Building a fuzzy logic information network and a decision-support system for olive cultivation in Andalusia

In Southern Spain, olive ( Olea europaea L.) growing is an important part of the economy, especially in the provinces of Jaén, Córdoba and Granada. This work proposes the first stages of an Information and Decision-Support System (IDSS) for providing different types of users (farmers, agricultural engineers, public services, etc.) with information on olive growing and the environment, and also assisting in decision-making. The main purposes of the project reported in this paper are to process uncertain or imprecise data, such as those concerning the environment or crops, and combine user data with other scientific-experimental data. The possibility of storing agricultural and ecological information in fuzzy relational databases, vital to the development of an IDSS is described. The information will be processed using knowledge extraction tools (fuzzy data-mining) that will allow rules on expert knowledge for assessing suitability of land to be developed and making thematic maps with the aid of Geographic Information Systems. Flexible querying will allow the users to collect information interactively from databases, while user information is constantly added. Flexible querying of databases, land suitability and thematic maps may be used to help in decision-making.


Introduction 1
Decision-support in agriculture affects both crops in all their phases (planting, management, costs, etc.) and distribution and organization of production systems.Decision-Support Systems (DSS), as expert systems, were originally developed to facilitate the application of crop models in a system approach to agriculture research.They were also motivated by a need to integrate knowledge about soil, climate, crops and management for making better decisions about transferring production technology from one location to others where soil and climate differed (Jones et al., 2003).The DSS approach provides a framework to research and understand how the system and its components function.This knowledge is then fed into models that predict the system's behavior for given conditions (Jones et al., 2003).Land evaluation based on physical and socioeconomic data is therefore a vital tool for decisionmaking, since it assesses the suitability of different agro-ecological systems in an area by analyzing the relationships between the variables affecting these systems (Sys et al., 1991).
One design task required for a fully operative DSS is the creation and management of databases able to integrate different information and knowledge resources in a decision-making context (Harrison, 1991;Marakas, 1999;Haag et al., 2000;De la Rosa et al., 2004).In this sense, it is very recommendable to involve users in the development and implementation of DSS tools (Matthies et al., 2007).User (such as the farmer) and technician or researcher's knowledge can be used as a basis for modeling tasks and integrating models created while designing a DSS.
However, to infer relevant knowledge from this information, knowledge discovery (knowledge extraction) with data-mining techniques (extraction rules), which have been demonstrated to be the best tools for agricultural and environmental systems, are used (Hoshi et al., 2000;Poch et al., 2004;Kawano et al., 2005).
Furthermore, in an agro-environmental context, the information derived from the vagueness of human reasoning, while experiencing and interpreting the complexity of the real world is uncertain and imprecise (Bragato, 2004).This is especially critical in the case of the farmer's knowledge of his/her agricultural system.
Fuzzy logic provides an effective decision-making tool for dealing with this kind of imprecision and uncertainty (Groenemans et al., 1997;McBratney and Odeh, 1997;Center and Verma, 1998;Kuo and Xue, 1998;Evsukoff et al., 2000;Geneste et al., 2003).First proposed by Zadeh in his Theory of Fuzzy Subsets (Zadeh, 1965), fuzzy logic works with rough data and linguistic values of variable and imprecise relationships to find the best solution or alternatives to complex problems (Panigrahi, 1998), such as those in agricultural systems.Fuzzy logic has been used successfully as a prediction tool in agriculture, for example, in sugar production (Petridis and Kaburlasos, 2003) and livestock raising (Lacroix et al., 1998).
The use of fuzzy data-mining could increase the flexibility and adaptability of rule-based DSS, and inference from fuzzy sets could be an alternative to current methods (De la Torre et al., 2005).
The purpose of this study is to establish the basis for creation of a future rule-oriented, integrated Information and Decision-Support System (IDSS) for olive (Olea europaea L.) crop management in the region of Andalusia, based on optimum deployment of resources according to their potential, e.g., sustainable land management systems.The paper summarizes the information collected (highlighting its imprecise nature), its representation in fuzzy relational databases, how flexible queries are made, different kinds of knowledge are merged and knowledge is extracted (using fuzzy data-mining).It describes the main features of this particular implementation, and establishes the basis for the development of a future complete IDSS.

Premises for the development of the IDSS
For development of a suitable IDSS, it is assumed that optimum cultivation is a system with the highest productivity compatible with maintaining the suitability of the soil for cultivation.To achieve this, a number of objectives must be reached: a) maximum profit; b) products of maximum quality; c) minimum cost, for which totally mechanized olive cultivation is required; d) absence of significant limitations for the crop, so that this is not in a marginal situation; e) the use of good agricultural practices, understood as being the necessary cultural maintenance without cutting profits or increasing environmental risks; f) land use which protects the soil.
The development of this type of expert system requires the operations shown in Figure 1.Data collection for current and potential systems of olive cultivation may be carried out through the bibliography or directly.Working with agronomic systems involves the use of different types of knowledge such as scientific knowledge and that derived from users.Knowledge in the form of bibliographical and direct data on cultivation systems is stored in soil and cultivation databases and in digital maps.
Expert knowledge rules are found by extracting knowledge from the information.An example of rule extraction by means of fuzzy data-mining is explained in the last section of the manuscript.These will allow establishing a knowledge tree based on fuzzy logic and use an expert evaluation system, thus creating a computerized suitability classification based on fuzzy rules.
From the databases and the digitalized maps, homogeneous soil-cultivation areas called «agropedological units» (AU) are established by merging soil and cultivation knowledge.These units are assessed using suitability classification, thus allowing classification of the suitability of the current and potential systems for olive cultivation.
Using Geographical Information Systems (GIS) all the results from the assessment and everything held in databases and digital maps may be expressed spatially (thematic maps).Its usefulness in land-use planning is greatest and allows model outputs, such as constraints on agriculture and site-specif ic best-management practices, to be identified in a spatially explicit manner (Smith et al., 2000).

Description of the information Type, source and spatial representation of the information
For an agricultural IDSS, in addition to socioeconomic information, information on soils (in the wide sense, including climate, topography, etc.) and crops is required.The former consists of pedological (soil considered in the strict sense of «individual soil»), climatic, topographical and geological data and can be found in the bibliography in the form of thematic maps, databases, digital terrain models, etc.This information is represented spatially as soil unit maps, climatic areas with the same temperature or rainfall, isocline zones, lithostratigraphic units, etc.The information on crops is of an agronomic (varieties of plants cultivated, irrigation systems, fertilizers, etc.) and socioeconomic nature.The cultivation data are generally compiled directly through interviews with farmers and indirectly from agricultural reports.Surveys function as a result of rural people in many developing countries having a rich understanding of their resources (Thrupp, 1989;Warren, 1989).
The survey used in this IDSS consisted of three sections: A) location and general data (a first section with survey control information together with geographical data); B) crop management (including attributes of land area, number and characteristics of plants, treatment and care, and production data) and C) soil data (general characteristics, such as soil depth, hardened layer depth, mean slope, soil texture, etc.).In the survey, the spatial unit of information on the crop is the plot or farmer's property, while other data are given by district or even province (local or regional level).

Inaccuracy and uncertainty of the information
The first source of uncertainty in the farmer survey is due to the fact that only a small percentage of the farmers in the province of Granada was interviewed (210 surveys); Rodríguez et al. (1998) state that the minimum percentage of interviews is a function of the variability of the agricultural systems under study.Since knowledge about this variability requires surveys a vicious circle is created.
Most of the farmers interviewed have highly fragmented properties; the survey is carried out on the most representative area which may lead to a distortion in size of the property.The survey may also create uncertainty or inaccuracy as a result of linguistic problems since some terms are not common for the whole of the area surveyed.The cost of many agricultural activities is difficult to calculate, especially when these tasks are carried out by the farmer and his family.Farmers are also sometimes uncertain about the prices of fertilizers and other products.Another problem is the tendency, for socio-cultural reasons, neither to disclose the real agricultural practices carried out, nor their cost.It should always be remembered that the farmer will have a different concept about the soil to that held by the soil scientist.
Soil databases store two essential types of attributes regarding their obtention: bibliographical and experimental, and, within the latter, morphological and analytical one (soil horizons, physical, chemical, etc).In Figure 2 the types of data are related to the factors and their uncertainty and inaccuracy.It should also be noted that much of the information regarding soils is qualitative and is thus exposed to the researcher's subjectivity and, furthermore, its mathematical processing is difficult (Webster, 1977;Webster and Oliver, 1990).The analysis of an object, soil, whose variability is produced in a «continuum», is diff icult, since it requires the use of new methods based on fuzzy logic and fuzzy groups (McBratney and De Gruijter, 1992).

Imprecise information representation using fuzzy logic
One of the most widespread methodologies in information organization and processing is the rela-Building a decision-support system in olive cultivation in Andalusia 255 Bibliographical: -Thermopluviometrical -Topographical -Geological -Vegetation, etc.

Data types Uncertainty and imprecision factors Uncertainty and imprecision degree
Analytical: -Physical -Chemical -Physical-chemical -Minerological -Researcher's experience/readiness degree -Realization and interpretation of the analytical method -Precision and repeatibility of the measure -Inherent difficulty for the method -Availability of lab resources -Variability of the analyzed feature -Researcher's experience/readiness degree -Lack of previous knowledge of the study area -Elapsed time in soil describing -Difficulty degree of every described feature -Inherent difficulty degree of studied soils (natural complexity and variability degree of each feature) Originally, the relational model did not allow imprecise or uncertain information to be handled.However, in recent years, researchers have tackled the problem of how to make this model more flexible in order to allow this feature.Fuzzy logic is thought to be able to deal with this kind of problem both effectively and efficiently.
Over the last two decades various improvements have been made to the relational model so that it may manage imprecision and uncertainty using fuzzy logic.Different models have been proposed.The GEFRED model (Medina et al., 1994) is generally the most wellknown as it handles imprecision and uncertainty simultaneously.It also proposes a special approach in the query process to the Fuzzy Relational Database Management System (FRDBMS), in which any information request to the database is reduced to a query in relational terms.The model considers nearness relations over attribute domains and also the possibility of setting different fulf illment degrees for any of the attributes in the query.

Fusion of the information on crops and soils. Spatial representation of the data (GIS)
The fusion of data on a spatial level was resolved by establishing AUs for the province of Granada.The AU is an abstract concept defined by soil typologies and other physical parameters (slope and temperature) which have a decisive influence on olive cultivation.The AUs were established by superimposing three different maps: soils, mean annual temperature and slope.The aim of this division of the land is to establish areas with a suitable size in order to reduce the degree of variability of the conditions for this crop and to ensure that the number of data for each kind of unit is sufficient.The databases for the soils and the interviews were spatially associated within the AUs.
The soil map used was that for the province of Granada on the 1:200,000 map scale (Pérez-Pujalte, 1980).This map is based on geology and physiography, which results in the grouping of soil typologies which have different classifications but with great similarities in terms of morphology and properties (soil texture, organic carbon content, pH, etc.).
The temperature map was established using the temperature/altitude correlation, with this province having very high correlation coefficients.According to these correlations, the limits for olive cultivation were -7°C for the mean minimum absolute temperature of the coldest month and 13°C for the annual mean, giving the following thresholds: higher than 1,250 m, low suitability; from 1,250 down to 800 m, reasonable suitability; from 800 down to 275 m, optimum suitability, and, lower than 275 m, reasonable suitability.
Slope conditions affect certain aspects of cultivation such as: soil management (mainly irrigation and mechanization) and risks of degradation (mainly hydric erosion).Using the limits established by the Food and Agricultural Organization (FAO) for the slope, four classes of limitation can be established, these being the four map units of slope on the map: less than 6% slope, no limitation; from 6 to 13%, moderate limitations; from 13 to 25%, severe limitations, and, more than 25%, very severe limitations.
The AU can be considered as a «unit of information», that is, it cannot be defined by itself but by its attributes.It is not intended to be a «natural» unit of the landscape but rather a division which mixes the soil map unit, based on a natural soil classification (FAO, 1974), with artif icial units for the landscape established using utilitarian criteria (risk of frost, latent period, possible mechanization, etc.).However, since these utilitarian criteria are expressed as thresholds of altitude or slope, natural variability is also involved.Thus, the AU is a division of the landscape which is essentially natural; it is not exactly a «unit of land type» as defined by FAO (1974).
Figure 3 shows how the AU is interpreted graphically.Once the AU is established, the agronomic information (surveys of farmers) and information on soils and other databases are incorporated.The AU can be reinterpreted from the point of view of olive cultivation (including utilitarian information) with the aid of the suitability classification.The suitability of these AUs can be used for decision-making, geographically speaking, using the systems for flexible querying.
The incorporation of GIS as tools for modeling and spatial analysis is vital for any project where georeferenced information must be integrated (Longley et al., 1999).With a view to handling the mapping information and establish the AUs, ARC/INFO v.7.02 for Windows NT was used owing to its capacity for the integration of database systems external to the system, such as Oracle, and with special software for the construction of models of land assessment, such as the ALES system.The initial mapping information was incorporated into the system by digitalizing the corresponding maps or was acquired directly in digital formats compatible with ARC/INFO.In some cases this information was modeled directly as vectorial covers, for example, the maps for soil and soil uses.In other cases raster models in a grid or lattice format of ARC/INFO were used, as is the case of the digital terrain model with a pixel size of 20 m.The climatic data (mean annual temperature and mean annual precipitation) were initially modeled on a digital terrain model using linear correlation equations obtained from 40 meteorological stations in or around the province of Granada.Different raster models were constructed for the different climatic parameters modeled.Vectorial covers were generated from this model that corresponds with the limits of the climatic parameters employed in the mapping.
Slope maps in the form of vectorial covers were also generated from the digital terrain map, using the slope margins listed previously.
The databases associated with the vectorial covers were constructed and handled in Oracle, with the covers being connected through the corresponding feature attribute tables.
Once all the information had been integrated into the GIS, spatial analysis tools such as topological analysis (spatial join, proximities, buffering, etc.) of vectorial covers together with logical operations on the feature attribute tables were used to obtain maps of agropedological units, thematic maps, etc.

Flexible querying and knowledge extraction Operations on imprecise data
Many different criteria have been considered in order to decide whether to model information as precise or imprecise.On the one hand, there are numerical values.When these are obtained from a reliable source, such as an analytical study, it is not necessary to store them as fuzzy data, for instance, % of nitrogen in soil.On the other hand, there are no exact measurements for the products used in fertilizing or fumigation and the quantities are based on the farmer's experience.This is an additional source of imprecision, and data must be handled and stored appropriately.Information on other attributes such as land slope is supplied using the linguistic labels (flat, moderately sloping, sloping) which are closer to natural language, although the underlying domain may be numerical.
These can therefore be modeled as fuzzy values, using trapezoidal or interval functions as possibility distributions.Even when the attributes are handled as labels, a query can be made regarding the numerical (fuzzy) values.In order to do so, information about the match between linguistic labels and their associated value sets must be stored in the FRDBMS Catalog, more precisely, in the Fuzzy Meta-Knowledge Base (FMB).
There are also other types of imprecise attributes, which are those represented as a set of scalar nonnumerical values by means of a nearness relationship.For example, the attribute «land tractorability» has a class set, defined as (high, medium, low), following an intuitive order relationship.The nearness matrix between these categories is stored in the FMB.
In the soil database, most of the attributes are defined on a categorized domain due to the standards proposed by the FAO.As existing databases are modeled in this way it is impossible for imprecision or uncertainty to be handled.However, it is possible to handle these features.Although stored data is «crisp», uncertainty can be added when flexible queries are performed, thereby improving the model so that information may be obtained.
Thus, it may be said that two types of fuzzy data processing are performed.One of these is called a priori handling, in which there is fuzzy data since this is how it was stored in the DB.The other is called a posteriori handling and takes place in the query process.Here, data that have really been stored as precise can be fuzzified.

The flexible query
A typical system for handling bases provides procedures for the storage, access and modification of the information.Through the facility of access to the information the user can obtain information contained in the database (DB).It is even possible to specify the criteria with which the information is selected; or to demand that only the information which fulfills a certain condition (degree of fulfillment) is returned to the user.
Furthermore, when part of the information is fuzzy it is necessary to provide methods which give good access to this information.One way to achieve this is to flexibilize the query.This gives the language for querying the DB the ability to handle imprecise information and to express fuzzy data in natural language terms.
For example, a typical query addressed to the DB: «Tell me which farms have an annual production of more than 3000 kg», would read, «SELECT farm_name, annual_production FROM farm WHERE annual_production>= 3000».
Supposing that it is needed to know which farms have a «high» annual production, the flexible query would be as follows: «SELECT farm_name, annual_ production FROM farm WHERE annual_production = HIGH».
The common term «high» is extremely imprecise.How many olives should be obtained for production to be considered «high»?Internally, the DB system could model this characteristic in different ways, although this processing should be hidden from the user, who is only interested in finding out which farms have high production.It should also be noted that this flexible query can also be carried out both on attributes of a fuzzy nature and on those with a perfectly def ined value.
Different models of DB have considered the problem of flexible query, one of which, the GEFRED model (Medina et al., 1994), mentioned previously, is the most widely known one.
In the process of implementation of the information, the modules for data input and flexible query should be examined.

Data input module
Using this first module, users can easily enter new data.It consists of two clearly different parts.The visible part is a set of HTML forms (Fig. 4), each associated to a table in the database.Users can access them via internet using any web browser.Under this, there is a Java-based application, comprising a servlet (current replacements for CGI, an usual technology employed in web servers) and an insertor class.This part processes the sentences and communicates with the FRDBMS.The servlet application manages the web pages, takes data from forms and informs the users about the success or failure of the transactions performed.In order to accomplish these tasks it transmits the content of the forms to the insertor class, and receives and manages the success or error messages sent by the server.
Data is processed within the insertor class in order to build the insertion sentence.Communication with the FRDBMS is also maintained here, and any sentences sent and any resulting messages received.
The module is sufficiently general to be adapted to any given set of tables, which may or may not contain fuzzy values.For this case, two versions have been implemented.The f irst is for the survey database and the second for the soil database.Another added feature is that of allowing simple queries and on-line updates of previously stored data using the same interface.If more complex queries are to be performed, e.g.flexible queries, another tool must be used, as explained below.

Flexible query module
This module is responsible for displaying information, which has previously been stored in the database, to the user according to a flexible (or non-flexible) query specified by the user.Although two different applications were originally used in this module, these were later integrated into one.The original program, FQ (Fuzzy Queries) (Galindo et al., 1998), was improved with the name of FQBuilder (Fig. 5).This was programmed in Visual Basic.Firstly, a separate Java-based application was developed in order to improve FQ.Following a short period of experimentation, a more complete and powerful application was developed.FuzzyQueries2 (Fig. 6) takes the best of the previous releases and adds new features.Using this application, users can easily edit and build flexible queries in FSQL.
FuzzyQueries2 can display a set of reserved words, common operators and objects in the database (tables, columns and linguistic label sets), which users can take advantage of.Using FuzzyQueries2 users can query the system catalog, and, within this, the Fuzzy Meta-Knowledge Base (FMB).In particular, FuzzyQueries2 displays the existing matches between linguistic labels and trapezoidal distributions, both defined on fuzzy attributes in the databases.Users can define their own possibilistic distributions using the graphic interface.Once the sentence has been built, it is sent to the FRDBMS, where it is analyzed and any syntax errors are found, before being executed.The results are then displayed to users.

Knowledge extraction
Using the query (flexible or not) on the database being used it is possible to obtain extrinsic or direct information immediately.However, one of the main reasons for compiling data on olive cultivation and soils in Granada is to obtain intrinsic or hidden information and to find relationships which are not evident between attributes which are not known in advance.To obtain this new class of information, knowledge extraction and data-mining techniques are used (Cubero et al., 1995;Carrasco et al., 2000;Delgado et al., 2000).
The tools for knowledge extraction are extremely powerful but must be refined beforehand in order to orientate the search without increasing excessively experimental time while extracting genuinely useful information for the user.It is thus necessary to carry out a first exploratory analysis of the data to select the attributes which best define the information.A variety of useful statistical tools are available for this purpose.
The first step is to carry out a multivariate analysis by means of a principal components analysis (PCA) which restricts the initial group of attributes to a subgroup which is easier to handle.Once these principal components are known, providing initial information on the relevance of the attributes, the elements can be grouped into clusters or classes according to these principal components (Calero et al., 2005).These groups are an additional source of information, in this case for the similarities between elements (Delgado et al., 1999).
From this first exploratory analysis of the data were obtained a restricted group of relevant attributes and the clusters in which the elements were divided.A knowledge extraction technique can now be used.One option is to make use of the classes obtained in the second step to use an algorithm based on classification trees, such as C4.5 or ID3 (Carrasco et al., 1998).Another option is to search for intrinsic relationships between attributes using a data-mining tool such as rules of association (Delgado et al., 1999(Delgado et al., , 2000)).The results obtained can then be compared with the expected results, providing a basis for further study.

Example of fuzzy data-mining application over user knowledge in olive cultivation
As described earlier, the information was obtained from the knowledge that farmers supplied (user knowledge) by means of surveys about farm management, soils and the environment.The user data obtained from a survey with 126 variables carried out on 210 olive grove farmers in the Province of Granada (southern Spain) was stored in a GEFRED model database and processed using data-mining.This information was pre-processed in order to select the most relevant attributes.The resulting data were fuzzified, by defining linguistic labels over numeric attributes [for each numeric attribute, is defined a set of linguistic labels (high, medium, low) over the numeric domain] and fuzzy relations over scalar attributes.Then, a fuzzy association rule extraction process was applied in order to obtain user evaluation rules (UER).
Using 34 variables selected following a cleaning process, 1420 fuzzy association rules were obtained in which the antecedent is a variable of management, soil, or environment, and the consequent was: olive fruit production (kg ha -1 ), % of oil in olive fruit, or acidity of fruit on tree (olive oil quality).Only 182 of these rules were selected, either because the value of their certainty factor (CF) (assurance measure) was high, or because they were most interesting from the point of view of expert knowledge.From the 182 rules, 85 were deemed to be user evaluation rules indicating the suitability for olive cultivation according to the different attributes, and they were drawn up on the basis of user knowledge.Some of these rules were corroborated with the knowledge regarding olive cultivation found in the bibliography, others contradicted the said knowledge and others revealed relationships not previously described.This procedure would be deemed to be an evaluation method based on empirical data.
As follows, are commented some rules that might be of interest following the scheme: (Antecedent) → (Consequent) CF, comparing them with expert rules taken from the bibliography.The consistency of the method for obtaining fuzzy association rules by means of fuzzy data-mining was confirmed with the following rule: (Mean minimum temperature in the coldest month = high) → (Mean altitude = low) CF = 1.The association between these two variables has universal validity.The results are focused on rules in which the consequent is a production variable (olive fruit production in kg ha -1 ), that is the most significant one.
The following rules, related with agronomical management, would be deemed to be highly interesting.This is the case for: (% planting frame of less than 10 m = medium) → (production = low) CF 0.60.This rule, according to the bibliography, matches the one found in Barranco et al. (1999).The authors consider the optimal planting frame to be between 200 and 300 trees ha -1 , i.e., an approximate distance between trees of 10 m or less.
Some UER are obvious, as is the case of (% of young olive trees = high) → (production = low) CF 0.70 or (% of olive trees in full production = low) → (production = low) CF 0.44.
Other UER seem to be contradictory according to other sources of knowledge.This is the case of the negative relationship between irrigation and production [(irrigation = with irrigation) → (production = low) CF 0.35], that should be positive, as described in the bibliography (Barranco et al., 1999), even more considering the severe summer drought in the region.This contradictory association might be explained using a third variable: the Granada olive grove system is undergoing a phase of modernization in which irrigation and the rejuvenation of the olive trees are included amongst the improvements, therefore, the olive trees with irrigation are usually the youngest ones, which are not yet in full production.
There are also UER that reveal knowledge that is difficult to interpret, although the information that they hide is interesting.For example: the rule that relates high production with a low number of olive trees [(number olive trees = low) → (production = high) CF 0.40] and its complementary rule that is the one which associates low production with a large farming area [(farming area = high) → (production = low) CF 0.49].It should be borne in mind that the larger the farming area, the larger the number of olive trees.More research would be required to provide clearer answers and in order to validate this rule.
Some UER related with soil characteristics may also be considered very interesting.(erosion = with erosion) → (production = low) CF 0.30.These association rules reveal that the characteristics of the soil which influence agricultural practices, the handling of the soil or the development of the roots, are all clearly perceived by the farmer.In most cases when these soil characteristics are unfavourable (including the erosion variable), they are related to low production.The soil colour perceived and defined by the farmers also produces interesting UER.For example: (soil colour = dark grey) → (production = low) CF 0.59.This rule is difficult to explain and needs further field research.
The latter and other UER, demonstrate that the method for knowledge extraction described in this section is of an exploratory nature and so allows new hypotheses and studies to be proposed.

Conclusions
The purpose of this study was to establish the basis for a data system to assist in decision-making in the scope of olive production in the region of Andalusia (Southern Spain).To do so, intense work has been done to integrate information with different degrees of imprecision and uncertainty gathered from different experimental and empirical sources.In addition, the goal is to create an interactive system that will provide information, while simultaneously receiving information, for the different system users, by implementing a procedure for flexible querying.
An important part of the work is the design of fuzzy databases and the use of fuzzy logic for storing and processing the environmental data (mainly on soils) and agricultural data.Current database models are too strict to store imprecise and uncertain data of this kind.Moreover, fuzzy databases modeled by fuzzy logic allow flexible querying, which is an essential tool in the creation of an IDSS.
The system which it is intended to create is still in the development stage.Several stages have been completed and others are still underway.Compilation of information and database implementation is already finished.An exploratory analysis of the data has been carried out and Agropedological Units have been established using GIS.Expert knowledge rules based on the farmer survey have been extracted using fuzzy data-mining techniques (fuzzy association rules and fuzzy approximate dependencies).
Some problems have come up in the user information compilation stage and in flexible iterative querying by those users.Specifically, the number of items in the farmer survey was excessive and reiterative, for example, plot size and number of olive trees.Some of the questions asked the farmer e.g., about soil colour, are difficult to interpret as well.The query process also baffles some of the users, who lack training, since they are usually elderly and have small farms.So to solve these problems, a new grower survey is now being designed with fewer items, the questions are being rewritten, graphs and photographs are included and a more user-friendly interface is being implemented for flexible querying (both for data entry on and retrieval from the web page through an extensive client application).

Figure 1 .
Figure 1.Operations of the information and decision-support system.

Figure 2 .
Figure 2. Imprecision and uncertainty factors in soil data.

Figure 5 .
Figure 5. Construction of a query with FQBuilder.