al. 2013)and reduction in water quality (Lv et al. 2013) will further exacerbatethe challenges farmers face in irrigated regions. Improvements in cropmanagement and breeding are needed to secure future wheat produc-tion increases.

TheYaqui Valley is oneof themost importantwheat producing areasin Mexico, accounting for approximately 40% of national wheat produc-tion (Schoups et al. 2006). It comprises approximately 225,000 ha of ir-rigated cropland cultivated mainly during the winter season (Ortiz-Monasterio and Raun 2007). More than 50% of this cropland is usedfor wheat (Lobell et al. 2004). The Yaqui Valley climate is arid with anaverage annual precipitation of 300 mm; most of which falls betweenJune and September (Schoups et al. 2006), outside the wheat growingseason. Maximum air temperature can exceed 34 °C at the end of thewheat growing season (April–May). Further, Yaqui Valley agro-climaticconditions are representative of about 40% of wheat production areas indeveloping countries (Verhulst et al. 2011). These regions face many ofthe samewater challenges, temperature conditions (Asseng et al. 2011;Lobell et al. 2012), and environmental issues (Seifert et al. 2011).

77M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

As in other parts of the world, farmers in the Yaqui Valley make re-source allocation decisions before the beginning of the season and cur-rently must do so without the aid of information on climate conditionsof the coming season (Asseng et al. 2012; Lobell et al. 2004; Lobell et al.2005; Moeller et al. 2008). Without such climate information, farmerscannot tailor their management practices, such as crop rotation(Cossani et al. 2007) or fertilizer rates and timings (Asseng et al.2012), to the climate of the coming season. Suchmismatches in agricul-tural practices and seasonal climate result in increased yield gaps anddecreased resource use efficiency (Cossani et al. 2010; Sadras et al.2003). Hence, seasonal forecasts can provide an important tool for im-proving the productivity and efficiency of cropping systems (Meinkeand Stone 2005). However, such a tool is only valuable if the availableinformation leads farmers to change their decision-making process(Hammer 2000). Seasonal forecasts are widely used in rainfed condi-tions for different applications, such as optimizing nitrogen (N) man-agement under seasonal variability (Asseng et al. 2012; Moeller et al.2008; Yu et al. 2008), predicting stored soil water at planting(Hammer et al. 1996) or making decisions about cropping systems(Carberry et al. 2000). Recent studies on short-term (Mishra et al.2013) and long-term (Calanca et al. 2011)weather forecasts have inves-tigated optimizing crop irrigation management and predicting soilwater availability. In the Yaqui Valley, weather forecasts are used to pre-dict daily crop evapotranspiration (ET) for water management(Abdelghani et al. 2008) However, due to short lead times, these studieslack information about the inter-annual climate variability (Hunt andHirst 2000), limiting the overall optimization of the system.

Using seasonal forecasts to make decisions about N management,sowing, and supplementary irrigation has not been widely explored inirrigated and supplementary irrigated cropping systems. Hence, the ob-jectives of this study were to evaluate potential and actual seasonalforecasts for assisting crop management decisions in fully irrigated(FIS), supplementary irrigated (SIS), and rainfed cropping systems(RFS).

2. Methods and materials

This methodology tests how climate information could potentiallyimprove nitrogen management decisions in an arid environment con-sidering three different cropping systems. These systems differ in theamount of irrigated water available during the wheat growing season.The nitrogen recommendations are based on a crop model that deter-mines the optimal seasonal N amount according to the irrigated wateravailable for the wheat growing season and the climate condition fore-casted for the wheat growing season at the beginning of the season. Aseasonal forecast thus allows farmers with different irrigated water re-strictions tomanage nitrogen fertilizer applications season-type specificbased on the seasonal rainfall forecast.

Fig. 1 is a flow diagram of the different modules that comprise themethodology. Historical weather data and agricultural practices of theYaqui Valley were entered into a crop model. The model was calibratedand simulationswere run for 27 years of data. The cropmodel simulatedwheat growth and returned an expected yield for each N fertilizer man-agement option in each year. This procedure was repeated for each ofthe three cropping systems. Season-specific tercile categories werebuilt using seasonal forecast data and historical weather information.

Fig. 1. Flow diagram of the different modules that comprise the methodology.

The skill of the forecast was then computed to evaluate the ability ofthe forecast to predict the tercile category correctly. Finally, based onthe yields and recommended agricultural practices determined foreach season-specific category, an economic analysis was performedcomparing the net returns for a farmer following the forecast to thoseof a farmer not following the forecast.

2.1. APSIM Nwheat model

The widely adopted and tested wheat model, the Agricultural Pro-duction System Simulator (APSIM; Keating et al. 2003) Nwheat model(Asseng et al. 1998; Asseng et al. 2001a; Asseng and Milroy 2006;Asseng et al. 2001b) was used in this study. The APSIM Nwheat modelincludes modules that simulate growth, development, and yield ofwheat crops, as well as soil water, N, and carbon dynamics. The cropmodule accounts for crop development and growth,water and nitrogenuptake, and considers various stress conditions of awheat crop (Keatinget al. 2003). The model calculates an attainable yield for a specific envi-ronment, limited by temperature, solar radiation, rainfall, water, and Nsupply (Asseng 2004; Lobell et al. 2009). The Nwheat model also in-cludes a temperature stress algorithm to capture the effect of tempera-ture increases on wheat growth processes such as leaf growth andphotosynthesis (Asseng et al. 2011).

2.2. Model calibration

APSIM-Nwheat was calibrated using experimental data collectedduring the 2011–2012 wheat-growing season in the Yaqui Valley atthe CIMMYT experimental station in Obregon, Sonora (27°25′N, 109°54′W) with an elevation of 38 m asl and a Hyposodic Vertisol soil type(Calcaric, Chromic) (Verhulst et al. 2009). The experiment consisted ofa set of wheat genotypes sown in 3.5 m long and 1.6 m wide plotseach containing two raised beds with two rows per bed. These weregrown under four different treatments designed to represent a rangeof temperature and soil moisture conditions. The treatments includeda fully irrigated treatment (I) with sowing date November 24, a droughttreatment (D) with two irrigation applications and a sowing date of De-cember 11, a heat stress treatment (H), fully irrigated with sowing dateFebruary 27, and an extreme heat stress treatment (EH) during grainfilling, and fully irrigated with sowing date March 30. Experimentalplots weremanaged intensively to ensure crop productionwas not lim-ited by nutrient availability or biotic stress (including weeds, diseases,and insect pests). One representative genotype was selected (CIMMYTGID 5180708) for calibration and subsequent simulations.

2.3. Simulations

Simulations used historic climate data (1982–2009) of a Yaqui Val-ley weather station (27°11′N, 109°32′W) obtained from the Centro deInvestigaciones Agrícolas del Noroeste (CIANO). Gaps in solar radiationdata were filled from the AgMERRA historical climate forcing dataset(Ruane et al. 2013). A common planting date of mid-November waschosen for all simulations (Lobell and Ortiz-Monasterio 2008). Aseeding rate of 120 kg ha−1 was considered consistent with currentpractices (Lopes et al. 2013; Reynolds et al. 1994).

Specific management practices were tested in each system. For thefully irrigated system (FIS), automatic irrigation was applied everytime the available soil water was below a critical fraction of availablesoil water (0.5) to avoidwater stress. The typical concentration of nitro-gen as NO3 and NH4was 0.0457mgN l−1, equivalent to 0.0457 kg ha−1

for 100 mm of irrigation applied. Seven N treatments were used in theirrigated system, from 60 to 240 kg N ha−1 in increments of30 kg N ha−1 split into two applications, one at sowing and another40 days after sowing (DAS). For the supplementary irrigation system(SIS), a limited irrigation of 100 mm was applied per season, split intotwo applications, one at sowing and another at 80 DAS. 120 kg N ha−1

78 M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

was also applied and split into two applications, one at sowing and an-other at 40 DAS. For the rainfed system (RFS), no irrigationwas applied,and 60 kg N ha−1 was applied and split in two applications, one at sow-ing and another at 40 DAS (Table 1).

The soil used in the simulationwas classified as a Hyposodic Vertisol(Calcaric, Chromic) in the World Reference Base (Verhulst et al. 2009).Maximum rooting depth was considered to be 180 cm for simulationsand the average maximum soil moisture capacity of the profile was120 mm. The soil is a mainly coarse sandy clay mixed with montmoril-lonitic clay (Riley et al. 2001), with b1% organic matter (Lobell et al.2002). The soil parameters (pH, organic carbon, and bulk density)used as input in the simulationwere from a soil characterization carriedout at the CIMMYT research station, Dr. Norman E. Borlaug, in the YaquiValley, Obregón, Sonora, block 810 (Verhulst et al. 2009). The soil waterproperties were taken from theWorld Inventory of Soil Emission Poten-tials (WISE) database (Romero et al. 2012).

2.4. Seasonal forecast

A re-forecast procedure comparing past seasonal forecasts (alsocalled hindcasts) with measured historical weather data was used toquantify the value of the seasonal forecast for wheat management inthe Yaqui Valley in Obregón, Sonora. The three seasonal forecasts eval-uated were an always-correct-season-type forecast (potential of a cli-mate forecast) (ACF), a Global Circulation Model-based forecast(GCMF), and an El Niño Southern Oscillation-based forecast (ENSOF).

The selectedGCMFwas the Climate Forecast System (CFS), a coupledocean-atmosphere-land seasonal prediction system which became op-erational at the NOAA's National Center for Environmental Prediction(NCEP) in August 2004 (Saha et al. 2006). The CFS version 2 (CFS-V2)has 28 years of hindcast (1982–2009); hindcast initializing startsevery month, with a 4 times daily integration and up to 9 months leadtime (Saha et al. 2010; Yuan et al. 2011). The latitude and longitudepoint extracted from the hindcast data was (27°52′N, 109°41′W) forthe Yaqui Valley. Average temperatures and precipitation of thehindcast were calculated for each year. The same classification wasdone with the historical weather data.

The CFS-V2 hindcast data in the study encompassed 27 simulatedwheat-growing seasons (November–May). A GCM forecast (orhindcast) provides daily climate information. These were summarizedacross the growing season (e.g. cumulative rainfall over the growingseason, November to May). The summarized seasonal climate datawere then sorted across the 27 years (e.g. from highest to lowest rain-fall) and grouped into terciles (i.e. wettest 33% of years, average 33%of years and driest 33% of years). To test if the GCM forecasted a tercilecorrectly, the category of the forecasted year was compared with thecategory of the measured historical rainfall. If the forecast was withinthe same category (i.e. tercile) as the measured historical rainfall cate-gory, the forecast was defined as correct. When the forecasted rainfallcategory was different from the measured historical rainfall category,the forecast was defined as incorrect. The skill of the forecast was de-fined as the percentage of correct forecasts out of the total 27 years(Moeller et al. 2008).

Seasonal forecast skill was evaluated on the likelihood of predictingthe tercile-category of the year correctly. Categorical variableswere cre-ated for the 27wheat growing years taking the value of 1 if the seasonal

Table 1Management practices tested in a fully irrigated system (FIS), in a supplementary irrigated sys

SystemsIrrigation

Total amount (mm) Application

FIS Full At thresholdSIS 100 Sowing, 80 DASa

RFS None None

a Split in two equal applications.

forecast data and historical climate data ranked that year in the top 1/3tercile category and 0 otherwise. The forecast skill was assessed usingthe information of these two categorical variables to test the Accuracy,Heidke Skill Score and S% verification measurements.

In the FIS, water was assumed not limiting for wheat growth. Theweather variables, driving yield variability, were solar radiation andtemperature. Despite the fact that solar radiation measurements werenot available from theGCMF, therewas a strongnegative correlation be-tween solar radiation and precipitation, as higher precipitation is asso-ciated with more cloudy days. This negative correlation made itpossible to create a “1/3 driest coolest years category (or season-type).” This category is referred to as “1/3 higher solar radiation sea-sons” and is used to study those years that encompass the potentiallyhigher-yielding seasons (i.e., seasons with higher solar radiation andcooler, more suitable temperature conditions) compared to the otheryears. For the SIS and RFS, water was a limiting factor for wheat growth,so the top tercile of wettest seasons was used to classify the years intothe 1/3 wettest season category compared to the other years.

The ENSOF was created using the three-month mean anomaliesfrom August to October, previous to the sowing date (November 15).The El Niño 3.4 index seasonal forecast values were used to categorizethe years into nine El Niño seasons (N0.5 °C), eight La Niña seasons(b−0.5 °C), and ten Neutral seasons (Table 2). The updated El Niño3.4 index three months anomalies is publicly available (NOAA 2014).

Table 2 shows the upper category (tercile) harvest years used tobuild the rainfall category (SIS and RFS) or the dry and cool category(FIS). The El Niño and La Niña year categories are also shown for theSIS and RFS, and the FIS, respectively. The data were grouped into sea-sonal forecast tercile-categories to evaluate the performance of the dif-ferent seasonal forecasts in each cropping system. The skill wascomputed using the Accuracy (ACC) and S% verification measurements(Table 2) (Jolliffe and Stephenson 2003). The S%measurewas calculatedassuming a random forecast obtained by a coin toss. The probability ofsuch random forecast correctly forecasting years in the wettest tercileis 1/6 (1/2 ∗ 1/3), and the probability of it correctly forecasting the re-mainder of years is 2/6 (1/2 ∗ 2/3). Considering this, the probability ofthe random forecast being correct is 50% (1/6 + 2/6). The S% formulais then defined as:

S% ¼ %correct forecast−%correct forecast due to chance100%−%correct forecast due to chance

To evaluate the ENSOF performance, data were split between thebest performing El Niño Southern Oscillation (ENSO) phase (i.e. highestsimulated yields) and the other two phases tomake comparisons. In theFIS, the best performing phase was La Niña, but in the SIS and RFS, thebest-performing phase was El Niño.

The three seasonal forecasts in the irrigated system were evaluatedin terms of net return andN fertilizer cost. The optimumNmanagementfor each forecast category was calculated using cross validation (the“leave one out” technique). For this technique, each year was left outonce, and the optimumNmanagement for the remaining years was cal-culated. Then the average of the cross-validated N was computed. Theseasonal forecasts in the SIS were evaluated in terms of sowing a cropunder limited supplementary irrigation (100 mm irrigation and120 kg N ha−1) as well as not sowing a crop with different initial

tem (SIS), and in a rainfed system (RFS).

Nitrogen

Total amount (kg N ha−1) Application

60, 90, 120, 150, 180, 210, 240 Sowing, 40 DASa

120 Sowing, 40 DASa

60 Sowing, 40 DASa

Table 2Upper category (tercile) harvest years used to build the GCM-based forecast (GCMF) categories for wet, dry and dry-cool weather conditions and the ENSO-based forecast (ENSOF) cat-egories for El Niño and La Niña. S% seasonal forecast skill obtained for rainfall, temperature dry-cool and ENSO phases.

Forecast System Variable S% Upper category years

GCM SIS, RFS Rainfall 0.26⁎ 1986 1987 1991 1992 1994 1995 2002 2003 2005FIS DryCool 0.26⁎ 1983 1985 1990 1993 1997 1998 2001 2004 2007

ENSO SIS, RFS El Niño 1.00⁎⁎ 1983 1987 1988 1992 1995 1998 2003 2005 2007FIS La Niña 0.70⁎⁎ 1984 1986 1989 1996 1999 2000 2001 2008

FIS fully irrigated system.SIS supplementary irrigated system.RFS rainfed system.⁎ 10% significance level.⁎⁎ 5% significance level.

79M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

plant-available stored soil water conditions (PAWC) at sowing of 0%,50%, and 100%. The seasonal forecasts in the RFS were evaluated interms of sowing a crop under rainfed conditions (sowing and

Fig. 2.Historical water availability, temperature and solar radiation (1983–2009) during the wh(mm) is shown as bars with the 1/3 wettest tercile years in full bars; meanmaximum temperatSeasonal rainfall (mm) (November–May) B) Mean cumulative rainfall (mm) between Novembaverage rainfall (mm) during the 1/3 wettest tercile years (full bars) and all other years (open(CDF) of simulated plant-available stored soil water at sowing for the 1/3 wettest tercile years

60 kg N ha−1) as well as not sowing a crop with different initialPAWC at sowing of 0%, 50%, and 100%. In both the SIS and RFS, thevalue of the forecasts was seasonal, calculated considering all years

eat-growing season (November–May) in the Yaqui Valley in Sonora, Mexico. Precipitationure (°C) is shown as dotted lines; and mean daily solar radiation (MJ/m2) as solid lines. A)er 1 and May 31 for the 1/3 wettest tercile years compared to all other years. C) Monthlybars). Vertical bars show ± one standard deviation. D) Cumulative distribution functioncompared to all other years.

Fig. 4. Simulated versus observed grain yields for four contrasting growing conditions:irrigated (full circle), drought (open circle), heat stress (full triangle) and extreme heatstress (open triangle) in the Yaqui Valley, Sonora, Mexico for three cultivars. RMSE =0.9 Mg ha−1.

Fig. 3.Monthly averageheat days (Tmax N 34 °C) in the Yaqui Valley, Sonora,Mexico duringthe 1/3 wettest tercile years (full bars) and all other years (open bars); monthly averagemaximum temperature in the 1/3 wettest years (full line above) and other years(dashed line above); monthly average minimum temperature in the 1/3 wettest years(full line below) and in other years (dashed line below); and monthly average solarradiation in the 1/3 wettest years (full gray line) and in other years (dashed gray line).Vertical bars show ± one standard deviation.

80 M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

(1983–2009). Following this approach, we could compute the averagenet return of the season and compare the value of always sowing tothe value of only sowing in years with a favorable forecast.

2.5. Economic analysis

To determine the wheat management production cost in the YaquiValley, Sonora, we used a farmer's survey (2008–2010) developed bythe International Maize and Wheat Improvement Center (CIMMYT) incollaboration with Stanford University as well as a wheat managementreport published by the Secretary of Agriculture in México (SAGARPA2008). To calculate the net returns for a specific year (NRi) a wheatprice of 250 USD Mg−1 (Py), and an N cost of 1.25 USD kg N−1 applied(Pf) were assumed. Fixed costs (Cd) were calculated considering soilpreparation cost, sowing cost, harvest cost, and irrigation cost in the sys-tems that used irrigation. The fixed costs considered for each systemwere 449 USD ha−1 for FIS, 324 USD ha−1 for SIS, and 242 USD ha−1

for RFS (CIMMYT and Stanford-University 2011; SAGARPA 2008).

NRi ¼ Y � Py−Cd−F � P f

Marginal net returns were calculated for changes in N application toevaluate the net return of each dollar invested in fertilizer. For rainfedfarms, the assumptionwasmade that for a farmer to invest an addition-al unit (20 kg N ha−1) of fertilizer, the marginal net return had to betwice as much as the original investment for the additional N fertilizerapplication (i.e., 2 US$ for each dollar invested) (Asseng et al. 2012).Even though this conservative approach accounts for the farmer's risk-averse behavior, this assumption does not hold in irrigated farms.Farmers in irrigated regions donot follow this type of risk-averse behav-ior as strongly as under rainfed conditions. Rather, irrigation farmerstend to over-fertilize to guarantee the highest achievable yields usingfertilizer subsidies (Christensen et al. 2006; Lobell andOrtiz-Monasterio 2006; Lobell et al. 2004). For that reason, in the FISwe assumed farmers optimized the N fertilizer application in situationswhere the net returns equal zero.

The value of seasonal forecast was calculated comparing the averagenet return for a farmer following the same management practice allyears regardless of the climate conditions (NRopt) with that of a farmerusing season-type specific management based on the climate predic-tions of the seasonal forecast (NRf). The mathematical expression forcomputing the value of seasonal forecast would then be:

ValueofSeasonalForecast ¼ ∑ni¼1NRf

n−

∑ni¼1NRopt

n

where i represents the number of years from 1 to n.

3. Results

3.1. Characterization of the region

3.1.1. PrecipitationPrecipitation in the Yaqui Valley is extremely low duringmost of the

wheat-growing season. Seasonal rainfall ranged from 2 to 344 mm be-tween 1983 and 2009. Average seasonal rainfall was 167 mm in the 1/3 wettest years and 33 mm in the remaining years (Fig. 2a). The maindifferences in rainfall patterns were observed early in the wheat season(Fig. 2b), especially during December and January (Fig. 2c). In the YaquiValley, most rainfall occurs during the summer. Variability in summerrainfall creates differences in initial soil moisture for wheat croppingfrom season to season. Simulations were carried out to determine thepre-planting soilmoisture and in N50% of the 1/3wettest years, the sim-ulated initial soil available water at sowing was N110 mm, which isgreater than half the field capacity of 250 mm (Fig. 2d). The oppositewas observed during all other years. Due to the variability in pre-plant-ing soil moisture the three initial PAWC at sowing considered in the SISand RFS were 0%, 50% and 100%

3.1.2. Temperature and solar radiationTemperature and solar radiation also varied from year to year (Fig.

2a). In the 1/3 wettest tercile years, months averagemaximum temper-ature were lower (0.83 °C less) but months average minimum temper-aturewere higher (by 0.76 °C) in comparison to other years. Differencesin number of heat stress days (Tmax N 34 °C)were also present. In the 1/3wettest years the crops experienced on average 50% less heat stressdays compared to the other years during November and April and

81M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

about 25% less heat stress days during May. On average, less solar radi-ation (seasonal daily mean of 14.7 MJ m2) was observed during the 1/3wettest years early in thewheat-growing season (November–February)compared to the other years (seasonal daily mean of 15.6 MJ m2) (Fig.3).

3.1.3. Model calibrationThe APSIM Nwheat model was manually calibrated for one cultivar

under the four environments described in Section 2.1. It was consideredsuitable to use in the seasonal forecast evaluation for this region (Fig. 4)with a root mean square error (RMSE) of 2 days for days to anthesis and0.90 Mg ha−1 for yield. The climate conditions for the year 2011–2012.I.e. the year used to calibrate themodel is comparable to the average cli-mate conditions of the dry years with total seasonal rainfall of 51 mm,with most of the precipitation occurring early in the season.

3.2. Value of seasonal forecast in the FIS

In the fully irrigated system (FIS), the APSIMNwheatmodel simulat-ed yields without water limitations. A relative large difference in year-to-year yields with ±1.2 Mg ha−1 from a mean of 7 Mg ha−1 was sim-ulated, despite no effect of water stress (Fig. 5a). On average, the 1/3wettest years had lower simulated yields (6.7 Mg ha−1) compared to

Fig. 5. Cropmodel output for A) Simulated wheat yields (1983–2009) for the 1/3 wettest tercilradiation tercile years (cross) and all other years (open square) in a fully irrigated system atSimulated wheat yields B) versus cumulative seasonal rainfall C) seasonal mean maximum tem

the other years (7.2 Mg ha−1); this was also shown by a negative corre-lation of precipitationwith yield (−0.24). Hence, wet seasons indicatedlower yield potential in FIS. When considering the negative correlationbetween precipitation and solar radiation (data not shown), this obser-vation could be explained by a reduction in intercepted solar radiationcaused by greater cloud cover during years with high precipitation(Fig. 5b, c, and d). Average maximum temperature also showed a nega-tive correlation (−0.18) with simulated yields but was not statisticallysignificant at 95% confidence level. Dry, cool years resulted in, on aver-age, the highest simulated yields in irrigated systems. For instance, the1/3 higher solar radiation season years had on average 7.4Mg ha−1 sim-ulated yields compared to the other years, which had on average6.9 Mg ha−1.

Climate conditions also affected the N response of yields. Fig. 6shows the average net returns of N response for the 1/3 higher solar ra-diation tercile years compared to the remaining years. Considering thatfarmers in FIS aim to maximize net returns, the optimal N fertilizer ap-plication for the 1/3 higher solar radiation tercile years would be150 kg N ha−1 and 120 kg N ha−1 for the other years. However, currentfarmers in this irrigated region apply approximately 240 kgNha−1, wellabove the level needed for maximum net returns. Reducing the currentN fertilizer rate to the maximum effective rate would increase the netreturns in FIS. Reducing N management from 240 kg N ha−1 to

e years (full square), the 1/3 driest coolest year category referred to as the 1/3 higher solarthe Yaqui Valley in Sonora, Mexico sown on November 15 with 240 kg N fertilizer ha−1.perature and D) seasonal daily mean solar radiation.

Fig. 7. Average net return and N cost under the current N (CN) fertilizer practices of240 kg N ha−1, reduced N (RN) fertilizer of 150 kg N ha−1 and of season-specific Nfertilizer applications: using an always-correct-season-type forecast (ACF) category toapply 150 kg N ha−1 in the 1/3 higher solar radiation tercile years and 120 kg N ha−1

for all other years; using a Global Circulation Model-based forecast (GCMF) to apply150 kg N ha−1 only when the high solar radiation condition is forecasted and120 kg N ha−1 otherwise; and, using a 3-month El Niño Southern Oscillation-basedforecast (ENSOF) of August–September–October (ASO) to apply 150 kg N ha−1 during aLa Niña phase, the highest-yielding phase, and on average 128 kg N ha−1 otherwise.

Fig. 6.Average net return of N response for the 1/3 higher solar radiation tercile years (fullsquare) and all other years (open square). Average error bars show ± one standarddeviation from mean.

82 M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

150 kg N ha−1 in all years generated a net return gain of 102 USD ha−1

and a N fertilizer cost savings of 112 USD ha−1 (Fig. 7). The use of a sea-sonal forecast could beused to further lower N rates based on forecastedseason type for additional cost savings.

The value of the seasonal forecast was determined by calculating thenet returns andN fertilizer cost for each forecast category. Using a hypo-thetical always-correct-forecast (ACF) based on actualweather data, ap-plying 150 kg N ha−1 for the 1/3 higher solar radiation years and120 kg N ha−1 for other years generated an annual net return of1117 USD ha−1 and an annual N fertilizer cost of 162 USD ha−1.

In the case of the General Circulation Model-based forecast (GCMF),when applying 150 kg N ha−1 in the forecasted higher solar radiationyears and 120 kg N ha−1 for other years, annual net returns were1112 USD ha−1 with annual N fertilizer cost of 162 USD ha−1. Whenusing the GCMF, the forecast correctly predicted 4 of 9 years for the 1/3 higher solar radiation category.

For the ENSO-based forecast (ENSOF),when applying 150 kg N ha−1

during a La Niña phase and 128 kg N ha−1 for other phases, the annualnet returns were 1112 USD ha−1 and the annual N fertilizer cost was168 USD ha−1 (Fig. 7). The ENSOF predicted correctly 7 of 10 La Niñayears.

3.3. Value of seasonal forecast in the SIS

Fig. 8 shows the net returns of the supplementary irrigated system(SIS) for three different initial stored soil water contents at sowing, in-cluding 0%, 50%, and 100% of total plant available soil water content(PAWC). With an initial 0% PAWC, sowing a crop every year resultedin net losses in almost all years. Hence, some initial soil water was crit-ical for profitability of the SIS. However, always sowing a crop generateda positive net return in most years with initial soil water at 50% PAWCand in all years for 100% PAWC.

3.3.1. Zero percent initial soil water contentDeciding to sow a crop based on ACF in the SIS could result in higher

net returns, but only with dry initial soil water conditions (0% PAWC). Ifirrigation was limited to 100mm ha−1 season−1, applying this amountof irrigation togetherwith 120 kgN ha−1 on a initially dry soil in the 1/3wettest rainfall seasons and not sowing otherwise could result in an an-nual net return of 65 USD ha−1 season−1 with a hypothetical ACF.

When using a GCMF, the forecast predicted 4 of 9 years of the 1/3wettest season-type category. Several times this forecast predicted a

wet season when the season was dry, and the wrong forecast wouldhave resulted in net losses. The simulated seasonal net return with aGCMF was −10 USD ha−1 compared to −104 USD ha−1 if the cropwas always sown on dry soil.

For the ENSOF, the El Niño seasons captured only 4 of 9 years of the1/3 wettest category, meaning a correct forecast of an El Niño phase didnot necessarily guarantee a wet wheat-growing season. The simulatednet return using an ENSOF was −4 USD ha−1 season−1 compared to−104 USD ha−1 season−1 if the crop was always sown on dry soil.

3.3.2. Fifty percent initial soil water contentFollowing a seasonal forecast instead of always sowing with an ini-

tially half-full soil water profile had no added value in the SIS. The netreturn for always sowing on an initially half-full soil water profile(298 USD ha−1 season−1) was higher than those obtained from follow-inganACF(209USDha−1 season−1), aGCMF(113USDha−1 season−1),and an ENSOF (122 USD ha−1 season−1) (Fig. 10)

3.3.3. One hundred percent of initial soil water contentAlways sowing with an initially full soil water profile resulted in

higher net returns than following a seasonal forecast. The net return(465 USD ha−1 season−1) for always sowing on an initially full soilwater profile was higher than those obtained from following an ACF(253 USD ha−1 season−1), a GCMF (171 USD ha−1 season−1), and anENSOF (178 USD ha−1 season−1) (Fig. 10).

3.4. Value of seasonal forecast in the RFS

Fig. 9 shows the net returns for the three different initial PAWC inthe rainfed system. When there was no soil water at sowing, sowing acrop every year in the RFS resulted in net losses in all years exceptone. As expected, cultivating crops without any supplementary irriga-tion is not feasible in this arid environment because of insufficient rain-fall in most seasons. However, always sowing a crop generated a

83M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

positive net return for half of years when initial soil water was 50%PAWC and in all years when initial soil water was 100% PAWC.

3.4.1. Zero percent initial soil water contentDry soil at the beginning of the season increased the risk of crop fail-

ure in an ACF (−20 USD ha−1 season−1), in a GCMF (−43 USD ha−1-

season−1), and in an ENSOF (−36 USD ha−1 season−1), but followingthe forecasts prevented losses of 224 USD ha−1 season−1 on averagecompared to always sowing a crop on dry soil.

3.4.2. Fifty percent initial soil water contentIn rainfed conditions, the ACF had added value, but only when

sowing a crop and applying 60 kg N ha−1 on a half-full soil waterprofile in the 1/3 wettest seasons and not sowing otherwise. TheACF could produce a simulated net return of 123 USD ha−1 season−1

compared to always sowing (109 USD ha−1 season−1). The GCMFreturned a positive net return with a half-full soil water profile of55 USD ha−1 season−1, but the net return was still less than the al-ways sowing strategy (109 USD ha−1 season−1). The ENSOF net re-turn on a half-full soil water profile, 62 USD ha−1 season−1, wasalso less than the always sowing strategy (109 USD ha−1 season−1)and therefore had no added value.

Fig. 8. Net returns from always sowing a crop (100 mm irrigation and 120 kg N ha−1) (open b(cross), a Global Circulation Model-based forecast (GCMF) (full circle) and an El Niño Southernirrigation and 120 kg N ha−1) only in the 1/3 wettest seasons for different initial plant-availab

3.4.3. One hundred percent initial soil water contentWith a full stored soil water at the beginning of the season a forecast

in the RFS also resulted in no added value. The net return of the ACFwas194 USD ha−1 season−1 compared to 317 USD ha−1 season−1 with al-ways sowing (Fig. 10). The net return of a GCMF was 122 USD ha−1-

season−1, and the net return of an ENSOF was125 USD ha−1 season−1 compared to 317 USD ha−1 season−1 with al-ways sowing (Fig. 10).

4. Discussion

Several caveats bear mentioning. First, the results obtained in thisstudy are based on simulations run using a single representative soil(Verhulst et al. 2009). Although the soils in the Yaqui Valley are relative-ly uniform, the results obtained in this analysis might have been differ-ent if performed for other soil types (Lobell et al. 2002). Similarly, thevalue of seasonal forecast was calculated considering a constant priceand the results might have change if a different price structure hadbeen considered in the analysis (Asseng et al. 2012; Mauget et al.2009). Market feedbacks were also not considered as wheat prices aremostly determined at a global scale. Other adaptation options thatwere not considered include changing planting dates and cultivars,

ars), using an always-correct-season-type forecast (ACF) for 1/3 wettest seasons categoryOscillation-based forecast (ENSOF) (full diamond) to decide on sowing a crop (100 mm

le stored soil water at sowing with A) 0%, B) 50% and C) 100%.

84 M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

which are less responsive to seasonal variability in irrigated croppingsystems. Finally, a change in the crop mix as a management strategywas not considered in this analysis as wheat is the main crop grownin this region and farmers would normally plant only once a year dueto water limitations. But, as found in other studies (Sultan et al. 2010)a change in the crop mix can be used to mitigate the negative effect ofclimate variability and can also lead to a change in the value of climateinformation (Choi et al. 2015).

Such caveats notwithstanding, the present study showed that sea-sonal forecasts could potentially benefit farmers in both rainfed and ir-rigated systems. Despite no water limitations in FIS, the simulatedwheat yield variability in the arid environment of the Yaqui Valley islarge, as confirmed by field experiments (Sayre et al. 1997); therefore,it is possible to target higher-yielding seasons with season-specificcropmanagement. The value of seasonal forecasts comes from correctlypredicting the years when weather conditions result in potentiallyhigher-yielding seasons. Under rainfed conditions, seasonal rainfall pre-dictions help target “good” or wetter seasons (Asseng et al. 2012; Joneset al. 2000;Moeller et al. 2008). In the present analysis, seasonal predic-tions for temperature and solar radiation resulted in potential benefitsto farmers. Similar utility of temperature forecasting was observed forfrost damage (Castellanos et al. 2009; Kala et al. 2009).

Fig. 9.Net returns from always sowing a crop (rainfed and60 kgN ha−1) (openbars), using an aCirculation Model-based forecast (GCMF) (full circle) and an El Niño Southern Oscillation-baseonly in the 1/3 wettest seasons for different initial plant-available stored soil water at sowing

The present analysis reported non-viable cropping activitieswithoutforecasting in supplementary and rainfed conditions for the Yaqui Val-ley; exceptions to this are seasons with a full initial soil water profile,which is a scenario that tends to decrease as water becomes scarcer.However, following a skillful seasonal forecast and monitoring the ini-tial plant-available soil water at sowing could help farmers achieve apositive net return under limited irrigation.

The price of water is often subsidized in irrigated systems, andfarmers in these systems often apply an excessive amount of N to guar-antee a high yield (Christensen et al. 2006; Lobell and Ortiz-Monasterio2006; Lobell et al. 2004). This approach is different from rainfed farmersof Australia. In these cases, N fertilizer is applied as long as a 2:1 returnto investment ratio is guaranteed (Asseng et al. 2012). Carberry et al.(2000) found crop rotation management decisions vary with thefarmer's objectives. Different crop rotation systems are chosen basedon the concerns of the farmer (e.g. gross margin, risk, cash flow, or soilloss). On the other hand, water requirements vary if the farmer is opti-mizing water productivity or crop yield (Chen et al. 2010; Zhao et al.2013). In this study, we found that N fertilizer application could be re-duced by N90 kg N ha−1 if the farmers were interested in maximizingnet returns as well as reducing risk. A risk-mitigating incentive wouldinclude reducing input costs, such as N cost. Matching N with water

lways-correct-season-type forecast (ACF) for 1/3wettest seasons category (cross), a Globald forecast (ENSOF) (full diamond) to decide on sowing a crop (rainfed and 60 kg N ha−1)with A) 0%, B) 50% and C) 100%.

85M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

availability could reduce profit risk and improve resource use efficiencyat the same time (Cossani et al. 2009; Cossani et al. 2010; Sadras et al.2003; Sadras 2004). Improvements in yields and water use efficiencyhave already been achieved in farmers' fields using models for wateravailability and predictions for optimized N applications in time andamount under rainfed Mediterranean systems (Cossani et al. 2011). Askillful seasonal forecast in irrigated regions could help farmers to re-duce inputs or improve input use efficiency. As seasonal forecasts in-crease in skill, they will become important tools for mitigating climaterisk and managing resources more efficiently. However, seasonal fore-casts are imperfect (Hammer 2000), and farmers willing to use suchtools will need to take this into account when making managementdecisions.

Little difference was found between alternative forecast systems infully irrigated systems. This implies that the forecast skill was not a de-terminant of net return under no water limitations, but was critical inreducing N fertilizer applications. The value of a ACF in the Yaqui Valleycompared to a fixed N application of 150 kg N ha−1 across all years was2 USDha−1; this is in the lower range of forecast benefits in rainfed con-ditions (Asseng et al. 2012; Jones et al. 2000), but includes N-fertilizersavings of 26 USD ha−1 when compared to the simulated optimized Nfertilizer application of 150 kg N ha−1. Much higher forecast benefitsare shown when compared with current N fertilizer practices of irriga-tion farmers. GCMs currently predict precipitation and temperature(Drosdowsky and Allan 2000; Goddard et al. 2003). In this study, the in-teraction between low temperature and high solar radiation resulted inpotentially higher yields, as is expected because of higher photosynthet-ic activity and biomass production. Because GCMs do not output solarradiation, the negative correlation between solar radiation and precipi-tation can be used to identify potentially high-yielding seasons byselecting the driest-coolest seasons as proxy for the higher solar radia-tion seasons, as shown in the present study. However, the skill obtainedfrom the GCM for the higher solar radiation season category was lowercompared to rainfall categories in other regions (Moeller et al. 2008). Ifseasonal forecasts are applied in the irrigated regions, predictions of

Fig. 10. Average net return in A) a supplementary irrigation system (SIS) (100 mmirrigation and 120 kg N ha−1) and B) a rainfed system (RFS) (no irrigation and60 kg N ha−1) for always sowing a crop (square), sowing a crop in the 1/3 wettestseasons only according an always-correct-season-type forecast (ACF) (circle), usinga Global Circulation Model-based forecast (GCMF) (triangle) and an El NiñoSouthern Oscillation-based forecast (ENSOF) (diamond). Average error bars areshown (±1 s.d. of the mean).

cloudiness (Meinke and Stone 2005) and solar radiation could help pre-dict those potentially high-yielding seasons more accurately.

This paper discussed both potential and actual seasonal forecasts, in-dicating that the latter still has a limited value in this region because oflow forecast skill. These forecasts have an S% value of 0.41 for tempera-ture and 0.26 for precipitation, which is below the skill found in Austra-lia for precipitation (Moeller et al. 2008). Similar to other reports, an ACFresulted in higher value than the best actual GCMF and ENSOF (Jones etal. 2000) in the three analyzed cropping systems. Even though theENSOF had high skill levels in predicting the ENSO category, predictingthe tercile season-type correctly generated more value than predictingthe ENSO phases correctly.

In this study, the performance of the seasonal forecast was based onthe tercile season-type correct prediction. However, different benefitscould be obtained using a different forecast or employing a higher reso-lution of weather clustering.

In water-limited conditions, the correct prediction of seasonal rain-fall is critical. Under restricted water conditions, seasonal rainfall andstored soil water at sowing are the key sources ofwater for crop growth,biomass production, and yield, as observed previously in North China(Chen et al. 2010). Additionally, Lobell and Ortiz-Monasterio (2006)suggested reducing irrigation in the Yaqui Valley has a negative effecton yield and depends mainly on the initial soil water conditions; theirfinding is supported in this study and also validated by several studiesin other environments (Asseng et al. 2012; Carberry et al. 2000;Hammer et al. 1996).

The value of seasonal forecasts to assist in potentially higher-yield-ing seasons will gain importance as mix rainfed-irrigation systems be-come more common in irrigated regions (Shiferaw et al. 2013). So far,many irrigated regions use fixed seasonal water allocations regardlessof the seasonal conditions, and farmers tend to use more water thanthey needwhen they do not have to pay for it (Mishra et al. 2013). How-ever, as water demands shift from agriculture to industry and house-holds (Zhao et al. 2013), and wheat cropping moves from irrigated torainfed production (Shiferaw et al. 2013), farmers will have to managetheir resources more efficiently. As access to water becomes more re-stricted, farmers will have to rely more on in-season rainfall, and thevalue of predicting a wet season will be increased.

5. Conclusion

The present study demonstrated the potential of applying a seasonalforecast in irrigated cropping systems to maximize net returns and re-duce input costs. In irrigated cropping systems with restricted wateravailability, seasonal forecasts for sowing decisions must be linked tostored soil water information at the beginning of the season. The useof seasonal forecasts as a decision-support tool in irrigated regionswill gain importance as resources (e.g. water) become scarcer. Howev-er, its potential benefits for farmers will only be achieved if the currentforecast skill can be increased.

Acknowledgements

We would like to thank Conacyt (México) for the financial supportto Melissa A. Ramírez Rodrigues and for support from the CGIAR Re-search Program on Climate Change, Agriculture and Food Security(CCAFS) and the South East Climate Consortium (SECC).

References

Abdelghani, C., et al., 2008. Using remotely sensed data to estimate area-averaged dailysurface fluxes over a semi-arid mixed agricultural land. Agric. For. Meteorol. 148(3), 330–342.

Asseng, S., 2004. Wheat Crop Systems — A Simulation Analysis. CSIRO Publishing.Asseng, S., et al., 1998. Use of the APSIM wheat model to predict yield, drainage, and

NO3′(−) leaching for a deep sand. Aust. J. Agric. Res. 49 (3), 363–377.

86 M.A. Ramírez-Rodrigues et al. / Agricultural Systems 147 (2016) 76–86

Asseng, S., Fillery, I.R.P., Dunin, F.X., Keating, B.A., Meinke, H., 2001a. Potential deep drain-age under wheat crops in a Mediterranean climate. I. Temporal and spatial variability.Aust. J. Agric. Res. 52 (1), 45–56.

Asseng, S., Foster, I., Turner, N.C., 2011. The impact of temperature variability on wheatyields. Glob. Chang. Biol. 17 (2), 997–1012.

Asseng, S., McIntosh, P.C., Wang, G.M., Khimashia, N., 2012. Optimal N fertilizer manage-ment based on a seasonal forecast. Eur. J. Agron. 38, 66–73.

Asseng, S., Milroy, S.P., 2006. Simulation of environmental and genetic effects on grainprotein concentration in wheat. Eur. J. Agron. 25 (2), 119–128.

Asseng, S., Turner, N.C., Keating, B.A., 2001b. Analysis of water- and nitrogen-use efficien-cy of wheat in a Mediterranean climate. Plant Soil 233 (1), 127–143.

Balwinder, S., Gaydon, D.S., Humphreys, E., Eberbach, P.L., 2011. The effects of mulch andirrigation management on wheat in Punjab, India-evaluation of the APSIM model.Field Crop Res. 124 (1), 1–13.

BeVier, G., 2012. Global food systems: feeding the world. Reprod. Domest. Anim. 47,77–79.

Calanca, P., Bolius, D., Weigel, A.P., Liniger, M.A., 2011. Application of long-range weatherforecasts to agricultural decision problems in Europe. J. Agric. Sci. 149, 15–22.

Carberry, P., Hammer, G., Meinke, H., Bange, M., 2000. The potential value of seasonal cli-mate forecasting in managing cropping systems. In: Hammer, N.N.a.C.M.G.L. (Ed.),Applications of Seasonal Climate Forecasting in Agricultural and Natural Ecosystems.The Australian Experience. Kluwer Academic Publishers, The Netherlands.

Castellanos, M.T., Tarquis, A.M., Morato, M.C., Saa, A., 2009. Forecast of frost days based onmonthly temperatures. Span. J. Agric. Res. 7 (3), 513–524.

CIMMYT and Stanford-University, 2011. Wheat Production and Farmer's Practices in theYaqui Valley. Sonora, Mexico. http://www.cimmyt.org/en/about-us/job-opportunities/1457-12,242-maize-seed-systems-specialist-for-latin-america.

Cossani, C.M., Savin, R., Slafer, G.A., 2007. Contrasting performance of barley and wheat ina wide range of conditions in Mediterranean Catalonia (Spain). Ann. Appl. Biol. 151(2), 167–173.

Cossani, C.M., Slafer, G.A., Savin, R., 2009. Yield and biomass in wheat and barley under arange of conditions in a Mediterranean site. Field Crop Res. 112 (2-3), 205–213.

Cossani, C.M., Slafer, G.A., Savin, R., 2010. Co-limitation of nitrogen andwater on yield andresource-use efficiencies of wheat and barley. Crop Pasture Sci. 61, 844–851.

Cossani, C.M., Thabet, C., Mellouli, H.J., Slafer, G.A., 2011. Improving wheat yields throughN fertilization in Mediterranean Tunisia. Exp. Agric. 47 (03), 459–475.

Chen, C., Wang, E., Yu, Q., 2010. Modelling the effects of climate variability and watermanagement on crop water productivity and water balance in the North ChinaPlain. Agric. Water Manag. 97 (8), 1175–1184.

Choi, H.S., et al., 2015. Potential effects of perfect seasonal climate forecasting on agricul-tural markets, welfare and land use: a case study of Spain. Agric. Syst. 133, 177–189.

Christensen, L., Riley, W.J., Ortiz-Monasterio, I., 2006. Nitrogen cycling in an irrigatedwheat system in Sonora, Mexico: measurements and modeling. Nutr. Cycl.Agroecosyst. 75 (1–3), 175–186.

Drosdowsky, W., Allan, R., 2000. The potential for improved statistical seasonal climateforecasts. In: Hammer, N.N.a.C.M.G.L. (Ed.), Applications of Seasonal Climate Forecast-ing in Agricultural and Natural Ecosystems. The Australian Experience. Kluwer Aca-demic Publishers, The Netherlands.

Food and Agriculture Organization, 2012. FAOSTAT: Food balance Sheets. Seefaostat.fao.org.

Goddard, L., Barnston, A.G., Mason, S.J., 2003. Evaluation of the IRI's “net assessment” sea-sonal climate forecasts 1997–2001. Bull. Am. Meteorol. Soc. 84 (12), 1761.

Hammer, G., 2000. A general systems approach to applying seasonal climate forecasts. In:Hammer, N.N.a.C.M.G.L. (Ed.), Applications of Seasonal Climate Forecasting in Agri-cultural and Natural Ecosystems. The Australian Experience. Kluwer Academic Pub-lishers, the Netherlands.

Hammer, G.L., Holzworth, D.P., Stone, R., 1996. The value of skill in seasonal climate fore-casting to wheat crop management in a region with high climatic variability. Aust.J. Agric. Res. 47 (5), 717–737.

Hunt, B., Hirst, A.C., 2000. Global climatic models and their potential for seasonal climateforecasting. In: Hammer, N.N.a.C.M.G.L. (Ed.), Applications of Seasonal Climate Fore-casting in Agricultural and Natural Ecosystems The Australian Experience. Kluwer Ac-ademic Publishers, the Netherlands.

Jolliffe, I.T., Stephenson, D.B. (Eds.), 2003. Forecast Verification: A Practitioner's Guide inAtmospheric Science. John Wiley & Sons Ltd., West Sussex, England.

Jones, J.W., Hansen, J.W., Royce, F.S., Messina, C.D., 2000. Potential benefits of climate fore-casting to agriculture. Agric. Ecosyst. Environ. 82 (1–3), 169–184.

Kala, J., Lyons, T.J., Foster, I.J., Nair, U.S., 2009. Validation of a simple steady-state forecastof minimum nocturnal temperatures. J. Appl. Meteorol. Climatol. 48 (3), 624–633.

Keating, B.A., et al., 2003. An overview of APSIM, a model designed for farming systemssimulation. Eur. J. Agron. 18 (3–4), 267–288.

Licker, R., et al., 2010. Mind the gap: how do climate and agricultural management ex-plain the ‘yield gap’ of croplands around the world? Glob. Ecol. Biogeogr. 19 (6),769–782.

Liu, X.W., Shao, L.W., Sun, H.Y., Chen, S.Y., Zhang, X.Y., 2013. Responses of yield and wateruse efficiency to irrigation amount decided by pan evaporation for winter wheat.Agric. Water Manag. 129, 173–180.

Lobell, D.B., Cassman, K.G., Field, C.B., 2009. Crop yield gaps: their importance, magni-tudes, and causes, annual review of environment and resources. Annu. Rev. Environ.Resour. 179–204.

Lobell, D.B., Ortiz-Monasterio, J.I., 2006. Evaluating strategies for improved water use inspring wheat with CERES. Agric. Water Manag. 84 (3), 249–258.

Lobell, D.B., Ortiz-Monasterio, J.I., 2008. Satellite monitoring of yield responses to irriga-tion practices across thousands of fields. Agron. J. 100 (4), 1005–1012.

Lobell, D.B., Ortiz-Monasterio, J.I., Addams, C.L., Asner, G.P., 2002. Soil, climate, and man-agement impacts on regional wheat productivity in Mexico from remote sensing.Agric. For. Meteorol. 114 (1–2), 31–43.

Lobell, D.B., Ortiz-Monasterio, J.I., Asner, G.P., 2004. Relative importance of soil and cli-mate variability for nitrogen management in irrigated wheat. Field Crop Res. 87 (2–3), 155–165.

Lobell, D.B., Ortiz-Monasterio, J.I., Asner, G.P., Naylor, R.L., Falcon, W.P., 2005. Combiningfield surveys, remote sensing, and regression trees to understand yield variations inan irrigated wheat landscape. Agron. J. 97 (1), 241–249.

Lobell, D.B., Sibley, A., Ortiz-Monasterio, J.I., 2012. Extreme heat effects on wheat senes-cence in India. Nat. Clim. Chang. 2 (3), 186–189.

Lopes, M.S., et al., 2013. QTL for yield and associated traits in the Seri/Babax populationgrown across several environments in Mexico, in the West Asia, North Africa, andSouth Asia regions. Theor. Appl. Genet. 126 (4), 971–984.

Lv, Z.F., Liu, X.J., Cao,W.X., Zhu, Y., 2013. Climate change impacts on regional winterwheatproduction in main wheat production regions of China. Agric. For. Meteorol. 171,234–248.

Mauget, S., Zhang, J., Ko, J.H., 2009. The value of ENSO forecast information to dual-pur-pose winter wheat production in the US Southern High Plains. J. Appl. Meteorol.Climatol. 48 (10), 2100–2117.

Meinke, H., Stone, R., 2005. Seasonal and inter-annual climate forecasting: the new toolfor increasing preparedness to climate variability and change in agricultural planningand operations. Clim. Chang. 70 (1-2), 221–253.

Mishra, A., Siderius, C., Aberson, K., van der Ploeg, M., Froebrich, J., 2013. Short-term rain-fall forecasts as a soft adaptation to climate change in irrigation management inNorth-East India. Agric. Water Manag. 127, 97–106.

Moeller, C., Smith, I., Asseng, S., Ludwig, F., Telcik, N., 2008. The potential value of seasonalforecasts of rainfall categories — case studies from the wheatbelt in WesternAustralia's Mediterranean region. Agric. For. Meteorol. 148 (4), 606–618.

NOAA, 2014. http://www.cpc.ncep.noaa.gov./products/analysis_monitoring/ensostuff/ensoyears.shtml.

Ortiz-Monasterio, J.I., Raun,W., 2007. Reduced nitrogen and improved farm income for ir-rigated spring wheat in the Yaqui Valley, Mexico, using sensor based nitrogen man-agement. J. Agric. Sci. 145, 215–222.

Reynolds, M.P., Balota, M., Delgado, M.I.B., Amani, I., Fischer, R.A., 1994. Physiological andmorphological traits associated with spring wheat yield under hot irrigated condi-tions. Aust. J. Plant Physiol. 21 (6), 717–730.

Riley, W.J., Ortiz-Monasterio, I., Matson, P.A., 2001. Nitrogen leaching and soil nitrate, ni-trite, and ammonium levels under irrigated wheat in Northern Mexico. Nutr. Cycl.Agroecosyst. 61 (3), 223–236.

Romero, C.C., et al., 2012. Reanalysis of a global soil database for crop and environmentalmodeling. Environ. Model Softw. 35, 163–170.

Ruane, A.C., Winter, J.W., Prabhat McDermid, S., 2013. AgMIP Climate Datasets and Sce-narios, ASA, CSSA and SSSA International Annual Meeting Symposium-Perspectiveson Climate Effects on Agriculture: The International Efforts of AgMIP, Tampa, Florida.

Sadras, V., Roget, D., Krause, M., 2003. Dynamic cropping strategies for risk managementin dry-land farming systems. Agric. Syst. 76 (3), 929–948.

Sadras, V.O., 2004. Yield and water-use efficiency of water- and nitrogen-stressed wheatcrops increase with degree of co-limitation. Eur. J. Agron. 21 (4), 455–464.

SAGARPA, 2008. Costos de Producción de Trigo Ciclo 2007–2008.Saha, S., et al., 2010. The NCEP Climate Forecast System reanalysis. Bull. Am.Meteorol. Soc.

91 (8), 1015–1057.Saha, S., et al., 2006. The NCEP climate forecast system. J. Clim. 19 (15), 3483–3517.Sayre, K.D., Rajaram, S., Fischer, R.A., 1997. Yield potential progress in short bread wheats

in northwest Mexico. Crop Sci. 37 (1), 36–42.Schoups, G., Addams, C.L., Minjares, J.L., Gorelick, S.M., 2006. Sustainable conjunctive

water management in irrigated agriculture: model formulation and application tothe Yaqui Valley, Mexico. Water Resour. Res. 42 (10).

Seifert, C., Ortiz-Monasterio, J.I., Lobell, D.B., 2011. Satellite-based detection of salinity andsodicity impacts on wheat production in the Mexicali Valley. Soil Sci. Soc. Am. J. 75(2), 699–707.

Shiferaw, B., et al., 2013. Crops that feed the world 10. Past successes and future chal-lenges to the role played by wheat in global food security. Food Secur. 5 (3), 291–317.

Sultan, B., Barbier, B., Fortilus, J., Mbaye, S.M., Leclerc, G., 2010. Estimating the potentialeconomic value of seasonal forecasts in West Africa: a long-term ex-ante assessmentin Senegal. Weather Clim. Soc. 2 (1), 69–87.

Verhulst, N., Deckers, J., Govaerts, B., 2009. Classification of the soil at CIMMYT's experi-mental station in the Yaqui Valley near Ciudad Obregón, Sonora, Mexico, CIMMYT Re-port Mexico, D.F. CIMMYT.

Verhulst, N., et al., 2011. Wheat yield and tillage-straw management system x year inter-action explained by climatic co-variables for an irrigated bed planting system innorthwestern Mexico. Field Crop Res. 124 (3), 347–356.

Yu, Q., Wang, E.L., Smith, C.J., 2008. A modelling investigation into the economic and en-vironmental values of ‘perfect’ climate forecasts for wheat production under con-trasting rainfall conditions. Int. J. Climatol. 28 (2), 255–266.

Yuan, X., Wood, E.F., Luo, L., Pan, M., 2011. A first look at Climate Forecast System version2 (CFSv2) for hydrological seasonal prediction. Geophys. Res. Lett. 38 (L13402), 1–7.

Zhao, Z., Huang, X., Xu, J., 2013. The effects of irrigation on wheat yield on a global scale.Appl. Eng. Agric. 29 (3), 335–342.