1. Page 1 LAITS Laboratory for Advanced Information Technology and Standards Ontology-driven Automatic Web Service Composition for Geoscience Automation Liping Di Center for Spatial Information Science and Systems (CSISS) George Mason University 6301 Ivy Lane, Suite 620 Greenbelt, MD 20770 ldi@gmu.edu 301-982-0795

2. Page 2 LAITS Laboratory for Advanced Information Technology and Standards Introduction Geospatial data is the major type of data that human beings has collected. –more than 80% of the data are geospatial data. Image/gridded data is the dominant form of geospatial data in terms of volume. –Most of those data are collected by the Earth Observing (EO) community. Geospatial data will grow to ~exabyte very soon. –NASA EOSDIS has more than two petabytes of data in archives and is added more than 2 terabytes of new data per day. –Application data centers: 10’s of terabytes of geospatial data. –Tens of thousands of datasets on-line now. Geospatial data has to be converted to user-specific information and knowledge before they become useful. How to effectively, wisely, and easily use the geospatial data is the key geospatial technology issue that we have to solve.

3. Page 3 LAITS Laboratory for Advanced Information Technology and Standards The Problems Currently, the conversion from Geospatial data to information and knowledge requires the users at their local site having: –Significant amount of knowledge Domain knowledge for information/knowledge extraction from raw data Domain knowledge on the geospatial data processing/formats –Significant amount of computer hardware and software resources. –As a result, the use of geospatial data is currently very expensive Most geo-imagery will never be directly analysed by humans –Human attention is the scarce resource, insufficient to analyze petabytes of geospatial data. –Many datasets never have been analysed before they are archived. The fundamental problem is that current data and information systems at data providers’ site can only provide on-line ordering, and access at best, of data and standard products, not the user-specific information and knowledge. –If a product that the user want does not exist, the system has no way to create it on demand. –Rich in geospatial data but poor in up-to-date geospatial information and knowledge. Instead of only being used by experts, geospatial data should also be used easily by other people, from students to decision-makers, and be transformed easily into higher level information and knowledge.

4. Page 4 LAITS Laboratory for Advanced Information Technology and Standards Ultimate Goal-Intelligent System for Geoscience Automation An intelligent geospatial knowledge system for automating the process from geospatial data to information to knowledge based on users’ requirements –choreographed intelligent web services for geospatial knowledge discovery; –the dissemination of personalized, on-demand geospatial data, information and knowledge. The system can answer many “what if” questions through –Automatic service composition: intelligently chaining individual service modules to form a complex geospatial model; –matching the input data with the model; –executing the model to deliver the answer to the users.

5. Page 5 LAITS Laboratory for Advanced Information Technology and Standards Geospatial Web Services and Service Composition Geospatial Web Service –A modular Web application that acts on geospatial data, information, or knowledge. Composition of Geospatial Web Services –Assembling individual geospatial Web services to create high-level geospatial processes. Geospatial web services are changing the way in which geospatial applications are designed, developed, deployed and executed. –From centralized to distributed, from isolated implementation to large-scale sharing, and from private solution to standardized solution

6. Page 6 LAITS Laboratory for Advanced Information Technology and Standards “Intelligent” Mechanism for Composition Facilitating information discovery and integration over the network in web service environment. Automating the assembly of service chains. –Involving Universal Description Discovery and Integration (UDDI), Web Service Description Language (WSDL) and Simple Object Access Protocol (SOAP) –Using AI Planning method –Driven by geospatial domain knowledge expressed in ontology

7. Page 7 LAITS Laboratory for Advanced Information Technology and Standards Solve complex geospatial problems by using distributed geospatial services: The Geo-object and Geo-tree Concepts Archived data (geo-object) User (requested) geo-object Intermediate geo-object Without serviceWith service Modeling and virtual data services User request User received Data transformation services (format transformation, reprojection, regriding, etc) Data access services (spatial/temporal/parameter subsetting, mapping, etc) High level services (classification, geophysical parameters, modeling, etc)

8. Page 8 LAITS Laboratory for Advanced Information Technology and Standards “Geo-object” concepts Geo-Object –A granule of geospatial information. –Consisting of data itself, a set of data attributes (data metadata), and optionally a set of description of methods that can act on it (service metadata). Archived Geo-Object –Physically exists in a data system or archive Virtual Geo-Object –A geo-object that doesn’t physically exist but an information system knows how to create it on-demand. User Geo-Object –a geo-object that user wants.

9. Page 9 LAITS Laboratory for Advanced Information Technology and Standards “Geo-Tree” Model “Geo-Tree” –A composition of chained services and geo-object to enable the process of creating “User Geo-Object” from “Archived Geo-Objects” and “Virtual-Objects”. –The service chain represents a geospatial process model Geospatial service chain where, for each adjacent pair of services, the output of the first service is part of the inputs of the second service. –It represents the geospatial process knowledge The intelligent geospatial web services are all about how to automatically create the geotree that can generate the user geo- object. –How to do that: through AI and semantic web technologies

10. Page 10 LAITS Laboratory for Advanced Information Technology and Standards Relationship between Geo-tree and Service Chain Geo-tree is a conceptual geospatial process model representing the step-by-step how a geo-object (representing either GI or GK) is derived. –representing the domain knowledge needed for analyzing geospatial data to derive user-specific GI or GK. –A library of Geo-trees represents the enterprise/community knowledge since it is contributed by the members of them. The root of a geo-tree is a virtual product that the Geo- tree can derive, and all sub-trees are also virtual products. A service chain is the instantiation of the Geo-tree. If a Geo-tree can be instantiated, then all virtual products in the tree can be produced on demand.

11. Page 11 LAITS Laboratory for Advanced Information Technology and Standards Automatic “Geo-Tree” Composition Composition-AI planning problem –Goal: “User Geo-Object” –Initial state: “Geo-Object” availability status –Planning operator: geospatial service –Dependence and constraints: precondition and effect of service –Searching: backward Composition Ingredients –Semantic description of Services and Data Supporting effective “Geo-Object” discovery, seamless resource integration and reuse. –Domain knowledge Ontology to describe the relationships among services, data, and the association between service and data Creating new “Geo-Object” and suggesting what should be done following the new “Geo-Object” and which “Geo-Object” should be chosen next. “Geo-Tree” Representation –BPEL4WS –OWL-S

12. Page 12 LAITS Laboratory for Advanced Information Technology and Standards Modeling “Geo-Object” with Semantics (1) Semantic “Geo-Object” –Geospatial Ontology Conceptualization of geospatial domain. Including Service Ontology and Data Ontology –OWL-S A OWL-based Web service ontology –“profile” describes what a geospatial web service does by specifying its input and output, preconditions, effects and other non-functional properties, –“process” describes how the service works by specifying either atomic process or composition process, –“grounding” describes how service is invoked by specifying the details of communication protocol.

13. Page 13 LAITS Laboratory for Advanced Information Technology and Standards Modeling “Geo-Object” with Semantics (2) With semantics, we can –Reason about the functionalities that service performs –Reason about preconditions of using the service and effects after using the service –Understand the relationships and rules of geospatial computing –Understand the meanings and order of exchanged messages –Compose “Geo-Objects” to create a new more complex “Geo- Object”. The key components are discussed in the following slides

14. Page 14 LAITS Laboratory for Advanced Information Technology and Standards Service Ontology Group service types into hierarchy based on scientific meaning of the services. –For example, Landuse classification service type has many subtypes, include IGBP_classification, Olson_classification, etc, Define the relationship between the service types. Each service instance has to declare which service type it belongs. It is essential for model construction, accurate discovery of service instances, automatic/semi-automatic service chaining, and both model and service interoperability.

15. Page 15 LAITS Laboratory for Advanced Information Technology and Standards Data Ontology Group data types into hierarchy based on scientific meaning of the data. Define the relationship between data types Each data instance has to declare which data type it belong to. It is essential for model construction, accurate discovery of data instances, automatic/semi-automatic service chaining, model interoperability and match of services with data.

16. Page 16 LAITS Laboratory for Advanced Information Technology and Standards Ontology-enabled Catalog/Registry Service Catalog/Registry Service –Playing a ‘directory’ role “Geo-Objects” and associated services advertise the availabilities of their resources using metadata in the catalog users can query the metadata in the catalog to discover interesting resources. –OGC Catalog Service for Web (CS/W) ogcRIM (Registration Information Model) –Based on ebRIM –Supporting the publish, discovery and access of geospatial information

17. Page 17 LAITS Laboratory for Advanced Information Technology and Standards Ontology-enabled Catalog/Registry Service Add-on Ontology –Adding an ontology layer on top of the ogcRIM to support semantic matching. –Assuming that the output of a geospatial service is a Geo-Object Go and the request is R, the “flexible semantic matching” can be exact, if Go = R, then Go and R are exact equivalent; plug in, if Go subsumes R than Go could be plugged in place of R; subsume, if R subsumes Go, then Go just completes part of R and R needs other Go to implement the other part of R or whole R; fail, there is no relationship between Go and R.

18. Page 18 LAITS Laboratory for Advanced Information Technology and Standards “Geo-Tree” Composition Composition Process ① Finding a “Geo-Object” in the archive that matches the user Geo-Object R through the catalog/registry service. ② If “yes”, let such one be R. If not, do step 3. ③ Find a service Si that can create such a “Geo-Object”. If not, the requested “Geo-Object” cannot be produced. ④ If “yes”, plug the service into the “Geo-Tree” and instantiate its inputs, outputs, preconditions and effects based on domain knowledge. ⑤ Lets assume the service Si requires n “Geo-Obejct” inputs, labeled as I1, I2,.. In. Check if all inputs of Si are available. If an input In is not already available, go to step 3. ⑥ Iterate the above process until all inputs are available.

19. Page 19 LAITS Laboratory for Advanced Information Technology and Standards Steps for getting user’s query answered ① User queries the system using natural language. ② The system converts the query into a formal description of the user geo-object R. ③ Find a “Geo-Object” in archives that matches the user Geo- Object R through the catalog/registry service. ④ If “yes”, answer to users query is pre-exist. Return the answer to the user. If not, do step 5. ⑤ Find a service Si that can create such a “Geo-Object”. If not, the requested “ User Geo-Object” cannot be produced. ⑥ If “yes”, plug the service into the “Geo-Tree” and instantiate its inputs, outputs, preconditions and effects based on domain knowledge. ⑦ Lets assume the service Si requires n “Geo-Obejct” inputs, labeled as I1, I2,.. In. Check if all inputs of Si are available. If an input In is not already available, go to step 5. ⑧ Iterate the above process until all inputs are available. ⑨ Execute the Geo-tree to create the product for the user.

20. Page 20 LAITS Laboratory for Advanced Information Technology and Standards Implementation Architecture

21. Page 21 LAITS Laboratory for Advanced Information Technology and Standards Providing an overall knowledge about the system and answering relevant queries. Implementation Architecture Intelligent user interface that assists the user in formulating queries and results Providing semantic “match-making” Producing a “Geo-Tree” using AI planning Algorithm Translating a abstract workflow into a concrete execution plan. Wrapping the underlying atomic or composite geospatial services and advertises their capabilities

22. Page 22 LAITS Laboratory for Advanced Information Technology and Standards Natural Language Query Ask a Geosptial Question using the English Language What is the possibility of landslide in the Dimond Canyon, California, United States on January 10, 2005?

23. Page 23 LAITS Laboratory for Advanced Information Technology and Standards Parse the Question using Components from Snark * Understanding the query * Snark is developed by SRI, http://www.ai.sri.com/tutorial.html (Stickel, Waldinger, Chaudhri)http://www.ai.sri.com/tutorial.html

24. Page 24 LAITS Laboratory for Advanced Information Technology and Standards Form the user geo-object described by XML Derive standard XML Description from the Snark Parsing Results

25. Page 25 LAITS Laboratory for Advanced Information Technology and Standards Construction of Geo-Tree/Service Chain Construct Composite Service Model in OWL-S

26. Page 26 LAITS Laboratory for Advanced Information Technology and Standards Execution of the Chain to Obtain Answer Execute the Composite Model and Obtain the Answer

27. Page 27 LAITS Laboratory for Advanced Information Technology and Standards Automatic Modeling Process (6) The Resultant Landslide Susceptibility Map

28. Page 28 LAITS Laboratory for Advanced Information Technology and Standards RedImage LandslideSusceptibility SlopeAspectNDVI Landcover (WICS) NIRImage Training Image ETMImage DEM Training (WICS) WCS DEM Slope Aspect Landcover NDVI Training Parameter Training Parameter ETMImage NIRImageREDImage LandslideSusceptibility Training Image ETMImage The Data and Service Involved in this Example

29. Page 29 LAITS Laboratory for Advanced Information Technology and Standards Data Type Definition Based on GCMD Science Theme

30. Page 30 LAITS Laboratory for Advanced Information Technology and Standards Service Type Definition Based on GCMD Service Types

31. Page 31 LAITS Laboratory for Advanced Information Technology and Standards The OWL-S Engine Architecture  Semantic Schema: ServiceType, DataType, Association  OWL-S library: OWL-Ss for Available Services  Registry: Catalog Service for Web  Reasoning/Matching: ServiceType, DataType  OWL-S Manager: OWLSPower

32. Page 32 LAITS Laboratory for Advanced Information Technology and Standards The OWL-S Match Flow  Data type driven reasoning currently implemented  Reasoning process facilitated by ServiceType/DataType associations  Four types of matches implemented: exact, subsume, relaxed, and all

33. Page 33 LAITS Laboratory for Advanced Information Technology and Standards Automatic Service Chaining Resultant service chain managed in two approaches: OWL-S engine and BPEL engine

34. Page 34 LAITS Laboratory for Advanced Information Technology and Standards Three major components in the XML:  Temporal information: Convert Snark temporal keywords to the UTC format as noted in http://www.w3.org/TR/NOTE-datatime.  Spatial information: Derive latitude and longitude bounding box from Alexandria Digital Library (ADL)* Gazetteer Protocol.  Object ontology: Combine Wordnet* with OWL GeoData description to generate ontology- represented GeoObject. http://www.laits.gmu.edu:8099/QuestionProcess/nl/nl_index.jsp Convert User Question to XML * ADL is developed by a project team headquartered at University of California at Santa Barbara **WordNet is developed by the Cognitive Science Laboratory at the Princeton University

35. Page 35 LAITS Laboratory for Advanced Information Technology and Standards Convert User Question to XML (Cont.) Temporal Information

36. Page 36 LAITS Laboratory for Advanced Information Technology and Standards Spatial Information Convert User Question to XML (Cont.)

37. Page 37 LAITS Laboratory for Advanced Information Technology and Standards Ontology Information Convert User Question to XML (Cont.)

38. Page 38 LAITS Laboratory for Advanced Information Technology and Standards Convert User Question to XML (Cont.) The resultant XML description is ready to be submitted to the chaining process

39. Page 39 LAITS Laboratory for Advanced Information Technology and Standards Automatic Process Composition

40. Page 40 LAITS Laboratory for Advanced Information Technology and Standards Results to Answer User’s Query The answer to the user question is a landslide susceptibility map of the area and time concerned.

41. Page 41 LAITS Laboratory for Advanced Information Technology and Standards The OWL-S Manager  OWL-S file management:  Set data/service ontology  Set association  DataType and ServiceType match  Exact  Subsume  Relaxed  All  Composite service chain construction

42. Page 42 LAITS Laboratory for Advanced Information Technology and Standards Register Data and Service OWL OWL schema can be registered either using URLs or typing/pasting the files directly.

43. Page 43 LAITS Laboratory for Advanced Information Technology and Standards OWL-S can be registered either using URLs or typing/pasting the files directly. Deploy OWL-S

44. Page 44 LAITS Laboratory for Advanced Information Technology and Standards Manipulate Deployed OWL-S The deployed OWL-Ss can be viewed using the GetCapabilities function. OWL-Ss can be undeployed using the UnDeploy function

45. Page 45 LAITS Laboratory for Advanced Information Technology and Standards Manipulate Deployed OWL-S (Cont.) Add associations between service type and data type to facilitate the reasoning.

46. Page 46 LAITS Laboratory for Advanced Information Technology and Standards OWL-S to BPEL Conversion There are two ways to execute the composite service process, through the OWLSPower engine or the BPELPower engine. Both engines are developed in LAITS. The following is the OWL-S to BPEL conversion page.

47. Page 47 LAITS Laboratory for Advanced Information Technology and Standards Screen shot of OWL-S converted BPEL OWL-S to BPEL Conversion (Cont.)

48. Page 48 LAITS Laboratory for Advanced Information Technology and Standards Screen shot of WSDL OWL-S to BPEL Conversion (Cont.)

49. Page 49 LAITS Laboratory for Advanced Information Technology and Standards Manage BPEL in PBELPower The BPEL converted from OWL-S can be deployed, managed, and executed in BPELPower

50. Page 50 LAITS Laboratory for Advanced Information Technology and Standards The DEMtoSlope Service Process The diagram shows the activities of the Slope service BPEL process, converted from OWL-S, deployed in BPELPower.

51. Page 51 LAITS Laboratory for Advanced Information Technology and Standards The DEMtoSlope Service Process (Cont.) The full diagram view the activities of the Slope service BPEL process, converted from OWL-S, deployed in BPELPower.