Tm

Na

b

ARRA

KACPWBG

I

dmoowBvraitlDt

0h

Land Use Policy 35 (2013) 119– 130

Contents lists available at SciVerse ScienceDirect

Land Use Policy

jou rn al hom epage: www.elsev ier .com/ locate / landusepol

he impact of climate and price risks on agricultural land use and cropanagement decisions

iklaus Lehmanna,∗, Simon Brinera, Robert Fingerb

Institute for Environmental Decisions, Agrifood & Agri-environmental Economics Group, ETH Zurich, Sonneggstrasse 33, CH-8092 Zurich, SwitzerlandAgricultural Economics and Rural Policy Group, Wageningen University, Hollandseweg 1, NL-6706 KN Wageningen, The Netherlands

a r t i c l e i n f o

rticle history:eceived 16 July 2012eceived in revised form 9 May 2013ccepted 16 May 2013

eywords:gricultural land uselimate changerice riskshole-farm model

ioeconomic modellingenetic algorithm

a b s t r a c t

This article aims to investigate the impacts of climate change and of lower and more volatile crop pricelevels as currently observed in the European Union (EU) on optimal management decisions, averageincome and income risks in crop production in Western Switzerland. To this end, a bioeconomic whole-farm model has been developed that non-parametrically combines the crop growth model CropSyst withan economic decision model using a genetic algorithm. The analysis focuses on the farm level, whichenables us to integrate a wide set of potential adaptation responses, comprising changes in agriculturalland use as well as crop-specific fertilization and irrigation strategies. Furthermore, the farmer’s certaintyequivalent is employed as objective function, which enables the consideration of not only impacts onaverage income but also impacts on income variability.

The study shows that that the effects of EU crop prices on the optimal management decisions as well ason the farmer’s certainty equivalent are much stronger than the effects of climate change. Furthermore,our results indicate that the impacts of income risks on the crop farm’s optimal management schemes

are of rather low importance. This is due to two major reasons: first, direct payments make up a largepercentage of the agricultural income in Switzerland which makes Swiss farmers less vulnerable to mar-ket and climate volatility. Second, arable crop farms in Switzerland have by law to cultivate at least fourdifferent crops. Due to these diverse cropping systems and high government direct payments risk doesneither under climate change, market liberalization nor combinations thereof, play a very decisive rolein arable farming in Switzerland.

ntroduction

Production and price risks are important aspects in farmers’ecision-making (Saunders et al., 1997; Angus et al., 2009). Whilearket or price risk reflects the variations in prices of agricultural

utputs and inputs (Harwood et al., 1999), production risks mainlyccur because crop growth highly depends on its environment (e.g.,eather conditions and pest pressure) that can rapidly change.oth production and market risks, however, affect the incomeariability in agriculture. To cope with production and marketisks, farmers typically have several on-farm, self-insuring optionsnd risk-mitigation measures to protect against income volatil-ty. One of the most important on-farm risk-reducing strategies iso diversify farm activities, for example by expanding the portfo-

io of different agricultural land uses (Mishra and El-Osta, 2002).iversification strategies not only mitigate price risks but also fluc-

uations of overall farm outputs due to production risks (Mishra and

∗ Corresponding author. Tel.: +41 44 632 28 32; fax: +41 44 632 10 86.E-mail address: nlehmann@ethz.ch (N. Lehmann).

264-8377/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.landusepol.2013.05.008

© 2013 Elsevier Ltd. All rights reserved.

El-Osta, 2002). Besides such large-scale strategies, also adjustmentsof crop-specific management decisions potentially mitigate incomevariability (Sandmo, 1971). In general, risk-averse decision makersare expected to invest less in inputs if the returns from these invest-ments are more uncertain and thus increase income variability. Forinstance, higher nitrogen application on grassland tends to increaseyield variability (for discussions and examples, see Finger, 2012). Incontrast, more intensive use of irrigation decreases the variability ofcrop yields, thereby reducing production risks (Finger et al., 2011;Lehmann et al., 2013). Responses of farmers to changing market andproduction conditions are highly relevant for agricultural and envi-ronmental policy makers because the induced changes in land useas well as changes in input allocation have direct impacts on foodsupply, environmental loads from agriculture and the landscape.

Whole-farm models are appropriate tools to assess the impactof price and climate scenarios on farmers’ management strategies,average income and income variability (Pannell et al., 2000). This is

because the full potential of adjusting crop-specific managementschemes for risk management is only tapped if all activities of afarm are considered simultaneously. In contrast, single-crop inves-tigations may over-estimate the role of production and price risks

1 Use Po

ifgfoeR2oco2

ctadasncTaTtadpspe2mtbonmeoiCttie

cofvMtbmptepSFiFiiG

variables, which potentially lead to higher objective values. Thenagain, all sets of decision variables comprised in the new populationare used as input variables in CropSyst and the economic decision

20 N. Lehmann et al. / Land

n agricultural decision-making. In addition, the assessment at thearm level is also of great importance since risk management strate-ies often are dependent on specific constraints with regard toarm resources (e.g., land and working time) and environmentalbligations (e.g., nutrient balances). Most available studies, how-ver, focus on single-crop management decisions (Rosegrant andoumasset, 1985; Rajsic et al., 2009; Finger, 2012; Lehmann et al.,013). Other studies use whole-farm models but address only theptimal land allocation among different crops without consideringrop-specific management decisions such as nitrogen fertilizationr irrigation intensities (Chavas and Holt, 1990; Sckokai and Moro,006; Musshoff and Hirschauer, 2009).

Based on this background, we combine the process-basedrop growth model CropSyst with an economic decision modelo develop a whole-farm model that accounts not only for landllocation but also for crop-specific management decisions. Theeveloped bioeconomic whole-farm model is used to maximize

farmer’s utility while optimizing farm scale management deci-ions under different climate and price scenarios. Thus, we use aormative approach based on the neoclassical theory, which per-eives economic agents as utility optimizers (Buysse et al., 2007).he outcome of the economic decision model is therefore a man-gement scheme that results in the highest utility levels for farmers.o express farmers’ utility levels, the certainty equivalent (CE) athe farm level is used. The CE depends not only on the total aver-ge farm income but also on income variability that accounts forifferent sources of risk. In previous research, the combination ofrocess-based crop growth models with economic models has beenuggested to investigate the influence of climate change (CC) androduction risks in cropping systems (for discussions, see Challinort al., 2009; Reidsma et al., 2010; Finger et al., 2011; Olesen et al.,011). One of the main advantages of process-based crop growthodels is their ability to simulate plant growth under scenarios

hat exceed the current conditions (Finger, 2009). Thus, process-ased crop growth models are suitable tools for the simulationf crop yields under CC scenarios. Yet, crop models generally doot consider market- and policy-driven adaptive responses to cropanagement (Risbey et al., 1999). By linking crop growth mod-

ls with economic decision models, however, adaptation decisionsf farmers to changing market and policy conditions can be takennto account. In this study, the linkage of the crop growth modelropSyst with the economic decision model and the optimiza-ion routine is facilitated by a genetic algorithm (GA). To analyzehe influence of changes in climate and market prices on farmers’ncome, income volatility and farm management decisions, differ-nt climate and price scenarios are considered.

The developed model is applied to a representative arablerop farm located in the Broye watershed in the Western partf Switzerland. This region already faces high variability of rain-all within the growing season, which leads to a high crop yieldariability and triggers the frequent use of irrigation (Robra andastrullo, 2011). The frequent use of irrigation causes environmen-

al problems, such as low water levels in the region’s surface waterodies (Mühlberger de Preux,2008). Land use and crop-specificanagement decisions taken by the region’s farmers are thus of

articular relevance for policy makers. This policy relevance is fur-hermore underlined by the fact that significant changes in riskxposure of Swiss farmers are expected. Currently, average croprices are much higher, and crop price volatility is much smaller inwitzerland than in other European countries (El Benni et al., 2012;inger and El Benni, 2012). For instance, the average price of wheatn Switzerland is about three times higher than in Germany or

rance (Finger and El Benni, 2012). The relative wheat price volatil-ty (expressed as coefficient of variation) in Switzerland, however,s about fifty percent smaller than those observed in France andermany (Finger and El Benni, 2012). In the future, trade of

licy 35 (2013) 119– 130

agricultural products between Switzerland and the European Unionmight be liberalized, leading to lower and more volatile prices ofagricultural goods in Switzerland. Moreover, significant changes inproduction risks in Swiss crop production are expected due to CC(Torriani et al., 2007). These changes, however, are expected to beheterogeneous across different crops (Lehmann, 2010).

The objectives of the presented study are threefold: First, wedevelop a whole-farm model that is used to identify optimal man-agement decisions for a representative arable farm in the Broyewatershed and compare our modelling results with observationsfrom the study region. Second, we assess the impacts of CC andcrop price scenarios on the optimal management decisions. Finally,we quantify the impact of CC and crop price scenarios on farm-ers’ income and income risks while adjustments in the optimalmanagement decisions are taken into account.

Methods

In order to optimize agricultural management decisions relatedto land-use and crop-specific nitrogen fertilization and irrigationintensities, a bio-economic whole-farm model is used. Thisbio-economic whole-farm model comprises three different sub-models: the generic weather generator LARSWG (Semenov andBarrow, 1997; Semenov et al., 1998), the mechanistic crop growthmodel CropSyst (Stöckle et al., 2003) and an economic decisionmodel at farm scale. In addition, a genetic algorithm (GA) is usedas optimization technique. More details on the component modelsand the settings of the GA are presented in the following subsec-tions.

The structure of the modelling approach and the linkagesbetween the sub-models are given in Fig. 1.

First, a population of candidate solutions is randomly generatedby the GA (see upper right part in Fig. 1). Each candidate solu-tion comprises a specific set of considered decision variables (i.e.,nitrogen fertilization amount, irrigation strategy and crop acreage),which are taken as potential solutions for an optimal (i.e., util-ity maximizing) farm management scheme. These sets of decisionvariables are passed in a next step to CropSyst (middle panel ofFig. 1), where they are used as management input variables for cropyield simulations. To represent production risks due to uncertainweather conditions, 25 variable weather years are generated withthe stochastic weather generator LARSWG.1 Thus, crop yields aresimulated for each crop and each set of management decisions for aperiod of 25 weather years. The 25 simulated yields of all crops arethen fed into the economic model (bottom-right panel of Fig. 1) tocompute the whole-farm return and the related production costs(e.g., fertilization amount, irrigation and drying costs). Besides pro-duction risks, also price risks are taken into account, and a set ofstochastic crop prices is generated for the 25 years of simulations(details are presented below). Finally, the whole-farm return andproduction costs are used to calculate the certainty equivalent (CE)(representing the utility of a risk-averse decision maker) at thefarm scale, which is the objective value in the optimization pro-cess. Once the objective values (i.e., CE) of all candidate solutionsin the initial population are derived, the GA is used to select themost promising candidate solutions (i.e., candidate solutions whichlead to the highest CE) and to create applying the genetic operators(i.e., mutation and crossover) a subsequent population of decision

1 Following Jame and Cutforth (1996), crop growth simulations should be con-ducted during at least 25-30 weather years in order to account for the risk associatedwith unpredictable weather conditions.

N. Lehmann et al. / Land Use Policy 35 (2013) 119– 130 121

put fa

mtdta

S

trawed(BwBtr2(dCnbsSttaw

Fig. 1. Overview of the modelling approach. The scenario-related in

odel in order to evaluate for each solution its objective value ando create a subsequent population. Thus, the processes of the GAescribed above (i.e., evaluation of the candidate solutions, selec-ion and application of genetic operators) are repeated until thelgorithm converges to an optimal solution.

tochastic weather generator

Daily weather input variables (daily minimum and maximumemperature, rainfall occurrence and amount, and daily total solaradiation) for crop yield simulations with CropSyst are gener-ted for present and future climate conditions using the stochasticeather generator LARSWG (Semenov and Barrow, 1997; Semenov

t al., 1998). For the calibration of LARSWG, historical daily weatherata of the period 1981–2010 from the climate station PayernePAY, 6◦57′ E, 46◦49′ N, 490 m a.s.l.), which is located within theroye watershed, is used. After calibration, 25 years of syntheticeather data are generated for a Baseline and two CC scenarios. The

aseline scenario, representing current climatic conditions, refers tohe period 1981–2010. The CC scenarios (ETHZ-CLM and SMHI-Had)epresent climate conditions for the nominal timeframe between036 and 2065, assuming the IPCC SRES emission scenario A1BNakicenovic et al., 2000). For both CC scenarios, boundary con-itions were obtained from global simulations with the Hadleyentre global climate model HadCM3. Moreover, the ETHZ-CLM sce-ario was conducted with the regional climate model maintainedy the Swiss Federal Institute of Technology, while the SMHI-Hadcenario was completed with the regional climate model of thewedish Meteorological and Hydrological Institute. Most impor-

antly, both CC scenarios indicate a significant temperature increasehroughout the year. Furthermore, the ETHZ-CLM scenario is char-cterized by strong precipitation decreases in summer months,hile under the SMHI-Had scenario, precipitation increases in

ctors (i.e., climate variables and market prices) are coloured in grey.

winter and decreases in spring and summer months. Further infor-mation on the employed climate scenarios and the downscalingapproaches is presented in Lehmann et al. (2013) and in Table A1in Appendix A.

Crop growth model

We use CropSyst (Version 4.13.09) to simulate climate andmanagement-dependent yields for six crops: winter wheat(Triticum spp. L.), winter barley (Hordeum vulgare L.), winterrapeseed (Brassica napus L.), grain maize (Zea mays L.), potatoes(Solanum tuberosum L.) and sugar beets (Beta vulgaris L.). CropSystis a process-based crop growth model that simulates biological andenvironmental aboveground and belowground processes of a sin-gle land block fragment using daily weather data and informationabout soil and crop characteristics as well as a specific manage-ment scheme at a daily scale. Stöckle et al. (2003) provide a detailedoverview on the model and its components as well as on its appli-cations. CropSyst has already been applied in different studies toestimate the impact of climate change on Swiss crop production(Torriani et al., 2007; Finger et al., 2011; Lehmann et al., 2013).

For this study, a calibration of CropSyst for the study region byKlein et al. (2012) was used based on yield records from the SwissFarm Accountancy Data Network (FADN). This approach had theadvantage that CropSyst was calibrated against yield records com-ing from farm observations and not only from field trials, whichallowed calibrating CropSyst closer to the real-world situation.Further information on the CropSyst calibration approach usedin this study is given in Klein et al. (2012). Furthermore, identical

initial soil conditions, with a soil texture of 59.8% sand, 11.3% clayand 28.9% silt, are assumed for each simulation year in this study.The soil’s initial content of organic matter is set for the top layerat 2.8% and at 2% for the other layers (Klein et al., 2012). For all soil

1 Use Po

law

E

maTtm

�

wcdtctimlnstTgm

r

Swiss farmers available, a sensitivity analysis with regard to theassumed level of risk aversion has been conducted. In doing so,we repeat the optimization procedure for all considered scenar-ios using risk aversion parameters of � = 0 (risk neutral) and � = 5(highly risk averse). This consideration of different levels of risk

TR

p

pe

o

22 N. Lehmann et al. / Land

ayers, an initial concentration of 5 kg N ha−1 in the form of NO3nd 5 kg N ha−1 in the form of NH4 per 0.1 m soil depth is assumed,hich is in line with Weisskopf et al. (2001).

conomic decision model

To integrate the previous modelling steps within the economicodel, first variable and fixed costs related to a chosen set of man-

gement decisions are specified for each of the 25 simulation years.hen, revenues based on the resulting crop yields are calculated forhe corresponding years. Combining these figures, the annual gross

argin at the farm level is calculated as shown in Eq. (1):

=N∑

i=1

ai · (�i + DPi + cfix,i + cirrig,i − cvar,i) (1)

here � is the annual gross margin at farm level (CHF), ai is theultivated surface (ha), �i is the revenue (CHF ha−1) and DPi are theirect payments depending on the crop i (CHF ha−1). cfix,i stands forhe fixed costs (excluding irrigation systems), cirrig,i for the fixedosts of the irrigation systems and cvar,i for the variable costs ofhe crop i (all expressed in CHF ha−1). Note that cirrig,i = 0 if norrigation is applied. The variable costs cvar,i depend on the chosen

anagement decisions (cp. Table 4) and the resulting crop yieldevels. Fixed costs include costs that depend on the crop but areot subject to management decisions, such as expenditures foreeds, pesticides and agricultural machinery. More details abouthe assumptions on revenues and costs employed are given inable 1. The expected gross margin E(�) and the variance of the

ross margin �2

� at farm level are derived from the 25 annual grossargins.Both expected gross margin and its variability are used to

epresent farmers’ decision-making. More specifically, they are

able 1evenues and costs.

Winter wheat Winter bar

RevenueCrop price levels (in CHF t−1). Averages of the

period 2002–2010 (Standard deviation inparentheses)a,b ,c

506 (41) 372 (39)

Direct paymentDirect payment (CHF ha−1)d 1680 1680

Fixed costsSeed (CHF ha−1)d 218 143

Plant protection (CHF ha−1)d 265 265

Plant growth regulant (CHF ha−1)d 41 41

Contract work and machinery costs (CHF ha−1)d 783 783

Fixed irrigation costsIrrigation system costs (CHF ha−1)e 447 447

Variable costsNitrogen fertilizer (CHF kg−1 N−1)d 1.4 1.4

Other fertilizer costs (CHF kg−1 N−1)d 0.72 0.73Hail insurance (% of crop yield revenue)d 2.4 2.4

Cleaning, drying costs (CHF t−1)d 39.5 32.5

Other costs (CHF t−1)d 6.7 1.2

Variable irrigation costs (CHF m−3)e 0.1 0.1

Interest rate (%)d,f 3.0 3.0

a Source: FAO (2011).b It has been assumed that 75% of the total potato harvest is sold as table potatoes (AGR

otato harvest is sold as feed potatoes (AGRIDEA and FIBL, 2010).c Since the sugar market has been liberalized in 2009, which caused a decrease of Swiss

rice distributions. In order to account for higher prices levels of agricultural products insures that mean prices and coefficients of variation remain as observed in Switzerlandd Source: AGRIDEA and FIBL (2010).e Source: Spörri (2011).f The interest claim is computed as product of the interest rate and the invested capita

f 6 months.

licy 35 (2013) 119– 130

combined in a certainty equivalent (CE) maximization approach torepresent the utility maximization problem of a risk-averse farmer.The CE is defined as the sure sum of money that has the same util-ity as the expected utility of a risky alternative (Keeney and Raiffa,1976) and is defined as follows:

CE = E(�) − RP (2)

where E(�) is the expected gross margin at farm level and RP is therisk premium, both expressed in CHF. The RP is the sure amountof money the decision maker is willing to pay to eliminate riskexposure (Di Falco et al., 2007). According to Pratt (1964), the RPcan be approximated by Eq. (3):

RP ≈ 12

· �

E(�)· �2

� (3)

where � is the coefficient of relative risk aversion and �2� is the vari-

ance of the gross margin at farm level �. For this study, we assume� to be 2, which corresponds to a moderate risk-averse decisionmaker and implies decreasing absolute risk aversion (Di Falco andChavas, 2006). Since there are no estimates of the risk aversion of

aversion, including risk-neutral moderate risk-averse and ratherstrong risk-averse behaviour, further represents the heterogeneityof risk preferences among farmers (Rosenzweig and Binswanger,1993).

ley Winter rapeseed Grain maize Potatoes Sugar beets

788 (96) 371 (53) 456 (29) 66 (8)c

2680 1680 1680 3580

108 268 3585 407250 220 800 525

0 0 0 0787 844 2591 1409

447 447 447 447

1.4 1.4 1.4 1.4 0.94 1.54 3.49 1.41

5.6 3.6 2.4 2.458.5 71.3 1.5 016.3 0 0.5 12

0.1 0.1 0.1 0.13.0 3.0 3.0 3.0

IDEA and FIBL, 2010) at the average price given in table. The remaining 25% of the

reference prices by more than 40%, we used German sugarbeet prices to representn Switzerland we multiplied the German prices by a factor of 1.5. This procedure.

l (fixed costs, fixed irrigation costs and variable costs) for an average commitment

N. Lehmann et al. / Land Use Policy 35 (2013) 119– 130 123

Table 2Summary statistics of national crop prices in Switzerland and France in the period 2002–2010.

Crop Switzerland France

Average price (CHF t−1)a Coefficient of variation (%) Average price (CHF t−1)a Coefficient of variation (%)

Wheat 506 8.2 192 28.7Barley 372 10.6 177 26.8Rapeseed 788 12.1 401 22.8Grain maize 371 14.4 201 26.8Potatoes 456 6.4 270 21.1Sugar beetsb 66 11.9 54 18.4

P

tiStelE

crddolaect(

b5na

wcaMgava

dr32iaehttt

tn

Table 3Overview of the applied scenarios.

Crop price scenario

Swiss crop prices EU crop prices

Climate scenarioBaseline Baseline-CH Baseline-EU

a Source: FAO (2011).b See Table 1 for further details.

rice scenarios

Two scenarios with regard to crop price distributions are used:he scenario CH considers average price levels and price volatil-ty currently observed in Switzerland. Because crop prices inwitzerland are much higher and price volatilities are much lowerhan in neighbouring countries (Finger and El Benni, 2012), wemploy a second price scenario (EU) that assumes the crop priceevels and crop price variability that is currently observed in theuropean Union.

For both scenarios, we use averages, variances and covariance ofrop prices in Switzerland (CH scenario) and France (EU scenario),espectively, in the period 2002–2010 obtained from the FAOSTATatabase (FAO, 2011) (see Table 2 for the averages and standardeviations and Table B1 in Appendix B for the correlation structuref the observed crop prices). Thus, we assume that future expectediberalization of agricultural markets in Switzerland will convergeverage crop prices and increase crop price volatility to the lev-ls currently observed in France. Crop prices in France have beenhosen to represent the EU scenario, since France is the most impor-ant producer of cereal, oilseed crops and sugar beets in the EU-27Eurostat, 2011).

Table 2 shows that the average crop price levels in France areetween 18% and 60% lower and the crop price volatility is between0% and 350% higher than in Switzerland. Thus, under the EU sce-ario, crop prices are not only significantly smaller, but the farmerlso faces much higher price risks.

In order to generate 25 simulation years of volatile crop prices,hich represent scenario-specific average crop price levels and

rop price variability, we apply a multivariate normal distributionpproach (Ripley, 1987). More specifically, we use the R packageASS 7.3-16 available from CRAN (http://cran.r-project.org) and

enerate 25 stochastic prices for both scenarios and each crop. Thispproach ensures not only the mean crop price levels and crop priceolatility, but also correlations between prices of the different cropsre represented in the decision process.

Besides market prices of agricultural products, governmentirect payments are also of high importance for Swiss farmers. Cur-ently, government direct payments make up on average almost0% of total farm revenue in Swiss agriculture (Finger and Lehmann,012). However, since projections of future direct payment levels

nvolve a high degree of uncertainty, we follow Briner et al. (2012)nd assume that direct payments are kept constant on today’s lev-ls. On the one hand, government support for the agricultural sectoras a very long tradition in Switzerland (Haller, 2010). On the other,

he liberalization of agricultural markets may not necessarily leado an adoption of the entire EU agricultural policy framework byhe Swiss government.2

2 For instance, the Swiss cheese market has been liberalized (i.e., there are norade barriers or distortions with the European Union), but direct payments haveot been affected by this liberalization step.

ETHZ-CLM ETHZ-CH ETHZ-EUSMHI-Had SMHI-CH SMHI-EU

Table 3 presents all possible combinations of the three consid-ered climate and two employed crop price scenarios.

The farm

The proposed modelling approach is applied to a representa-tive arable farm in the Western part of Switzerland, located in theBroye watershed (for details, see Lehmann et al., 2013). In terms ofproduction activities, the model considers the six most importantarable crops in Swiss agriculture: winter wheat, winter barley, win-ter rapeseed, grain maize, potatoes and sugar beets. The developedwhole-farm model is used to optimize crop acreage, as well as thecrop-specific nitrogen fertilization intensity and irrigation strategy(cp. Table 4). To account for restrictions with regard to crop-specificagronomic limitations as well as with regard to limitations imposedby the agricultural policy in Switzerland (i.e., cross-compliancerequirements), the following constraints are implemented into themodel:

• The total farm acreage amounts to 30 ha, representing the averagesurface of arable farms located in the region of Payerne.3

• To ensure an adequate crop rotation, cross-compliance obliga-tions limit the maximum share of several crops: winter wheatis limited to a maximum acreage of 50%; the sum of all cereals(without grain maize) is limited to 66%; the maximum crop shareof grain maize is 40%; and the maximum share of winter rape-seed, potatoes and sugar beets is 25% of total arable land (BLW,2011). Furthermore, we restrict the sum of the winter rapeseedand sugar beet surface due to rotational restrictions to 40% of thetotal arable land (Vullioud, 2005).

• The farmer is obliged to cultivate a minimum of four differentcrops to fulfil cross-compliance requirements (BLW, 2011).

• According to the cross-compliance obligations, the farm has tocomply with a balanced nitrogen supply and demand at farmlevel as revealed by the official Swiss nutrient balance method“Suisse Bilanz” (AGRIDEA and BLW, 2012). In this nutrient bal-

ance approach, a yield-dependent maximum nitrogen amount isspecified for each crop, whereas the nitrogen demand and supplyhas to be balanced at the farm level.

3 The average surface of arable farms located within a 15-km radius around Pay-erne amounts to 33 ha (BLW, 2010).

124 N. Lehmann et al. / Land Use Policy 35 (2013) 119– 130

Table 4Considered management variables.

Decisionvariable

Crop Management variable and unit Range (min–max)considered in themodelling approach

Variable incrementconsidered in themodelling approach

Number ofalternatives

1 Winter wheat Crop acreage in % of total arable surface 0–50 1 512 Winter wheat Nitrogen fertilization amount in kg ha−1 0–200 10 213 Winter wheat Irrigation strategy (trigger point of irrigation)a 0–1 0.1 114 Winter barley Crop acreage in % of total arable surface 0–66 1 675 Winter barley Nitrogen fertilization amount in kg ha−1 0–200 10 216 Winter barley Irrigation strategy (trigger point of irrigation)a 0–1 0.1 117 Winter rapeseed Crop acreage in % of total arable surface 0–25 1 268 Winter rapeseed Nitrogen fertilization amount in kg ha−1 0–200 10 219 Winter rapeseed Irrigation strategy (trigger point of irrigation)a 0–1 0.1 11

10 Grain maize Crop acreage in % of total arable surface 0–25 1 2611 Grain maize Nitrogen fertilization amount in kg ha−1 0–200 10 2112 Grain maize Irrigation strategy (trigger point of irrigation)a 0–1 0.1 1113 Potato Crop acreage in % of total arable surface 0–25 1 2614 Potato Nitrogen fertilization amount in kg ha−1 0–150 10 1615 Potato Irrigation strategy (trigger point of irrigation)a 0–1 0.1 1116 Sugarbeet Crop acreage in % of total arable surface 0–25 1 2617 Sugarbeet Nitrogen fertilization amount kg ha−1 0–150 10 16

ation)

cally t

•

•

•

tcmic

O

toooswtmrs

The yields and the economic viability (expressed as comparablegross margin5) of different crops are compared with farm surveydata obtained from the farm accountancy data network (FADN)

18 Sugarbeet Irrigation strategy (trigger point of irrig

a The trigger point of irrigation represents the level of soil moisture that automati

The farmer’s maximum available working time is assumed toamount to 2800 h per year. Following AGRIDEA and FIBL (2010),the required total amounts of labour (including management andfield work) are: 41 h per hectare for winter wheat, winter barleyand winter rapeseed; 37 h per hectare for grain maize and 258and 67 h per hectare for potatoes and sugar beets, respectively.In addition, we also account for the fact that field work islimited by weather conditions to half the days of the vegeta-tion period (Luder, 1996) with a maximum daily working timeof 10 h (Musshoff and Hirschauer, 2009). For current and futureclimate conditions, the vegetation period is assumed to last 220and 250 days, respectively (Calanca and Holzkämper, 2010). Therequired field working time per crop is defined as follows: win-ter wheat, winter barley: 16 h per hectare; winter rapeseed: 18 hper hectare; grain maize: 11 h per hectare; potatoes: 218 h perhectare and sugar beets: 27 h per hectare (all following AGRIDEAand FIBL, 2010).Finally, we restrict nitrogen intensity for potatoes and sugarbeets to a maximum amount of 150 kg N ha−1 and 130 kg N ha−1,respectively. Higher nitrogen fertilization dosages are not con-sidered in practice because they would have a negative impacton crop quality (A. Zimmermann, personal communication).

The choice of management variables shown in Table 4 illustrateshe wide range of risk mitigation and CC adaptation responsesonsidered in our approach. Farmers can change crop-specificanagement decisions (i.e., nitrogen fertilization and irrigation

ntensity), alter the land allocation across crops or may even removerops entirely from their crop mix.

ptimization routine

Due to the discrete nature of the decision variables (cp. Table 4),he maximization of the CE can be interpreted as a combinatorialptimization problem, which is characterized by a finite numberf feasible solutions. Nevertheless, more than 1023 combinationsf different sets of decision variables would be theoretically pos-ible, and the evaluation of each of these possible combinationsould be too time consuming. To overcome this problem, we use in

his study a GA as optimization technique. GAs are a heuristic opti-ization technique and mimic the biological concept of genetic

eproduction (Mayer et al., 2001), following the concept of “theurvival of the fittest” (Aytug et al., 2003). A GA starts with the

a 0–1 0.1 11

riggers irrigation, and ranges from 0 (permanent wilting point) to 1 (field capacity).

generation of an initial population of individuals, each represent-ing a possible solution for a given problem (e.g., crop share of winterwheat or the nitrogen fertilization amount for winter wheat). Thedecision variables in GAs are coded as binary strings of genes on achromosome (=individual) that represent a potential solution of theoptimization problem. The initial population of possible solutions(=chromosomes) evolves over time by selecting the best individualsin each generation and reproducing offspring for the next gener-ation applying recombination, mutation and crossover until thealgorithm converges to an optimum (Gen and Cheng, 2000). Inthis study, this optimum represents the farm level managementstrategy maximizing the farmer’s CE. We use the C++ based GAlibrary package GAlib (Wall, 1996) and apply a steady-state GA. Thesteady-state GA uses overlapping populations, whereas the usercan specify how much of the population should be replaced in eachgeneration (Wall, 1996). The control parameters in the GA are setas follows: genome size = 8 bits; population size = 5000; proportionof replacement = 0.2; selection routine = roulette wheel; mutationprobability = 0.25; crossover probability = 0.5; and the GA is termi-nated when a population’s best fitness value does not change fora number of 3000 generations. Since GAs do not guarantee thatthe global optimum solution will be reached, each optimizationrun is repeated three times using different randomly generated ini-tial populations. For all scenarios in this study, the repeated runshave led to the same optimal solutions, which are thus presentedas global optima.

Model evaluation

To evaluate the developed whole-farm model, we compare themodelling results obtained under the Baseline-CH scenario, whichrepresents current climate conditions and Swiss price levels, withdifferent observed reference data sets. The Swiss agricultural infor-mation system (AGIS)4 has been used to estimate typical crop plans(expressed as land use decisions) of arable farms around Payerne.

4 The AGIS database is compiled by the Federal Office for Agriculture and recordsevery farm in Switzerland (BLW, 2010).

5 The comparable gross margin is defined as the sum of the revenues (withoutdirect payments) minus the sum of all variable costs (Mouron and Schmid, 2011).

N. Lehmann et al. / Land Use Policy 35 (2013) 119– 130 125

Table 5Comparison of simulated (Baseline-CH scenario) and observed management decisions.

Crop Modelledoptimumcrop landshare

Observed average cropland shares in the year2009 (±SD)a

Modelled optimumnitrogen fertilizationintensity (kg N ha−1)

Recommendednitrogen fertilizationintensity (kg N ha−1)b

Modelled optimumirrigation intensity(mm ha−1)

Observed range ofirrigation intensity inthe Broye watershed(min–max) (mm ha−1)c

Winter wheat 45% 40% (±8%) 160 160 0 0Winter barley 0% 4% (±6%) nad 130 na 0Winter rapeseed 11% 13% (±9%) 160 140 0 0Maize 10% 7% (±9%) 120 110 105 20–140Potato 9% 5% (±10%) 150 120 108 5–200Sugarbeet 25% 14% (±10%) 130 110 125 13–120

a Sample from the Swiss agricultural information system (AGIS) (BLW, 2010) of the year 2009 considering farms located in municipalities with a geographic centroid withina 15-km radius around the climate station of Payerne. In order to ensure the comparability of the modelling results with the observed data, only farms without any livestockand with a minimum surface of arable land (i.e., grasslands are not considered) of at least 25 ha have been selected. In total 52 farm records have been used. The crop sharesrefer to the total arable surface.

b Source: AGRIDEA and BLW (2012).c Source: Robra and Mastrullo (2011).d Crop is not included in the optimal crop rotation.

Table 6Comparison of modelled and observed crop yield levels and comparable gross margins.

Crop Modelled averagecomparable grossmargin (CHF ha−1)

Observed comparable grossmargin (75%-percentile)(CHF ha−1)a,b

Modelled average cropyield (t ha−1)

Observed crop yield(75%-percentile)(t ha−1)a

Number offarm records

Winterwheat 2581 2752 8.0 7.4 123Winterbarley nac 2344 na 8.2 107Winter rapeseed 1643 1859 3.6 3.6 91Grain maize 2520 2472 14.0 10.4 33Potato 9157 10,349 44.6 42.5 58Sugarbeet 3548 5628 87.9 79.9 106

a Source: Sample of FADN farms located in municipalities with a geographic centroid within a 15-km radius around the climate station of Payerne. Farm records of theperiod 2005–2009 have been used as reference.

b Since sugarbeet market has been liberalized in 2009 leading to a sharp decrease in sugarbeet prices, FADN data only of the year 2009 (13 farm records) has been used fort garbe

(f(dai

vnor8mtserim

innbonfa

as

normative optimal management schemes (Buysse et al., 2007), theoptimization results obtained under the Baseline-CH scenario areclose to real-world observations. The developed whole-farm model

he derivation of the observed 75%-percentile of the comparable gross margin of suc Crop is not included in the optimal crop rotation.

Mouron and Schmid, 2011). Observed irrigation practices are takenrom a survey conducted in the year 2011 in the Broye watershedRobra and Mastrullo, 2011). Finally, we use recommended yield-ependent nitrogen fertilization amounts in Switzerland (AGRIDEAnd BLW, 2012) as references for the obtained nitrogen fertilizationntensities.

Table 5 provides a comparison between the optimum decisionariables obtained from the model applying the Baseline-CH sce-ario and observations made in the Broye watershed. The simulatedptimal land shares of all crops except sugar beets lie within theange of what can be observed in reality. On average more than3% of the arable land is occupied by wheat, barley, rapeseed, grainaize, potatoes and sugar beets, which outlines the high impor-

ance of these crops in the study area. Nevertheless, we find theugar beet area to be overestimated by our model. This can bexplained by the fact that sugar beet production in Switzerland isestricted by quotas issued by the manufacturing company. Thus,n reality the free choice for sugar beet cultivation assumed in our

odel may not be available for all farmers.Regarding nitrogen fertilization, the modelled optimum fertil-

zation intensities are for all crops close to the yield-dependentitrogen fertilization recommendations. The higher modelleditrogen fertilization intensities for grain maize, potatoes and sugareets can be explained by the fact that these crops are irrigated inur model, which increases the average yield levels and the relateditrogen demand (Di Paolo and Rinaldi, 2008). The recommended

ertilization levels reported in AGRIDEA and BLW (2012), however,

re based on rainfed crop production.

Finally, the model outputs show that irrigation is only profit-ble under the Baseline-CH scenario for grain maize, potatoes andugar beets. This outcome is in line with the results of the survey

et.

conducted by Robra and Mastrullo (2011). Moreover, all modelledirrigation intensities lie within (or close to) the observed range.

Finally, Table 6 compares modelled and observed comparablegross margins and crop yields. In order to compare the modellingresults with only well managed farms, the upper 75%-percentileof the considered observed farm data has been taken as reference.The modelled average yields tend to be for all crops except win-ter rapeseed higher than the observations. This difference can beexplained by the fact that CropSyst does not account for pests andweeds, which lead to lower crop yield levels in practice. Further-more, grain maize, potatoes and sugar beets are irrigated in themodelling results, while not all farmers irrigate these crops in real-ity. Although the simulated crop yields are slightly overestimated,the obtained average comparable gross margins are lower than theFADN observations for all considered crops except grain maize. Thisis mainly due to the fact that most crops in Switzerland have aquality-related price structure. For instance, the sugar beet price inSwitzerland depends on the harvest’s sugar content. The FAO pricesapplied to the farm model, however, reflect only basic price levelswithout consideration of quality aspects; that is, they are lowerthan the prices reported in the FADN records.6

Although the presented whole-farm model is based on anormative mathematical programming approach that does notnecessarily simulate farmers’ actual behaviour, but rather results in

6 Note that since crop yield quality aspects are not integrated in CropSyst, wefocused in this study on crop yield quantity only.

126 N. Lehmann et al. / Land Use Policy 35 (2013) 119– 130

45%

0%

11%

10%

9%25%

Baseline−CH49%

0%

15%

0%11%

25%

ETHZ−CH49%

0%

15%

0%11%

25%

SMHI−CH

0%

48%

15%

4%8%

25%

Baseline−EU

11%

49%

15%0%0%

25%

ETHZ−EU

0%

49%

15%

11% 0%

25%

SMHI−EU

Winter wheatWinter barley

Winter rapeseedGrain maize

PotatoSugar beet

er dif

cd

R

L

sfa

aaSUtattsturcncciwt

lo

Fig. 2. Optimal crop land allocation und

an thus be judged to be appropriate to simulate the managementecisions of an arable farm at Payerne.

esults and discussion

and allocation

It is of particular interest how CC and the applied crop pricecenarios affect the management decisions on the modelled arablearm. To this end, we first investigate how the farmer’s land use isdjusted under the considered scenarios (Fig. 2).

Under the Baseline-CH scenario, winter wheat and sugar beetsre the most dominant crops (Fig. 2). The remaining surface isllocated to winter rapeseed, potatoes and grain maize. Assumingwiss prices, CC impacts on the optimal crop plan are rather small.nder both future climate scenarios, grain maize disappears from

he optimal crop mix, while the acreages of winter wheat, potatond winter rapeseed are increased. Lehmann et al. (2013) showhat the cultivation of grain maize at Payerne is very sensitive tohe expected changes in climate conditions. More specifically, theirtudy indicates that even if irrigation is considered as an adapta-ion strategy, crop yields of grain maize will significantly decreasender future climate scenarios, causing lower profitability. Winterapeseed gains in importance since direct payment levels for thisrop are high (cp. Table 2). This guarantees a high and stable eco-omic profitability of winter rapeseed even under scenarios whererop yields decrease. The same is true for sugar beets, which areultivated at maximum possible crop shares (25%) in all six scenar-os. Furthermore, the share of potatoes is increased since more field

ork days are possible under CC which promotes the cultivation of

his field work-intensive crop.

Lower and more volatile crop prices (i.e., EU price scenarios)ead to stronger changes in the farm’s optimal land allocation thanbserved for the CC scenarios (see Fig. 2). Assuming EU prices, we

ferent climate and crop price scenarios.

find that winter barley replaces winter wheat as most dominantcrop in the optimal crop mix. Thus, lower crop prices reduce theself-supply of bread cereals (i.e., winter wheat) in Switzerland. Thisshift can be explained by the lower relative price decrease (cp.Table 2) and the lower production costs of winter barley comparedto winter wheat. In addition, the assumption of EU prices leads toa decrease of grain maize production under Baseline climate condi-tions by 60%. Furthermore, grain maize is not cultivated anymoreunder the ETHZ-EU scenario. Nevertheless, in contrast to the Swissprice scenarios, the optimal crop share of grain maize increasesunder the SMHI-EU scenario. Whereas decreasing yields and higherirrigation costs make grain maize less profitable than winter wheatcultivation under the ETHZ–EU scenario, the production of rainfedgrain maize is still more profitable than the production of winterwheat under the SMHI-EU scenario (Fig. 2). As in the case of winterbarley, the relative price decrease of grain maize under the EU pricescenarios is smaller than the relative price reductions of winterwheat. Furthermore, assuming EU prices, the cultivation of pota-toes is profitable only under current climate conditions. Besides thelower crop prices, CC additionally decreases the relative profitabil-ity of potato production, causing lower crop yields and increasedirrigation requirements. Due to the high levels of crop-specificdirect payments, which have not been modified throughout thescenarios, sugar beets and winter rapeseed are cultivated under allEU price scenarios at the upper limits set by the cross-complianceregulations and crop rotation restrictions, respectively.

Crop yields and production risks

Irrespective of the market price scenarios, future expected

climate conditions have a negative impact on average crop yields(cp. Table 7). This decrease is most pronounced for winter wheatand potatoes. For instance, potato yields decrease under theETHZ-CH scenario by 19% compared to the Baseline-CH scenario.

N. Lehmann et al. / Land Use Policy 35 (2013) 119– 130 127

Table 7Average crop yields (t ha−1) and coefficient of variation (in parentheses) for all crops and scenarios.

Crop Baseline-CH ETHZ-CH SMHI-CH Baseline-EU ETHZ-EU SMHI-CH

Winterwheat 8.0 (9.1%) 6.8 (8.9%) 7.0 (7.4%) na 6.2 (14.1%) naWinterbarley naa na na 8.0 (9.4%) 5.3 (15.4%) 6.3 (9.9%)Winter rapeseed 3.6 (7.9%) 3.4 (11.0%) 3.3 (6.7%) 2.9 (10.0%) 2.8 (12.3%) 2.3 (8.9%)Grain maize 14.0 (3.1%) na na 8.9 (18.1%) na 7.3 (24.3%)Potato 44.6 (9.3%) 36.4 (9.3%) 39.1 (6.1%) 44.6 (9.3%) na na

.7 (3.

Nltwwrum

iiiiwdtHid

N

pil

Fa

Sugarbeet 84.4 (6.5%) 79.0 (4.4%) 81

a Crop is not included in the optimal crop rotation.

evertheless, CC has only small negative impacts on average yieldevels of winter rapeseed and sugar beets. This can be traced backo the fact that sugar beets are expected to benefit from globalarming and a longer growing season, provided that sufficientater is available (Olesen and Bindi, 2002). In the case of winter

apeseed, the earlier harvest due to the shorter vegetation periodnder CC reduces the risk of heat and drought stresses, which mayainly take place in summer months.Furthermore, our results show that CC may not necessarily

ncrease production risks (cp. Table 7). Decreasing yield variabilitys found in this study for winter wheat, sugar beets and potatoesf adjustments in crop-specific management schemes are takennto account. The production of winter-sown crops, such as winter

heat, tends to benefit from more stable yields due to warmer andryer climatic conditions. In contrast, CC actually increases produc-ion risks for spring-sown crops, such as potatoes or sugar beets.owever, CC also gives farmers incentives to use irrigation more

ntensively to mitigate increasing climate risks, finally resulting inecreasing production risks for potatoes and sugar beets.

itrogen fertilization and irrigation

Besides the composition of the optimal crop mix, CC and EUrices also affect the optimal crop-specific management schemes,

.e., nitrogen and irrigation strategies (cp. Fig. 3). Both, EU crop priceevels and CC, lead to a reduction of nitrogen fertilization levels

Winter wheat

050

100

150

200

015

025

037

550

0

N F

erti

lizat

ion

(kg⋅ h

a−1

) Winter bar

050

100

150

200

Grain maize

050

100

150

200

015

025

037

550

0

N F

erti

lizat

ion

( kg⋅h

a−1

) Potat

050

100

150

200

Baseline−CH

Baseline−EU

ETHZ−

ETHZ−

ig. 3. Optimal irrigation and nitrogen fertilization intensity. The optimal nitrogen fertilizare depicted by the black triangles (right y-axis). Note that a missing black triangle indica

1%) 84.4 (6.5%) 79.0 (4.4%) 81.7 (3.1%)

for all crops except potatoes and sugar beets. Aggregated on thefarm level and assuming Swiss prices, the total farm level nitrogendecreases from 4428 kg N in the Baseline-CH scenario to 3492 kg Nunder the ETHZ-CH scenario. The implementation of EU pricesfurther enhances this effect. More specifically, the total appliednitrogen amount at the farm level amounts to only 1614 kg N underthe ETHZ-EU scenario.

Due to warmer and dryer climatic conditions in the CC scenarios,the optimal irrigation intensity is increased for all irrigated crops(i.e., grain maize, potatoes and sugar beets). Under the assumptionof EU prices, however, the optimal irrigation intensity increasesonly for sugar beets. Potatoes are no longer cultivated under CC,while grain maize is produced without the use of irrigation underthe Baseline-EU and SMHI-EU scenarios and disappears from theoptimum crop mix under the ETHZ–EU scenario. For winter wheat,winter barley and winter rapeseed, which have their main grow-ing periods in autumn and spring, irrigation is not profitable in anyof the scenarios. This is in particular due the fact that CC does notlead to very distinct changes in climate conditions within the grow-ing seasons of these crops. Total farm water consumption increasesfrom 15,458 m3 under the Baseline-CH scenario to 30,965 m3 and to19,918 m3 under the ETHZ-CH scenario and the SMH-CH scenario,

respectively. Since lower crop prices decrease the economic bene-fit of irrigation under the EU price scenario, the irrigation demandfor farms increases only from 11,982 m3 under the Baseline-EUscenario to 24,402 m3 and to 15,885 m3 under the assumption of

ley

015

025

037

550

0 Winter rapeseed

050

100

150

200

015

025

037

550

0

Irri

gat

ion

(mm

)

o

015

025

037

550

0 Sugar beet

050

100

150

200

015

025

037

550

0

Irri

gat

ion

(mm

)

CH

EU

SMHI−CH

SMHI−EU

tion intensities are represented by bars (left y-axis), the optimal irrigation intensitiestes that a crop is not included in the optimal crop mix.

128 N. Lehmann et al. / Land Use Po

CEAvg

Gross Ma rginCV RP

020

4060

8012

016

0

02

46

810

1214

16

CV

an

d R

P (

%)

CE

an

d G

ross

Mar

gin

(10

00 C

HF

)

Baseline−CH

Baseline−EU

ETHZ−CH

ETHZ−EU

SMHI−CH

SMHI−EU

Fig. 4. Climate change and price risks impacts on the farmer’s certainty equivalent(ip

ttss(o

C

mttovpt(

uieriif2paati

minLSwsfa

CE), average gross margin, income variability (CV) and risk premium (RP). The CVs defined as the standard deviation divided by the average gross margin. The riskremium (RP) is derived from Eq. (3).

he ETZH-EU and the SMH-EU scenario, respectively. Nevertheless,hese results show that irrigation requirements in the Broye water-hed will increase under CC independent from the chosen pricecenario. Because water already is a scarce resource in the regionMühlberger de Preux, 2008), CC will further intensify the conflictf water use.

ertainty equivalent, average gross margin and income risks

As shown in Fig. 4, CC decreases farmers’ CE and average grossargin less than changes in the crop prices. Whereas CC decreases

he CE under both price scenarios by 8–12%, a change from Swisso EU prices leads in all climate scenarios to a decrease of the CEf about 50%. Furthermore, CC slightly decreases the farm incomeariability (i.e., the coefficient of variation, see CV in Fig. 4) for bothrice scenarios, whereas under the more volatile EU crop prices,he income risks are increased between 43% (ETHZ-EU) and 53%Baseline-EU).

Although farm level income volatility is expected to increasender the EU price scenarios, the relative risk premiums (see RP

n Fig. 4) at the farm scale remain at low levels for all consid-red scenarios. In none of the considered scenarios, the relativeisk premiums exceed 2.4% of the expected gross margin, whichs far lower than the risk premiums found in other studies. Fornstance, Kim and Chavas (2003) estimate relative risk premiumsor different farm types in Wisconsin to amount between 2% and0% of the expected profit. Groom et al. (2008) find the relative riskremium of cereal and vegetable farmers in Cyprus to amount tobout 20% of the expected profit. Furthermore, using a single-croppproach, Finger (2012) finds the risk premium to account for morehan 16% of the expected gross margin in grain maize productionn Switzerland.

There are two main reasons for the rather low relative risk pre-ium found in all scenarios of this study: first, Swiss farmers’

ncome depends to a high degree on direct payments, which areot subject to seasonal weather or price variations (see Finger andehmann, 2012). Second, due to cross-compliance requirements in

witzerland, farmers have to cultivate at least four different cropshich reduces income risks at the farm level. Thus, the conclu-

ion may change if the model scale moves from the field to thearm scale, i.e., if whole-farm adjustment processes are taken intoccount. For instance, the coefficient of variation (CV) of the gross

licy 35 (2013) 119– 130

margins of grain maize amounts in this study under the Baseline-EUscenario to 31%. But still, the CV of the total gross margin at farmlevel does not exceed 10%. Therefore, the study’s results show thatthe use of diverse cropping systems, as is common practice in Swissagriculture, provides a valuable risk management instrument. Thisdiversity should be maintained and encouraged by governmen-tal regulations because it will be essential to cope with expectedincreases in climate and market risks.

Sensitivity analysis

In order to analyze the sensitivity of the assumed risk aver-sion on the modelling results, we also run the optimization modelapplying a risk aversion of � = 5 (i.e., highly risk-averse decisionmaker) and a risk aversion of � = 0 (risk-neutral decision maker).The assumption of a risk-neutral decision maker (� = 0) leads exclu-sively under the Baseline-CH and the Baseline-EU scenario to smallchanges in the optimal crop mix. More specifically, under theBaseline-CH scenario, the optimal crop share of winter wheat isincreased from 45% to 47% at the expense of the grain maize andwinter rapeseed acreage, while under the Baseline-EU scenario,grain maize (4%) is replaced with winter wheat. If the relative riskaversion is increased to � = 5, no changes in the optimal land usepatterns are found. The crop-specific management decisions withregard to irrigation are unaffected by the assumed level of risk aver-sion. Moreover, the assumption � = 5 of or � = 0 affects the optimalcrop-specific nitrogen intensities for all crops only in a range of±10 kg ha−1.

The fact that the assumption of the decision maker’s risks aver-sion has almost no impact on the optimal management schemesindicates that risk does, even under future expected climate condi-tions and more volatile crop prices, not play a very decisive rolein Swiss arable farming. On the one hand, it is known that riskaversion is only one of possible reasons for diversification of farmactivities (Pannell et al., 2000). Other reasons such as different soiltypes, resource constraints (i.e., workforce, machinery, land) andbenefits of rotation sequences (e.g., disease control, nitrogen fix-ation by legumes, soil fertility) may have stronger incentives fordiversification than the decision maker’s risk aversion. On the other,direct payments make up a large proportion of the total agriculturalincome in Switzerland which generally decreases the importanceof production and market risks in Switzerland.

Notwithstanding, it is important to underline that risk has beenincluded in this study in a static framework. This means that theidentified optimal management schemes are in all 25 simulationyears identical. For farmers, however, it is also important how torespond tactically and dynamically to unfolding threads and oppor-tunities in market and climate conditions (Pannell et al., 2000). Thiskind of risk management, however, has not been taken into accountin this study.

Conclusions

In this paper, we developed a bioeconomic whole-farm modelthat combines non-parametrically a crop growth model with aneconomic decision model using a genetic algorithm (GA). The use ofthe farmer’s certainty equivalent (CE) as objective function enabledaccounting not only for changes in average climate and price levels,but also for the climate and price risks. This modelling approachwas used to investigate impacts of likely changes in climate andcrop prices on an arable farm’s responses with respect to land useas well as on crop-specific fertilization and irrigation strategies inWestern Switzerland.

The application of the whole-farm model to the six consid-ered scenarios showed that changing crop prices from Swiss toEU average and volatility levels has much stronger impacts on theoptimal management decisions than CC. These changes in optimal

Use Po

maaEdClf(lfoFgecc

sCr(aac

N. Lehmann et al. / Land

anagement patterns are almost exclusively due to decreasedverage price levels. Because of high levels of direct payments and

diversified crop portfolio, the increased price volatility under theU price scenario has only small impacts on optimal managementecisions. Thus, even under more volatile crop prices and underC, the matter of risk is small in Swiss arable farming. Neverthe-

ess, the assumption of EU crop prices led to reductions in thearmer’s certainty equivalent by up to 51%. However, Tran et al.2012) show that global warming will increase world crop priceevels, which might also raise the crop price levels in the EU. Thus,uture research should not only consider crop price levels currentlybserved in Europe, but also design scenarios of crop price trends.urthermore, CC is expected to offer the possibility of the emer-ence of new crops and new varieties in Northern Europe (Ewertt al., 2005; Olesen and Bindi, 2002). Thus, not only crops currentlyultivated in the Broye watershed but also the introduction of newrop species should be considered in future studies.

The impacts of CC on farmers’ utility and income are muchmaller (8–12%). Nevertheless, under both, CH prices and EU prices,C increases, the arable farm’s irrigation requirements. Since theegion already suffers water scarcity in dry and hot summer monthsMühlberger de Preux, 2008), the region’s policy makers have to

djust the current irrigation water policy (e.g., implementation of

water price and water quota) in order to reduce the region’s agri-ultural water consumption.

Table A1Applied changes in climate variables for the ETHZ-CLM and SMHI-Had scenario.

Month ETHZ-CLM

�Tmin (◦C) �Tmax (◦C) �Rad (%) �Precip

January +2.51 +2.51 −3 −4

February +1.82 +2.00 −4 −2

March +1.91 +2.14 −4 −2

April +2.06 +2.15 −2 −3

May +1.85 +2.07 +2 −6

June +2.18 +3.08 +7 −18

July +2.82 +4.23 +9 −30

August +3.11 +4.39 +8 −28

September +2.78 +3.41 +3 −11

October +2.29 +2.36 +0 −1

November +2.28 +2.23 +0 −4

December +2.69 +2.6 −2 −4

Table shows the absolute applied changes in the monthly average minimum temperaturechanges in the monthly average radiation (�Rad) and in the monthly average precipitati

Table B1Correlation matrix of crop prices.

Wheat Barley Rapeseed

SwitzerlandWheat 1 0.84 0.41

Barley 1 −0.02

Rapeseed 1

Grain Maize

Potatoes

Sugarbeets

FranceWheat 1 0.96 0.87

Barley 1 0.85

Rapeseed 1

Grain maize

Potatoes

Sugarbeets

Table shows the correlation matrix of crop prices in the period 2002–2010 in Switzerlanthe Pearson correlation coefficient.

licy 35 (2013) 119– 130 129

Although the modelling results obtained under current climateand price conditions showed good agreement with real-worldobservations, future research may also consider non-rationalbehaviour of decision makers (Mack et al., 2011) and to takeinto account, that utility optimization is only one of the farmers’goals (Greiner et al., 2009). Another limitation of the presentedwhole-farm model is that we used a static modelling approach.Static modelling approaches cannot be used to assess how farm-ers respond tactically and dynamically to unfolding threats andopportunities. In order to integrate tactical adjustments, which areknown to increase the importance of risk management in agricul-ture (Pannell et al., 2000), the presented whole-farm model shouldbe used as basis for a stochastic programming framework withrecourse.

Acknowledgements

This work was supported by the Swiss National Science Founda-tion in the framework of the National Research Programme 61. Wewould like to thank MeteoSwiss, the Research Station AgroscopeReckenholz-Tänikon ART and the Federal Office for Agriculture forproviding climate, FADN and AGIS data. Furthermore, we thank theanonymous reviewers for helpful comments on an earlier versionof the manuscript.

Appendix A.

SMHI-Had

(%) �Tmin (◦C) �Tmax (◦C) �Rad (%) �Precip (%)

+2.33 +1.74 −6 +14+1.90 +1.34 −4 +6+1.31 +1.11 −3 +2+1.03 +1.07 −2 −2+1.48 +1.59 +0 −7+2.00 +2.13 +1 −8+2.08 +2.15 +0 −3+2.00 +1.98 −2 −1+1.67 +1.61 −2 +4+1.46 +1.32 −5 +16+1.86 +1.56 −8 +24+2.34 +1.92 −8 +22

(�Tmin), in the monthly average maximum temperature (�Tmax), and the relativeon sum (�Precip) for the applied CC scenarios ETHZ-CLM and SMHI-Had.

Appendix B.

Grain maize Potatoes Sugar beets

0.77 −0.33 0.710.97 −0.35 0.727

−0.09 0.15 0.291 −0.25 0.67

1 −0.501

0.92 0.52 −0.510.77 0.40 −0.380.77 0.58 −0.731 0.65 −0.54

1 −0.571

d and France obtained from the FAOSTAT database (FAO, 2011). All values refer to

1 Use Po

R

AAA

A

B

B

B

B

C

C

C

D

D

D

E

EE

FF

F

F

F

F

G

G

G

H

H

J

K

K

K

L

L

Wall, M., 1996. A C++ Library of Genetic Algorithm Components. Technical Report.

30 N. Lehmann et al. / Land

eferences

GRIDEA and BLW, 2012. Wegleitung Suisse-Bilanz, Technical Report. AGRIDEA.GRIDEA and FIBL, 2010. Deckungsbeiträge 2010, Technical Report. AGRIDEA.ngus, A., Burgess, P.J., Morris, J., Lingard, J., 2009. Agriculture and land use: demand

for and supply of agricultural commodities, characteristics of the farming andfood industries, and implications for land use in the UK. Land Use Policy 26,230–242.

ytug, H., Khouja, M., Vergara, F.E., 2003. Use of genetic algorithms to solve pro-duction and operations management problems: a review. International Journalof Production Research 41 (17), 3955–4009.

LW, 2010. Agricultural Information System (AGIS). Swiss Federal Office for Agri-culture (BLW), Bern, Switzerland.

LW, 2011. Weisungen und Erläuterungen über die Direktzahlungsverodnung andie Landwirtschaft. Federal Office for Agriculture (BLW), Bern, Switzerland.

riner, S., Elkin, C., Huber, R., Grêt-Regamey, A., 2012. Assessing the impacts of eco-nomic and climate changes on land-use in mountain regions: a spatial dynamicmodeling approach. Agriculture, Ecosystems & Environment 149, 50–63.

uysse, J., Van Huylenbroeck, G., Lauwers, L., 2007. Normative, positive andeconometric mathematical programming as tools for incorporation of mul-tifunctionality in agricultural policy modelling. Agriculture, Ecosystems &Environment 120 (1), 70–81.

alanca, P., Holzkämper, A., 2010. Agrarmeteorologische Bedingungen im SchweizerMittelland von 1864 bis 2050. Agrarforschung Schweiz 1 (9), 320–325.

hallinor, A.J., Ewert, F., Arnold, S., Simelton, E., Fraser, E., 2009. Crops and climatechange: progress, trends, and challenges in simulating impacts and informingadaptation. Journal of Experimental Botany 60, 2775–2789.

havas, J.P., Holt, M.T., 1990. Acreage decisions under risk: the case of corn andsoybeans. American Journal of Agricultural Economics 72 (3), 529–538.

i Falco, S., Chavas, J.P., 2006. Crop genetic diversity, farm productivity and themanagement of environmental risk in rainfed agriculture. European Review ofAgricultural Economics 33 (3), 289–314.

i Falco, S., Chavas, J.P., Smale, M., 2007. Farmer management of production risk ondegraded lands: The role of winterwheat varieties in the Tigray region, Ethiopia.Agricultural Economics 36, 147–156.

i Paolo, E., Rinaldi, M., 2008. Yield response of corn to irrigation and nitrogen fertil-ization in a Mediterranean environment. Field Crops Research 105 (3), 202–210.

l Benni, N., Finger, R., Mann, S., 2012. Effects of agricultural policy reforms and farmcharacteristics on income risk in Swiss agriculture. Agricultural Finance Review72 (3), 301–324.

urostat, 2011. Agriculture and Fisheries Data. Eurostat, Luxemburg.wert, F., Rounsevell, M.D.A., Reginster, I., Metzger, M.J., Leemans, R., 2005. Future

scenarios of European agricultural land use: I Estimating changes in crop pro-ductivity. Agriculture, Ecosystems & Environment 107 (2), 101–116.

AO, 2011. FAOSTAT Statistical Database, http://faostat.fao.org/inger, F., 2009. Climate change impacts and adaptation in Swiss cereal

production—integrating biophysical and economic modeling. Dissertation ETHZurich. No. 18191.

inger, R., 2012. Nitrogen use and the effects of nitrogen taxation under consider-ations of production and price risk. Agricultural Systems 107, 13–20.

inger, R., El Benni, N., 2012. A note on price risks in Swiss crop production –Empirical results and comparisons with other countries. In: Yearbook of Socioe-conomics in Agriculture., pp. 131–151.

inger, R., Lehmann, N., 2012. The influence of direct payments on farmers’ hailinsurance decisions. Agricultural Economics 43, 343–354.

inger, R., Hediger, W., Schmid, S., 2011. Irrigation as adaptation strategy to cli-mate change: a biophysical and economic appraisal for Swiss maize production.Climatic Change 105 (3–4), 509–528.

en, M., Cheng, R., 2000. Genetic Algorithms and Engineering Optimization. Wiley-Interscience, USA.

reiner, R., Patterson, L., Miller, O., 2009. Motivations, risk perceptions and adop-tion of conservation practices by farmers. Agricultural Systems 99 (2–3), 86–104.

room, B., Koundouri, P., Nauges, C., Thomas, A., 2008. The story of the moment: riskaverse cypriot farmers respond to drought management. Applied Economics 40(3), 315–326.

aller, T., 2010. Agrarschutz und öffentliche Agrarausgaben in der Schweiz: Was willdie Bevölkerung? In: Yearbook of Socioeconomics in Agriculture., pp. 329–361.

arwood, J., Heifner, R., Coble, K., Perry, J., Somwaru, A., 1999. Managing Risk inFarming: Concepts, Research, and Analysis, Agricultural Economic Report No.US Department of Agriculture, Washington.

ame, Y.W., Cutforth, H.W., 1996. Crop growth models for decision support systems.Canadian Journal of Plant Science 76, 9–19.

eeney, R.L., Raiffa, H., 1976. Decisions with Multiple Objectives: Preferences andValue Tradeoffs. Wiley, New York.

im, K., Chavas, J.P., 2003. Technological change and risk management: an appli-cation to the economics of corn production. Agricultural Economics 29 (2),125–142.

lein, T., Calanca, P., Holzkämper, A., Lehmann, N., Rösch, A., Fuhrer, J., 2012. Anefficient crop model calibration procedure for use with farm accountancy data.Agricultural Systems 111, 23–33.

ehmann, N., 2010. Regional crop modeling: how future climate may impact cropyields in Switzerland. Master Thesis. Swiss Federal Institute of Technology (ETH),

Zurich, Switzerland.

ehmann, N., Finger, R., Klein, T., Calanca, P., Walther, A., 2013. Adapting crop man-agement practices to climate change: modeling optimal solutions at the localscale. Agricultural Systems 117, 55–65.

licy 35 (2013) 119– 130

Luder, W., 1996. Wetterrisiko und verfügbare Feldarbeitstage in der Schweiz.Arbeitswirtschaftliche Grundlagen aus meteorologischen Messreihe. FAT-Berichte 490.

Mack, G., Möhring, A., Zimmermann, A., Gennaio, M.P., Mann, S., Ferjani, A., 2011.Farm entry policy and its impact on structural change analysed by an agent-

based sector model. In: EAAE 2011 Congress Change and Uncertainty Challengesfor Agriculture, Food and Natural Resources, Zurich, Switzerland, August 30 toSeptember 2, 2011.

Mayer, D.G., Belward, J.A., Burrage, K., 2001. Robust parameter settings of evolution-ary algorithms for the optimisation of agricultural systems models. AgriculturalSystems 69 (3), 199–213.

Mishra, A., El-Osta, H., 2002. Risk management through enterprise diversification:A farm-level analysis. In: AAEA Meetings in Long Beach, CA, USA, July 28–31,2002.

Mouron, P., Schmid, D., 2011. Zentrale Auswertung von Buchhaltungsdaten –Grundlagenbericht 2011. Agroscope Reckenholz-Tänikon Research Station ART,Tänikon.

Mühlberger de Preux, C., 2008. Broye: fish or chips? In: Umwelt 2/08, Dossier 26(Landwirtschaft).

Musshoff, O., Hirschauer, N., 2009. Optimizing production decisions using a hybridsimulation – A genetic algorithm approach. Canadian Journal of AgriculturalEconomics 57 (1), 35–54.

Nakicenovic, N., Alcamo, J., Davis, G., de Vries, B., Fenhann, J., Gaffin, S., Gregory, K.,Grbler, A., Jung, T.Y., Kram, T., La Rovere, E.L., Michaelis, L., Mori, S., Morita, T.,Pepper, W., Pitcher, H., Price, L., Raihi, K., Roehrl, A., Rogner, H.H., Sankovski, A.,Schlesinger, M., Shukla, P., Smith, S., Swart, R., van Rooijen, S., Victor, N., Dadi, Z.,2000. IPCC Special Report on Emissions Scenarios. Cambridge University Press,Cambridge.

Olesen, J.E., Bindi, M., 2002. Consequences of climate change for European agricul-tural productivity, land use and policy. European Journal of Agronomy 16 (4),239–262.

Olesen, J.E., Trnka, M., Kersebaum, K.C., Skjelvag, A.O., Seguin, B., Peltonen-Sainio,P., Rossi, F., Kozyra, J., Micale, F., 2011. Impacts and adaptation of Europeancrop production systems to climate change. European Journal of Agronomy 34,96–112.

Pannell, D.J., Malcolm, L.R., Kingwell, R.S., 2000. Are we risking too much? Perspec-tives on risk in farm modelling. Agricultural Economics 23 (1), 69–78.

Pratt, J.W., 1964. Risk aversion in the small and in the large. Econometrica 32 (1–2),122–136.

Rajsic, P., Weersink, A., Gandorfer, M., 2009. Risk and nitrogen application levels.Canadian Journal of Agricultural Economics 57 (2), 223–239.

Reidsma, P., Ewert, F., Lansink, A.O., Leemans, R., 2010. Adaptation to climate changeand climate variability in European agriculture: the importance of farm-levelresponses. European Journal of Agronomy 32, 91–102.

Ripley, B.D., 1987. Stochastic Simulation. Wiley, New York.Risbey, J., Kandlikar, M., Dowlatabadi, H., Graetz, D., 1999. Scale, context, and

decision making in agricultural adaptation to climate variability and change.Mitigation and Adaptation Strategies for Global Change 4 (2), 137–165.

Robra, J.P., Mastrullo, J., 2011. Evaluation d’irrigation dans la Broye des besoins eneau. Technical Report (In French).

Rosegrant, M.W., Roumasset, J.A., 1985. The effect of fertilizer on risk: a het-eroscedastic production function with measurable stochastic inputs. AustralianJournal of Agricultural Economics 29 (2), 107–121.

Rosenzweig, M.R., Binswanger, H.P., 1993. Wealth, weather risk and the compositionand profitability of agricultural investments. Economic Journal 103, 56–78.

Sandmo, A., 1971. On the theory of the competitive firm under price uncertainty.American Economic Review 61 (1), 65–73.

Saunders, L.S., Webb, T.H., Barringer, J.R.F., 1997. Integrating economic data withspatial biophysical data to analyse profitability and risks of winterwheat pro-duction on a regional basis. Agricultural Systems 55 (4), 583–599.

Sckokai, P., Moro, D., 2006. Modeling the reforms of the common agricultural policyfor arable crops under uncertainty. American Journal of Agricultural Economics88 (1), 43–56.

Semenov, M.A., Barrow, E.M., 1997. Use of a stochastic weather generator in thedevelopment of climate change scenarios. Climatic Change 35, 397–414.

Semenov, M.A., Brooks, R.J., Barrow, E.M., Richardson, C.W., 1998. Comparison ofthe WGEN and LARS-WG stochastic weather generators for diverse climates.Climate Research 10, 95–107.

Spörri, M., 2011. Economic efficiency of irrigation strategies for arable cropsunder current and future climate scenarios. In: Bachelor Thesis. ETH Zurich,Switzerland.

Stöckle, C.O., Donatelli, M., Nelson, R., 2003. CropSyst, a cropping systems simulationModel. European Journal of Agronomy 18 (3–4), 289–307.

Torriani, D.S., Calanca, P., Schmid, S., Beniston, M., Fuhrer, J., 2007. Potential effectsof changes in mean climate and climate variability on the yield of winter andspring crops in Switzerland. Climate Research 34 (1), 59–69.

Tran, A.N., Welch, J.R., Lobell, D., Roberts, M.J., Schlenker, W., 2012. Commodityprices and volatility in response to anticipated climate change. In: AAEA AnnualMeeting, Seattle, Washington, August 12–14, 2012.

Vullioud, P., 2005. Optimale Fruchtfolgen im Feldbau (3. Auflage). Land-wirtschaftliche Beratung, Datenblätter Ackerbau, Agroscope RAC Changins,Eidgenössische landwirtschaftliche Forschungsanstalt, Changins, Switzerland.

Massachusetts Institute of Technology, Mechanical Engineering Department,Cambridge, USA.

Weisskopf, P., Zihlmann, U., Walther, U., 2001. Einfluss der Bewirtschaftung auf dieStickstoffdynamik im Bodenwasser. Agrarforschung Schweiz 9, 348–353.