Ann Oper Res DOI 10.

1007/s10479-011-0902-7

A portfolio theory approach to crop planning under environmental constraints

Marius R dulescu Constanta Zoie R dulescu a a Gheorghia Zb ganu t a

Springer Science+Business Media, LLC 2011

Abstract This paper presents a multiobjective model for crop planning in agriculture. The approach is based on portfolio theory. The model takes into account weather risks, market risks and environmental risks. Input data include historical land productivity data for various crops, soil types and yield response to fertilizer/pesticide application. Several environmental levels for the application of fertilizers/pesticides, and the monetary penalties for overcoming these levels, are also considered. Starting from the multiobjective model we formulate several single objective optimization problems: the minimum environmental risk problem, the maximum expected return problem and the minimum nancial risk problem. We prove that the minimum environmental risk problem is equivalent to a mixed integer problem with a linear objective function. Two numerical results for the minimum environmental risk problem are presented. Keywords Crop planning Portfolio theory Multiobjective model Environmental risk Environmental level Mixed integer model

M. R dulescu ( ) a Institute of Mathematical Statistics and Applied Mathematics, Casa Academiei Romane, Calea 13, Septembrie nr. 13, 050711 Bucharest 5, Romania e-mail: mradulescu.csmro@yahoo.com C.Z. R dulescu a National Institute for Research and Development in Informatics, 8-10 Averescu Avenue, 011455 Bucharest 1, Romania e-mail: radulescucz@yahoo.com C.Z. R dulescu a e-mail: radulescu@ici.ro G. Zb ganu a Faculty of Mathematics and Computer Science, University of Bucharest, Academiei 14, 010014 Bucharest, Romania e-mail: zbagang@fmi.unibuc.ro

Ann Oper Res

1 Risk management in agriculture Agriculture is one of the worlds largest industries with billions of dollars/euros invested in industrial crop and livestock production. It is subject to various risks. We mention here few of them: weather risks, pest risks, disease risks, market risks, the interaction of technology with other farm and management characteristics risks, genetics risks, machinery efciency risk and the quality of inputs risks. Increasingly, farmers from all over the world are being exposed to unpredictable competitive markets for inputs and outputs, so that price or market risk is often signicant and may increase over time. Market risk includes the risks generated by the unpredictable currency exchange rates. There are several instruments to control the nancial risks, such as hedging, insurances, long term contracts etc or the adoption of different strategic positions such as value aggregation of products, vertical supply chains, certicates of origin etc. Agricultural insurance is an important risk management tool for uncertainties. It supports the management of the large losses and adverse developments that can occur. Sound planning also determines a favorable outcome or yield for the farmer and mitigation of the risks. One of the most important decisions faced by farmers is the selection of crops they want to grow. Prospective growers must know how to use risk management strategies in order to select the crops that best suit their needs. One of the most popular approaches to managing risk is to reduce risk exposure through diversication. There exist several types of diversication ranging from crop production to livestock production. Farmers income diversication, see Barrett et al. (2001), is a result of them. Farmers have to decide how much diversication is enough to capture most of the potential gains from expanding their enterprise mix. The effects of crop production diversication are reected in the relationship between absolute risk levels and the number of crops included in a portfolio. They are expected to be similar to those for stock market portfolios. Thus risk is reduced signicantly as additional assets are added to a single product portfolio. In other words, risk reduction may be largely achieved by including several assets in a portfolio. This means that adding another crop to an existing rotation or creating an entirely new portfolio may be an effective risk management strategy. The intensication of agricultural practices, in particular the growing use of fertilizers and pesticides, and the specialization and concentration of crop and livestock production, has had an increasing impact on water quality. The main agricultural water pollutants are nitrates, phosphates, and pesticides. Rising nitrate concentrations threaten the quality of drinking water, while high level of pesticide use contributes substantially to indirect emissions of toxic substances. Increasing levels of nitrates and phosphates in surface waters reduce their ability to support plant and animal life and make them less attractive for recreation. Adverse environmental impacts of agricultural practices are of great concern in many countries. Many governmental programs and policy instruments have been implemented or proposed to reduce agricultural pollution and to achieve certain environmental standards. However, strict and certain compliance with environmental standards is difcult to achieve due to the diffuse nature and the uncertainty of agricultural non-point source pollution. In this present paper we formulate a multiobjective model for crop planning in agriculture that takes into account weather risks, market risks and environmental risks. In order to consider the environmental risk several environmental levels are introduced. These levels are called desirable levels. No pollution taxes are paid by the farmer if the quantities of agricultural chemicals used are under the desirable levels. If these quantities exceed the desirable levels then monetary penalties proportional with the amount of the chemical input that exceed these limits shall to be paid by the farmer. The environmental risk of a crop plan

Ann Oper Res

is dened as the overall monetary penalties incurred paid by a farmer. It is a nonsmooth function of the crop plan. The multiobjective model for crop planning aims to minimize the environmental risk, to maximize the expected return and to minimize the nancial risk in the presence of a set of constraints such as: the demand constraints, the budget constraints and the environmental constraints. Starting from the multiobjective model for crop planning we formulate several single objective optimization problems: the minimum environmental risk problem, the maximum expected return problem and the minimum nancial risk problem. We prove that the minimum environmental risk problem is equivalent to a mixed integer problem with a linear objective function and linear and quadratic constraints. A numerical case is analyzed for the minimum environmental risk problem. In the next section several methods for diversication of agricultural production are presented. Our model covers the rst two types of production diversication. The methods developed in this paper can be easily extended in order to also include the third type of diversication that is temporal diversication.

2 Diversication of agricultural production There are several denitions: see Ellis (1998, 2000). Farmers use three types of production diversication. The most common type is diversication across products. This is a strategy derived from portfolio theory developed in the stock market. This strategy can be applied by any farmer with some knowledge of how to cultivate more than one crop, including farmers with small and/or contiguous parcels of land. The goal of this type of diversication is to reduce variance in sale revenues by participating in more than one product market. To be successful, the product markets must have low or negative levels of correlation in their prices. The second type of diversication, across locations, has also been well known for some time (Goland 1993), but practiced less often because it requires operating two or more parcels that are geographically separated, a requirement which could be infeasible for some farmers (Nartea and Bany 1994). Under this spatial diversication, a farmer must scatter crop production across locations sufciently far apart to have low levels of correlation in their weather extremes. Thus, because the focus is on reducing yield variance, this strategy can be applied by farmers specialized in growing one single crop. Finally, cultivar diversity is a form of temporal diversication, but it incorporates aspects of each of the other two diversication strategies. The common goal of cultivar diversity is to have portions of total acreage (either contiguous or scattered) reach the harvest stage at different times of the year (Park and Florkowski 2003). By selecting cultivars of a single crop that are not highly correlated in their growth schedules, farmers can both: (a) reduce average yield variability by reducing weather risk exposure (a feature of geographical diversication), and (b) raise average price received and/or lower price variance by being able to sell output in more than one market season (similar to product diversication). The practice of crop diversication complicates both production and marketing but it can increase prots. Diversication is a favorite risk-management tool of many farmers. For example, a survey in California found most farmers use some type of diversication as a risk management strategy, while few producers use the available nancial risk management tools (Blank et al. 1997). That study reported only 23.4%, 6.2%, and 24.4% of farmers in the state used forward contracting, hedging, and crop insurance, respectively,

Ann Oper Res

as risk reduction strategies, suggesting these tools may be ineffective. Farmers described diversication across crops as an easily implemented and effective strategy for managing revenue risk. This view was supported in an earlier study (Blank 1990), where the author showed there was an optimal amount of crop diversication among crop portfolios, and that this risk management strategy was always preferable to specialization.

3 Mathematical models applied in agriculture Increasing environmental problems in agriculture urge policy-makers to develop instruments to reduce and control the pollution caused by current intensive farming practices. These measures should be both effective from the environmental point of view, which is a public goal, as well as acceptable at the farm level with regard to private goals such as income and continuity at the farm. Operations research is an effective tool for modeling the complex interaction of production intensity, environmental aspects and farm income. Agriculture is one of the elds where mathematical models of operations research were rst used and also where they have been most widely applied. The number of mathematical models in agriculture has rapidly grown in the last decades, due to the impressive development of personal computers and commercial software programs. One of the rst papers that apply linear programming to agricultural decision making situations is Heady (1954). The early models were formulated at the farm level. One of the rst books dedicated to the application of linear programming to agriculture was Beneke and Winterboer (1973). For historical notes regarding the applications of operations research models to agriculture and other applications we recommend (Weintraub et al. 2001, 2007). Simulation models have also been used in the last decades at the farm level to assess the economic impact of several agricultural policies. Until recently, most of the farmers have made intensive use of chemical inputs in order to obtain higher yields at low costs. Unfortunately this policy had produced many negative side effects (or, as economists put it, negative externalities) for the environment. At present there exists a trend to develop mathematical models that are able to evaluate the economic impact of environment damages and thus to contribute to the development of a sustainable agriculture. The great majority of the problems connected to sustainable agriculture have a multicriteria character. The book (Romero and Rehman 2003) is a comprehensive monograph for the state of the art of multi-criteria analysis in agriculture until 2003. The concept of sustainable agriculture supposes harmonization or simultaneous realization of the objectives connected to economic growth, environment and social development. The mathematical models that take into account both objectives mentioned above are scarce. We mention a few of them below. In Wossink et al. (1992) an extension of the linear programming optimization model is employed in farm economics with an environmental component to analyze and evaluate the effects of alternative environmental policy instruments for agriculture. The application presented concerns the potential role of technical innovations and of input levies to reduce biocide use in crop production. In Johnson et al. (1991) a crop simulator model, called CERES, was linked to a dynamic optimization model in order to determine the optimum application of water and fertilizers that maximize the gross margin of yield. In Zekri and Herruzo (1994) the authors use a combination between a crop simulator model and a mixed multiobjective programming model in order to assess the effects of an increase of nitrogen prices and drainage water reduction. In Teague et al. (1995) the authors use the EPIC-PST model to predict the environmental risks from the use of pesticide and nitrates. A combination of these results with a Target MOTAD optimization model are used for the assessment

Ann Oper Res

of the trade-offs that exist between income and an index that linked both risks associated to the use of pesticides and nitrates. In Annetts and Audsley (2002) a multiple objective linear programming model that allows maximization of the return and environmental outcomes is presented. A method that has been successfully applied to crop planning is the Interactive Multiple Goal Linear Programming (IMGLP). In Nidumolu et al. (2007) an IMGLP model was developed that considers objectives of multiple stakeholders, i.e. different farmer groups, district agricultural ofcers and agricultural scientists for agricultural land use analysis. The analysis focuses on crop selection; considering irrigated and non-irrigated crops such as rice, sugarcane, sorghum, cotton, millet, pulses and groundnut. Interests of the most important stakeholders, farmers, policy makers and water users association are investigated. Important objectives of the farmers are increased income and retaining paddy area; of the policy makers (Agricultural Department) increased farmers income, maintaining rural employment, improve water-use efciency, reduce fertilizer and biocide use and discourage farmers from cultivating marginal lands; of the water users association optimizing water use. Scenarios have been constructed by combining objectives and constraints. In C.T. Hoanh et al. (2000) a multiple goal linear programming (MGLP) model was included in a land use planning and a DSS system (LUPAS). It was used to investigate the consequences for land use of various decisions taken under the economic reform. The incorporation of methods and tools for multi-decision-level (eld, farm, and region) analysis made LUPAS a powerful and versatile DSS for the integrated resource management and policy design. In general these models do not take into account simultaneously uncertainty and environmental issues. In general, decision processes in agriculture are complicated by the uncertainty surrounding important components such as weather, product prices, and government policies. Biological and economic management processes are basic to the success of agricultural enterprises. Many applications of risk analysis in agriculture show the impact of uncertainty on agricultural decision making. The uncertainty from the agriculture problems is modeled with the help of probability theory. Many of the practical problems that occur in agriculture are stochastic programming problems with multiple objectives. In practice, in the process of mathematical modeling, one cannot take into account all the factors that have an impact to agricultural production. The number of such factors is large and the growth of their number determines the rapid growth of the complexity of the models. Among the mathematical models applied in agriculture one can quote prediction models, crop planning models, optimal selection of fertilizer/pesticide models, crop rotation models etc. An important mathematical instrument which was successfully applied to modeling the problems from agriculture was portfolio theory. The above mentioned theory was developed as a result of the research in the domain of nancial management. Its aim is the elaboration of a quantitative analysis of how investors can diversify their portfolio in order to minimize risk and maximize returns. The theory was introduced in 1952 by University of Chicago economics student Harry Markowitz, who published his doctoral thesis, Portfolio Selection in the Journal of Finance (Markowitz 1952). The application of portfolio theory for nding an optimal allocation of agricultural land is popular in the literature. Since the 1950s, agricultural economists have adapted many popular nancial portfolio selection rules to the farm managers land allocation problem. The rst application of the portfolio theory to crop planning goes back to Freund (1956). In his approach Freund dened the risk of an agricultural enterprise through the variability of its returns, measured by variance. This implies the use of quadratic programming for nding the optimal crop patterns. In Hazell (1971) the author had changed the risk measure for the farm return. The variance was replaced by the mean absolute deviation. As a result, Hazell showed

Ann Oper Res

that the crop planning problem is equivalent to a linear programming problem. Single index portfolio models were used to farm planning in the papers (Collins and Barry 1986; Turvey et al. 1988). In Newbery and Stiglitz (1981), Schaefer (1992), Hardaker et al. (2004), Hazell and Norton (1986), Blank (2001) were presented or applied various variants of portfolio theory to the land allocation decisions. In Collender (1989), Reeves and Lilieholm (1993), Romero (1976, 2000) were studied several models for resources allocation in agriculture that are taking into account specic risks. Mathematical models that take into account farmers decisions and climate change were studied in Lewandrowski and Brazee (1993). In Rafsnider et al. (1993) portfolio analysis, using nonlinear risk programming, was applied to identify the risk-efcient crop portfolios for a group of subsistence farmers. In Fafchamps (1992), a simple theoretical model of crop portfolio choice when the revenue of individual crops is correlated with consumption prices is studied. In Roche and McQuinn (2004) the authors investigate the efcient allocation of land in a mean-variance sense. They use portfolio theory in order to examine the potential implications on the land allocation decision of the 2002 EU Commissions proposed mid-term reform of the Common Agricultural Policy related to decoupled payments. Applications of portfolio theory to biodiversity conservation were studied in Figge (2004). For other references regarding applications of portfolio theory to agriculture see R dulescu et al. (2006, 2010). a

4 A crop planning model for sustainable agriculture In this section a mixed integer programming model with multiple objectives for crop planning in agriculture is formulated. The model takes into account weather risks, market risks and environmental risks. Input data include historical land productivity data for various crops and soil types, yield response to fertilizer/pesticide application, unitary costs of fertilizers/pesticides, costs for crops cultivation (without using fertilizers/pesticides) on the plots. The application of fertilizers/pesticides to agricultural crops is desirable since they contribute to the growth of agricultural production. On the other side the application of fertilizers/pesticides in great quantities, over some levels bring damages to environment and human health. In order to protect the environment and of course the peoples health we shall take into account two kinds of levels: desirable levels and maximum admissible levels for the application of chemical inputs (fertilizers/pesticides). If the quantity of the chemical input is under the desirable level then no penalty is paid by the farmer. If the quantity of the chemical input lies between the desirable level and the maximum admissible level then monetary penalties proportional to the amount that it overcomes the desirable level must be paid by the farmer. Consider that an agricultural region is divided into several agricultural subregions. Suppose that: the soil quality of a subregion is homogeneous the climate in a subregion is the same historical data on land productivities in each agricultural subregions is available for a given set of crops For example a subregion may have non-irrigated good land quality, other region may have irrigated medium land quality, etc. Consider a farm located in an agricultural region as described above, which has an agricultural land divided into several plots. Let P1 , P2 , . . . , Pm be the plots from the farms land.

Ann Oper Res

Suppose that each plot Pj belongs to an agricultural subregion and consequently the soil quality of each plot is homogeneous. We consider that if a plot is cultivated then it is cultivated with the same crop. Denote by Sj the area of the plot Pj , j = 1, 2, . . . , m. We consider that the farmer has to choose a crop plan from n crops C1 , C2 , . . . , Cn , that is to make an allocation of crops to plots. In order to obtain high yields the farmer uses fertilizers/pesticides. Denote by F1 , F2 , . . . , Fk the set of the k fertilizers/pesticides used by the farmer. For the fertilizer/pesticide Fr denote by q1r the desirable level for the application of Fr . A crop plan that uses a quantity of fertilizer/pesticide under the desirable level is not subjected to monetary penalizations. An overcome of the desirable level is subjected to the application of environmental taxes. These taxes and the desirable and the maximum admissible levels may be established at various levels of decision making: local level, regional level, country level etc. Their amount depends on the existing level of pollution and on the community aspiration for maintaining a clean environment. In our paper will shall consider appropriate monetary penalties for the application of fertilizers/pesticides that are proportional to the amount that exceeds the desirable level. We shall denote by q2ir the maximum admissible level for the application of Fr per one hectare cultivated with crop Ci . The introduction of the pollution levels in the management of risk environment goes back to Qiu et al. (2001). For mathematical models that use the pollution levels in production planning see R dulescu et al. (2009). a Let Ui be the Cartesian product of intervals [0, q2ir ], that is: Ui = [0, q2i1 ] [0, q2i2 ] [0, q2ik ]. Consider the probability space ( , K, P ). Denote I = {1, 2, . . . , n}, J = {1, 2, . . . , m}, K = {1, 2, . . . , k}. For every i I, j J we dene the plot productivity functions cij : Ui R+ and the market price functions bi : R+ . Thus if q is a vector from Ui then the function q () = cij (, q), is a random variable. Analogous all the functions bi are random variables. cij (, q) represents the quantity (in kg) of crop Ci that can be produced on one hectare of plot Pj and bi the market price of one kg of crop Ci . In this paper the effect of crop rotation is neglected. That is if one cultivates the same crop repeatedly on the same plot over the time, the yield decrease. The effect of crop rotation is taken into account in the papers (Alfandari et al. 2009; Castellazzi et al. 2008; Detlefsen and Jensen 2007; Dogliotti et al. 2003; El-Nazer and McCarl 1986; Haneveld and Stegeman 2005; Mimouni et al. 2000; Santos et al. 2008; Vizvri et al. 2009). One can easily see that in our model the plot productivity functions do not depend on the crops cultivated previously on that plot. Let aij be the sum of money used by the farmer in order to cultivate one hectare of plot Pj with crop Ci without using fertilizers/pesticides. For every i I , j J , r K denote by: dr the cost of one kg of fertilizer/pesticide Fr . wr the penalization coefcient for the case the quantity of fertilizer/pesticide Fr overcomes the desirable level q1r . yij r the decision variable representing the quantity of fertilizer/pesticide Fr applied to one hectare of plot Pj cultivated with crop Ci . Denote yij = (yij 1 , yij 2 , . . . , yij k ). xij the decision variable that takes the value 1 if the crop Ci is cultivated on plot Pj and takes value zero if the crop Ci is not cultivated on the plot Pj . [M1 , M2 ] the range for the sum of money available for investment. Qi the inferior bound for the expected yield of crop Ci necessary to be obtained. It may be thought as the expected demand for crop Ci .

Ann Oper Res

In our model we shall make the following assumption:

k

cij (, q) = cij 0 () +
r=1

cij r ()qr

where q = (q1 , q2 , . . . , qk ) Ui and

The meaning of the coefcients cij r is the following: cij r the additional quantity (in kg) of crop Ci obtained from one hectare of plot Pj when one kg of fertilizer/pesticide Fr is used. The above assumption is a simplication since, in reality, the impact of an additional quantity of fertilizer Fr does depend on the levels of other fertilizers Fr . In some cases this assumption could be restrictive as there can be synergy of fertilizers. cij 0 the quantity of crop Ci that it is obtained from one hectare of plot Pj when no fertilizer/pesticide are used. We shall call a crop plan a couple (x, y) where x = (xij ) is a n m matrix with entries 01 and y = (yij ) where yij = (yij 1 , yij 2 , . . . , yij k ) Ui , i I , j J . The environmental constraints in the model are represented by the inequalities: 0 yij r q2ir xij , i I, j J, r K. (1)

This means that the quantity of fertilizers/pesticides is smaller than an upper limit q2ir if the plot Pj is cultivated with crop Ci . The quantity of fertilizers/pesticides is equal to zero if the plot Pj is not cultivated with crop Ci . When some crop does not need the fertilizer/pesticide Fr one can take q2ir = 0. From the above inequality one can easily see that yij r = xij yij r , i I, j J, r K. (2)

Indeed, if xij = 0 then from inequality (1) it follows that yij r = 0 hence equality (2) holds. In the case xij = 1 the equality (2) obviously holds. The yield of crop Ci obtained from the plot Pj when the decisions are given by the crop plan (x, y) is equal to
k k

xij Sj cij 0 +
r=1

cij r yij r

= Sj cij 0 xij +
r=1

cij r yij r .

Take yij 0 = xij for every i I , j J . The yield of crop Ci when the decisions are given by the crop plan (x, y) is equal to
m k

cij r yij r Sj .
j =1 r=0

The sales revenue when the decisions are given by the crop plan (x, y) is equal to the random variable
n m k

f1 (x, y, ) =
i=1 j =1 r=0

bi cij r Sj yij r .

Ann Oper Res

The cultivation cost for the crop plan (x, y) is equal to:
n m k n m k

f2 (x, y) =
i=1 j =1

aij +
r=1

dr yij r Sj xij =
i=1 j =1

aij yij 0 +
r=1

dr yij r Sj .

For every real number a denote by a+ the positive part of a that is a+ = max(a, 0) = |a|+a . 2 We dene the environmental risk as the monetary penalizations paid by the farmer. It is equal to
k n m

f3 (x, y) =
r=1

wr
i=1 j =1

yij r Sj q1r
+

One can easily see that there exists a penalization proportional to the amount of fertilizer/pesticide applied in excess over the desirable level. The environmental risk is a nonsmooth function of the crop plan (x, y). We dene the gross margin (respectively the net return) obtained by the farmer when the crop plan (x, y) is applied as follows: f4 (x, y, ) = f1 (x, y, ) f2 (x, y) (respectively) f5 (x, y, ) = f1 (x, y, ) f2 (x, y) f3 (x, y). Note that
n m k

f4 (x, y, ) =
i=1 j =1

(bi cij 0 aij )yij 0 +

r=1

(bi cij r dr )yij r Sj

and
n m k

E(f4 (x, y, )) =
i=1 j =1

(E(bi cij 0 ) aij )yij 0 +

r=1

(E(bi cij r ) dr )yij r Sj .

The nancial risk associated to the crop plan (x, y) is dened as the variance of the gross margin or of the net return. For every i, I , j, J , r, K denote ij r = E bi cij r b c E bi cij r E b c . Then the variance of the net return is equal to Var(f4 (x, y, )) = Var(f5 (x, y, )) = Var(f1 (x, y, ))
n m k n m k

=
i=1 j =1 r=0 =1 =1 =0

ij r yij r y Sj S .

Ann Oper Res

The demand constraints are represented by the inequalities:

m k

E(cij r )yij r Sj Qi ,
j =1 r=0

i I.

The budget constraints are represented by the inequalities: M1 f2 (x, y) M2 . The restrictions n xij 1, j J show that every plot is cultivated with at most one i=1 crop. The farmer intends to obtain optimal production plans that minimize the environmental penalties, minimize the nancial risk and maximize the expected return. According to the above requirements the multiobjective mixed integer programming model for the crop planning is the following: min min max
k n m

wr
r=1 n m i=1 j =1 k n

yij r Sj q1r
+ m k

ij r yij r y Sj S ,
i=1 j =1 r=0 =1 =1 =0 n m k

(E(bi cij 0 ) aij )yij 0 +

i=1 j =1 r=1

(E(bi cij r ) dr )yij r Sj ,

M1 f2 (x, y) M2 , m k E(cij r )yij r Sj Qi , i I, j =1 r=0 0 yij r q2ir yij 0 , i I, j J, r K, n yij 0 1, for every j J, i=1 yij 0 {0, 1}, i I, j J. Mathematical programming problems of the above type are very difcult to solve since they exhibit exponential complexity resulting from the presence of integer variables. Traditional approaches that apply in pure integer programming are not very helpful since the existence of continuous variables complicates their solution. Recently, simulation-based methods and heuristic algorithms have been successfully used for solving such problems. In the literature there is a small number of papers dealing with mixed integer quadratic programming problems (Lazimy 1982; Tziligakis 1999). In Lazimy (1982) is described an approach for solving such programs based on the generalized Benders decomposition. A new equivalent formulation that renders the program tractable is developed, under which the dual objective function is linear in the integer variables and the dual constraint set is independent of these variables. In Tziligakis (1999) an approach based on the relaxation method is developed for solving mixed integer quadratic programs. A heuristic algorithm is built for providing tight lower and upper bounds for the mixed integer problem. Starting from the above multiobjective programming model for the crop planning we formulate several single objective problems. More precisely we shall consider a minimum environmental risk problem, a maximum return problem and a minimum return risk problem.

Ann Oper Res

The models mentioned above are very useful in practice since they help farmers to make optimal decisions in complex situations in which weather and economic risks as well as environmental constraints may be involved. The application in practice of the models supposes that there are records (historical data) on the land productivities for various crops, plots and the records on the impact of fertilizers/pesticides on the crops. One of the key assumptions is that a sufcient amount of (historical) data is available. In reality, it is unlikely that this assumption holds. In order to solve this limited data problem one can use bootstrapping methods that increase the number of data through simulation. Another problem is the problem of establishing the environmental levels. In an agricultural region the values of the environmental levels are supposed to be xed by the region or by the state environmental protection department. It strongly depends on the degree of pollution in the agricultural region. The value of this level may decrease in a polluted area and increase in a low polluted area.

5 The minimum environmental risk problem In the framework of this problem the farmer tries to nd an optimal allocation of crops to plots and an optimal plan for fertilizer and pesticide application that minimize the environmental risk taking into account that: the nancial risk, that is Var(f5 (x, y, )), is smaller than a prescribed level the expected return is greater than a given level W the cost of cultivation and application of the fertilizers and pesticides lies in the range [M1 , M2 ] the demand constraints are satised k n m min wr yij r Sj q1r , r=1 i=1 j =1 + Var(f5 (x, y, )) , k n m E(f4 (x, y, )) wr yij r Sj q1r r=1 i=1 j =1 M1 f2 (x,y) M2 , m k E(cij r )yij r Sj Qi , i I, j =1 r=0 0 yij r q2ir yij 0 , i I, j J, r K, n yij 0 1, for every j J, i=1 yij 0 {0, 1}, i I, j J.

W,
+

With an introduction of some additional variables the objective function can be transformed in a linear map. Thus the minimum environmental risk problem is equivalent to the following mixed integer quadratic problem with linear objective function and linear and quadratic

Ann Oper Res

constraints: k min wr zr , n mr=1 yij r Sj q1r zr , r K, i=1 j =1 Var(f5 (x, y, )) , k E(f (x, y, )) wr zr W, 4 M f (x, y) M , 1 2 2 m k E(cij r )yij r Sj Qi , i I, j =1 r=0 0 yij r q2ir yij 0 , 0 zr , i I, j J, r K, n yij 0 1, for every j J, i=1 yij 0 {0, 1}, i I, j J. Recall that the land productivity coefcients cij r are random variables. They incorporate the weather risks. Historical data on the response rate to fertilizer/pesticide application, that is the numbers cij rt can be considered as realizations of the random variables cij r . The coefcients bi whose meaning is market price of crops are random variables that incorporate the market risk. Historical data on market price of crops, that is the numbers bit = the market price for one unit of crop Ci at moment t can be considered as realizations of the random variables bi . Taking into account the above things, we shall approximate the variance of the return
n m k n m k r=1

Var(f5 (x, y, )) =
i=1 j =1 r=0 =1 =1 =0

ij r yij r y Sj S

with the sample variance, that is with the number

n m k n m k

R(x, y) =
i=1 j =1 r=0 =1 =1 =0

ij r yij r y Sj S

where we denoted: ij r = 1 T
T

bit cij rt bt crt

t=1

1 T2

bit cij rt
t=1 t=1

bt crt .

We shall approximate the means E(bi cij r ), E(cij r ) by their sample means. Here T is the number of moments of time in the time horizon considered. We shall approximate the minimum environmental risk problem with a similar problem in which the variance of the return is replaced by its sample variance and the means by their sample means.

Ann Oper Res

6 The maximum return problem In the framework of the maximum return problem the farmer tries to nd crop plan (x, y) that maximize the expected net return taking into account that the nancial risk, that is Var(f5 (x, y, )), is smaller than a prescribed level the cultivation cost and application of the fertilizers and pesticides lies in the range [M1 , M2 ] the demand constraints are satised k n m max E(f (x, y, )) , wr yij r Sj q1r 4 r=1 i=1 j =1 + Var(f (x, y, )) , 5 M f (x,y) M , 1 2 2 m k E(cij r )yij r Sj Qi , i I, j =1 r=0 0 y q y , i I, j J, r K, ij r 2ir ij 0 n yij 0 1, for every j J, i=1 yij 0 {0, 1}, i I, j J.

7 The minimum nancial risk problem In the framework of this problem the farmer tries to nd an optimal crop plan (x, y) that minimizes the nancial risk taking into account that the expected net return is greater than a given level W the cultivation cost and application of the fertilizers and pesticides lies in the range [M1 , M2 ] the demand constraints are satised min(Var(f5 (x, y, ))), k n m E(f4 (x, y, )) wr yij r Sj q1r W, r=1 i=1 j =1 + M1 f2 (x, y) M2 , m k E(cij r )yij r Sj Qi , i I, j =1 r=0 0 yij r q2ir yij 0 , i I, j J, r K, n yij 0 1, for every j J, i=1 yij 0 {0, 1}, i I, j J. With an introduction of some additional variables the minimum nancial risk problem can be transformed into a mixed integer problem with a quadratic objective function and

Ann Oper Res

linear constraints. min(Var(f (x, y, ))), 5 k E(f (x, y, )) wr zr W, 4 r=1 n m yij r Sj q1r zr , r K, i=1 j =1 M f (x, y) M , 1 2 2 m k E(cij r )yij r Sj Qi , i I, j =1 r=0 0 y q y , 0 z , i I, j J, r K, ij r 2ir ij 0 r n yij 0 1, for every j J, i=1 yij 0 {0, 1}, i I, j J. 8 The decision process for the minimum environmental risk problem An important problem in the decision-making process of the choice of a crop plan is the existence of some indicators that will guide the farmer in the selection of the parameters of the problem: M1 , M2 , Qi , i = 1, 2, . . . , n, and W . These indicators can be computed by solving several mixed integer problems. Denote by X1 the set of all y = (yij r )iI,j J,rK{0} such that its entries satisfy the last three constraints of the minimum environmental risk problem. Consider the problem y (P1 ) max{f2 ( ) : y X1 }.
max Denote by M2 its optimal value. This value represents an upper bound for the sum of money the farmer can invest in some crop plan. The farmer may choose the parameter M2 max max smaller or equal than M2 . Denote by M1 the optimal value of the problem

(P2 )

max{f2 ( ) : y X1 , f2 ( ) M2 }. y y

For every i I consider the problem

m k

(P3 (i))

max
j =1 r=0

E(cij r )yij r Sj : y X1

and denote by Qmax its optimal value. The farmer can choose the parameters Qi , i i = 1, 2, . . . , n such that Qi Qmax , i = 1, 2, . . . , n. Denote i
m k

X2 = y X1 :
j =1 r=0

E(cij r )yij r Sj Qi for every i I .

Of course the farmer has to choose the parameters Qi , i = 1, 2, . . . , n such that the set X2 is non-empty. This choice may be done as a result of a sequence of trial error experiments.

Ann Oper Res

Consider the problem (P4 ) min{f2 ( ) : y X2 }. y

min Denote by M1 its optimal value. Then the farmer has to choose the parameter M1 in the max min interval [M1 , M1 ]. Suppose that the farmer has chosen the parameters M1 , M2 and Qi , i = 1, 2, . . . , n and wants to determine a range in which he must choose the parameter . Consider the problems:

(P5 ) (P6 )

min{Var(f5 ( )) : y X2 }, y max{Var(f5 ( )) : y X2 }. y

Denote by min (respectively by max ) the optimal values of the problems (P5 ) (respectively (P6 )). Then the farmer has to choose the parameter in the interval [ min , max ]. Suppose now that the farmer has already chosen the parameter in the above mentioned interval. Consider the problems:
k n m

(P7 ( ))

min E(f4 (x, y, ))

r=1 k

wr
i=1 j =1 n m

yij r Sj q1r
+

: y X2 , Var(f5 ( )) , y

(P8 ( ))

max E(f4 (x, y, ))

r=1

wr
i=1 j =1

yij r Sj q1r
+

: y X2 , Var(f5 ( )) y

and denote by W min ( ) and W max ( ) their optimal values. The farmer has to choose the parameter W in the interval [W min ( ), W max ( )]. Suppose that the farmer has already chosen the parameter W in the above mentioned interval. Consider the problem
k n m

min
r=1

wr
i=1 j =1

yij r Sj q1r
+ k n m

: y X2 , Var(f5 ( )) , y

E(f4 (x, y, ))
r=1

wr
i=1 j =1

yij r Sj q1r
+

W .

This problem is the minimum environmental risk problem. We have thus described a procedure for the selection of the parameters of the problem. This procedure allows the farmer to select parameters of the model in such a way that the solutions of the model are feasible. Consequently this procedure brings an argument to the model applicability and validity.

9 Numerical example for the minimum environmental risk problem In the following we shall analyze a numerical example for the minimum environmental risk problem. The input data are represented by: the matrix A = (aij ) of cultivation costs the matrix B = (bit ) of historical market prices for crops the four-dimensional matrix C = (cij rt ) of crop yields response rate to fertilizer/pesticide application

Ann Oper Res

the matrix Q2 = (q2ir ) of maximum admissible levels for fertilizer/pesticide application the vector w = (w1 , w2 , . . . , wk ) of monetary penalties for fertilizer/pesticide application the vector q1 = (q11 , q12 , . . . , q1k ) of desirable levels for fertilizer/pesticide application the vector Q = (Q1 , Q2 , . . . , Qn ) of demand levels for crops the vector S = (S1 , S2 , . . . , Sm ) of areas of plots the vector d = (d1 , d2 , . . . , dk ) of fertilizer/pesticide prices

The user parameters are: the budget limits M1 and M2 the lower limit for the expected return (the number W ) the upper limit for the nancial risk (parameter ) The decision variables are the allocation matrix x = (xij ) whose entries are 0 and 1 and the matrix y = (yij ) where yij = (yij 1 , yij 2 , . . . , yij k ) Ui , i I , j J . The objective map of this problem is the environmental risk, that is the overall monetary penalizations paid by the farmer for his crop plan. It is equal to
k n m

f3 (x, y) =
r=1

wr
i=1 j =1

yij r Sj q1r
+

The number of variables in the above problem is equal to mnk + mn + k and the number of constraints is equal to 2mnk + m + n + 2k + 3. The number of binary variables is equal to mn. We consider 3 crops and two fertilizers urea and triple superphosphate. Crop C1 is wheat, crop C2 is corn, crop C3 is barley. The demand levels (components of vector Q) are: Q1 = 17000 kg wheat, Q2 = 7000 kg corn, Q3 = 10000 kg barley. We consider 5 plots. The areas of the plots are S1 = 9 ha, S2 = 15 ha, S3 = 12 ha, S4 = 11 ha, S5 = 10 ha. The period of time considered is 19952006. By solving two mixed linear integer problems (see the preceding paragraph) we commax max min min pute M2 and M1 . We nd M2 = 44131 euros and M1 = 18369 euros. We choose max min M1 = M1 and M2 = M2 . By solving two other mixed integer problems (see the preceding paragraph) we compute min and max . For 5 values of that is for 1 = 13824600, 2 = 23982700, 3 = 34140800, 4 = 44298900, 5 = 54457000 are computed W min (i ) and W max (i ), i = 1, 2, . . . , 5. For 6 values of W chosen in the intervals [W min (i ), W max (i )] is solved the minimum environmental risk problem. The results are displayed in Table 1. In the fourth column of Table 1 are displayed Varcal, the values of the risk computed for the optimal crop plan obtained with the help of the parameters and W from the columns 2 and 3. Analogously in the fth and the sixth column are displayed Wcal, the values of the return computed for the optimal crop plan obtained with the help of the parameters and W , and Mcal, the sum invested in the optimal crop plan. In the last column of Table 1 are displayed the optimal values of the minimum environmental risk problem that is the environmental penalties paid for the crop plan. For = 54457000 the optimal allocations for the various values of W are displayed in Table 2. The optimal values of the objective function are displayed in the seventh column. In the eighth column are displayed the expenses made with the fertilizers computed for the optimal crop plan. In the last column are displayed the rates of return (the ratio between the return and the cultivation costs) computed for the optimal crop plan. The cells that are empty in the table mean that the plots are not cultivated (fallow). From Table 2 one can note that for the rst three values of W , the values of the objective function are zero since the no fertilizer is used. An analysis of the monetary penalizations

Ann Oper Res Table 1 The optimal values of the environment risk problem for various values of the parameters and W Nr. crt. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 13824600 13824600 13824600 13824600 13824600 13824600 23982700 23982700 23982700 23982700 23982700 23982700 34140800 34140800 34140800 34140800 34140800 34140800 44298900 44298900 44298900 44298900 44298900 44298900 54457000 54457000 54457000 54457000 54457000 54457000 W 1066.80 251.49 563.81 1379.12 2194.43 3009.74 1199.59 272.23 655.13 1582.49 2509.84 3437.20 1199.59 110.28 979.02 2068.32 3157.63 4246.93 1199.59 44.69 1110.22 2265.12 3420.02 4574.93 1199.59 18.12 1235.83 2453.54 3671.25 4888.96 Varcal 11949412 10806292 13474417 12071211 13264525 13824600 22924320 9308453 10833172 12404341 16475684 23301521 22924320 9590329 11375299 15911657 31298718 33297769 22924320 8879292 11600301 16160842 32113475 44298900 22924320 8986053 11818648 28973060 32573501 54212999 Wcal 1066.80 251.49 563.81 1379.12 2194.43 3009.74 1199.59 272.23 655.13 1582.49 2509.84 3437.20 1199.59 110.28 979.02 2068.32 3157.63 4246.93 1199.59 44.69 1110.22 2265.12 3420.02 4574.93 1199.59 18.12 1235.83 2453.54 3671.25 4888.96 Mcal 30127.50 24440.39 24168.77 22485.46 22699.51 22000.33 37707.50 21917.98 22329.94 22533.28 24063.14 21409.16 37707.50 21994.67 22399.51 23940.48 30562.25 38510.86 37707.50 20926.40 22427.70 23995.15 30873.77 39752.16 37707.50 20956.87 22454.68 30313.50 31000.94 40517.36 Env. risk 0.00 8.77 33.29 71.52 122.49 196.06 0.00 2.07 34.50 82.91 141.83 171.75 0.00 6.10 51.06 112.62 175.53 379.58 0.00 7.52 57.77 125.64 206.11 380.71 0.00 9.13 64.19 136.84 236.38 392.71

for environmental pollution (seventh column) shows that: the greater is the return desired by the farmer, the greater is the amount of fertilizer used, therefore the greater are the monetary penalties paid for the environment pollution. However the rate of return is in general increasing with respect to the parameter W . The efcient frontier of the minimum environmental risk problem is displayed in Fig. 1. The efcient frontier is the graph of the function that associates to values of W the optimal value of the objective function. It shows how the variation of the target prot inuences the amount of the monetary penalties paid for overcoming the environmental levels. The graph above is the graph of the fertilizer expenses versus the parameter W . max max min The numbers M2 , M1 , M1 and Qmax , i I that dene the ranges of parameters i M1 , M2 and Qi , i I were computed with ILOG-CPLEX package for MIP. The numbers min , max , W min ( ) and W max ( ) that dene the ranges of parameters and W were com-

Ann Oper Res Table 2 Optimal allocations of crops to plots and optimal values of the objective function for various values of parameter W W 1199.59 895.16 590.73 286.31 18.12 322.55 626.98 931.40 1235.83 1540.26 1844.69 2149.11 2453.54 2757.97 3062.40 3366.82 3671.25 3975.68 4280.11 4584.53 4888.96 barley barley barley barley barley barley barley barley barley Plot1 corn wheat wheat Plot2 wheat corn corn barley barley barley barley barley barley barley barley corn corn corn corn corn corn corn corn corn corn corn corn corn corn wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat wheat Plot3 corn Plot4 corn Plot5 barley barley barley corn corn corn corn corn corn corn corn corn corn corn corn corn corn corn corn corn corn Env. Risk 0 0 0 1.73 9.10 18.71 33.06 48.63 64.19 80.18 99.87 117.97 136.84 152.87 168.90 199.69 236.38 264.08 300.68 343.32 392.71 Exp. Fert. 0 0 0 49.1 189.05 371.78 461.39 526.79 592.18 659.31 742.01 825.42 919.50 986.83 1054.15 1452.84 1606.94 1785.18 2036.84 2215.91 2423.36 Rate return 0.03 0.02 0.02 0.01 0.00 0.01 0.03 0.04 0.06 0.07 0.08 0.09 0.08 0.09 0.10 0.11 0.12 0.10 0.11 0.11 0.12

Fig. 1 The efcient frontier for the minimum environmental risk problem

puted with DICOPT2X-C for MINLP. The optimization problems were solved on a computer with 2 GHz Intel Core 2 Duo processor and Windows XP operating system. Numerical examples for various dimensions were performed (n varied from 5 to 7 crops, m varied from

Ann Oper Res Table 3 Optimal allocations of crops to plots for 5 values of parameter W Area Fob Exp. fert. plot1 plot2 plot3 plot4 plot5 plot6 plot7 plot8 plot9 plot10 plot11 plot12 plot13 plot14 plot15 plot16 plot17 plot18 plot19 plot20 plot21 plot22 plot23 plot24 plot25 plot26 plot27 plot28 plot29 plot30 9 15 12 11 10 13 12 15 10 15 10 9 15 12 11 10 13 12 15 10 15 10 9 15 12 11 10 13 12 15 wheat corn wheat corn corn wheat corn sunower corn corn sunower corn sunower corn corn corn wheat sunower corn wheat corn sunower wheat sunower corn corn sunower wheat sunower corn sunower wheat wheat corn sunower barley sunower sunower sunower sunower sunower wheat barley corn corn sunower sunower sunower rapeseed sunower wheat corn sunower corn corn barley corn barley sunower W1 = 6520 0 0 rapeseed corn barley W2 = 8540 102 1970 rapeseed sunower barley W3 = 23601 569 4908 barley sunower barley W4 = 38662 1137 9200 barley sunower corn sunower sunower corn sunower sunower corn sunower sunower rapeseed sunower sunower corn sunower sunower corn sunower corn sunower wheat sunower sunower corn sunower wheat sunower sunower sunower W5 = 53723 2073 13512 barley sunower corn sunower sunower sunower sunower sunower sunower sunower sunower barley sunower sunower sunower sunower sunower corn sunower corn sunower sunower sunower sunower corn sunower sunower sunower sunower wheat

10 to 120 plots and k varied from 2 to 4 fertilizers/pesticides). The CPU time for obtaining solutions of optimization problems varied from one minute to 35 minutes. Some of the solutions obtained were suboptimal. The numerical experiments performed for the minimum environmental risk problem showed that GAMS solvers are suitable for the resolution of realistic instances of the problems presented in the paper. We consider in the following another numerical example with 5 crops: corn, wheat, barley, sunower and rapeseed, 30 plots, two fertilizers (urea and triple superphosphate) and two herbicides (Banvel 480 S and Dual Gold 960 EC). We take M1 = 10000 euro and M2 = 70000 euro. For = 859234771300 the optimal allocations of crops to plots for various values of W are displayed in Table 3 in columns 37. The area of plots is displayed in

Ann Oper Res

the second column. The cells that are empty in the table mean that the plots are not cultivated (fallow). In the second row are displayed the optimal values of the objective function, that is the penalties paid for the use of fertilizers and pesticides. In the third row are displayed the expenses made with fertilizers and pesticides. From Table 3 one can note that when the rate of return increases, the number of fallows decreases and also the area cultivated with sunower increase. An explanation for this thing is the fact that the market price for sunower is higher than the other prices, whence the cultivation of sunower is more protable.

10 Conclusions Agriculture is a risky business and a source of pollution for the environment. In this paper we have presented a multiobjective programming model that tries to solve the tradeoff between the economic and environmental objectives of the agricultural production. The aim of the model is to nd an optimal allocation of crops to plots and an optimal application rate of the chemicals such that the nancial risk and monetary penalties paid for the environmental pollution are minimized and expected return of the agricultural production is maximized. Starting from the multiobjective model several single objective models are formulated: the minimum environmental risk model, the maximum expected return model and the minimum nancial risk model. By introducing some supplementary variables we have shown that the above models are equivalent to mixed integer quadratic models. For the minimum environmental risk model two numerical examples are analyzed.
Acknowledgements suggestions. We especially thank to four anonymous referees for helpful comments and critical

References
Alfandari, L., Lemalade, J. L., Nagih, A., & Plateau, G. (2009). A MIP ow model for crop-rotation planning in a sustainable development context. Annals of Operation Research. doi:10.1007/s10479-009-0553-0. Annetts, J. E., & Audsley, E. (2002). Multiple objective linear programming for environmental farm planning. The Journal of the Operational Research Society, 53, 933943. Barrett, C., Reardon, T., & Webb, P. (2001). Nonfarm income diversication and household livelihood strategies in rural Africa: concepts, dynamics and policy implications. Food Policy, 26(4), 315331. Beneke, R. R., & Winterboer, R. (1973). Linear programming applications to agriculture. Ames Iowa: Iowa State University Press. Blank, S. C. (1990). Returns to limited crop diversication. Western Journal of Agricultural Economics, 15, 204212. Blank, S. C., Carter, C. A., & McDonald, J. (1997). Is the market failing agricultural producers who wish to manage risks? Contemporary Economic Policy, 15, 103112. Blank, S. C. (2001). Producers get squeezed up the farming food chain: a theory of crop portfolio composition and land use. Review of Agricultural Economics, 23, 404422. Castellazzi, M. S., Wood, G. A., Burgess, P. J., Morris, J., Conrad, K. F., & Perry, J. N. (2008). A systematic representation of crop rotations. Agricultural Systems, 97(12), 2633. Collender, R. N. (1989). Estimation risk in farm planning under uncertainty. American Journal of Agricultural Economics, 71, 9961002. Collins, R., & Barry, P. (1986). Risk analysis with single-index portfolio models: an application to farm planning. American Journal of Agricultural Economics, 68, 152161. Detlefsen, N. K., & Jensen, A. L. (2007). Modelling optimal crop sequences using network ows. Agricultural Systems, 94(2), 566572. Dogliotti, S., Rossing, W. A. H., & Van Ittersum, M. K. (2003). ROTAT, a tool for systematically generating crop rotations. European Journal of Agronomy, 19(2), 239250. Ellis, F. (1998). Household strategies and rural livelihood diversication. Journal of Development Studies, 35(1), 138.

Ann Oper Res Ellis, F. (2000). Rural livelihoods and diversity in developing countries. Oxford: Oxford University Press. El-Nazer, T., & McCarl, B. A. (1986). The choice of crop rotation: a modeling approach and case study. American Journal of Agricultural Economics, 68(1), 127136. Fafchamps, M. (1992). Cash crop production, food price volatility, and rural market integration in the Third World. American Journal of Agricultural Economics, 74(1), 9099. Figge, F. (2004). Bio-folio: applying portfolio theory to biodiversity. Biodiversity and Conservation, 13, 827 849. Freund, R. J. (1956). The introduction of risk into a programming model. Econometrica, 24, 253263. Goland, C. (1993). Agricultural risk management through diversity: eld scattering in Cuyo, Peru. Culture & Agriculture, 45/46, 813. Haneveld, W. K., & Stegeman, A. W. (2005). Crop succession requirements in agricultural production planning. European Journal of Operations Research, 166(2), 406429. Hardaker, J. B., Huirne, R. B. M., Anderson, J. R., & Lien, G. (2004). Coping with Risk in Agriculture (2nd ed.). Oxfordshire: CABI Publishing. Hazell, P. B. R. (1971). A linear alternative to quadratic and semivariance programming for farm planning under uncertainty. American Journal of Agricultural Economics, 53, 5362. Hazell, P. B. R., & Norton, R. D. (1986). Mathematical Programming for Economic Analysis in Agriculture. New York: Macmillan. Heady, Earl O. (1954). Simplied presentation and logical aspects of linear programming technique. Journal of Farm Economics, 36(5), 10351048. Hoanh, C. T., Lai, N. X., Hoa, V. T. K., et al. (2000). Can Tho Province case study: land use planning under the economic reform. In R. P. Roetter, H. Van Keulen, A. G. Laborte, H. H. Hoanh, & C. T. Van Laar (Eds.), SysNet Research Paper Series: No. 3. Synthesis of methodology development and case studies (pp. 4752). Johnson, S., Adams, R., & Perry, G. (1991). The on-farm costs of reducing groundwater pollution. American Journal of Agricultural Economics, 73, 10631073. Lazimy, R. (1982). Mixed-integer quadratic programming. Mathematical Programming, 23(1), 332349. Lewandrowski, J. K., & Brazee, R. J. (1993). Farm programs and climate change. Climatic Change, 23(1), 120. Markowitz, H. M. (1952). Portfolio selection. The Journal of Finance, 7(1), 7791. Mimouni, M., Zekri, S., & Flichman, G. (2000). Modelling the trade-offs between farm income and the reduction of erosion and nitrate pollution. Annals of Operation Research, 94(14), 91103. Nartea, G., & Bany, P. J. (1994). Risk efciency and cost effects of geographic diversication. Review of Agricultural Economics, 16(3), 341351. Newbery, D. M. G., & Stiglitz, J. E. (1981). The theory of commodity price stabilization: a study in the economics of risk. Oxford: Clarendon Press. Nidumolu, U. B., van Keulen, H., Lubbers, M., & Mapfumo, A. (2007). Combining multiple goal linear programming and inter-stakeholder communication matrix to generate land use planning options. Environmental Modelling & Software, 22, 7383. Park, T. A., & Florkowski, W. J. (2003). Selection of peach varieties and the role of quality attributes. Journal of Agricultural and Resource Economics, 28(1), 138151. Qiu, Z., Prato, T., & McCamley, F. (2001). Evaluating environmental risks using safety-rst constraints. American Journal of Agricultural Economics, 83(2), 402413. R dulescu, M., R dulescu, S., & R dulescu, C. Z. (2006). Mathematical models for optimal asset allocation. a a a Bucureti: Editura Academiei Romne (in Romanian) s R dulescu, M., R dulescu, S., & R dulescu, C. Z. (2009). Sustainable production technologies which take a a a into account environmental constraints. European Journal of Operational Research, 193, 730740. R dulescu, M., R dulescu, C. Z., Turek Rahoveanu, M., & Zb ganu, Gh. (2010). A portfolio theory approach a a a to shery management. Studies in Informatics and Control, 19(3), 285294. Rafsnider, G. T., Driouchi, A., Tew, Bernard V., & Reid, D. W. (1993). Theory and application of risk analysis in an agricultural development setting: a Moroccan case study. European Review of Agricultural Economics, 20, 471489. Reeves, L. H., & Lilieholm, R. J. (1993). Reducing nancial risk in agroforestry planning: a case study in Costa Rica. Agroforestry Systems, 21, 169175. Roche, M. J., & McQuinn, K. (2004). Riskier product portfolio under decoupled payments. European Review of Agricultural Economics, 31(2), 111123. Romero, C. (1976). Una aplicacin del modelo de Markowitz a la seleccin de planes de variedades de manzanos en la provincia de Lrida. Revista de Estudios Agro-sociales, 97, 6179. Romero, C., & Rehman, T. (2003). Developments in Agricultural Economics: Vol. 11. Multiple criteria analysis for agricultural decisions (2nd ed.). Amsterdam: Elsevier.

Ann Oper Res Romero, C. (2000). Risk programming for agricultural resource allocation: a multidimensional risk approach. Annals of Operation Research, 94, 5768. Santos, L. M. R., Michelon, P., Arenales, M. N., & Santos, R. H. S. (2008). Crop rotation scheduling with adjacency constraints. Annals of Operation Research. doi:10.1007/s10479-008-0478-z. Online First. Schaefer, K. C. (1992). A portfolio model for evaluating risk in economic development projects, with an application to agriculture in Niger. Journal of Agricultural Economics, 43, 412423. Teague, M. L., Bernardo, D. J., & Mapp, H. P. (1995). Farm-level economic analysis incorporating stochastic environmental risk assessment. American Journal of Agricultural Economics, 77(1), 819. Turvey, C. G., Driver, H. C., & Baker, T. G. (1988). Systematic and nonsystematic risk in farm portfolio selection. American Journal of Agricultural Economics, 70(4), 831836. Tziligakis, C. N. (1999). Relaxation and exact algorithms for solving mixed integer-quadratic optimization problems. Thesis (S.M.), Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center. Vizvri, B., Lakner, Z., Csizmadia, Z., & Kovcs, G. (2009). A stochastic programming and simulation based analysis of the structure of production on the arable land. Annals of Operation Research. doi:10.1007/s10479-009-0635-z. Weintraub, A., Romero, C., Bjrndal, T., & Epstein, R. (Eds.) (2007). International Series in Operations Research & Management Science: Vol. 99. Handbook of Operations Research in Natural Resources. Berlin: Springer. Weintraub, A., Romero, C., Bjrndal, T., & Lane, D. (2001). Operational research models and the management of renewable natural resources: a review. Bergen Open Research Archive, Institute for Research in Economics and Business Administration (SNF), Working Paper 11/2001. http://bora.nhh.no/ handle/2330/1371. Wossink, G. A., de Koeijer, T. J., & Renkema, J. A. (1992). Environmental-economic policy assessment: a farm economic approach. Agricultural Systems, 39, 421438. Zekri, S., & Herruzo, A. C. (1994). Complementary instruments to EEC nitrogen policy in non-sensitive areas: a case study in southern Spain. Agricultural Systems, 46, 245255.