“Metadata DownUnder” Moving ISO & OGC standards into the Semantic Web Presented at “Metadata DownUnder” 11th Open Forum on Metadata Registries, Sydney,

Moving ISO & OGC standards into the Semantic WebPresented at “Metadata DownUnder”“Metadata DownUnder”11th Open Forum on Metadata Registries, Sydney, NSW, Australia

Laurent Lefort

22 May 2008

Water For a Healthy Country

CSIRO Moving ISO TC 211 & OGC standards into the Semantic Web “Metadata DownUnder”: 11th Open Forum on Metadata Registries Sydney, NSW, Australia

Outline

• Ontologies and water data standards• Transforming ISO TC 211 and OGC

standards into ontologies• Work on OWL versions for multiple standards

• Findings on the transformation methods• Findings on the resulting ontologies and on how to build better ontologies

• What is the added value of Semantic Web technologies?


Context: water resources management for Australia

Water Resources Observation Network (WRON) program

• One of CSIRO’s Water for a Healthy Country Flagship themes

• Support to major research alliance between Bureau of Meteorology & CSIRO (WIRADA) to deliver mission-critical R&D

Specific Activity on Water Data standards

Hydrometricdata

Geospatialdata

Usage andentitlement

data

Models

Source: Vertessy 2006: Australia’s water resources information imperative and the role of the Water Resources Observation Network (WRON)


Interest in water data standards

Water budget combining 4 sub-domains• Atmospheric Water (& Climate)• Surface water• Groundwater (& Geology)• Human use of water

Need to manage features and observations

Complex cross-domain interactions • e.g. transfer between surface water and groundwater

Need for a consistent standard basis (& method)• Data and Metadata


Generations of “standards” & integration complexity

ASCII-based

DB-based

Registries

XML

Model-driven generation of XML schemas

Custom XSL transfo. & web services

Distributed systems with same db schema

UML & XML schemas

XML schemas Reuseable XML schema stack

Master Data Managt

OWL ontologies Semantic integration

EPA STORET

EPA WQX

GWML

WOML

WFD schemas

eWater (EU)

SANDRE

SANDRE XML

Surface water & groundwater “standards”

Integration support

Stan

dard

use

rsSt

anda

rd d

evel

oper

s

ODM

WaterML (CUAHSI)


Semantic Web technologies for standards development

• RDF (Resource Description Format) for the “web of data”: annotations and links

• Value: flattened, web-compatible method to manage and link data into set of triples

• OWL (Web Ontology Language) for the web of (data) models

• Several variants based on description logic with different expressivity / scalability ratios

• Value: reasoning support to build class hierarchy and verify logical consistency


Building expectation that OWL can be useful

• Past and present efforts to create and use OWL versions of standards

• Drexel University (HydroSeek) • Uni of Muenster (ACE-GIS, SWING, EDINA)

• Discussions at the Water Resources Information Model Workshop (Canberra, Sep 2007)

• Recognition of the ontological value of some standards e.g. OGC Observations and Measurements

• Finney: Australian Marine Ontology, WALIS Forum 2008• Brodaric & Probst: DOLCE Rocks AAAI Spring Symp. 2008


Reasons to share our experience in building ontologies

• High demand for OWL versions of standards• Transition and ramping up period from a manual

process to a semi-automated one• Recently developed methods (ODM) and tools

(TopBraid) to create ontologies from UML models or from XML schemas

• Re-evaluation of current standard development practice

• Push for harmonisation of spatial standards (INSPIRE)

• Development of OGC Model driven approach• ISO 19150 Ontology group, led by Jean Brodeur

• Can SW help ISO TC 211? Can ISO TC 211 help SW?


What we present today

• Work on OWL versions for multiple standards• ISO and OGC standards• Standards based on ISO and OGC standards defined for the

water domain

• Findings on the transformation method• Comparison of ontology generation tools from UML models

and from XML schemas

• Findings on the resulting ontologies• Tactics to build better ontologies


Standards to transform into OWL

•Focus on water standards describing Features & Observations because of our interest in:• Reference datasets (continental scale)

• Identification of water features and of their topological and hydrological relationships

• Data exchange language for individual and aggregated observations


Two key building blocks to build features and observations standards

• ISO 19109: Geographic information -- Rules for application schema

• Defines a method to specify features know as the General Feature Model (GFM)

• OGC Observations and Measurements (O&M)• Refines the GFM method to manage observations

• Supported by common schema generation technologies (UML to XML schemas)

• To implement UML patterns out of “stereotypes”• To create definitions on top of existing schemas• Example of tools: ShapeChange, FullMoon (CSIRO)


Common principles used for standards based on GFM and O&M

• A core model defines the main classes forming the standard

• Through their relation to other specified classes or to generic spatial definitions

• Extra design flexibility is given in three areas • Attachment of properties to features, • Introduction of externally managed code lists • Provision for alternative usage (union)

• Specific restrictions on the applicability of the definitions can be added with a constraint language, such as Schematron


Added value of the Observations & Measurements standard

• Two user-managed class hierarchies in GFM-based specs:

• Feature and FeaturesCollection: a Feature-type is characterized by a specific set of properties

• Up to five user-managed class hierarchies in O&M-based specs

• Observation, SamplingFeature, PropertyType, Procedure and Result

• An Observation is an Event whose result is an estimate of the value of some Property of the Feature-of-interest, obtained using a specified Procedure

• Stronger ontological value for O&M• More branches and separation of concern: • Example: Difference between Feature and SamplingFeature

• Feature for the real world objects e.g. an aquifer• SamplingFeature to characterise how a measure is done e.g.

along a borehole


Standards transformed into OWL

• Application standards based on • ISO TC 211 General Feature Model • OGC Observations & Measurements

• Corresponding ISO/OGC standards from two origins:

• UML model grouping all the ISO TC 211 standards

• from the Harmonized Model Maintenance Group• XML schemas from OGC (schemas.opengis,net)

• Including GML, SensorML, …


Selected ontology generation methods

• XSL-based approaches• XO (CSIRO-developed) from UML 2.0 or XML

schemas to OWL• Rhizomik.net xsd2owl.xsl (open source but

restricted to non commercial usage)

• TopBraid Composer (commercial tool)• Transformation from UML 2.0 and XML schemas

to OWL• Enterprise Architect files can be pre-processed with an

EA-specific openArchitectureWare plugin


Generated ontologies

OpenGIS schemasGML, SensorMLSOS, WFS, WCSOther schemas

GeoSciML, WML, CSML

Rhizomik xsd2owl or

XO

TopBraid or XO

EA modelsHMMG

HollowWorldGeoSciML

GWML, WOMLCSML

DHS-GDM

one ontology file per source

except for TopBraid

Pre-processing to regroup all the schemas

Pre-processing to UML2.0

Generated ontologies


Example 1: om:Observation from XML schemas (TopBraid)


Example 2: om:Observation from UML model (TopBraid)

• Long URIs based on package names• hasFeatureOfInterest:

<http://ogc.uml/Model/Model/Externally-governed-packages/HollowWorld/CommonUsagePackages/ISO-19110/ISO-19115-Metadata/Metadata-entry-set-information/MD_Metadata>[0..1]


Example 3: om:Observation from UML model (XO)


Example 4: om:Observation from XSD (Rhizomik)


Important findings

• XML schemas easier to transform than UML models• As long as the transformation tool is capable to process tricky

xsd:include and xsd:import cases

• Modularity schemes in place for UML or XML schemas are not necessarily directly applicable in OWL

• Suggested alternative is to simply use the XML namespace scheme to group together schemas sharing the same namespace into one or a limited number of modules

• The method to define URIs works better with XML schemas than with UML models

• XSL-based approaches better handle low quality (or incomplete) UML input

• Known problems with UML/XMI files


XML schemas easier to transform than UML models

• UML models• High variability in the usage of stereotypes• Risk of problem if the UML model is not fully validated or

messy

• XML schemas• Availability of validation tools even for multi-part schemas• Less work to interpret the modelling intent • Always available directly or after generation from UML• Tighter management of successive versions

• Being able to generate the same output from both types of input for the same standard is critical to strengthen the transformation process


Modules (files), namespaces (prefix) and URIs (IDs) in OWL

Sub-Module 1filename gml.owl

Namespace: gmlhttp://opengis.net/gml

URI for each classes (and properties)

http://opengis.net/gml#AbstractObservation

Top Module

owl:import modules

Sub-Module 2filename om.owl

Namespace: omhttp://opengis.net/om

URI for each classes (and properties)

http://opengis.net/om#Observation

Difference with XML: can not have same namespaces in different modules


Method to define ontology modules

• OMG ODM recommendation to replicate UML package is inapplicable in our view

• Too many modules and the wrong ones • TopBraid’s UML import operation creates 184 OWL files for

the O&M model (which includes the ISO TC 211 standards)

• Recommendation• For XML schemas, group together schemas sharing the

same namespace into one or a limited number of modules• Define a method producing the same results for UML

models• Record the source module or schema as an annotation

property for traceability and/or round trip purposes


Method to define URIs

• Using UML package names to create URIs is not recommended

• See example 2

• Keeping the original XML schema namespace works well in practice

• Maybe two generation options are needed • To create separate definitions for different versions of the

same source• To merge definitions from different versions of the same

source


Three central issues with the current OMG ODM specification

• Modules derivation from packages • Impossible to apply in practice (too many

modules)

• Naming conventions to disambiguate property names

• Can lead to an explosion of the number of properties often not required

• Does not discuss the union & substitution group patterns which are widely used in ISO/OGC standards


Building better ontologies

•Assumption• Ontologies can help standards amateurs to understand them without reading the documentation or learning how they have been created

•This discussion• Tactics to capture the semantic essence of ISO/OGC & derived standards


Tactic 1: Stick to the original definitions

•Rendition of ISO standards which mirrors the original UML model • Drexel University team

• ISO and OGC ontologies in OWL-Protégé 2.1 (2004-05)

• ISO 19103, 19107-12, 19115, OGC Spatial referencing by Coordinates and GML

•Success factor: traceability to the origin of definitions (often overlooked)


Tactic 1: benefits of traceability

• Handle multiple definitions of Observations• OM1_Observation: published OGC O&M spec. (part 1)

version 1.0 • OM: Draft version of O&M• GML: gml:Observation


Tactic 2: Modularise, Winnow, Align with Upper Ontology

• University of Muenster (and EU projects partners)

• ACE-GIS: OWL-Protégé 1.2 (2004), SERES: OWL-Protégé 2.2 (2005), SWING: WSML (2008)

• Spatial representation (19107), Location (19111-19112), O&M (alignment with DOLCE and SWEET)

• Generally based on a costly manual process • Match what the end user wants• Weaker traceability to the sources of definitions


Tactic 3: Try to do both

• Replace the manual process by a smarter transformation designed to normalise the ontology skeleton

• Define the right branches at the top• Isolate unambiguous primitives (e.g. units)• Use modules/namespace/URIs to position source-

specific definitions against common ones

• Specific effort needed to• Reduce the number of root classes• Create deeper class & property hierarchies• Handle ambiguous property definitions


Example of normalised ontology skeleton

Define the right branches at the top

Isolate unambiguous primitives (e.g. units)

Use modules/namespace/URIs to position source-specific definitions against common ones


Conclusions

• Better ontology generation tactics can help to satisfy the demand for OWL versions of (groups of) standards

• Three priority areas have been identified• Systematically develop parallel transformation

chains from UML and XML schemas to enable cross-checking of outputs

• Develop more convenient and more robust modularity, namespace and URIs schemes

• Give feedback to ISO/OGC Policy group on the compatibility of their approach with OWL


Inputs for ISO 19150 Ontology group

• Can SW help ISO TC 211 (and OGC)? • Modelling and reasoning power of OWL

• Sub-properties in v. 1.0 and role composition in v. 2.0• Top level class hierarchy skeleton: normalised form of

ontologies, alignment to upper ontologies

• Can ISO TC 211 (and OGC) help SW?• Method to define a standard as a derived product

of another one• Transposable experience on how to extend or restrict a

specification• Use cases to inform SW work on ontologies and rules


Acknowledgements

Thanks to:• Ross Ackland, WRON Theme Leader, CSIRO• Simon Cox, Research scientist, CSIRO and OGC• Amit Parashar, CSIRO and Australian W3C office

And also to:• TopQuadrant for TopBraid Composer • Rhizomik.net for xsd2owl.xsl• Rick Jelliffe et al: XSL pre-processing of XML

schemas

CSIRO ICT CentreLaurent LefortSenior Research Engineer (Ontologies)

Phone: +61 2 6216 7046Email: [email protected]: wron.net.au

Contact UsPhone: 1300 363 400 or +61 3 9545 2176Email: [email protected] Web: www.csiro.au

Thank you

Backup slides


Standards

• GeoSciML (Geoscience Markup Language) • GFM-based, first standard to partially leverage O&M

• GWML (Groundwater Markup Language), WOML (Water Observation Markup Language)

• Two preliminary efforts based on O&M to create groundwater and surface water standards:

• CSML (Climate sciences Modelling Language) • Adapting & completing O&M for Met/Ocean data

• DHS-GDM (Department of Homeland Security Geospatial Data Model)

• Huge compilation of standards for homeland security applications


Summary of the 4 methods

Transfo.method

URI managt

Modules based on

OWL variant

Comments

TopBraid XSD

Namespace Namespaces using specific conventions

DL Best result in gal

TopBraid UML

Package too complex URIs

Package too many modules

DL Import not always successful

Rhizomik XSD

Unique Unique EL+ Adapted to handle multi-part schemas

XO UML Unique Unique EL+ Stereotypes (esp. Unions)

Documents

“Metadata DownUnder” Moving ISO & OGC standards into the Semantic Web Presented at “Metadata DownUnder” 11th Open Forum on Metadata Registries, Sydney,