Data/Metadata Data and Data Transformations

Transcription

1 A Framework for Classifying Scientic Metadata Helena Galhardas, Eric Simon and Anthony Tomasic INRIA Domaine de Volcea - Rocqencort 7853 Le Chesnay France First-Name.Last-Name@inria.fr Abstract The scientic commnity, pblic organizations and administrations have generated a large amont of data concerning the environment. Thereisaneed to allow sharing and exchange of this type of information byvarios kinds of sers inclding scientists, decision-makers and pblic athorities. Metadata arises as the soltion to spport these reqirements. We present a formal framework for classication of metadata that will give a niform denition of what metadata is, how it can be sed and where it mst be sed. This framework also provides a procedre for classifying elements of existing metadata standards. Introdction Nowadays, scientic data is considered of prime importance. Governments, organizations, the scientic commnity and the pblic in general are more and more concerned on bilding applications that handle scientic data. Nevertheless, scientic applications have some properties that make them diclt to be handled. The amont ofscientic data that needs to be shared by scientists is very large. It is arond two orders of magnitde above the size of the data involved in bank on-line transaction systems [5]. Scientic applications have a mlti-disciplinary characterastheycover, for example, scientic calcls, statistical data analysis, decision spport, management of natral risks and spport to reglations. Since a wide variety of commnities is involved, scientic data comprises many dierent types of data. Ths, we canhave: satellite images, maps, time series, descriptive text Fonded by \Institto Sperior Tecnico" - Technical University of Lisbon and by a JNICT fellowship of Program PRAXIS XXI (Portgal) docments or management data. Scientic data is sally heterogeneos and often distribted over heterogeneos software and hardware platforms, becase it is prodced by many dierent sorces. Moreover, dierent levels of sers (scientists, decision makers or pblic agencies and organizations) with distinct skills access to these distribted data sets. Conseqently, data is sed in mltiple and distinct ways. Another important characteristic of most of this data is that it is rled by scientic laws and it can be transformed by mathematical models. Some of the data transformation methods are reqired to be re-sed by dierent sers in the same commnity orby sers that belong to a dierent commnity. Users of scientic data have to face some problems de to the above mentioned properties of scientic applications. The rst task sers have toac- complish is to nd relevant data according to their interests. This might be diclt either becase data is not properly referenced by data sppliers, is dplicated and not consistently maintained or becase it is inscient and reqire additional processing. Secondly, the extraction of the reqired information from the data sorces might beadif- clt process. The main reasons for sch an obstacle are the heterogeneos natre of the nderlying data and systems (reqiring format translations), inherent administrative procedres (data mightbe private), and the costs associated to pre-processing data before it can be sed. Once retrieved, data may be hard to be sed. Detolacking of data classication by data sppliers, data sets may be incompatible or inconsistent. Finally, the qality of extracted data may be diclt to evalate. It is often hard to compare data prodced by distinct sorces and sing dierent scientic models with no docmentation abot the data prodction process.. Metadata A possible soltion to eliminate or redce the ser problems previosly mentioned is to pblish axiliary information in addition to the one accessible in data sets. Some examples of this additional information will follow. Searching sefl

2 data sorces for answering a ser qestion is easier if each data sorce pblishes its location and a smmary of its contents. Retrieving data from the selected sorces is possible only if one is aware of its strctre (e.g. schema of a database). The ser mst be able to choose the appropriate scientic model to apply to the retrieved data to prodce new data. So, information abot the interfaces of available transformation models and its ses mst also be pblicly available. A scientist may want to repeat the same experiment several times and record its reslts. Ths, historical information mst be stored and be available for interpretation. Finally, scientists also need to decide whether data has enogh qality to be sed. Information abot data accracy and the methods sed to prodce it will spport that decision. This axiliary information is sally called metadata. Althogh there is an intitive explanation of what metadata is - strctred dataabot data - there is no clear denition that describes what the notion of metadata encompasses. Many people dene metadata with dierent meanings. In the database domain, metadata is all the information that is stored in the database dictionary. In Digital Libraries, metadata provides information abot a data sorce content in order to spport the ecient search and retrieval of docments while releasing sers from being aware of the entire data sorce content. Within the commnity of environmental scientists, reference [4] claims that, besides its seflness in the process of data sorce discovery, metadata is also essential to the eective se of discovered data sorces, by providing the mechanisms for interoperability across protocol domains. For the rest of this paper, we se environmental scientist commnity as a prototypical commnity for scientic data management. Even within the environmental commnity, there is not a common denition and dierent metadata formats emerged from distinct disciplines. From an analysis of the existing standards, one concldes that there is not and there will never exist a common metadata format. This is mainly de to the heterogeneos natre of environmental commnities. Nevertheless, a common denominator metadata format as UDK model [4] or CEO metadata gidelines [3] exists. UDK model proposes a class hierarchy of metadata based on environmental data types, and CEO gidelines spply aminimm set of metadata to be sed by several environmental commnities. In parallel, a qalitative classication of the metadata sed dring the environmental data prodction process, namely to spport data location and evalation, can be fond in [2]. Despite these eorts, there still does not exist a way to formally explain to data prodcers and sers what metadata is, how can it be sed to solve ser problems and when it shold be sed dring the environmental data prodction and management process. To address these three points, we propose a framework for the formal classication of metadata. This framework shold be described orthogonally to the data types involved. In addition, this formal denition shold provide an nambigos method for comparing and evalating metadata standards. By accessing to sch a classication of existing metadata standards, sers and prodcers will then be sre of sing the more appropriate ones..2 Contribtions The main contribtions of this paper are: Denition of a framework for classication of metadata Application of this framework to the prodction of environmental data Denition of a procedre for classifying elements of a metadata standard. Application of the framework and procedre against one existing metadata standard and one metadata model. The next section will describe formally the framework for classication of metadata. Section 3 explains what types of metadata, from the framework, are sed throgh the environmental data prodction and management. The forth section introdces a methodology for classifying metadata elements from metadata standards. This methodology is exemplied for a metadata format (the US Federal Geographic Committee [6]) and a metadata model (UDK model [4]). 2 Framework for classication of metadata Or classication of metadata is based on rstorder logic. We se as example a small extract of the domain of oceanography based in [7] where we restrict orselves to relational data sets. The base relations are wind-data(id, speed, direction) and wave-data(id, height, direction, period). In or knowledge domain, information abot the sea conditions (wave-data) is obtained throgh the application of a scientic model (called predicted-wave) to the wind speed and direction. We briey introdce the basic notions of a rst-order logic [] that will be sed. A rst-order logic is composed of a theory and an interpretation. A theory consists of an alphabet, a rstorder langage L, a set of axioms and a set of inference rles. The alphabet is composed of: variables (x y z...), constants (a b c :::), fnction symbols (f g :::), (p q :::), connectives (_ ^ and :), qantiers (9 8) and pnctation symbols ( \ ", \(" and \)"). A well-formed formla is a collection of symbols from the theory that behave according to the rles of a rst-order logic. A fact is a well-formed formla applied to constants of the alphabet. The rst-order langage L, given

3 by an alphabet, is composed by thesetofallwellformed formlas bilt from the symbols of the alphabet. L is concerned with the syntactic or strctral aspects of the corresponding rst-order logic. An interpretation gives a meaning to constants, fnction symbols and predicate symbols. It is a domain of discorse composed by elements that are assigned to constants variables range over them fnction symbols are mapped into them and are assigned to relations between them. Interesting interpretations are those for whichawellformed formla expresses a tre statement. The main idea nderlying the classication of metadata sing rst-order logic is to write wellformed formlas and give them an interpretation. Dring the modeling of metadata, we will need to introdce new rst-order logics. Before their introdction, we explain the notation sed and some assmptions taken. The basic rst-order theory has a langage that is called L 0. Constants in L 0 are symbols like, \wave information", and that are interpreted as the real data vales. Ths, we se the classical denition of logical interpretation. We then constrct from L 0 a new rst-order logic (with langage L ) that is obtained by a restricted form of reication. The interpretation of this new rst-order logic states that and fnction symbols from the L 0 alphabet become constants of the new rst-order theory alphabet. L may contain other new constants, however it cannot contain any constant of L 0. So, is not a constant bt Integer will be a constant indicating type information. The same reasoning is applied to L to constrct a new rst-order logic L 2. Ths, the constants of L 2 are the predicate symbols of L. We avoid an innite seqence of L 0, L, L 2... by sing reication to redce the langage hierarchy to a limited nmber of levels. To handle the expression of facts from mltiple langages, we dene the nion of rst-order langages as follows. L is a new rst-order logic that is the nion of the langages of L 0 and L. The interpretation of L is the nion of the interpretations of L 0 and L. Ths, the constants of L are the nion of the constants of L 0 and the constants of L. Similarly, L is the nion between L 2 and L. The set of constants of L is then composed by the constants of L 0,of L and of L Modeling In the sbseqent sections we se the above denitions to model standards of data. Figre shows the general strategy of this modeling. The horizontal plans on the right side represent data and data transformations. The colmn on the left hand side shows the metadata associated with each plan and The formla 8x8y y(x) is a well-formed formla in a second-order logic, becase it qanties over predicate names. To reify this formla, predicate names are mapped into constants and the formla is rewritten as 8x 8y p(y x). the fragments of logic sed to model this metadata. The theory level of data corresponds to environmental data sets. It corresponds to a rst-order theory and its langage L 0. The interpretation level represents the data that is really stored. It is the interpretation of the previos theory. We assme that the interpretation incldes text docments, databases, images and les at the same level as integers and reals. The activity level represents the se of environmental data by applications. At this level, we consider that both data (basic data B or derived data D) and data transformations (tf and tf2) can be searched and sed by dierent sers. For example, facts 0 are: Example 2. wind-data(, 50, ``Soth'') wave-data(, 2, ``Soth-East'', 5) Ths, in langage L 0,we represent data. These facts match the base relations wind-data(id, speed, direction) and wave-data(id, height, direction, period). 2.2 Modeling Metadata Now, we want to model the representation of the data. This means we want to represent some information abot the schema of a database. (In sch a case, we will be talking abot metadata that is localized on the left side rectangle of Figre.) We dene a rst-order logic called L. Following on or example, facts are: Example 2.2 (wind-data) attribtes(wind-data, id, Integer) attribtes(wind-data, speed, Integer) attribtes(wind-data, direction, String) (wave-data) attribtes(wave-data, id, Integer) attribtes(wave-data, height, Integer) attribtes(wave-data, direction, String) attribtes(wave-data, period, Integer). Throgh the and attribtes 0, we are modeling the strctre of the data which corresponds to the denition of part of the meta-schema. The notion of index can also be represented as. Indexing wave-data by direction, for instance, corresponds to facts 0 of the form: Example 2.3 direction-wave-data(``soth'', 209), where \209" is the identication of a wavedata record that has direction \Soth". In contrast to other denitions of index (that consider it

4 Predicates, e.g.: hierarchy of classes Predicates, e.g.: data transformation B tf B D tf2 D Activity level Predicates, e.g.: DB schema indexes Predicates, e.g.: location Time-series Maps Satellite Images Relational DB Docments Theory level Predicates 0 correspond to data facts Exectable programs Interpretation level /Metadata and Figre : and Metadata plans as metadata), by dening an index we are jst describing one more data property. Bt, if we want to add some knowledge abot its strctre, then again we have tomovetol and add the following facts: Example 2.4 (direction-wave-data) attribtes(direction-wave-data, direction, String) attribtes(direction-wave-data, waveid, Integer) indexes(direction-wave-data,wave-data,direction) These facts correspond to. The only dierence is that and attribtes are \nary" with respect to directionwave-data and indexes is a \binary" predicate with respect to direction-wave-data and wave-data, that are both 0. The former ones represent the strctre of the index and the later denes the index. At the activity level of Figre, classifying data within a hierarchy makes it easier for sers to search for interesting data sets. In or example, let s assme that both data abot waves and abot wind are considered indicators for weather forecasting. This can be modeled as facts : Example 2.5 indicators(wave-data) indicators(wind-data). Bt, we may also say that weather-forecasting is a more generic class of data that encloses indicators. This wold imply to dene the predicate: Example 2.6 weather-forecasting(indicators) same thing and avoid 2 is to reify indicators and weather-forecasting and treat them as constants. So, instead of Example 2.6, we se a new predicate sbtype to represent the following facts : Example 2.7 sbtype(weather-forecasting, indicators) sbtype(indicators, wave-data) sbtype(indicators, wind-data). From the analysis of the examples, notice that there are several at L that model aspects of the same data. For instance, wave-data properties are modeled,, throgh the :, attribtes, indexes and sbtype. Althogh they are all, we can distingish two levels of : those that are minimm describe the strctre of data (like and attribtes) and the others that add some more knowledge abot the properties of data (like indexes and sbtype). Minimm in L represent the minimm model needed to reconstrct the lower level, L 0. The additional describe some semantics associated to data. 2.3 Mixing and Metadata Another kind of metadata that we call demi-metadata [8] arises from the fact that we need to associate data vales to L 0. Referring to Figre, we are at the theory level. If we want to state that all data abot waves can be obtained from a certain URL, we will add the fact: Example 2.8 location(wave-data, that is a predicate 2, becase indicators is a predicate. A dierent way of describing the This fact is neither a fact 0 nor a fact becase it ses wave-data, a constant inl,

5 and a constant 0. To deal with this sitation, we need to go to L and location isafactinl. We have demimetadata every time we associate constant vales 0 to 0. Intitively, this happens each time we want to describe properties that are independent of the data itself. 2.4 Modeling Referring again to the activity level in Figre, we model the transformation of data by the signatres of the programs that accomplish these transformations. Independently of its type, data is classied as basic or derived. Basic data is transformed throgh data transformation activities and generates derived data. Transformation activities, in general, have inpt argments, retrn an otpt and correspond to the exection of a scientic model or to some hman interpretation. Each transformation activity, applied to some data (basic or derived) will prodce new data called derived data. In or example, wave conditions can be calclated from wind conditions throgh the application of a prediction model called predicted-wave. This corresponds to the following fact : Example 2.9 predicted-wave(wind-data, wave-data) and to the fact: Example 2.0 data-transformation(predicted-wave, wind-data, wave-data). -transformation models properties of a predicate (predicted-wave) and two constants in L. In other words, data-transformation ses a constant 2 and two constants, and ths belongs to L. The semantics of the particlar data transformation predicted-wave is modeled throgh its interpretation which isthe exectable program. L models the way the data transformation is applied. An alternative way of doing it wold be to reify predicted-wave and to consider it as a constant. The predicate data-transformation wold then be a predicate in L. We chose the rst approach ofbeingat L as it seemed a more natral representation. 2.5 Framework for classifying Metadata Throgh the presented seqence of examples, we were able to dene and characterize several formal types of metadata. Figre 2 illstrates a grid where the x axis contains the formal types of metadata (what is metadata) and the y axis represents the kinds of knowledge modeled by those types of metadata. The y axis has informal examples on the right side of the gre. Each point in the grid models Semantics Strctre Description Vales Example 2. Example Example 2.9 Example 2.0 Example 2.4 Example 2.7 Example 2.2 Example 2.4 Example 2.8 scientific models software programs simlation models indexes relationships between data sets: hierarchy of classes "schema" of the data set indexes attribtes of data sets like: location vales from a data set /Metadata Figre 2: Framework for classifying metadata corresponds to one or more examples given in this section. For each kind of informal data/metadata in y, one nds the formal way to represent it as data/metadata in the x axis. For example, the collection of indexes, i.e., the predicate indexes,provides data semantics abot the relationship between data elements. This predicate is classied as data semantics on the y axis and as a predicate on the x axis. As another example, the strctre of each index is also modeled, i.e., and attribtes. These are classied as data strctre on the y axis and as on the x axis. We will se the same grid-based framework to explain when the dierent types of metadata shold be sed in the environmental data prodction and management and to apply the classication against metadata standards. 3 Application of the framework to environmental data prodction Figre 3 smmarizes the activities that spport the environmental data management and prodction process [8]. In this gre, environmental data prodced in data sorces is docmented and pblished by data prodcers. This way, sers (data consmers) locate and extract the appropriate data sorces to answer their reqirements. consmers access and se the data to eventally prodce new data or to aggregate it at a higher level of abstraction. Users can also analyze the retrieved data and draw conclsions on its tility orinter- pret it and generate new \vale-added" data. that is prodced as a reslt of ser interactions is also pblished. These activities are accomplished throgh the se of metadata. Or prpose in this section is to analyze the types of metadata that havetobesed dring each of the activities: locating, extracting, aggregating and analyzing data. For this analysis, we se the procedre described in Section 2.

6 Consmers Locating Extracting 4 Application of the framework for classifying metadata standards Aggregating Pblishing Prodcers The procedre followed for classifying the elements of metadata formats and models, according to or grid-based framework, is: Given a part of a model or standard: Analyzing Figre 3: Environmental data prodction and management activities models Aggregation Aggregation model it in a rst-order logic classify it according to the types of metadata in the x axis look, in the y axis, for the prpose of that fragment of metadata ll in a given (x y) coordinate. Semantics Strctre Description Aggregation Extraction Analysis We will apply or framework against the UDK metadata model and the FDGC metadata format sing this procedre. Firstly, we will give a brief overview of each of these approaches. Then, we will follow the classication procedre to characterize the associated metadata. Vales Analysis 0 /Metadata Figre 4: Use of metadata dring data prodction and management activities To locate a certain data set, we may se several types of data. We can se a description of the data set to locate data on a local server (for example an attribte called location that has an URL assigned to it). This location qery is modeled by location(x y) ^ localserver(y) which isa well formed formla. Ths, since the locating activity sesthelocation predicate which isa data description predicate, this activity isclassi- ed as L on the x axis and data description on the y axis. Other locating activities se an appropriate index over the data set (data strctre and data semantics over L ) or the sal transformation that is sed to obtain it (data transformations over L ). To extract data, one mst be aware of its strctral details represented by. The aggregation of data is done throgh the exection of data transformations by sing and L. analysis is needed in order to evalate the tility of data. For that, one may se descriptors of data (like the description of the measrement techniqes sed to obtain it) that are represented as or samples of data that are represented as facts 0. Figre 4 smmarizes this classication. 4. UDK model UDK (Umwelt-Datenkatalog) [4] or Environmental Cataloge is a meta-information system and navigation tool that docments collections of environmental data prodced by the German states and other sorces. Three types of objects are distingished in UDK: environmental objects, environmental data objects and UDK (metadata) objects. A (real-world) environmental object corresponds to a physical entity like a river and is described by a collection of environmental data objects. Each environmental data object is associated with UDK (metadata) objects that describe its format and contents. Environmental data objects can be organized throgh a hierarchy of classes with inheritance of attribtes. There are seven important classes: project data that corresponds to environmental impact stdies, empirical data that incldes measring series, data abot facilities (which factories or bildings are involved), maps, expertises and reports, prodct data and model data that corresponds to simlation reslts. Orthogonally, UDK metadata objects and environmental data objects are linked throgh semantic graphs called cataloges. These cataloges describe partof and responsible of relationships between metadata objects and environmental data objects. The procedre followed for identifying and classifying UDK metadata is the one described above. By sing it, we will be able to represent L 0, L, L and L for the UDK model. The examples sed here are based on reference [5]. An environmental data object describes, for instance a set of measrements made to captre the concentration of oxygen in a river. This corresponds to the fact 0 :

7 Example 4. oxygen-concentration(``thames'', 50, Jan/97) where the constants stand for the river name, the oxygen concentration level and the sampling date. UDK environmental data objects belong to classes and each class has a set of attribtes. Facts correspond to instances of classes. So, oxygenconcentration is an environmental data class. It is represented by the following : Example 4.2 environmental-data-classes(oxygen-concentration) attribtes(oxygen-concentration, river-name, String) attribtes(oxygen-concentration, conct-level, Integer) attribtes(oxygen-concentration, date, Date) These facts describe the strctre of data. To state that oxygen-concentration is a sbclass of the class measrements and that measrements is a sb-class of empirical-data, we can se the following facts : Example 4.3 sb-class(empirical-data, measrements) sb-class(measrements, oxygen-concentration) that reslt from reifying empirical-data and measrements as constants. The sbclass predicate describes semantics of data. UDK metadata objects add information abot environmental data objects and they can exist at several levels of aggregation of environmental data objects. For example, it may be sefl to store which is the name of the organization responsible for measring the concentration of oxygen. This corresponds to the fact : Example 4.4 responsible(oxygen-concentration, ``Company OC'') if wewant to state that \Company OC" is the responsible for all measrements of oxygen concentration. Bt we will be also talking abot a UDK metadata object if we say that \Company OC"is the company that is responsible for the measre of oxygen concentration in the Thames river. And this corresponds to the fact 0 : Example 4.5 oxygen-concentration(``thames'', 50, Jan/97, ``Company OC'') This means that UDK metadata objects are mapped into 0 and. Both of them correspond to data description. Figre 5 resmes the classication that we have done by example. models Semantics Strctre Description Vales UDK Env. Objects (Example 4.) 0 Class hierarchy of UDK Env. Objects (Example 4.3) Classes of UDK Env. Objects (Example 4.2) UDK Metadata Objects (Example 4.5) UDK Metadata Objects (Example 4.4) /Metadata Figre 5: Classication of UDK model metadata 4.2 FDGC metadata format The FGDC (US Federal Geographic Committee) standard [6], as it is common known,isin fact called Content Standards for Digital Geospatial Metadata (CSDGM) and it was approved by FDGC in 994. This standard is composed by a set of terminology and denitions described by metadata descriptive elements. Its main prpose is to help to determine the availability, sitability and means to access to geospatial data (althogh it can also serve to other types of environmental data). The CSDGM metadata elements are organized in seven grops according to the information they provide: Identication, Qality, Spatial Organization, Spatial Reference, Entity and Attribte, Distribtion and Metadata Reference. A short example, taken from the standard, of the syntax and semantic description of the Identication grop is: Syntax: Identification_Information = Citation + Description +... Citation = Citation_Information Description = Abstract + Prpose +... Citation_Information = {Originator}n + Pblication_Date +... Semantics: Citation - information to be sed to reference the data set. Compond. Originator - the name of an organization or individal that developed the data set. If the name of editors or compilers are provided, the name mst be followed by "(ed.)" or "(comp.)" respectively. Type: text Domain: "Unknown" free text Pblication_Date - the date when the data set is pblished or otherwise made available for release. Type: date Domain: "Unknown", "Unpblished material" free date Description - a characterization of the data set, inclding its intended se and limitations. Compond. Abstract - a brief narrative smmary of the data set. Type: text Domain: free text Prpose - a smmary of the intentions with which the data set was developed. Type: text Domain: free text...

8 For classifying the associated metadata, we will follow the same procedre as we did for the UDK model. Let s sppose that one wants to store satellite images from beaches. The element Originator from the Identication grop of metadata, makes a correspondence between the data set (or a particlar data element) and a data vale. It is represented as a predicate (or by a predicate ): models Semantics Strctre FGDC Entity and Attribtes (example 4.7) Example 4.6 originator(beach-image, ``Unknown'') beach-image(``miami Beach'', X0FA6, ``Unknown'') other FGDC grops of metadata Description (example 4.6) FGDC Metadata Reference (example 4.7) Vales other FGDC FGDC Metadata grops of metadata Reference (example 4.6) (example 4.7) where X0FA6 is the identier of the image. The Entity and Attribte grop of metadata describes the entities, the attribtes and the attribte domains of the data set. This corresponds to the conceptal schema of a database. It is then represented by facts,like: Example 4.7 entities(beach-image) attribtes(beach-image, image-id, Integer) attribtes(beach-image, image, Image) The other distinct grop of metadata is Metadata Reference that describes the other metadata grops. We may want to store the freshness of metadata elements. For instance, the freshness of Originator is represented by the fact: Example 4.8 freshness(originator, Janary/97) which is a fact (if originator was considered as a fact ) or is a fact 0 (if originator was considered as being embedded in a fact 0 ). Entity and Attribte metadata elements describe data strctre. Metadata Reference elements still describe data, even thogh they describe metadata. The rest of the FGDC Metadata grops add some information abot data so they describe data. The whole metadata classication is illstrated in gre 6. 0 /Metadata Figre 6: Classication of the FGDC metadata standard References [] Jon Barwise. Mathematical Logic. North Holland, 977. [2] Kate Beard. A Strctre for Organizing Metadata Collection. In Third International Conference/Workshop Integrating GIS and Environmental Modeling, Santa Fe, New Mexico, USA, Janary FE CD- ROM/sf papers/beard kate/metadatapaper.html. [3] Centre for Earth Observation Programme of the Eropean Commission. Recommendations on Metadata, May997. [4] Oliver Genther and Agnes Voisard. Metadata in Geographic and Environmental Management. In W. Klas and A. Sheth, editors, Managing Mltimedia : Using Metadata to Integrate and Apply Digital. McGraw Hill, 997. [5] Olivier Genther. Environmental Information Systems. Springer-Verlag, 998. [6] NSDI. The Federal Geographic Comittee Home Page. Jne 997. [7] Eric Simon. Personal Commnication, December 997. [8] Anthony Tomasic and Eric Simon. Improving Access to Environmental sing Context Information. SIGMOD Record, Conclsions To smmarize what was done throgh this paper, we rstly bilt, by-example, a formal classication of metadata based on rst-order logic. We identied three types of metadata (, and ). Throgh a grid we dened exactly the meaning of these kinds of metadata. We also identied the types of metadata, from the framework, that are sed throgh the environmental data prodction and management. Finally, we applied the framework against an environmental metadata model and a standard format and dened a procedre for classifying them.