An Indicator to Evaluate the Environmental Impact ofOlive Oil Waste Water’s Shedding on Cultivated Fields

Nicola Silvestri*, Gianni Fila2, Gianni Bellocchi3, Enrico Bonari4

1Department of Agronomy and Agro-Ecosystem Management Via San Michele degli Scalzi 2, 56124 Pisa, Italy

2Institute for Industrial Crops, Bologna, Italy3Research Associate at the Research Institute for Industrial Crops, Bologna, Italy

4Sant’Anna Advanced School for University Studies and Doctoral Research

Received: 8 October 2003. Accepted: 10 March 2006

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

AbstractSeveral climatic, soil, topographic, etc. factors need to be considered when evaluating the impact of human actionson the environment. Such variables may be related in a complex way to environmental impact, thus making its eva-luation difficult. Problems of this type emerge when evaluating the risks olive oil waste water pose to the environ-ment when shed on cultivated soils. This paper proposes a fuzzy expert system to calculate a modular indicator,ICARO, which allows an evaluation of the potential environmental impact of the application of olive oil waste wa-ter in a field. Five modules were formulated, one (“Waste water”) reflecting the nature of the waste water, two(“Groundwater”, “Surface water”) reflecting the risk for the most sensitive agro-environmental compartments(groundwater, surface water), one (“Crop”) reflecting possible consequences on the cropping system adopted, andone (“Soil”) reflecting the soil aptitude to receive waste waters. The input variables are therefore waste water amountand properties, site-specific conditions, and characteristics of the application considered. For each input variable, twofunctions describing membership to the fuzzy subsets Favorable (F) and Unfavorable (U) have been defined. Theexpert system calculates the value of each module according to both the degree of membership of the input va-riables to the subsets F and U, and a set of decision rules. The five modules can be considered individually or canbe aggregated (again according to level of membership to fuzzy subsets F and U and a set of decision rules) intothe synthetic indicator ICARO. Outcomes of a sensitivity analysis are presented. The system is flexible and can beused as a decision aid tool to authorize waste water’s shedding or subordinate the distribution on fields to accep-tance of some limitations (amount, timing, site, etc).

Keywords: environmental impact, expert system, fuzzy logic, ICARO, olive oil, waste water.

243

* Corresponding Author: Tel.: +39 050 599215; Fax: +39-050-540633. E-mail address: nsilve@agr.unipi.it

1. Introduction

Oil extraction from olive pulp is performed sep-arating oil from the rest simply by pressure (tra-ditional system) or, alternatively, by centrifuga-tion (more modern system) using either a three-phase or two-phase decanter system. The cen-trifugation two-phase decanter system is not asource of liquid waste (ordinarily said waste wa-ter) which, on the contrary, originates with bothother technologies, although in different quan-tities (about 400-600 l per ton of olives with thetraditional system, about 800-1000 l per ton ofolives with the centrifugation system with three-phase decanter system, according to Di Gio-

vacchino, 2001). About 2x106 liters of waste wa-ter is produced yearly in Italy, and similar situ-ations can be found in the other countries whereolive oil is produced (Spain, Greece, northernAfrica and Middle-East countries) (Rana et al.,2003).

The composition of the waste water emul-sion could change very much in relation to cli-matic and soil conditions, agronomic manage-ment, cultivar choice, oil extraction method, etc.Nonetheless, olive oil waste waters are alwayspotentially polluting and constitute a majorsource of organic pollution in the Mediter-ranean region, due to the high organic carbon

content, substantial portion of which is presentas aromatic compounds with known bio-toxico-logical activity (phenols, polyphenols and tan-nins), and the considerable values of acidity andsalinity (Potenz et al., 1985; Zouari and Ellouz,1996; Ciafardini et al., 1998; De Simone et al.,1998). Environmental problems take place, aslarge quantities of waste waters are producedduring a short period of time.

Several chemical and biological technologiesare increasingly used in order to transform oliveoil waste water into more biodegradableresidues, some with success, but they are insuf-ficient to permit discharges of effluents into theenvironment, and not cost effective because ofthe seasonal character of production and thesmall size of extraction plants (Borja et al., 1992;Matzavinos et al., 1996; Benitez et al., 1997;Poulios et al., 1999; Coombs and Hall, 2000). Al-ternatively, shedding oil waste water on culti-vated fields might be a suitable solution, if per-formed under controlled regimes (Bonari andSilvestri, 2003). However, waste waters sheddingposes risks to environmental compartments likewater and soil and the optimization of this prac-tice is a serious need (Tomati, 2001).

The Italian legislation is based on the law n.574/96, pertaining to norms concerning agricul-tural utilization of waters from olive oil indus-try. In particular, it establishes: 1) the maximumamount of oil waste waters that is possible toshed on fields, 2) the categories of suitable landsfor the distribution, and 3) the norms for waterstorage. In general, the risk of phytotoxicity tothe crop and the environmental cost are themain factors considered when proposing wastewater’s shedding on cultivated fields. In partic-ular, the definition of the site-specific conditionsfor distribution (amount, modes, timing, and themost suitable equipments) according to agro-nomic, pedological, orographic and climaticcharacteristics is required, thus avoiding the on-set of either undesired (e.g., limited drainage,smell emissions, etc.) or quite dangerous (e.g.,groundwater pollution, superficial water bodiescontamination, etc.) phenomena. Both farmersand public decision-makers would like to beable to take the potential environmental impactof the waste water’s shedding into account.

Giving a solid judgment about the impact onthe environment associated to waste water’sshedding requires to simultaneously consider

several variables. Balancing different aspects in-volved in the discourse is a complex process.Hence, combining several variables into one ag-gregated measure is desirable in order to havea comprehensive assessment of the environ-mental impact. This could be achieved by ag-gregating variables by summation, multiplica-tion, or a combination of both. These ap-proaches pose mathematical and conceptualproblems (Keeney and Raiffa, 1993), since therelevant variables differ in their nature, dimen-sions and range of possible values. Consideringthe inadequacy of such methods, a different ap-proach to evaluate the environmental impact canrely in setting up a fuzzy expert system (Hall andKandel, 1991) using decision rules (fuzzification).This technique is robust on uncertain and im-precise data such as subjective judgments(Duboy and Prade, 1980; Bardossy et al., 1985;Goyal and Deshpande, 2001), and allows the ag-gregation of dissimilar variables in a consistentand reproducible way (Bouchon-Meunier, 1993).

Within the last decades methods of this typehave been proposed for estimating the environ-mental impact of xenobiotic substances. Somemethods assess potential xenobiotic environ-mental impact, as perceived by the “experts”that created them. We are not aware methodsof this type exist for evaluating the impact onthe environment of waste waters from olive oilindustry when they are shed on cultivated fields.This paper describes an expert system which isat the core of an agro-ecological indicator,ICARO, reflecting an expert perception of theimpact of waste water’s shedding on the envi-ronment.

2. Methods

2.1. Structure of the indicator and fuzzy expertreasoningIt is assumed here that a comprehensive as-sessment about the impact of waste water’sshedding on the environment should consider:i) the amount of waste water to be shed in afield and its composition, ii) the proneness towhich the waste water can leave the field onwhich it is applied to reach groundwater andsurface water, iii) the possible toxicity to crops,and iv) the effect on soil quality generated bythe waste water interaction with soil properties.

Silvestri N., Fila G., Bellocchi G., Bonari E.

244

Such assumptions do conform to a generaltenet (e.g., Bacci et al., 1989), by which proce-dures aimed at assessing the impact of a givenaction on the environment imply the identifica-tion of three key components, i.e. source, criti-cal path, and target. The source is the point ornon-point origin of the pollutant. The criticalpath is the array of mechanisms allowing thetransfer and spread of pollutants to the target;criticality lies on the efficiency this process isplayed. The target is the last or intermediate en-vironmental compartment receiving the impact.

In the problem of interest the source is giv-en by the waste water to be shed in a field. Themodes by which this potential pollutant spreadsout in the environment are associated to deepdrainage, surface runoff, crop absorption, and soilinteractions. Such issues may include factors of en-ergy, resistance, protection, and vulnerability. En-ergy factors foster impact transmission to target,resistance factors pose possible opposition to sucha transmission, protection factors attenuate the ef-fect of the impact on the target, and vulnerabilityfactors generate potential conditions for majordanger. The critical ways here recognized do in-deed identify corresponding environmental targets:groundwater, surface water, crop, soil.

According to such premises, we defined fiveindicator modules, named “Waste water” (“Re-fluo” in Italian), “Groundwater” (“Falda” inItalian), “Surface water” (“Corpi idrici” in Ital-ian), “Crop” (“Coltura” in Italian), “Soil” (“Suo-lo” in Italian). The value of each module de-pends on one or more input variables and a setof 2n decision rules, n being the number of in-puts in a module. For each module, a dimen-sionless value between 0 (no risk of environ-mental impact) and 1 (maximum risk of envi-ronmental impact) is calculated. The procedure,based on the multi-valued fuzzy set theory in-troduced by Zadeh (1965), follows the so-calledSugeno or Takagi-Sugeno-Kang method offuzzy inference (Sugeno, 1985).

It consists of four functional components: 1)a fuzzifier that transforms numerical inputs in-to fuzzy sets (fuzzification), 2) a set of controlrules governing the relationships between inputsand outputs, 3) an inference engine that per-forms the fuzzy reasoning based on the controlrules, and 4) a defuzzifier that transforms thefuzzy output into real numerical numbers (de-fuzzification).

This approach, computationally efficient andwell-suited for mathematical analysis, has beenapplied to a wide variety of problems. Relevantapplications were, for instance, the developmentof novel approaches to support decisions re-garding sustainable development (Cornelissenet al., 2001), the estimation of parameters inevapotraspiration models (Odhiambo et al.,2001), and the evaluation of radiation modelperformance (Bellocchi et al., 2002). In the con-text of environmental impact, a fuzzy-based in-dicator was designed by van der Werf and Zim-mer (1998) and enhanced by Roussel et al.(2000), reflecting an expert perception of thepotential environmental impact of the applica-tion of a pesticide in a field crop. The approachfollowed here is largely inspired to such contri-butions, and the reader may refer to them fordetails on computational issues of the fuzzy in-ference process. We only report hereafter somebasics of the fuzzy procedure we have followed.

Fuzzification was achieved by categorizingthe input data space for each input variable in-to three membership classes (or subsets): Fa-vorable (F), Unfavorable (U), and partial (orfuzzy) membership; the output data space wasattributed a decision rule in the form of an ex-pert weight. The control rules for estimatingmodule values were based on rational relation-ships between inputs and outputs, expressed inlinguistic terms by if-then statements. For ex-ample, when two input variables are aggregat-ed, four rules are formalized as follows:

premise conclusionif x1 is F and x2 is F then y1 is B1if x1 is F and x2 is U then y2 is B2if x1 is U and x2 is F then y3 is B3if x1 is U and x2 is U then y4 is B4

where xi is an input variable, yi is an outputvariable and Bi is a decision rule (or expertweight). A “minimum operator” was used toevaluate the conjunction and by taking the min-imum of the quantified fuzzy sets, as obtainedfrom complementary S-shaped distributioncurves. Figure 1 shows S-shaped curves for anhypothetical input variables ranging from 1.0(F) to 2.5 (U). The S-function is flat at valuesof the input variables beyond F and U.

The output fuzzy sets for all the rules wereaggregated into a single fuzzy set. The aggregateoutput fuzzy set encompasses a range of output

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

245

values, and was defuzzified in order to resolvea single crisp output value from the set. The cen-troid method was selected to obtain the repre-sentative non-fuzzy value for the output, ascommonly adopted in the Sugeno-type systems.

In this work several input variables were ag-gregated into modules, and the modules in thefinal indicator, using fuzzy-based logic rules. Theexpert reasoning runs as follows: if all inputvariables are F, the value of the module is 0 (norisk for the environment); if all indices are U,the value of the module is 1 (maximum risk forthe environment). In setting up the decisionrules for the other combinations (some of theinputs are F while some others U) we had todecide on the relative importance of each input,by attributing an expert weight to each of them.Computations are done by using weights whichare normalized on a scale from 0 to 1. A nor-malized weight (bj) is given by dividing its val-ue by the sum of the weights of all variables tobe aggregated. A decision rule is indeed the ex-pert weight (Bi) assigned to a given conjunction(… and …) of inputs. The terms Bi and bi arerelated as follows:

where i is the ith rule, jF is the jth input variablewhich enters the decision rule as F, nF is thenumber of input variables which enter the de-cision rule as F, bj

F is the normalized weight forone of the input variables which enter the de-cision rule as F. As an example, let’s take threeinput variables and two hypothetical rules, e.g.:

if x1 (b1 = 0.40) is F and x2 then y1 is B1 = (b2 = 0.53) is F and x3 1-(0.40+0.53) (b3 = 0.07) is U = 0.07

if x1 (b1 = 0.40) is U and then y2 is B2 = x2 (b2 = 0.53) is U and x3 1-0.07 = 0.93(b3 = 0.07) is F

When an input variable only enters the de-cision rule as U, the resulting value of Bi is therelative incidence of such a variable into its ownmodule. In the example above Bi=0.07 is the rel-ative incidence of x3.

Both limits to the fuzzy subsets F and U, andweights were, in general, established based onour expert judgment, working on data sets frommulti-year researches carried out in different lo-cations over the Region Tuscany (Bonari andCeccarini, 1991, 1993, 2001, Bonari et al., 2001).

2.2. The module “Waste water”

The indicator module “Waste water” reflectsquantity and quality of waste waters by means ofthe input variables listed in Table 1. The U limitsselected for the rate of application are those setby the Italian law 574/96, by which amounts high-er than 80 and 50 m3 ha-1, respectively for me-chanical pressure and centrifugation methods, areforbidden. Other limits are extracted from Ital-ian laws as well (legislative decree 152/99 forfloating points; deliberation of the Committee ofMinisters 04/02/1977 for SAR). Transition inter-val for pH tends to cover the range of values ob-served with olive oil waste waters (4.8 to 5.4 ac-cording to Fiorentini et al. (1985); 4.7 to 5.8 ac-cording to Pacifico (1989); other reference dataare values fluctuating between 4.3 and 5.2, pub-lished in “Reflui e sottoprodotti della trasfor-mazione delle olive”, available through the website of ARSIA-Toscana at http://www.arsia.toscana.it). Values limiting the transition intervalfor electric conductivity are derived by the per-missible limits for classes of irrigation water giv-en in James et al. (1982). In our experience, thequantity inputs assume more relevant impor-tance than quality inputs in the environmentalimpact assessment, thus fairly large weights wereattributed to rate of application and time fromlast shedding (10 and 6, respectively).

Chemical composition was also consideredin the short preliminary proposal of ICARO(Silvestri et al., 2003), including in the structure

Silvestri N., Fila G., Bellocchi G., Bonari E.

246

0.0

0.3

0.5

0.8

1.0

1.0 1.4 1.8 2.1 2.5

input variable

Degree of membership

U

F

Fig. 1. Graphical presentation of S-shaped curves for an hy-pothetical input variable.

∑

=

−=

F

F

n

j

Fji bB

1

1

of the module “Waste water” the concentrationof heavy metals such as manganese, zinc and cop-per present in the waste water. Following that,more in-depth studies led us through the deci-sion to simplify this module by no longer takingmetals into account, based on scientific, technicaland legislative assessments. Three main issues un-derpin the core scientific ground. First, the threemetals mentioned, although more frequently ob-served than others, correspond to a restrictedsub-set of a larger assortment of metals at timesdetected in olive oil waste water, possibly in-cluding lead, nickel, cobalt, etc. Second, the com-bined effect of environmental factors and multi-ple sources of contamination (including cropmanagement) is to generate complex and chang-ing patterns in the allocation of metal impuritiesto waste waters; as a result, a strong seasonalvariability of metal concentrations has been evi-denced in waste waters of apparently similar ori-gin (Di Giovacchino, 2001). Third, different com-binations of metals could have different implica-tions on various environmental compartments,therefore complex coefficient schemes should beconsidered where each combination of metalshas its own weight. However, determining a setof coefficients that assess the joint effect of twoor more metals on one or more environmental

targets is not a trivial task, and never agreeablysolved (e.g., Kuncheva et al., 2000). On a techni-cal level, it must be considered that concentra-tion values for metals in olive oil waste water arerarely available, as metal determination is notpart of a standard laboratory analysis, and is of-ten not done owing budgetary constraints (thiswould generate missing inputs to the indicator).Moreover, the law 574/96 does not say anythingabout metals, and typical values of metal con-centration in olive oil waste waters are seldomcompatible with the tolerance limits portrayed bynon-specific reference legislation (e.g., legislativedecree 152/99, defining general policy for waterprotection from pollution and processing of ur-ban waste waters; deliberation of the Committeeof Ministers 27/07/1984, pertaining to strategiesfor waste products disposal).

2.3. The module “Groundwater”

The indicator module “Groundwater” reflects thepotential of waste waters to reach groundwaterthrough leaching and to affect its potential useas a source of drinking water for humans. Thevalue of “Groundwater” depends on three inputvariables: 1) soil saturation rate (energy), 2) sat-uration conductivity (resistance), 3) the fielddistance from wells (vulnerability). Table 2 re-

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

247

Table 1. The indicator module “Waste water”: input variables, favorable (F) and unfavorable limits (U), expert weights. Fordetails, see the text.

Module “Waste water”

Option Input variable Limits Expert weight

F Uquantitycentrifugation method rate of application (m3 ha-1) 16 80 10mechanical pressure method rate of application (m3 ha-1) 10 50 10

- time past from the last shedding (years) 5 2 6

quality- floating impurities (mg l-1) 80 200 1- pH (-) 5.5 4.5 1- SAR (-) 10 15 2- electric conductivity (_S cm-1) 750 2000 2

Table 2. The indicator module “Groundwater”: input variables, favorable (F) and unfavorable limits (U), expert weights.For details, see the text.

Module “Groundwater”

Type of factor Input variable Limits Expert weight

F Uenergy soil saturation rate (d-1) 0.25 0.75 6resistance saturated hydraulic conductivity (mm h-1) 0.36 36 8vulnerability field distance from wells (m) 200 100 1

ports the module features. Both membershiplimits and weights attributed to the input vari-ables do conform to our expert judgment. Fa-vorable and unfavorable boundaries for satura-tion conductivity do conform to the categoriza-tion proposed by the Soil Survey Staff and Nat-ural Resource Conservation Service (1993). Thesaturated hydraulic conductivity, measuring thesoil permeability, is closely related to potentialgroundwater contamination, and therefore thisinput was assigned the highest weight (i.e., 8). Soilsaturation rate, expressed as a ratio betweenmonthly rainfall and soil water holding, is alsoquite important (expert weight equal to 6).

2.4. The module “Surface water”

The indicator module “Surface water” reflects thepotentials of waste waters to reach surface waterthrough runoff or drift and to harm aquatic or-ganisms. Its value depends on six input variables(Table 3): 1) the rainfall erosivity (energy), 2) asoil morphologic factor (energy), 3) the soil erodi-bility (resistance), 4) a cover factor (protection),5) conservation management (protection), 6) fielddistance from water bodies (vulnerability).

With the only exception of field distance

from water bodies, other inputs are based onthe USLE (Universal Soil Loss Equation) ap-proach (Wischmeier and Smith, 1978). Both lim-its and weights attributed to either input vari-ables reflect the authors’ experience. In gener-al, crop management (ground cover, tillage, etc.)is assumed as the main factor governing the riskto surface water (expert weights equal to 8 and6 to cover factor and conservation management,respectively). Both cover factor and conserva-tion management are coefficients on a rangefrom 0 (best) to 1 (worst). Cover factor accountsfor land use (presence/absence of ground cov-er, type of crop/crop residues), as in Table 4.

Conservation management accounts for pos-sible farmer operations (i.e., tillage, buffersstrips) which help protect surface water fromwaste waters (Table 5). When possible, tillage

Silvestri N., Fila G., Bellocchi G., Bonari E.

248

Table 4. Cover factors of the module “Surface water”.

Land use Cover factor (-)

tilled fallow 1.00untilled fallow 0.50fallow with crop residues 0.25tilled with weeds 0.75cereals 0.40mechanically weeded crops 0.50rotated grasses 0.10specialized olive trees 0.60mix-cropped olive trees 0.50straight row vineyards 0.75contoured vineyards 0.60permanent grasses 0.01

Table 5. Conservation management coefficients of the mo-dule “Surface water”.

Management Slope Coefficient

Buffer strips Land treatment (%) (-)

- 0 1.001-2 0.603-5 0.506-8 0.50

absence contoured 9-12 0.6013-16 0.7017-20 0.8021-25 0.90

straight row - 1.00

- 0 0.601-2 0.453-5 0.386-8 0.38

presence contoured 9-12 0.4513-16 0.5317-20 0.6021-25 0.68

straight row - 0.75

Table 3. The indicator module “Surface water”: input variables, favorable (F) and unfavorable limits (U), expert weights.For details, see the text.

Module “Surface water”

Type of factor Input variable Limits Expert weight

F Uenergy rainfall erosivity (MJ mm ha-1 h-1 yr-1) 1000 6000 4energy soil morphologic factor (-) 0.5 1.5 4resistance soil erodibility (t h MJ-1 mm-1) 0.03 0.08 4protection cover factor (-) 0.2 1.0 8vulnerability conservation management (-) 0 1 6vulnerability field distance from water bodies (m) 300 150 4

should follow the contour or align transverse tothe direction of the slope, because this will re-duce the power of water flowing down the hillslope. Combining tillage on the contour with al-ternating strips of grass provides even furthersoil protection (e.g., Dillaha et al., 1989; Mus-cutt et al., 1993).

2.5. The module “Crop”

The indicator module “Crop” reflects the po-tentials of waste waters to reach cultivated cropsand to have negative effects on their growth anddevelopment. Its value depends on one inputvariable only, i.e. vulnerability factor. This ac-counts for a number of different factors (appli-cation time in relation to seeding date, crop,stage of development), summarized in a coeffi-cient ranging between 0 (the highest vulnera-bility) and 1 (the lowest vulnerability), as inTable 6. Suggested data derive from either au-thor’s experience or interpretation of resultsfrom Italian researches.

2.6. The module “Soil”

The indicator module “Soil” reflects site’s suit-ability to take delivery of waste waters. Its val-ue depends on seven input variables (Table 7):1) macroporosity (resistance), 2) salinity (resis-tance), 3) pH (resistance), 4) exchangeable sodi-um percentage (resistance), 5) mean tempera-ture in the month of shedding (protection), 6)field distance from drinking water sources (vul-nerability), and 7) field distance from houses(vulnerability).

Soils with high values of macroporosity aredesirable because better aerated. Higher valuesof both salinity and exchangeable sodium per-centage imply ever-increasing risks to the envi-ronment, and are assumed to be the most in-

fluential module inputs (expert weights equal to6 for both). Basic soils have to be preferred, tocompensate for quite high acidity conveyed bywaste waters. It is also desirable to shed wastewaters when air temperature is rather high, as thehigher is the temperature the faster is the wastewater degradation. Unsuitable soils are those lo-cated near to either drinking water sources orhouses. The U limits for both these inputs are setby Italian law (law 574/96, article 5).

2.7. Aggregation of the modules

The five modules described above can be usedto compare alternative waste waters applicationoptions with respect to their potential environ-mental impact. The available options might be

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

249

Table 6. Vulnerability factor of the module “Crop”.

Land use Vulnerability factor (-)

Fallow period

fallow, 2 months before sowing 1.00fallow, 1.5 months before sowing 0.75fallow, 1 months before sowing 0.50fallow, 0.5 months before sowing 0.25fallow, < 0.5 months before sowing 0.00Growing seasonwinter cereals, tillering 0.70rapeseed, rosette 0.70maize, 2-3 leaves 0.60sunflower, 2-3 leaves 0.50sugarbeet 0.10potato, full canopy 0.40leguminoseae grass, regrowth 0.30gramineae grass, regrowth 0.30pasture, re-growth 0.30uncovered set-aside 0.40olive, dormancy 0.90wine grape, dormancy 0.80fruit trees, dormancy 0.60

Table 7. The indicator module “Soil”: input variables, favorable (F) and unfavorable limits (U), expert weights. For details,see the text.

Module “Soil”

Type of factor Input variable Limits Expert weightF U

resistance macroposority (%) 35 15 1resistance salinity (EC, mS cm-1) 2 4 6resistance pH (-) 7.3 5.9 2resistance exchangeable sodium percentage (ESP, %) 5 15 6protection mean temperature in the month of shedding (°C) 25 5 1vulnerability field distance from drinking water sources (m) 700 300 2vulnerability field distance from houses (m) 500 200 2

ranked, for instance, by means of a multicriteriaanalysis technique, using the modules as evalua-tion criteria (Girardin et al., 2000). An alterna-tive approach would be to aggregate the fivemodules (second-level aggregation) in some wayinto an overall indicator (i.e., ICARO), reflectinga global judgment about the impact of waste wa-ters on the environment, again on a 0 to 1 scale.This can be done by summation, multiplicationor a combination of both, according to aggrega-tion schemes. For example: ICARO = (Waste wa-ters + Soil) · (Groundwater + Surface water +Crop), or ICARO = (Waste waters + Soil) · max(Groundwater, Surface water, Crop), etc.

We advocate here another mode of aggre-gation of the five modules, which uses decisionrules, similar to that described above for the ag-gregation of input variables into the modules.The value of the indicator ICARO depends onthe modules “Waste waters”, “Groundwater”,“Surface water”, “Crop” and “Soil”, according

to a set of 25 = 32 decision rules (not shown),derived from expert weights assigned (see lat-er). The definition of the limits of the transitioninterval is the same for the five input modules:we assigned complete membership to F if thevalue of the module is 0 (i.e., optimal waste wa-ter amount/quality, no risk to either groundwa-ter, surface water, or crop, optimal soil proper-ties and position) and complete membership toU if the value of the module is 1.

In setting up the other decision rules we hadto establish the relative importance of eachmodule. In most indicators of environmental im-pact surface water is given much emphasis (e.g.,van der Werf, 1996). Considering this, we de-cided to give the largest weight to the module“Surface water” (i.e., 30) in the aggregation ofthe modules. Based on our experience, decreas-ing importance was given to the other modules.As the amount of waste waters applied to soilis directly related to the impact on the envi-

Silvestri N., Fila G., Bellocchi G., Bonari E.

250

Table 8. Relative incidence of each input variable on the value of the indicator, based on normalized weights.

Input variable Relative incidence on ICARO

Module “Waste water”rate of application 0.45 x 0.25 = 0.1136time past from the last shedding 0.27 x 0.25 = 0.0682floating impurities 0.05 x 0.25 = 0.0114pH 0.05 x 0.25 = 0.0114SAR 0.09 x 0.25 = 0.0227electric conductivity 0.09 x 0.25 = 0.0227

Module “Groundwater”soil saturation rate 0.40 x 0.15 = 0.0600saturation conductivity 0.53 x 0.15 = 0.0800field distance from wells 0.07 x 0.15 = 0.0100

Module “Surface water”rainfall erosivity 0.13 x 0.30 = 0.0333soil morphologic factor 0.13 x 0.30 = 0.0333soil erodibility 0.13 x 0.30 = 0.0333cover factor 0.27 x 0.30 = 0.0667field distance from water bodies 0.13 x 0.30 = 0.0333conservation management 0.20 x 0.30 = 0.0500

Module “Crop”vulnerability factor 1.00 x 0.10 = 0.1000

Module “Soil”macroporosity 0.05 x 0.20 = 0.0100salinity 0.30 x 0.20 = 0.0600pH 0.10 x 0.20 = 0.0200exchangeable sodium percentage 0.30 x 0.20 = 0.0600mean temperature in the month of shedding 0.05 x 0.20 = 0.0100field distance from drinking water sources 0.10 x 0.20 = 0.0200field distance from houses 0.10 x 0.20 = 0.0200

ronment, the module “Waste water” was con-sidered of relatively high importance (i.e., ex-pert weight equal to 25). Then, soil proper-ties/position (module “Soil”) have an effectiverole in influencing the mechanisms allowing thetransfer and spread of waste waters to the sen-sitive targets (i.e., expert weight equal to 20).Groundwater is taken into higher considerationthan crop in the assessment of environmentalimpact, thus larger weight was assigned to themodule “Groundwater” (i.e., 15) than to themodule “Crop” (i.e., 10).

The relative incidence of each input variableon the indicator, based on the weights only, canbe deduced by combining the normalizedweights of the input variables into their ownmodule with the ones of the modules into theindicator (Table 8). In practice, the effective in-cidence of an input variable on the final scoreis dependent on either the value assumed by the

variable itself, the values of the other inputs, andF and U limits attributed.

2.8. Sensitivity analysis to input variables

In order to illustrate the functioning of the sys-tem, we present the sensitivity showed by theindicator to variation of the four most influen-tial input variables, i.e., rate of application, sat-uration conductivity, time past from the lastshedding, cover factor (Table 8). Each inputvariable was varied over its transition interval,while the others were kept fixed either at theextremes of the transition interval, i.e. at F (Fig-ure 2, top) and U (Figure 2, bottom), or at themedian value (Figure 2, middle). The sensitivi-ty analysis reflects the functioning of the systemand provides some indication about the impor-tance of each input variable on the value ofICARO. However, one should be aware that theeffect of the variation of a particular input overits transition interval on the value of ICARO isdependent on the value of the other inputs.Therefore, results presented here should not beconsidered other than illustrations of the func-tioning of the system.

The four selected variables are not very in-fluential when all other inputs are F (Figure 2,top) or U (Figure 2, bottom). In both these cas-es ICARO varies within a narrow range, froma maximum range of about 0.103 for rate of ap-plication (from 0 to ~0.103 when the others areF; from ~0.897 to 1 when the others are U), toa minimum of about 0.037 for time past fromthe last shedding (from 0 to ~0.037 when theothers are F; from ~0.963 to 1 when the othersare U). This means that ICARO is mostly in-fluenced by the other inputs (taken together)under these conditions. More important is theinfluence of the selected variables on ICAROwhen all other inputs are kept at the middle val-ue of their transition interval (Figure 2, middle).Under this condition, ICARO fluctuates be-tween ~0.432 and ~0.568 over the range of rateof application, and between ~0.461 and ~0.539over the range of time past from the last shed-ding.

2.9. The ICARO Software

A MS Visual Basic program implements the ex-pert system illustrated above. The inputs are en-tered by a friendly interface (in Italian only inthe current version), structured in seven tabs:

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

251

0.000

0.022

0.044

0.066

0.088

0.110

0.00 0.20 0.40 0.60 0.80 1.00

transition interval

ICARO

other inputs are kept at F

0.400

0.450

0.500

0.550

0.600

0.00 0.20 0.40 0.60 0.80 1.00

transition interval

ICARO

other inputs are kept at the

median value of their

0.880

0.910

0.940

0.970

1.000

0.00 0.20 0.40 0.60 0.80 1.00

transition interval

ICARO

rate of application saturation conductivity

time past from the last shedding cover factor

other inputs are kept at U

�

�

�

�

�

�

o�t�h�e�r� �i�n�p�u�t�s� �a�r�e� �k�e�p�t� �a�t� �F�

��

��

��

��

��

o�t�h�e�r� �i�n�p�u�t�s� �a�r�e� �k�e�p�t� �a�t� �U

five tabs are relative to inputs and compute themodules, one tab is designed to compute the in-dicator, one tab is compiled with general dataand shows the conditions imposed by the Ital-ian law. Input data can be saved in ASCII files.A sample of files is included in the installationpackage. Utilities are also included allowingusers to easily process a large amount of data(batch analysis), and explore the relative inci-dence of the input variables on the indicatorscore (sensitivity analysis).

A database of weather inputs which are rel-evant for the calculations is incorporated in thesoftware, but only limited to locations of theprovince of Pisa in the current version.

The program is fully documented in theuser’s manual (Italian version only available inthe current version), which gives detailed de-scription of both the techniques being imple-mented and the underpinning science. The man-ual is provided with the software package andis available both on-line from the interface, andas a standalone help.

The software ICARO may be supplied to in-terested users on request. Use for non-com-mercial purposes is only allowed. The authorsrequire that in any publication arising from usefor research purposes the source of the programshould be properly acknowledged. The authorsalso appreciate to receive a copy of the publica-tion. A dedicated web site is under development(http://www.agr.unipi.it/daga/icaro/ICARO.htm).The help file is currently available for download-ing (http://www.sipeaa.it/tools/icaro/icaro.htm).

Some software features follow, illustrating howthe program deals with computational issues as-sociated to module and indicator calculations.

2.10. Module “Waste water”

The user may select the proper option from twoalternative oil extraction methods: traditional(mechanical pressure), centrifugation. This de-termines the U limit for the rate of application(see Table 1).

2.11. Module “Groundwater”

The user ma y want to compute soil satura-tion rate (Vs, d-1), based on two alternativeequations:– field capacity (FC): Vs=(p/n)·FC·H– bulk density (BD): Vs=(p/n)·[(2.65-

BD)/2.65)]·H

where p is rainfall in the month when waste wa-ters are shed in the field (mm month-1), n is thenumber of rainy day in that month (days month-

1), H is tillage depth (mm).If not available, values for FC (mm mm-1)

and BD (t m-3) can be estimated from texturevia pedotransfer functions (Saxton et al., 1986).A pedotransfer function by Campbell and Sh-iozawa (1992) is also implemented to estimatethe saturated hydraulic conductivity (Ks, mm h-

1) from texture.

2.12. Module “Surface water”

Energy factors, i.e. rainfall erosivity (R, MJ mmha-1 h-1 yr-1) and morphologic factor (SL), canbe calculated according to simplified forms ofthe USLE approach (Wischmeier and Smith,1978; Arnoldus, 1980):

R = [4.17·_(p2/P)-152]·17.02SL = (a/22.13)b·(0.065+0.045·i+0.0065·i2)

where p is rainfall in the month when waste wa-ters are shed in the field (mm month-1), P is av-erage yearly rainfall (mm year-1), a is the slopelength (m), i is the slope (%), and b is a dimen-sionless factor depending on the slope (0.5 ifi>5%, 0.4 if 3%<i≤5%, 0.3 if 1<i≤3%, 0.2 if i≤1%).

The computation of soil erodibility (K, t hMJ-1 mm-1) is also based on USLE (Wischmeierand Smith, 1978):

K = [2.1·TF1.14·10-4·(12-OM)+3.25·(SF-2)+2.5·(PF-3)]·0.1317/100

where TF is texture factor – TF = (%silt+%veryfine sand)·(100-%argilla) –, OM is soil organicmatter (%), SF is structure factor (ranging from1 for very fine granular structure to 4 for poly-edric structure), PF is permeability factor (rang-ing from 1 for Ks>127 to 6 for Ks<1).

Both cover factor and conservation manage-ment cannot be entered directly. They are se-lected from a list of options, as in Table 4 andTable 5, respectively.

2.13. Module “Crop”

The value for vulnerability factor cannot be en-tered directly. It can only be selected from a listof options (Table 6).

2.14. Module “Soil”

The mean air temperature for the month whenwaste waters are applied in a field can be en-

Silvestri N., Fila G., Bellocchi G., Bonari E.

252

tered directly or selected from the database in-cluding historical weather data for the locationof interest (the database is limited to theprovince of Pisa only in the current version).The macroporosity (MP, %) can be either en-tered in the relative domain or estimated, basedon its relation to field capacity (FC, mm mm-1)and BD (t m-3):

MP = [(2.65-BD)/2.65-CC]·100

As already mentioned above, both FC andBD may be estimated by texture data.

2.15. Indicator

The indicator can either be computed in itscomplete form by aggregating five modules, orin a partial form by aggregating 2, 3, or 4 mod-ules. Although partial, the last option is allowedto permit a conclusion on the environmental im-pact associated to waste water’s shedding whensome inputs are missing.

3. Discussion

In the design of a system to assess waste watersenvironmental impact two major questions haveto be answered: a) which input variables shouldbe taken into account, and b) how should theinput variables be aggregated. The method pre-sented in this paper proposes an answer to bothquestions, but its relevance lies primarily in theanswer it provides to the second question. Com-pared to other methods, this approach containstwo key elements: the use of a fuzzy set and theuse of decision rules. The use of a fuzzy set pro-vides a well-designed solution to the problem ofdeciding the cut-off values for input variables:e.g., the limit between F response, U response,and transition response. The use of decisionrules provides a “rational” aggregation of inputvariables in the related module: e.g., soil satu-ration rate, saturation conductivity and field dis-tance from wells in the module “Groundwater”.The combination of these two concepts (limitsin the response, mode of aggregation) in sets offuzzy rules is attractive because, although thecombinations of values of input variables arepotentially infinite, a single set of fuzzy rulesconnects them all.

In this application, the system is based on acompromise between operational suitability

(the assessment of the environmental impact)and flexibility (hierarchization of objectives andaggregation of preferences). It requires extend-ed corroboration, considering that the objectiveof an expert system is the simulation of a hu-man expert. The expert system is corroboratedif it displays, under a variety of conditions, thesame responses that a human expert would dis-play. Experts are therefore invited to commenton the set-up and results of this system. If thereis disagreement between expert perception ofenvironmental impact and the inputs of the sys-tem, the cause of this divergence will be exam-ined in view of: a) choice of input variables; b)choice of the limits of the transition interval; c)formulation of the decision rules; and d) for-mulation of the mode of aggregation of themodules. All of these points may be modifiedaccording to expert consensus, after an exten-sive testing of the methodology. The softwareICARO may help to test the indicator on a va-riety of conditions. Allowing the creation of re-usable modules and indicators (ASCII files),ICARO may serve as a convenient means to sup-port collaborative work among large, distributednetwork of either scientists or end-users involvedin the assessment of the environmental impactassociated to waste water’s shedding.

4. Conclusions

We propose a fuzzy expert system reflecting ourexpert judgment of potential waste water envi-ronmental impact. Providing usable values of in-put variables, the method allowed to build anaggregated indicator with a modular structure.The system takes into account three types of in-put variables: waste water properties, site-spe-cific conditions, waste water toxicity to crops.The resulting output can be used as a supportto rank or choose among alternative waste wa-ter application options.

Although a wide expert examination is stillrequired for a general consensus about weightsand limits applied to both input variables andaggregated modules, the fuzzy sets suggestedhere may represent a pragmatic approach to-wards a satisfactory solution.

This approach to the evaluation of potentialwaste water environmental impact is useful fora number of reasons:

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

253

– it allows interested parties to express math-ematically individual or collective values andpreferences (uncertainty factors);

– it highlights the degree of waste water’sshedding acceptance/rejection associatedwith each information source (e.g., waste wa-ter amount and composition, site-specificconditions, phytotoxicity);

– it allows comparing alternative options;– it facilitates structuring of various compo-

nents of the evaluation process;– it reduces several sources and levels of in-

formation into a single value;– it allows examination of operational equiva-

lence between several input variables andmodules.The modular structure also presents advan-

tages. In the first place, interested parties haveaccess to both a synthetic indicator reflectingoverall judgment and to each of the modules.This means a transparency of each step, and acontrol opportunity exists for anybody con-cerned by the process itself. Secondly, the modeof aggregation of modules can be changed andnew modules can be eventually added. The mul-ti-value nature of the issue we are dealing withis explicitly stated, the rules are easy to read,and the numerical scores used for their conclu-sion are easy to tune to match expert opinions.The method illustrated here is flexible and canbe easily modified to aggregating other vari-ables. Moreover, the same method could be con-veniently applied to assess the potential envi-ronmental impact associated to pollutants oth-er than waste waters from olive oil extraction,which might be shed in cultivated fields, e.g. in-dustrial waters.

ICARO provides an evaluation tool bywhich various people (scientists, decision-mak-ers, farmers), with different standpoints, can dis-cuss around a common problem (how evaluatethe environmental impact of waste water’s shed-ding in cultivated fields), following a transpar-ent and non-ambiguous methodology. The eval-uation of the expert system is an ongoing pro-ject. Most of the experts involved in its devel-opment supplied valuable remarks. Severalmodifications and additions were made, leadingto the version presented here. At operationallevel, certain precautions can be taken to pre-vent interested parties from shedding highly im-pacting waste waters. Of course, the lower the

value of the indicator the better the chance toshed waste waters in a field without impactingon the environment. The experience accumulat-ed so far suggested us that ICARO values canbe categorized in low (0-0.33), intermediate(0.33-0.66), and high (0.66-1.00) scores. The en-vironmental risk is high when high values ofICARO occur, thus the shedding should not beauthorized, unless consistent modifications inthe shedding strategy are done (i.e., lessening ofthe rate of application, change of the site, etc.)or carrying out supplementary surveys. Inter-mediate values of the indicator make possiblethe distribution, but suggest little modificationsin the shedding strategy, essentially the timingof execution or the cropping system choice. Nomodification in the strategy is required whenICARO falls within the range of low values.These guidelines will be corroborated by furthertesting. ICARO is about being tested on a num-ber of fields, starting from the province of Pisa,involving farmers and public decision-makers.The improvement and evaluation of ICARO isthus carried out in interaction with the inter-ested parties.

Acknowledgements

The authors would like to thank Donata Accogli, LuciaCeccarini, Laura Ercoli, Claudia Giannini, AlessandroMasoni, Tiziana Sabbatini, Marj Tonini, Cristiano Tozzi-ni for providing valuable contributions on the setup ofthe indicator. This research is under the auspices of theItalian National Research Council (CNR) and the Ital-ian Ministry of Education, University and Scientific Re-search (former MURST, presently MIUR), project “Ri-ciclo dei reflui del sistema agricolo-industriale. Linea Re-flui Oleari” (web page at: HTTP://USERS.UNI-MI.IT/~fsangio/cnr/olio1.htm).

References

Arnoldus H.M.J. 1980. An approximation of the rainfallfactor in the universal soil loss equation. In: De BoodtM., Gabriels D. (eds.): Assessment of erosion, 127-132. John Wiley and Sons, New York.

Bacci L., Conese C., Maselli F., Romani M., Zienna P.1989. Metodologia per la costruzione di mappe di ris-chio ambientale: un’applicazione per la Val di Cor-nia. I.A.T.A.-C.N.R., Florence.

Bardossy A., Bogardi I., Duckstein L. 1985. Compositeprogramming as an extension of compromise pro-gramming. In: Serfini P. (ed.): Mathematics of multipleobjective optimisation, 375-408. Springer-Verlag, Wien.

Silvestri N., Fila G., Bellocchi G., Bonari E.

254

Bellocchi G., Acutis M., Fila G., Donatelli M. 2002. Anindicator of solar radiation model performance basedon a fuzzy expert system. Agron. J., 94:1222-1233.

Benitez F.J., Beltran-Heredia J., Acero J., Pinilla M. 1997.Simultaneous photodegradation and ozonation plusUV radiation of phenolic acids-Major pollutants inagro-industrial waste waters. J. Chem. Technol. andBiotechnol., 70:253-260.

Bonari E., Ceccarini L. 1991. Spargimento delle acquedi vegetazione dei frantoi sul terreno agrario. L’In-formatore Agrario, 13:49-57.

Bonari E., Ceccarini L. 1993. Sugli effetti dello spargi-mento delle acque di vegetazione sul terreno agrario:risultati di una ricerca sperimentale. Genio Rurale,5:60-67.

Bonari E., Ceccarini L. 2001. Effetti dei reflui oleari sul-la produzione di alcune colture agrarie. In: I sotto-prodotti dei frantoi oleari, 3, 121-146. Progetto edi-toriale PANDA Edizioni L’Informatore Agrario,Verona.

Bonari E., Giannini C., Ceccarini L., Silvestri N., ToniniM., Sabbatici T. 2001. Spargimento delle acque di veg-etazione dei frantoi oleari su terreno agrario. L’In-formatore Agrario, suppl. 50:8-12.

Bonari E., Silvestri N. 2003. Acque di vegetazione deifrantoi oleari. In: AA.VV. (eds.): Linee guida per ilriutilizzo sul suolo di acque reflue depurate e per l’u-tilizzazione agronomica delle acque di vegetazione edelle acque reflue di aziende agro-alimentari. ANPA(Agenzia Nazionale per la Protezione dell’Ambi-ente). IGER srl, Roma. (in press).

Borja R., Martin A, Alba J., Fiestas J.A. 1992. Enhance-ment of the anaerobic digestion of olive mill waste-water by the removal of phenolic inhibitors. ProcessBiochem., 27:231-237.

Bouchon-Meunier B. 1993. La logique floue. PressesUniversitaires de France, Paris, France.

Campbell G.S., Shiozawa S. 1992. Prediction of hydraulicproperties of soils using particle-size distribution andbulk density data. In: Van Genuchten M.Th., Leij F.J.,Lund L.J. (eds.): Proceedings of the InternationalWorkshop of “Indirect methods of measuring soil hy-draulic properties”, 11-13 October. Riverside, CA,USA, 317-328.

Ciafardini G., Zullo B.A., Cioccia G. 1998. Effetti delleacque reflue dei frantoi su due batteri azotofissatori.L’Informatore Agrario, 15:35-37.

Cornelissen A.M.G., van den Berg J., Koops W.J., Gross-man M., Udo H.M.J. 2001. Assessment of the contri-bution of sustainability indicators to sustainable de-velopment: a novel approach using fuzzy set theory.Agr. Ecosyst. Environ., 86:173-185.

Coombs J., Hall K. 2000. N-F 2000 – AIR programmefinal project results disseminated under FAIR-CT96-1904. CPL Press, Newbury.

De Simone C., Raglione M., De Marco A., D’AmbrosioC. 1998. Analisi degli effetti genotossici delle acque

di vegetazione di frantoi oleari. Agricoltura Ricerca,173:75-80.

Di Giovacchino L. 2001. I sottoprodotti della lavo-razione delle olive. Caratteristiche qualitative e quan-titative. In : I sottoprodotti dei frantoi oleari, 3, 29-40. Progetto editoriale PANDA Edizioni L’Informa-tore Agrario, Verona.

Dillaha T.A., Reneau R.B., Mostaghimi S., Lee D. 1989.Vegetative filter strips for agricultural non-pointsource pollution control. Trans. of ASAE, 32:513-519.

Dubois D., Prade H. 1980. Fuzzy sets and systems: the-ory and applications. Academic Press, New York.

Fiorentini R., Andrich G., Loni A., Pacciardi A., Galop-pini C. 1985. Le acque reflue dei frantoi. Compo-sizione e tecnologia di smaltimento. Giornata di qual-ificazione dell’assistenza tecnica in agricoltura.Braccagni, Grosseto.

Girardin P., Bockstaller C., van der Werf H.M.G. 2000.Assessment of potential impact of agricultural prac-tices on the environment – the AGRO*ECO Method.Environ. Impact Assess. Rev., 20:227-239.

Goyal S.K., Deshpande V.A. 2001. Comparison of weightassignment procedures in evaluation of environmen-tal impacts. Environ. Impact Assess. Rev., 21:553-563.

Hall L.O., Kandel A. 1991. The evolution from expertsystems to fuzzy expert systems. In: Kandel A. (ed.):Fuzzy expert systems theory, 3-21. CRC Press, BocaRaton.

James D.W., Hanks R.J., Jurinak J.H. 1982. Modern irri-gated soils. John Wiley and Sons, New York.

Keeney R.L., Raiffa H. 1993. Decisions with multiple ob-jectives: preferences and value tradeoffs. CambridgeUniversity Press, New York.

Kuncheva L.I., Wrench J., Jain L.C., Al-Zaidan A.S. 2000.A fuzzy model of heavy metal loadings in Liverpoolbay. Environ. Modell. Softw., 15:161-167.

Matzavinos D., Hellenbrand R., Metcalf I.S., LivingstonA.G. 1996. Partial wet oxidation of p-coumaric acid:Partial oxidation intermediates, reaction pathwaysand implications for waste water treatment. Wat. Res.,30:2969-2976.

Muscutt A.D., Harris G.L., Bailey S.W., Davies D.B. 1993.Buffer zones to improve water quality: a review oftheir potential use in UK agriculture. Agr. Ecosyst.Environ., 45:59-77.

Odhiambo L.O., Yoder R.E., Yoder D.C. 2001a. Estima-tion of reference crop evapotranspiration using fuzzystate models. Trans. of ASAE, 44:543-550.

Pacifico A. 1989. Dossier acque di vegetazione. Agri-coltura ed Innovazione, notiziario dell’Enea e di Re-nagri, 11, 38.

Potenz D., Righetti E., Pellettieri A., Girardi F., Antonac-ci P., Calianno L.A., Pergolese G. 1985. Evoluzione del-la fitotossicità in un terreno trattato con acque refluedi frantoi oleari. Inquinamento, 5:49-55.

Poulios I., Makri D., Prohaska X. 1999. Photocatalytic

Ital. J. Agron. / Riv. Agron., 2006, 1:243-257

255

treatment of olive milling waste water: oxidation ofphotocatechuic acid. Global Nest the Int. J., 1:55-62.

Rana G., Rinaldi M., Introna M. 2003. Volatilisation ofsubstances after spreading olive oil waste water onthe soil in a Mediterranean environment. Agr.Ecosyst. Environ., 96:49-58.

Roussel O., Cavelier A., van der Werf H.M.G. 2000.Adaptation and use of fuzzy expert system to assessthe environmental effect of pesticides applied to fieldcrops. Agric., Ecosyst. Environ., 80:143-148.

Saxton K.E., Rawls W.J., Romberger J.S., Papendick R.I.1986. Estimating generalized soil-water characteris-tics from texture. Soil Sci. Soc. Amer. J., 50:1031-1036.

Silvestri N., Fila G., Bellocchi G., Bonari E. 2003.ICARO: un software per valutare l’idoneità dei ter-reni agrari alla distribuzione delle acque reflue deifrantoi oleari. Atti del XXXV Convegno della Soci-età Italiana di Agronomia, 16-18 September, Portici(NA), Italy, 281-282.

Soil Survey Staff, Natural Resource Conservation Ser-vice 1993. Soil survey manual. United States De-partment of Agriculture, Handbook No. 18, Govern-

ment Printing Office. Washington D.C.

Sugeno M. 1985. An introductory survey of fuzzy con-trol. Information Sciences, 36:59-83.

Tomati U. 2001. Impatto ambientale ed uso di compostdalle acque reflue di frantoio. In: Tedeschi P., D’An-dria R., Touzani A. (eds.): Atti del convegno “Ges-tione dell’acqua e del territorio per un’olivicolturasostenibile”, 24-28 September, Naples, Italy, 296-311.

Van der Werf H.M.G., Zimmer C. 1998. An indicator ofpesticide environmental impact based on a fuzzy ex-pert system. Chemosphere, 36:2225-2249.

Wischmeier W.H., Smith D.D. 1978. Predicting rainfallerosion losses. Agricultural Handbook 537. U.S. De-partment of Agriculture, Agricultural Research Ser-vice, Washington, DC.

Zadeh L.A. 1965. Fuzzy sets. Information and Control,8:338-353.

Zouari N., Ellouz R. 1996. Toxic effect of coloured olivecompounds on the anaerobic digestion of olive oilmill effluents in UASB-like reactors. J. Chem. Tech.Biotechnol., 66:414-420.

Silvestri N., Fila G., Bellocchi G., Bonari E.

256