ISPRS Commission II, Working Group 2, Cambridge, July 1998

GIS or environmental analysis tasks are specific to Information Communities. They are often well defined and raise unique spatial questions. All spatial data analysis systems, including image processing and GIS offer a large amount of algorithms and analytical operations to solve spatial problems. For non-experts, this variety as well as the complexity of operations to perform which is mostly technical in nature impedes fast digital solutions.

Necessary system decisions, different data formats, data structures, and data models as well as lack of compatibility are additional drawbacks to GIS employment for different kinds of spatially related tasks. On the other hand, a broad technical toolbox is wanted that is capable of supplying full functionality for a wide range of freely defined problems.

Consequently there are strong demands to design operators closer to the user, but still universal, which means independent of the Information Community, who wants to use them.

Embedded in a flowchart based Graphical User Interface (GUI) a set of 20 universal GIS operations has been created to fill the gap between the great variety in analytical GIS functions and their usability. This solution addresses several aspects necessary to guarantee the ease-of-use of an analytical GIS-Tool (Virtual GIS or VGIS)  (Albrecht 1996, 1997). A major step towards the construction of a truly hybrid (i.e. integrated) analysis tool is the hierarchically structured operator design containing approaches towards a hybrid analysis.

During the process of developing an analytical solution for application specific spatial questions within the VGIS tool, a sequence of different operators is created that can be regarded and treated as a specific form of a graph. The visual programming potential of VGIS makes these process- graphs executable similar to a graphical case tool To assure consistency within such a graph, VGIS needs a data control mechanism through the use of metadata and a process control mechanism. 

The design of a generic universal high level GIS operator is described in section 2.

Specifications are lined out which will be needed to realize interaction between different operators (section 3) before an approach towards metadata management is offered in section 4 and underlying mechanisms to realize hybrid processing will be explained (section 5).

The conclusion will focus on further research topics in the future.

The design of Analysis-Operators in GIS covers structure and behavior. Behavior has to be defined as the abstract function performed by the operator, independent of data structure.

Structure describes the way an operator is designed to perform data structure independent processing. An operator is able to analyze input data, to select an adequate operation depending upon the kind of input data, and to return output data. Inside the operator, data driven selection processes ensure that adequate algorithms are chosen to analyze the specific input data types.

To understand what happens inside an operator, the view of a multi-shell object just like an onion comes into one's mind, symbolizing to a certain degree a hierarchical structure. In case of universal operators, we see three different levels:

The management level is the outmost shell the user interacts with. At this level, a certain problem of an information community is tackled through the knowledge and experience of the user. The medium for the problem solution is offered in form of the above GUI with visual programming capabilities and universal operators. Users can use stored graphs, create new graphs, change and assign parameters to the operators and create their personalized solutions. Communication takes place through the use of metadata in data catalogues and the operators. Metadata are passed through the chain of activated operators.

Semantic of an operator is realized at the control level (mid level) where input data, metadata and the supplied parameters are analyzed and translated into the adequate processing chain (polymorphism).

Processing then is performed on a very low level (processing level), depending upon the distributed computing platform, operating system, and the software environment.

At high level, the semantics of the operation is defined. The data to be analyzed are selected from the data catalogue with their associated metadata. The operator is specified upon its parametrization which is also metadata driven. This level manages the whole operator. It is the constituent part of the analysis task. The user defines the operation upon his cognitive experience.

The data description at this level is independent of any spatial data structure (e.g. raster or vector) or data schema.

At mid level, the controlling of the operation is carried out. The input data, selected at high level, are linked to their corresponding data representations in a geometric and structural level. The operator analyzes the metadata, given at the management level, and chooses the algorithm that matches the data types and other requirements. Here, the requirements for hybrid analysis functionality arise because of the different data representations.

At this level, the cognitive semantics of the data and operation gets lost, the information is translated into syntactically structured information for the processing in which their cognitive context is unimportant.

The low level is the processing level. The algorithm processes the data numerically, all operations characteristics are processed and sent back to mid level.

A universal analytical GIS-operator is characterized by its functionality. It is important to look upon this functionality in the context of input and output data (see Figure 1).

The universal operator is able to analyze the input data to choose the appropriate algorithm for processing. As a result, new output data are generated. A High-Level-Operator therefore has to be metadata driven. For selecting data and functionality, only metadata will be used. The following survey introduces the concept of operators and their input and output data.

The operator is defined by a generic method. This is specified by its characteristic parameters. The operation is described by the operators metadata. The operator generates specific metadata as a protocol for the lineage. The underlying control has to select the correct algorithm that fits the types and structures of the input data. Each algorithm has its own profile. The process control executes an analysis of the data as an inquiry for the right algorithm. The operator is designed in a polymorph way which includes hybrid operations.

The functionality of a GIS-Analysis-Operator can be divided in the three categories geometric converter, semantic translator, and metadata processor.

All geographic data have a geometric component. One of the main goals of spatial analysis is to solve geometric (metric and topological) questions. Generally a GIS analysis produces data with new geometric information. Consequently a GIS- Analysis- Operator converts the geometric input into the new geometry of the output data.

The Operator is controlled and prepared for processing, based on the metadata of the input data. Metadata is translated to process control parameters. The process produces new metadata. The generation of the output metadata is controlled by the operations metadata.

The controlling of the operator includes the preconditions that are derived from the input data and the chosen operation with its implemented algorithms. For the output, the postconditions include the requirements of the output data. The operation has to be protocolled including the lineage and other metadata (Jung and Albrecht 1997).

Universal operators interact with different instances. Existing specifications, especially those that are about to become standards, have to be regarded and are at the same time a valuable foundation for the definition of universal high level operations. As a result for well structured and designed operators formal specifications will emerge.

Specifications of the OGC will have to be considered for the implementation of features, coverages, data catalogues and metadata (OGC 1996) . Specifications are also needed for the user interface of operators.

Based upon the operator design described above, the specification of a generic high level spatial analysis operator  (Meyer 1996) is necessary to create instances of an operator class "high level- operators". An instance of that class could be any high level operator, e.g. buffer, overlay, shortest path ... At the current state of development the specification introduced in this section does not claim to be complete or formal in mathematical terms. It rather serves as a more detailed description of components in order to approach a formal specification. It is the attempt to outline the general conditions necessary to create high level operators. Therefore it has to be regarded as a step towards the realization of operator design.

The variety of instances of high level operators, design and specification comprises different aspects. The data structure independence of the management level has to be implemented through a data type driven polymorphism at the control level. Due to the polymorphism a high level operator is able to execute different algorithms or sequences of algorithms at the processing level.

In addition, cognitive and semantic aspects can not be neglected in operator design. Yet the structure of data plays a central role in the building process of high level operators concerning different parts of the specification.

The user interface specifies what kind of interaction can take place between the user and the operator. User interaction includes any form of communication between the user and the operator, such as:

Plausibility control of an operator should be able to determine whether an operation proves to be meaningful or not. Representation of this advanced form of semantics has to be one level above the creation of the correct data type management. It comprises several aspects.

Assumed that spatial data are organised in data types,  it has to be exactly specified which data types can be processed by an operator and which data types will be produced by the operator after processing. Data types distinguish between different representations of geometry within spatial data.

Polymorphism stands for different algorithms related to different data types of the same high level operator. Program structures are created at the control level by decisions made according to data types, user parametrisation and semantic specifications. Certain conditions have to be met:

At processing level standard software development guidelines have to be regarded. It is planned to choose an object-oriented approach with certain advantages related to the described operator design.

Optional is an extension of high level operators towards the restricted and/or expanded use in certain information communities. That would require:

Metadata is used to manage all operators at the highest level of abstraction. At the interface or management level, only metadata control user interaction. Explicit and implicit metadata are analysed to control the operator at mid level. The metadata have different meanings at the three levels as described in chapter 2.

In the following, the processing of metadata for high level-GIS- operators is divided into two parts: metadata of spatial data (Ganter 1993) and metadata of the operator.

The metadata describe the data at a high level of abstraction in a data catalogue with its related information. The information is the link to the database and to its data schema in which all other information is stored implicitly.

The operator metadata control their spatial analysis process. They map and process the metadata of input and output and protocol the operation and the lineage.

Hybrid analysis is needed for the integration of GIS and remote sensing data (Ehlers 1993; Woodsford 1994; Hinton 1996; Wilkinson 1996), for terrain analysis (DTMs in raster format with overlay of vector data) etc.

Within VGIS hybrid analysis is enabled through the polymorphism of the universal operators. Depending upon the structure of input data different algorithms can be invoked to perform requests as well as the creation of new geometries and objects. Emphasis has to be placed upon the operations that combine input data-types with different data structure (e.g. raster with vector objects, raster with TIN objects etc.).

Possible results of a hybrid analysis are new raster data, vector data or both of them.

Some examples of hybrid analysis functionality

The complexity of the hybrid functionality may be shown by the following survey:

The full range of spatial analysis may be reached with operations that interact with raster and vector data without the need of conversion. The requirements of hybrid analysis are divided in geometric interaction, semantic analysis and metadata processing (Egenhofer 1993).

The concept of universal GIS analysis operators independent of Information Communities is outlined and specified. Universality in this context includes a domain independence as well as the option to work with different data structures. This can only be achieved through metadata driven user interaction.

To realize the data structure independence of universal operators, hybrid analyses have to be integrated. Through a currently increasing demand for the integration of remote sensing and GIS, hybrid analysis techniques have gained significance. Our future work will focus on a detailed specification of hybrid analysis operations and their prototypical implementation.