17
Ecological Economics 30 (1999) 189 – 205 METHODS ECECMOD: an interdisciplinary modelling system for analyzing nutrient and soil losses from agriculture Arild Vatn a, *, Lars Bakken b , Peter Botterweg c , Eirik Romstad a a Department of Economics & Social Sciences, Agricultural Uni6ersity of Norway, PO Box 5033, 1432 Aas, Norway b Department of Soil & Water Sciences, Agricultural Uni6ersity of Norway, PO Box 5028, 1432 Aas, Norway c Center for Soil & En6ironmental Research, 1432 Aas, Norway Received 2 December 1997; received in revised form 9 September 1998; accepted 22 October 1998 Abstract This article discusses a set of principles for policy analysis of environmental problems. The main focus is on integrating economic and ecological analyses through a mathematical modelling framework. The paper starts by developing a general model for the study of environmental issues. Principles for operationalizing the model are discussed, and ECECMOD (a new modelling system constructed to analyze pollution from agricultural systems on the basis of these principles) is introduced. Some of the results obtained by ECECMOD are presented to facilitate a discussion about the gains to be obtained by this kind of analysis. The study shows that it is of great importance to combine economic and ecological analyses at a fairly high level of resolution when studying environmental effects of complex systems. © 1999 Elsevier Science B.V. All rights reserved. Keywords: Eco-eco modelling; Interdisciplinary research; Environmental regulations; Nitrogen; Erosion. www.elsevier.com/locate/ecolecon 1. Introduction Environmental problems are characterized by interactions among societal institutions, resource use choices made by economic agents within these institutions, and the different natural processes these choices influence and are influenced by. In this light it is disturbing that most studies (even those addressing environmental policy issues) have been undertaken within single disciplines. Given the character of the problem, such a strat- egy runs the risk of producing research with low cross-disciplinary coherence and little policy rele- vance (Vatn et al., 1997). The background of the study presented here is the North Sea Treaty mandating that the coun- tries of the area reduce nutrient emissions to the sea by 50%. Given the problem, we were ‘forced’ to find practical and productive ways of integrat- * Corresponding author. 0921-8009/99/$ - see front matter © 1999 Elsevier Science B.V. All rights reserved. PII:S0921-8009(98)00116-5

ECECMOD: an interdisciplinary modelling system for analyzing nutrient and soil losses from agriculture

Embed Size (px)

Citation preview

Ecological Economics 30 (1999) 189–205

METHODS

ECECMOD: an interdisciplinary modelling system foranalyzing nutrient and soil losses from agriculture

Arild Vatn a,*, Lars Bakken b, Peter Botterweg c, Eirik Romstad a

a Department of Economics & Social Sciences, Agricultural Uni6ersity of Norway, PO Box 5033, 1432 Aas, Norwayb Department of Soil & Water Sciences, Agricultural Uni6ersity of Norway, PO Box 5028, 1432 Aas, Norway

c Center for Soil & En6ironmental Research, 1432 Aas, Norway

Received 2 December 1997; received in revised form 9 September 1998; accepted 22 October 1998

Abstract

This article discusses a set of principles for policy analysis of environmental problems. The main focus is onintegrating economic and ecological analyses through a mathematical modelling framework. The paper starts bydeveloping a general model for the study of environmental issues. Principles for operationalizing the model arediscussed, and ECECMOD (a new modelling system constructed to analyze pollution from agricultural systems onthe basis of these principles) is introduced. Some of the results obtained by ECECMOD are presented to facilitate adiscussion about the gains to be obtained by this kind of analysis. The study shows that it is of great importance tocombine economic and ecological analyses at a fairly high level of resolution when studying environmental effects ofcomplex systems. © 1999 Elsevier Science B.V. All rights reserved.

Keywords: Eco-eco modelling; Interdisciplinary research; Environmental regulations; Nitrogen; Erosion.

www.elsevier.com/locate/ecolecon

1. Introduction

Environmental problems are characterized byinteractions among societal institutions, resourceuse choices made by economic agents within theseinstitutions, and the different natural processesthese choices influence and are influenced by. Inthis light it is disturbing that most studies (even

those addressing environmental policy issues)have been undertaken within single disciplines.Given the character of the problem, such a strat-egy runs the risk of producing research with lowcross-disciplinary coherence and little policy rele-vance (Vatn et al., 1997).

The background of the study presented here isthe North Sea Treaty mandating that the coun-tries of the area reduce nutrient emissions to thesea by 50%. Given the problem, we were ‘forced’to find practical and productive ways of integrat-* Corresponding author.

0921-8009/99/$ - see front matter © 1999 Elsevier Science B.V. All rights reserved.

PII: S 0921 -8009 (98 )00116 -5

A. Vatn et al. / Ecological Economics 30 (1999) 189–205190

ing relevant sciences. It turned out to be a chal-lenge both to integrate and to maintain the qual-ity standards of the involved disciplines. Weexperienced (in a very concrete way) why interdis-ciplinary research is so rare.

To be able to communicate coherently acrossthe fields of study, we had to find a commonlanguage and scientific structure. We found thatmathematical modelling could facilitate that. Ourfirst step was to develop a general model forstudying policy choices concerning pollution anddegradation of natural resources such as water,air and soils. Next, we had to find ways of makingthe model operative and of developing an analysistool by which empirical studies in the field ofnutrient losses from agriculture could be under-taken.

In this paper we present the most importantsteps of the process of model construction, illumi-nating the various conflicts that arise when inte-grating across disciplines. Further, we introducethe interdisciplinary modelling system that wasfinally developed (ECECMOD) and present someof the results obtained using this system. Sinceour focus is methodological, we will concentrateon results illustrating the potential of integrationand the gains that can be obtained by using ratherdetailed analysis with high levels of resolution.For those interested in a more comprehensiveoverview of the results, we refer to Vatn et al.(1996, 1997).

2. A general and idealized model

From an economics point of view, pollutionproblems are considered to be incentive problems.Economic agents do not face all costs associatedwith their activities. For natural scientists, pollu-tion is primarily about changes in matter cycles,cumulative or disruptive changes in various eco-systems where the ‘actors’ are ions, bacteria, fish,etc., operating at very different scales comparedto the economic ‘actors’. In order for the potentialconsequences of certain policies to be foreseen,the different levels/subsystems involved need to bewell linked and the right levels of resolution cho-sen. The problem, as we see it, is that economists

in their analysis primarily utilize very simplifiedassumptions about the natural world, and thenatural scientists hardly take into account the factthat in formulating policies one has to considermotives, constraints, and institutions determininghuman choices.

From an economic perspective, a pollutionproblem is of interest if there exists a potential forwelfare improvement by institutional changes re-sulting in less pollution. Coached in a principal-agent framework and assuming the agents to beexpected profit maximizers, the policy problemmay be formulated in the following general way:Principal:

MAX{u}

W=W{ %T

t=1

b t[ %n

i=1

Eit(pit), rt( %n

i=1

dit, uit)]}

(1a)

where

uit=(h� %

n

i=1

rit�

(dit

(1b)

subject to: Agent:

MAX{fit}

Ei(pi)= %Ti

t=1

{b itEit [ fit(p, 7, zit, fit, uit)]}

(1c)

and where: u denotes a vector of environmentaltaxes; W denotes social welfare; t denotes a timeindex, t�{1,2,...,T}; i denotes an index over loca-tion specific firms, i�{1, 2,...,n}; E denotes anexpectations operator set prior to each productiondecision made by the firm; p denotes profits; rdenotes a vector characterizing the state of therecipient; d denotes a vector of emissions influenc-ing other economic agents through changes in thecharacteristics of the recipient; h denotes a func-tion describing the value of the recipient given itsstate; ø denotes a vector of the agent’s choicevariables; b denotes the discount factor; f denotesa multi production function; p denotes a vector ofproduct prices; 7 denotes a vector of variableinput factor prices; and z denotes a vector describ-ing the natural assets used in production.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 191

To simplify the exposition, all agents (i ) areconsidered to be firms, and each firm owns ahomogenous element (i ) of the natural assets z.The arguments in the social welfare function Wconsist only of the present value of agents’ in-comes (profits p) and the states of the recipient r.The emissions are here considered to be global(e.g. the location of the emission does not influ-ence its effect). Following the standard Pigovianpolicy solution, the principal’s choice variable is aset of firm-specific taxes on emissions equal totheir marginal environmental costs. The taxchanges the incentive structure the agents arefacing, as the costs they impose on others becomepart of their optimization calculus.

Eq. (1c) contains little information about thestructure of the agent’s decision problem and thedynamics of the environment. A more completeformulation is given in Eqs. (2a), (2b), (2c) and(2d):

MAX{x itbitqit}

E(pi)

= %Ti

t=1

{b it[p ’tEt [ fit(xit, bit, qit, git(zit))]−u t* ’dit

−7t ’xit−bit ]}+b iTikt(biTi

, ziTi) (2a)

where:

dit=mit(x1,t−1, bi,t−1, qi,t−1, zi,t−1) (2b)

subject to:

zit= lit(xi,t−1, bi,t−1, qi,t−1, zi,t−1, rt−1) (2c)

where:

rt−1=st( %n

i=1

di,t−2, rt−2) (2d)

and where: x denotes a vector of variable inputsused; b denotes a vector of fixed inputs in theform of man-made capital; q denotes a parameterfor the quality of the chosen technology; g de-notes a function for the productivity of the natu-ral assets z ; u* denotes a vector of optimalenvironmental taxes obtained through solvingEqs. (1a), (1b) and (1c); k denotes a function forthe terminal value of assets b and z ; l denotes afunction for the effects of agents’ choices and

previous states of z and r on the state of thenatural assets r ; m denotes a function for theeffects of agents’ choices and previous state of zon emissions; and s denotes a function for therelation between emissions and the state of therecipient.

Eq. (2a) is formulated as a standard profitmaximization problem for agent (i ) given that itfaces a set of environmental taxes (u*) and hasaccess to natural resources zi. The input vectors xand b and technology (q) are specifications of thevector ø of choice variables as depicted in Eq.(1c).

The dynamics of the natural system which theagent influences through choices about inputs (x,b) and technology use (q), are described by Eqs.(2b), (2c) and (2d). As to assets, a distinction ismade between man-made capital b and naturalassets z and r. Elements of z are used as inputs inproduction, while r describes the state of therecipient. The distinctions are purely analytical,and what is a recipient (r) for one agent may bethe resource (z) for another. Thus, the dynamicsof emissions, the state of the recipients, and thequality of the resources are influencing each otherin complex and time dependent ways. This istaken care of by Eq. (2c) where the elements of zat time t depend upon previous capital invest-ments (in b and z) and the changes in the recipientr (Eq. (2d)).

Through the specification of z and t, the modelincorporates variation both in space and time. Inthe case of biologically based production systems,the productivity of the resources may vary sub-stantially due to variations in the quality of soiland water and the variability in weather. Theweather will only be known with certainty ex post.The agent must therefore make decisions on ex-pected profits, justifying the use of the expectationoperator E in Eq. (2a).

3. Principles for operationalizing the generalmodel

One is faced with a long list of difficult choiceswhen trying to operationalize the general modelcovered by the equation systems (Eqs. (1a) to (1c),

A. Vatn et al. / Ecological Economics 30 (1999) 189–205192

and (2a) to (2d)). In general there are importanttrade-offs to be made between preserving somecomplexities while reducing others.

First of all, agents are not necessarily profitmaximizers, even though most production analy-ses assume this. They may both have a morecomplex preference structure (utility maximiza-tion, risk aversion ( Nakajima, 1986; Singh et al.,1986)), and they may not (always) maximize.They may use satisficing rules or follow specificbehavioral norms developed among practitionersin the field (Simon, 1979; Hodgson, 1988; Vedeld,1998). Profit maximization still has high merit, asit is chosen also because it is easier to handleanalytically, avoiding further complications in analready very complex model structure.

Environmental taxes in Eqs. (1a) to (1c), and(2a) to (2d), assumed to be assigned to emissions,as in the standard point-source pollution case.This is not an obvious solution, since emissionsoften are non-point and variable. In most caseswhere renewable natural resource use is involved,the non-point character of emissions renders thecosts of gathering necessary information to under-take emission based regulations prohibitivelyhigh. This introduces an extra complication forthe principal, since measures may either have tobe assigned to input factors (x or b), or to theway these factors are combined internally andwith z through defined technologies (q). In ourcase we found it necessary to address thesecomplications.

The complexity of the problem will often makeit difficult to determine optimal levels of the pol-icy control variables in u. It is sufficient to men-tion the problems of valuing environmentalfunctions (Hausman, 1993; Vatn and Bromley,1994). Thus a full cost-benefit analysis may bevery difficult and, even obscure the problem. Analternative then is to use a cost-effectiveness crite-rion, measuring costs for obtaining certain levelsof emission reductions. Technically speaking thisimplies omitting the optimization problem definedin Eqs. (1a) to (1c) and instead produce informa-tion for the principal through a scenario typemodelling. Here the effect of various levels of thepolicy vector u on p, z, d and r is exploredutilizing the equations in Eqs. (2a) to (2d) only.

We chose to use this type of sequential searchprocedure. This does not necessarily reduce infor-mation quality. Rather, it may make the resultsmore transparent, as it also makes it easier topreserve higher resolution especially in the naturalscience part.

Given the above simplifications, one is stillfaced with the problem of describing and linkingprocesses (and thus disciplines) demanding analy-ses at widely different scales. Moreover, one hasto find tractable ways of solving an optimizationproblem with the complicated dynamics and feed-backs described in Eqs. (2a) to (2d). Let us ad-dress these issues.

3.1. Integration across scales

While economists do their analyses at the levelof the firm, a sector or the society, natural scien-tists normally undertake their analyses within geo-graphically localized, often rather small-scale-specified, physical and/or biological conditions.These differences are caused both by variations inscope and by the type of processes studied (Antleand Capalbo, 1993).

A way to overcome the problem of varyingscales and foci is to utilize the fact that naturalsystems tend to be organized in hierarchies(O’Neill et al., 1986). Processes are nested, withthose at one level becoming elements of those athigher. The problem becomes one of maintainingthe necessary non-linear fine-scale variation asone moves upwards into more coarse-scale aggre-gations. Rastetter et al. (1992) discuss three meth-ods by which to accomplish this:a. partial transformation using an expectation

operator to incorporate fine-scale variability;b. partitioning—separating the coarse-scale ob-

jects (aggregates) into a manageable set ofrelatively homogeneous subgroups; and

c. calibration—recalibrating fine-scale data tocoarse-scale information.

Incorporating fine-scale variability by a statisti-cal expectations operator (method (a)) may yieldgood results, but is often difficult to utilize insystems with many elements and dimensions ofvariation. The covariation structure often must beassumed, and hence becomes ad hoc. The useful-

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 193

ness of method (c) rests on the availability of databoth at the fine and coarser scales, and is difficultto use if the aim is to study effects of changes ina system.

As Costanza et al. (1995), we find partitioningto be a good solution for analyses of land-basedsystems. Here the various resources or decisionunits are classified into homogenous groups atrelevant levels and integrated in a hierarchicalstructure. This way partitioning offers a sur-veyable platform for linking analyses that must beundertaken at different levels and studied withvarious resolutions.

Linking across levels means to a large degreelinking across disciplines. While natural hier-archies exist, it may still be difficult to identifyaggregates that serve both the ‘aggregate’ and the‘disaggregate’ sciences well. Partly, this reflectsdisciplinary traditions, but it may also reveal realdiscontinuities. These kinds of problems are espe-cially prone to occur as one crosses over from thenatural to the social sciences. We will return tothis issue later.

3.2. Sol6ing the optimization problem

The optimization problem in Eqs. (2a) to (2d) isformulated as an optimal control problem in dis-crete time. The complexities, especially in theunderlying structure of the natural processes, willhowever often prohibit solving Eqs. (2a) to (2d)simultaneously. Reducing complexity may not bejustifiable. Two partial solution concepts thatstand out are:1. Iteratively solving each equation in Eqs. (2a)

to (2d) for relevant sequences of time, updat-ing the equations with necessary informationfrom each other.

2. Modelling for the whole time period in twosequences—first agents’ choices, then the ef-fect of these choices on the natural processes.To cover important feed-backs, the naturalscience information necessary to solve agents’optimization problems are produced initiallyunder simplified assumptions about agents’choices/practices.

The merits of these two methods depend on theproblem, the type of dynamics and the role of

time. Method 1 takes best care of the interactionsbetween the different levels of the system — thedynamics between the economic and ecologicalparts. For large systems, iteration will tend to bevery resource demanding, and it will by definitionnot be possible to solve the economic optimiza-tion problems simultaneously for the defined pe-riod of analysis.

According to procedure 2, the agents’ optimiza-tion problem will be solved given rather coarsepre-estimated information about developments inthe natural processes. Thereafter, the equationsdescribing the natural asset dynamics are solvedat the most desirable level of resolution and onthe basis of information about the model agents’choices of production practices. Agents’ informa-tion about the development of the natural re-sources are often coarse and uncertain at the timeof decision. Procedure 2 actually mimics that fea-ture, as it also is the easiest to implement if highresolution in the natural science part is necessary.We have chosen method 2 in our analysis.

4. ECECMOD—The economics and ecologymodelling system

To demonstrate how the above principles maybe utilized, we will turn to a presentation ofECECMOD—a modelling system established tostudy the effect of different policy measures onmineral emissions from agriculture. The system isfully documented in Vatn et al. (1996) and con-sists of a set of process-based, interlinked models.Some are constructed specifically for the project,while it has been possible, in other cases, to buildon existing models.

Reviewing the literature on agricultural pollu-tion, we found few studies focusing on the dy-namics between the relevant economic andecological processes. As an example, a dominanttrend in policy oriented studies of nitrogen emis-sions is to look at the effect of various instru-ments on a calculated nitrogen surplus (Vatn etal., 1997). In our opinion a more sophisticatedtreatment of natural processes with a higher levelof resolution is warranted.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205194

4.1. The agronomic system and nutrient losses

Nitrogen and phosphorus are essential elementsof mineral fertilizers added to enhance plant pro-duction in agriculture. If lost to the environment,the same elements represent a potential pollutionproblem. They may cause eutrophication in freshand/or coastal waters. Nitrates may reducegroundwater quality, induce acidification and, inthe form of N2O, affect the ozone layer andglobal warming.

Phosphorus displays more local effects, whilenitrogen in different compounds influences theenvironment both at the local level and on aglobal scale. Nitrogen is a dynamic and elusiveelement in nature. Therefore, both the form andmagnitude of losses depend on a series of condi-tions such as weather, fertilizer levels, plant up-take, incorporation of organic matter into soils,utilization of manure N, etc. Supplies of mineralN during the growing season is a prerequisite forsuccessful plant production. Nitrate (NO3

−) is avery mobile ion, making it a preferred N-source inplant production, but also prone to leaching.Losses may be decreased by reducing the input ofmineral N. However, both plant uptake and soilprocesses affect losses greatly. This implies that tocontrol losses, one must look at the agronomicpractice as a whole.

Plant roots are by far the strongest mineral Nsink in the soil. The response of plants to Nfertilization is crucial in predicting the relation-ship between fertilizer levels and nitrate leaching(Vold et al., 1994). Crop failure due to droughtperiods in rainfed agriculture results in large Nlosses through leached residual nitrate out offseason (Uhlen, 1989).

The soil and the heterotrophic soil microflora(+ fauna) are other potential sinks. N incorpo-rated in organic soil compounds represents a largepool of nitrogen in the system—20–100 times theannual doses of mineral fertilizers. Dependingupon the supply of energy through plant residuesand/or the type of tillage system used, the mi-croflora functions either as a net sink or netsource of mineral N. Due to the stability of theorganic components involved, a net increase ordecline in the soil organic N pool may continue

for decades, depending on the agronomic practice(Christensen and Johnston, 1997).

The annual changes in soil organic N may bemoderate compared to the total amount in thesoil, but large in relation to the nitrogen inputs(fertilizers) and outputs (crops) from the system.Soil texture has a profound influence on the leveland changes in soil organic nitrogen (ibid). Thisrole of soil organic N invalidates N budgets as acriterion for the environmental performance of anagronomic system with normal N inputs. At thesame time it constitutes an important rationale forchoosing a process-oriented N modelling ap-proach with a high resolution both in time andspace when analyzing the effects of policy mea-sures for reducing nitrate leaching.

Phosphorus is much less elusive than N, as it isstrongly bounded to soil mineral fractions. Sincelosses are to a large degree the result of erosion,the input of mineral phosphorus through fertiliz-ers has little impact on losses, at least for shortand intermediate time spans. Plant cover, tillagepractices, etc. are far more important for erosionand resulting P-losses than input/output relations.Thus even this case shows how important it is tounderstand the dynamics of the agronomic prac-tice as a whole. Further, the impacts vary sub-stantially with differences in soil characteristicsand topographical conditions. In the case of soilerosion, losses are highly episodic, due to theimportance of precipitation.

4.2. System, le6els and processes

As previously emphasized, studying dynamicsof the kind described above requires careful con-sideration of system boundaries, choice of resolu-tion, and how to link various processes. Inconstructing ECECMOD we decided to use wa-tersheds or catchments as the largest unit of ana-lysis. Weather developments and water move-ments are the most significant natural process ofimportance for system demarcation. Certainlycatchment areas can be substantial. There is inprincipal no technical obstacles other than capac-ity and data requirements related to coveringsubstantial areas. As we have formulated themodelling system, though, weather conditions are

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 195

considered homogeneous within each catchment,in practice often implying necessary division intosubcatchments. While few economic units aredefined by watershed borders, they are capable ofsubsuming a number of production units fairlywell.

Using partitioning as the basic method for pre-serving the necessary variation and for nestinglevels, we observe that farms, soil types, andtopography are important spatial elements ofvariation. Farm size, type of production, andstocking rates are notable sources of diversity atthe farm level. Nitrogen turnover, plant growthand P losses vary with climate, soil types andtopography, the latter influencing water erosion.Based on this, we chose the following levels andpartitioning criteria:1. Watersheds/catchments. Partitioning criteria:

Weather conditions, topography and produc-tion. Modelling: Aggregated emissions to airand water.

2. Farm. Partitioning criteria: Farm size, type ofproduction, and stocking rate. Modelling:Farmers’ choices of agronomic practices.

3. Farm field. Partitioning criteria: Soil proper-ties and chosen agronomic practice. Mod-elling: Crop growth and farmers’ choices ofagronomic practice; hydrology, nutrient (N)turnover and leaching.

4. Plot/grid cell. Partitioning criteria: Topogra-phy, soil properties, slopes, and agronomicpractice. Modelling: Erosion and P-losses.

With regard to the time dimension, it was nec-essary to combine high resolution (episodic na-ture) with long simulation periods (variation). Thetotal length of the analysis period is set at 20years. Farmers’ decisions are dominantly mod-elled at the level of years or seasons (choice ofcrops, tillage practice etc.), but in some cases atthe level of days (sowing dates, etc.). Each ofthese decisions is modelled consistently with theinformation resolution of the corresponding deci-sion problem. The natural science modelling ismainly undertaken at the level of days. In theerosion analysis, the basis is rainfall events thatmay have durations from minutes up till severalhours.

4.3. The structure of the modelling system

ECECMOD consists of a set of linked submod-els covering the different parts of the system asdescribed in the above hierarchy. As the mod-elling system now works, the optimization prob-lem in Eq. (2a) is solved on a year by year basis.This is due to capacity constraints. It does, how-ever, represent a minor problem in our case, sincethe terminal value of the relevant assets (e.g. soiland soil N volume) is not considered to influencethe choices studied.1 Fig. 1 gives an overview ofthe main structure of ECECMOD.

The figure differentiates between external in-puts, (process based) models, and model outputs(intermediate and final states). The level of resolu-tion (scales) for the space and time dimensions isalso given. The analyses are driven by a set ofinput data defined by farm structures and soil andweather conditions in the actual catchment. Bychanging the institutional setting within whichfarmers’ decisions are to be made, it is possible toanalyze the effects on emissions and the costsimplied. Thus far the effects of the emissions ondown-stream water-courses is not integrated intothe model.

As to the different parts of the system, wewould emphasize the following:

ECMOD is an optimizing model operating atthe farm level. The optimizing problem to besolved follows in principle the formulation in Eq.(2a) with constraints related to acreage and labor.The policy variable is (as already emphasized)formulated differently though, as taxes or othertypes of regulations are attached to inputs (x orb) like fertilizers, or to the management practices(q) like plowing time, catch crop use2, storingfacilities for manure, etc. and not to emissions.

ECMOD consists of a set of modules solvingthe choice problems that are assumed to be the

1 For example, farmers are, under Norwegian conditions,not considered undertaking soil mining — a point where themodel formulation actually deviates from a strict net profitmaximizing rule.

2 Catch crops are in this case undersown rye grass in grain.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205196

Fig. 1. The structure of ECECMOD.

ones influencing emissions the most. These are thesetup of machinery, choice of crops and croprotation, fertilizing levels, manure practices (bothstoring and spreading), soil tillage, and sowingdates. The optimizing procedures are mainly non-linear and based on expectations where relevant.The model also calculates economic parameters tobe used when estimating the cost-efficiency ofdifferent strategies.

SOIL is a deterministic, one dimensional, lay-ered hydrology model simulating water content,water movements and temperature driven byweather input and soil characteristics (Jansson,1991; Botterweg, 1992). The information pro-duced here is used both as driving data for mod-elling nutrient turnover (N) and erosion and fordetermining sowing date (simplified pre-runsmade for use in ECMOD).

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 197

SOILN–NO is also a one-dimensional, layereddeterministic model describing nitrogen turnover(Johnsson et al., 1987; Vold, 1997). It producesinformation about loss of N as a function ofagronomic practice, plant growth, weather, andsoil characteristics (soil type and agronomic his-tory) expanded from point to field level assuminghomogenous fields. The modelling of the micro-bial N-transformations has been found to respondadequately to various physical and chemical fac-tors, and to various litter types (Vold et al., 1994;Vold, 1997). Parameterization of the plant N up-take function is based on regional agronomicexperiments (yield and N uptake) and regionalannual yield statistics. It is estimated as a set ofnon-linear functions with yield and available N asarguments with good fit (R2 for single cropsvaried from 0.82 to 0.92; Vold et al., 1999). Pre-runs are undertaken to estimate the effect onN-mineralization of changed manure practices,catch crop use, etc. for use in ECMOD.

EUROSEM describes erosion as a function ofagronomic practice, weather, soil characteristicsand local topography. It is an episode-oriented,process-based model substantially diverging fromthe more dominant tradition of regression-basedanalyses (Botterweg, 1998; Morgan et al., 1998).It is used to produce estimates for soil losses as afunction of soil characteristics, topography, agro-nomic practice, and weather conditions.

Landscape models. Finally, two systems ormodels for aggregating data from plot and/orfield level to the level of watersheds are developed.The aggregation of costs and N leaching estimatesis a fairly simple weighing based on the relativedistribution of productions, soils, etc. in land-scapes. For erosion, GRIDSEM (Leek, 1993) isused to estimate mass transport in landscapes,aggregating losses over whole watersheds as afunction of estimated losses at each plot/grid cell(from EUROSEM), the distance from plot tosurface waters, and topographic characteristicsover this distance (Botterweg et al., 1988).

ECMOD is run for a set of representativemodel farms covering the variation in each catch-ment. The actual farms in the area are partitionedinto groups defined by what is considered themost important variables in our case—e.g. type

of production, farm size and amount of manure(total and per ha). For each of these groups amodel farm is defined, generally based on themean values for various parameters for the under-lying real farms. Each model farm is divided intofields with representative soil characteristics. Thelandscape is divided into series of plots/grid cellshomogeneous in soil type and slope. To be able totransfer information about agronomic practicesfrom the analyses at the model farm level back tothe landscape level, each plot is attached to amodel farm field on the basis of soil propertiesand which partition of real farms it belongs to.The methodology chosen thus demands dataabout the location of each farm and extensivelandscape information.

4.4. Partitioning and interaction — crop growthas an example

The way crop growth is modelled illustrates themanner in which different levels of ECECMODinteract. The function f in Eq. (2a) is developed inthe following way:

Y= fijkt(N1t, N2t, N3t)*Vat�P (3)

where: Y denotes yield as dry matter; i denotesmodel farm field/type of soil; j denotes type ofcrop; k denotes climate zone; t denotes time (yearor season); N1t denotes nitrogen in mineral fertil-izers applied at time t ; N2t denotes nitrogen frommanure in mineral form (ammonia) applied attime t ; N3t denotes mineralized N from the or-ganic components of the soil at time t ; Vat denotesan operator covering the effects of different ele-ments of agronomic practice except fertilizing;and P denotes phosphorus in fertilizers.

The production function is partitioned by soil,crop, climate zone and year, the latter to captureweather dependent variation between years. Fur-ther, it is structured so that mineral N is the onlyargument in the production function, while theinfluence of other elements on f(·), is taken care ofby the operator Vat. The N component is dividedinto three parts, whereby mineral N from fertiliz-ers and manure are elements of the x-vector inEq. (2a) and mineralized N from the soil is anelement of git(zit). Other elements of the produc-

A. Vatn et al. / Ecological Economics 30 (1999) 189–205198

tion function of Eq. (2a) are either kept con-stant (e.g. incorporated into f ) or they are takeninto account by Vat.

The function fijkt is estimated on the basis oftrial data for an agronomic practice using falltillage and mineral fertilizers only. Changes intillage practice, sowing date, use of cover crops,etc. are taken into account by Vat inducingchanges in the parameters of fijkt relative to theirestimated effects in trials. This illustrates thatthe structure is chosen also to facilitate the useof data from factorial trials in a more dynamictype of modelling.3 As Eq. (3) is structured,changes in mineralization (N3) due to the use ofanimal manure, catch crops etc. is captured in away that also influences the optimal use of fer-tilizers in the form of N1 or N2. The level of N3

is established on a more coarse scale than in thefinal N modelling—e.g. the average values forthe whole 20-year period is used. The data isproduced through pre-estimations by runningSOILN–NO under different standardized catchcrop regimes.

The crop growth module illustrates a majordifference in needs between economics and thenatural sciences. While the plant growing pro-cesses must be modelled following the actualweather, the farmer makes decisions mainly onexpectations about the weather and subsequentcrop growth. This distinction is taken care of byestimating annual production functions to beused in the natural science part of the modelling( fijkt) and an average function used to representthe expectations of the farmer ( fijk). Thisdemonstrates that modelling economic choices ata coarser scale than the N dynamics may oftenbe the appropriate choice, representing no infor-mation loss compared to the real world situa-tion. In some cases (as in the study of a splitfertilizing regime) we allow the farmer to updatefijk with information about weather developmentbetween the different applications to secure con-sistency.

5. The gains of high resolution eco-eco modelling

To illustrate some of the gains of the type ofanalysis advocated here, we will turn to a selectedgroup of results from a study undertaken for twocatchments in southeastern Norway. A series ofresearch objectives has been formulated focusingon the cost-effectiveness of input regulations ver-sus regulating the agronomic practices. Since nat-ural conditions and types of production varysubstantially among farms and regions, it is ofspecial interest to analyze gains and losses relatedto uniform versus differentiated policy measures.

5.1. The study sites

The analysis covers the Mørdre catchment anda subsection of the Auli catchment in the countiesof Akershus and Vestfold, respectively. Theamount of arable land in the Mørdre catchment isapproximately 450 ha and in Auli 2.600 ha. Bothareas are dominated by grain production, whilethere is some animal production especially in theAuli area. The distribution of soil types is verydifferent in the two catchments (Figs. 2 and 3).

While clayey and sandy soils dominate in Auli,silty soils dominate in Mørdre. Auli has a highermean annual temperature (6.3°C compared to4.2°C) and precipitation (1009 mm compared to809 mm). In both areas, grain is the most impor-tant crop. However, Auli has a much larger pro-portion of animal husbandry. The amount ofmanure N per ha is on average ca. 75 kg in Auliand 18 kg in Mørdre.

5.2. Strategies for reducing nitrogen emissions.

Various types of regulations directed either to-wards inputs (x, b) or the farming practice (q) areavailable to the authorities. Analyzing this wedistinguish between ‘physical’ measures which arethe changes farmers undertake in the way theyfarm, and ‘policy’ measures which are the instru-ments the authorities use to motivate or compelfarmers to change practices. According to thedynamics of the plant-soil system (z) in grainproducing areas like ours, we grouped physicalmeasures into four categories:

3 It must also be emphasized that most of the data availablewas of this kind.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 199

� reduced total nutrient input (lower ‘intensity’,better manure practices, etc.);

� improved adjustment of nutrient input (espe-cially of N) to yearly growth condition (splitfertilizing, etc.);

� changed tillage practices and/or continuousplant cover (catch crops) etc. to reduce erosionand enlarge the plant N uptake period; and

� increased rates of N immobilization in the soilin periods where crops are not grown (incorpo-ration of catch crops, straw, etc.).Policy measures to accomplish the above

changes may be a tax on mineral N fertilizers(fertilizer intensity/manure handling), seasonalquota system (split fertilizing), mandatory or sub-sidized practices (changed tillage practices andcatch crops). The interactions covered by ECEC-MOD make it possible to account for a widerange of both direct and indirect effects in thesystem, related to changes in agronomic practicesand losses of nutrients. Many of these effectscould hardly have been studied through a moreaggregate procedure, and in some instances theeffect of preserving high resolution turns out to beprofound. Through the following presentation ofresults, we will try to illuminate the role of pro-duction type, choice of crops, choice of technol-ogy, soil characteristics, soil dynamics andtopography.

5.3. Comparing different policy measures

To compare policies, a set of scenarios wasconstructed. To validate the modelling system, a

Fig. 3. Distribution of production types in Auli and Mørdre,farms classified by dominant production.

scenario covering the actual policy conditions in1992 was run. Compared with available data, theresults were very encouraging both concerningchosen agronomic practices and leaching levels.The fit for yearly and average yields, fertilizerlevels and choice of crops was very good, withdeviations of only a few percent. The level ofspring tillage was somewhat over-estimated,whereas the opposite was the case regarding ma-nure storing capacities (for more specific docu-mentation see Vatn et al., 1996). As to leachingand erosion, we encountered some problems withacquiring spatially comparable observationaldata. Data both from smaller parts of the catch-ments and lysimeter experiments have been uti-lized. Except for the lysimeter trials, directcomparisons are difficult to do. For these trialsmodel fit was in general found to be very good,except for abnormally high fertilizer level treat-ments (Vold, 1997). Concerning erosion and Plosses comparisons with field observations wasutilized, showing deviations normally within arange less than 920–30%.4 Estimated level of Plosses in Auli were the ones deviating the most.For more specific documentation see Botterweg(1998) and Vatn et al. (1996).

Next, a Base scenario was run. Again the refer-ence was the policy conditions in 1992, but exist-ing environmental regulations were ignored toformulate the best basis for comparing various

Fig. 2. Distribution of soil types in Auli and Mørdre.

4 When evaluating this, one should remember the difficultiesin finding strictly comparable results and the high variability inerosion processes.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205200

Table 1The effect of various policy measures to reduce emissionsa

DS lossdScenario DN aircN leaching Soc.abatem. Soc.abatem.DP loss Costsmean mean costs Bcosts Amean farmermean

NOKe/hakg/ha (%)b kg/ha (%)b NOKe/ha NOKe/kg Nfkg/ha (%)b kg/ha (%)b

Base6.5 4.97 1180Auli 40.41.9 0.56 28129.4Mørdre

Tax1004.0 (38) 4.88 (2) 1171 (1)Auli 57034.5 (15) 77 131.1 (42) 0.56 (0) 281 (0) 61025.0 (15) 51Mørdre 12

Catch506.5 (0) 2.81 (43) 660 (44)Auli 19929.6 (27) 199 181.9 (0) 0.56 (0) 283 (−1) 29421.5 (27) 294Mørdre 37

Soil-sub6.5 (0) 2.25 (55) 505 (57) −370 138Auli 37040.0 (1)1.9 (0) 0.55 (2) 2.77 (1) −510 3029.5 (0) gMørdre

a Estimated costs and changes in emissions as compared with the Base scenario.b Percent reduction from Base.c Loss of ammonia-N in connection with manure spreading.d Soil loss.e NOK=Norwegian kroner.f Costs per Kg reduced N leaching.g Irrelevant since N leaching is increasing

policy options. We will start by comparing theresults from the following scenarios:� Tax100: 100% tax on N-fertilizers;� Catch50: 50% arable land requirement on catch

crops or grass cover; and� Soil-sub: A per hectare subsidy for spring/no

tillage (1992 level: 1000 NOK per hectare).While the tax has to be uniform for the entire

economic–political area (in our case the whole ofNorway) the other measures can be differentiatedbetween regions if desirable.

The results (including the Base scenario) aregiven in Table 1. The estimates are produced onthe basis of weather data for each area for theperiod 1973–1992. The table shows a fairly broadarray of environmental indicators, making it pos-sible to analyze trade-offs involved. Estimates aregiven for N leaching, NH3 losses to air, P lossesand losses of soil. A set of cost-efficiency mea-sures is also presented. ‘Farmers’ costs’ showsaverage reductions in profits per ha as measuredagainst the Base scenario. Two average socialcost-efficiency measures (defined as ‘Farmers’costs’ exclusive environmental taxes/subsidies) are

presented. Since there are several types of emis-sions with no obvious common yardstick, a per hameasure is given (‘social abatement costs A’). Inthe case of ‘social abatement costs B’, all costs arecarried by the N-leaching—the dominant form ofN emission. Costs related to policy administrationare not included. This will be addressed later.

Starting with the Base scenario, we see that theestimated agronomic practices and losses differbetween the areas. The higher level of leaching inAuli is mainly due to the level of animal manureand the rather large proportion of sandy soils,which are most prone to leaching. The low soilloss in Mørdre is explained by differences in slopecharacteristics, as well as the dominance of siltysoils. These soils are generally tilled in spring sincethat produces the highest yields. Spring tillagereduces erosion.

Turning to the other scenarios, we observe arather complex picture of changes. As one wouldassume, Tax100 mainly influences N leaching andammonia losses while Soil-sub influences P andsoil losses mostly. The effects vary substantiallybetween the catchments, though. The low effects

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 201

of Soil-sub in Mørdre is mainly explained bythe mechanisms keeping losses in the Base sce-nario down.

Catch50 reduces both N-leaching, P and soilloss. It could thus seem to be a good alternativein areas where all the measured types of lossesare of importance. The effect of Catch50 on N-leaching is quite substantial, illustrating thatlosses can be reduced considerably if one is ableto continue plant uptake for a longer period ofthe year. Reductions in inputs are thus not theonly interesting options for reducing N losses.For the most part, nitrogen taken up by catchcrops enters a pool of stable organic compoundsafter ploughing. This pool has a decay rate (firstorder decay) of about 0.01 per year (Vatn et al.,1996, pp. 55–67). This implies that the system isable to accumulate N in this pool for severaldecades before returns (as humus-derived min-eral N) becomes a significant contributor to themineral N supply. A separate model studyshowed that the catch cropping will reduce themineral N supply to the cash crop for a periodof some 30–40 years. This actually results inincreased optimal N fertilizer (N1+N2) levels.After this period the soil starts ‘paying back’and the level of fertilizers can then be reduced(Vold et al., 1995).

Looking at the costs, the tax entails the low-est social abatement costs if we only considerN-leaching. The reductions in P and soil lossesfollowing the two other measures, may still beconsidered large enough to outweigh the higherper hectare costs, especially in the case of catchcrops. We observe that farmers’ costs and socialcosts are identical in the case of Catch50, whilein the tax scenario, the tax counts for about90% of farmers’ costs. The observed incomegain (negative costs) for the farmer in the caseof Soil-sub, indicates that the subsidy was sethigher than necessary to motivate farmers to re-duce fall tillage. The table illustrates that thedistributional effects can be substantial depend-ing on the type of policy measures used (taxes,requirements or subsidies).

Catch crops are less effective in Mørdre thanin Auli. The social abatement costs are 50–100% higher than those of the latter. Several

factors cause this. The lower initial level ofleaching plays a role. The most significant mech-anisms, however, is lower catch crop yields inMørdre due to climatic factors, and the changesin tillage practices following the use of suchcrops. The silty soils of Mørdre will, in the ab-sence of catch crops, be tilled in spring. Usingcatch crops implies a switch towards fall tillage,causing decreased yields counteracting the posi-tive effect of catch crops on leaching. The yieldloss also influences the income from production,resulting in higher per ha costs, as shown inTable 1.

Table 1 covers only average costs. Calcula-tions show that while the marginal costs arefairly constant for catch crops over large inter-vals, they increase almost linearly with increasedtax levels. At the margin, catch crops seem tocompete with an N-tax under Auli conditions attax levels around 75%, even if we only considerN leaching. In the case of a tax, marginal costsincrease primarily because the plant productionfunctions are concave. For catch crops, the mar-ginal costs shift only dependent on which soiltypes are involved.

As mentioned earlier, administration costs arenot included in Table 1. The costs of adminis-tering the tax system is considered negligible.Presently, a low tax exists. Since there are fewwholesalers in Norway, the administration costsamount to just a few man months per year. Thecosts of a catch crop requirement will be higher.Estimates made (Vatn et al., 1996) show thatadministration costs will increase social costs byapproximately 5%.

5.4. Variation between productions

Production type is an important source ofvariability that Table 1 does not inform usabout. Table 2 shows the effects of a tax and acatch crop requirement for a selected group ofmodel farms in Auli. Here we only focus onN-leaching.

MF1 is one of three model farms representingmilk production in Auli. MF4 represents pro-ducers of pork/grain with fairly low manure lev-els, while MF7 covers the smallest grain farms.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205202

In the Base scenario these farms have an averageleaching varying between 37 and 42 kg N/ha.MF1 has the lowest and MF4 the highest losses.While the tax has the largest effect on the milkproducing model farm, the effect of a 50% catchcrops/grass land requirement is negligible here,since it already has nearly 50% of its acreage ingrass production. In the case of farms with grainas the dominant crop, the situation is the oppo-site.

We observe a much larger reduction in the useof fertilizer N in grass than in grain production—both absolutely and relatively. While the averageeffect of a 100% N tax is approximately 27% (40kg/ha) for grass, the figures for grain are 10% (11kg/ha). The difference is primarily due to substitu-tion between fertilizer N and clover in productionof roughage. We observe positive effects followingchanges in the manuring practices, too. In thecase of MF1, Tax100 induces increased storagecapacity up till a full year’s storage. This elimi-nates losses related to fall spreading. The greatesteffect of this change, however, is due to lesscompaction damages, hence improved conditionsfor plant growth.

Finally, we observe that social abatement costsvary rather substantially across the model farmsin the case of a 100% tax, while they are muchmore uniform for the catch crop regime. Theprivate costs are hardest to bear for the special-ized grain farm since per ha incomes are lowest inthis case.

5.5. High sensiti6ity for changes in the plant–soilsystem

5.5.1. The effect of different tillage practicesThe results are, as already observed, heavily

dependent upon a series of interacting agronomicfactors. One aspect of plant–soil interaction, theeffect of tillage practice on N dynamics, caughtour interest. Analyzing this more in detail weobserved:� Postponing fall tillage by about 5 weeks from

the end of August to early October gives, ac-cording to ECECMOD, a reduction in leachingapproximately at the level of a 100% N tax ingrain dominated areas like Auli and Mørdre.This is entirely an effect of N-uptake by weedsand germinating shed grains and its subsequentinfluence on the soil micro flora/fauna. Theestimates show higher effects in Auli than inMørdre. This difference relates mostly to thefact that 50% of the area in Mørdre is tilled inspring also in the Base scenario and is thus notinfluenced by this change. One should take intoaccount that delayed fall tillage induces noextra costs, except the risk that some areas willnot be tilled in time in years with rainy au-tumns and/or early frost.

� Increased spring tillage (mandatory or follow-ing from a subsidy scheme) turned out, as wehave already seen, to give very small effects onN leaching. This was a surprise. Even thoughthe mass of weeds/shed grain is increased by

Table 2Changes in N-leaching (D N) and costs as compared with the Base scenarioa

Model farmb Tax100 Catch50

Soc. abatem.D N Costs CostsSoc. abatem. D Ncosts kg/hafarmerkg/ha farmer costs

NOK/kg DNNOK/kg DNNOK/ha NOK/kg DN NOK/ha NOK/kg DN

−10.5 610 59MF1 18 −1.3 20 16 16−3.7 300 16 16510 139 46MF4 −18.3

−15.9 300 19 19650 129MF7 10−5.1

a Some model farm estimates, Auli.b MF1, milk/beef; MF4, pork/grain; MF7, grain.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 203

expanding the growth period for some extraweeks compared to late fall tillage, we observethat the effect of extra plant uptake is counter-acted by a decline in the subsequent main cropyield on sandy and clayey soils the followingseason.To comprehend the sensitivity of the system,

recognize that the average leaching in the Basescenario is about 20–25% of a total plant uptakeof about 150 kg N per ha and year. A change inthis uptake of about 3% will, other things beingequal, create a potential change in leaching at thelevel induced by a 100% N tax. On clayey soilslike the ones in Auli, yield losses following atransition from fall to spring tillage are at theaverage level of about 4%, explaining much of theabove results.

As mentioned, the modelled effect of delayedautumn tillage on nitrate leaching is attributableto N-assimilation by weeds and germinating shedgrains throughout the autumn. This has not beenconfirmed by direct observations of nitrate leach-ing in field experiments, however. The model as-sumes no extra mineralization of soil organic Nby tillage, despite the widespread notion thatplowing has such effects. However, this is anassertion for which there is very little evidence.Our modelling has pinpointed that nitrate leach-ing is sensitive to such possible direct effects ofplowing and to N-assimilation by weeds and ger-minating shed grains. This puts new questions onthe agenda for natural scientists about the specificmechanisms invoked by ploughing, and on theN-assimilation by weeds.

5.5.2. Split fertilizationThe assumed positive effect of split fertilization

is related to better adjustment of the fertilizerlevel to the actual growing conditions each year.A more complete analysis of the issue is given inRomstad and Rørstad (1995). Here we will em-phasize the following:� An effect of split fertilizing is lower fertilizer

levels in ‘bad’ years, but also increased levels in‘good’ years. To create an environmentally in-teresting scheme, the total level of N must bereduced even for split fertilizing.

� A system combining reduced and split fertiliza-tion can be obtained by a scheme of season-specific tradeable N quotas. Except for sandysoils, the results obtained from such a schemeare not very different from those obtained us-ing a uniform N-tax giving the same reductionsin the overall fertilizer levels. The costs forfarmers are considerably lower (no taxes),while the administration costs are expected tobe higher.The difference between sandy soils and other

soils was as expected. The results for clay and siltwere still surprising to us since yields also varysubstantially between years in these cases. Interac-tions between hydrology and the biological sinkmechanisms explain much of the observed effects.Further, the results depend strongly on the preci-sion of the crop growth modelling. To exemplify:in our analyses, average yield is reduced by ap-proximately 2% due to the extra trafficking fol-lowing a second round N application. As we havealready seen, this is of a magnitude influencingleaching rather substantially. The positive effectof increased fertilizing precision obtained by splitfertilizing is thus more or less nullified by thisseemingly minor effect on the yield for bothclayey and silty soils.

6. Conclusion

The results obtained demonstrate the necessityof interdisciplinary cooperation and a formalizedintegration between economic and natural scien-tific models when designing policy measures toreduce agricultural pollution. The fact that pollu-tion consists of several components and is con-trolled by a plethora of mechanisms, warrants acomplex dynamic modelling thus ensuring thatthe mechanisms and pollutants involved are ade-quately represented. The inclusion of the mostrelevant pollution components also ensures anexplicit treatment of the potential conflicts be-tween aims. This is of importance when studyingpollution from agriculture, as the principles dis-cussed here are equally relevant for studies inother fields with high levels of complexity.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205204

Adequate understanding of the basis for farm-ers’ choices, both from an agronomic and socialscience point of view, is important. We have useda profit maximization rule in our analysis, believ-ing that it is an adequate basis for ‘comparingchanges’ in adaptation under different policy sce-narios. We admit that it does not offer a completedescription of agents’ behavior. Still, progress isobtained by undertaking a detailed description ofthe setting and constraints under which choicesare made. Farmers respond differently to politicalmeasures, depending on their production system,warranting a careful economic analysis of suchchoices at a reasonably high resolution.

The interdisciplinary discourse is necessary toidentify and nurture new/more relevant researchwithin each discipline. The division of science intorather narrow subfields has been necessary toincrease the depth of the analysis. As the onlystrategy, however, it prevents the understandingof inter-linkages and dynamics at higher systemlevels. Practical relevance may be lost. Interdisci-plinary research may serve us greatly by justmaking the disciplines relate better to higher re-spective lower level entities.

ECECMOD is a structure that can be devel-oped in different directions. First of all otherpollution problems like pesticide use can be inte-grated, a process we have started on. Such anenlargement has proven to be relatively easy toundertake without having to make major changesin the system. This is both an effect of the wayECECMOD is structured, its process orientation,and the level of resolution chosen.

Acknowledgements

The project is financed by the Research Councilof Norway. The authors would like to thankHalstein Lundeby, Per Kr. Rørstad and ArildVold for their participation in the development ofECECMOD. We would also like to credit theapproximate 40 colleagues at the AgriculturalUniversity of Norway, Center of Soil and Envi-ronmental Research, Norwegian Crop ResearchInstitute, Norwegian Institute of Land Inventoryand Statistics Norway providing us with data and

different background analysis. Finally, we thankReidun Aasheim for assistance in preparing thefigures and three reviewers for constructivecomments.

References

Antle, J.M., Capalbo, S.M., 1993. Integrating economic &physical models for analyzing environmental effects ofagricultural policy on nonpoint-source pollution. In: Rus-sell, C.S., Shogren, J.S. (Eds.), Theory, Modeling andExperience in the Management of Nonpoint-Source Pollu-tion. Kluwer, Boston, MA, pp. 155–178.

Botterweg, P.F., 1992. European Soil Erosion Model. Winterroutines—possible solutions. Economics & Ecology/RMPA working paper no 10. Agricultural University ofNorway, 12 pp. (unpublished).

Botterweg, P.F., 1998. Snowmelt and frozen soils in simulationmodels. In: Boardman, J., Favis-Mortlock, D. (Eds.),Modelling Soil Erosion by Water. NATO ASI Series,Series I-55. Springer-Verlag, Berlin, pp. 365–376.

Botterweg, P., Leek, R., Romstad, R., Vatn, A., 1988. TheEurosem–Gridsem modelling system for erosion analysesunder different natural and economic conditions. Ecol.Modelling 108, 115–129.

Christensen, B.T., Johnston, A.E., 1997. Soil organic matterand soil quality: lessons learned from long term fieldexperiments at Askov and Rothamstead. In: Gregonich,E.G., Carter, M.R. (Eds.), Soil Quality for Crop Produc-tion. Elsevier, Amsterdam, pp. 399–430.

Costanza, R., Wainger, L., Bockstael, N., 1995. Integratedecological economic systems modelling: theoretical issuesand practical applications. In: Milon, J.W., Shogren, J.F.(Eds.), Integrating Economic and Ecological Indicators.Practical Methods for Environmental Policy Analysis.Praeger, Westport, pp. 45–66.

Hausman, J.A. (Ed.), 1993. Contingent Valuation. A CriticalAssessment. North-Holland, Amsterdam.

Hodgson, G.M., 1988. Economics and Institutions. PolityPress, Cambridge.

Jansson, P.E., 1991. Simulation model for soil water and heatconditions. Description of the SOIL model. Report 165.Swedish University of Agricultural sciences, Department ofSoil Sciences, Uppsala.

Johnsson, H., Bergstrom, L., Jansson, P.E., Paustian, K.,1987. Simulated nitrogen dynamics and losses in a layeredagricultural soil. Agric. Ecosyst. Environ. 18, 333–356.

Leek, R., 1993. Using Remote Sensing for Terrestrial Moni-toring and Prediction of Sediment Yield to Rivers. Reportno 35, report series Hydrology. University of Norway.

Morgan, R.P.C., Quinton, J.N., Smith, R.E., Govers, G.,Poesen, J.W.A., Chisci, G., Torri, D., 1998. The EU-ROSEM model. In: Boardman, J., Favis-Mortlock, D.(Eds.), Modelling Soil Erosion by Water. NATO ASISeries, Series I-55. Springer-Verlag, Berlin, pp. 389–398.

A. Vatn et al. / Ecological Economics 30 (1999) 189–205 205

Nakajima, C., 1986. Subjective Equilibrium Theory of theFarm Household. Elsevier, Amsterdam.

O’Neill, R.V., DeAngelis, D.L., Waide, J.B., Allen, T.F.H.,1986. A Hierachical Concept of Ecosystems. PrincetonUniversity Press, Princeton, NJ.

Rastetter, E.B., King, A.W., Hornberger, G.M., Cosby, B.J.,O’Neill, R.V., Hobbie, J.E., 1992. Aggregating fine scaleecological knowledge to model coarse-scale attributes ofecosystems. Ecol. Appl. 2 (1), 55–70.

Romstad, E., Rørstad, P.K., 1995. Policy Measures to InduceSplit Fertilization in Grain Production, IØS DiscussionPaper cD-18/1995, Agricultural University of Norway(unpublished).

Simon, H.A., 1979. Rational decision making in businessorganization. Am. Econ. Rev. 69 (4), 439–513.

Singh, I., Squire, L., Strauss, J., 1986. Agricultural HouseholdModels. Extensions, Applications and Policy. John Hop-kins Press, Baltimore, MD.

Uhlen, G., 1989. Nutrient leaching and surface runoff in fieldlysimeters on a cultivated soil. Nutrient balances 1974–81.Norwegian J. Agric. Sci. 68 (3), 33–46.

Vatn, A., Bromley, D.W., 1994. Choices without prices with-out apologies. J. Environ. Econ. Manage. 26, 129–148.

Vatn, A., Bakken, L., Bleken,. M.A., Botterweg, P., Lundeby,H., Romstad, E., Rørstad. P.K., Vold, A., 1996. Policiesfor reduced nutrient losses and erosion from Norwegianagriculture. Integrating economics and ecology. NorwegianJ. Agric. Sci., Supplement no 23.

Vatn, A., Bakken, L., Bleken, M.A., Botterweg, P., Lundeby,H., Romstad, E., Rørstad, P.K., Vold, A., 1997. Regulat-ing nonpoint-source pollution from agriculture—an inte-grated modelling analysis. Eur. Rev. Agric. Econ. 24,207–229.

Vedeld, P., 1998. Farmers and Fertilizers. A Study of Adapta-tion and Response to Price Change among NorwegianFarmers. Ph.D.thesis. Department of Economics and So-cial Sciences, Agricultural University of Norway.

Vold, A., 1997. Development and evaluation of a mathemati-cal model for plant N-uptake and N-leaching from agricul-tural soils. Doctor scientiarium thesis 1997:28, AgriculturalUniversity of Norway.

Vold, A., Breland, T.A., Bakken, L.R., 1994. Prediction ofshort- and long-term effects of green manuring. In: Linden,B. (Ed.), The Use of Catch or Cover Crops to ReduceLeaching and Erosion. NJF-utredning/rapport nr. 99,Proc. NJF-seminar no. 245, pp. 181–192.

Vold, A., Bakken, L.R., Vatn, A., 1995. The use of a simu-lated barley cropping system as a bioassay for estimatingnet effects of catch crop and manure application on theN-availability to subsequent crops. Economics & Ecology/RMPA Working paper no 40. Agricultural University ofNorway, 23 pp. (unpublished).

Vold, A., Bakken, L.R., Uhlen. G., Vatn, A., 1999. Use ofdata from long term experiments to model plant nitrogenuptake. Nutr. Cycling Agroecosyst., in press.

..