Dynamic and Mobile GIS: Investigating Changes in Space and Time - Chapter 5

Chapter 5 nen, A Process-oriented Data Model Femke Reitsma1 and Jochen Albrecht2 1Institute of Geography, School of Geosciences, The University of Edinburgh, Scotland 2Department of Geography, Hunter College, City University of New York, USA 5.1 Introduction Thus far, GIScience has lacked an appropriate data model to represent processes; processes such as erosion, migration and pollution dispersal. The need for extending geographic representations for processes has been recognised in GIScience literature (Peuquet, 2001; Raper, 2000; Worboys, 2001) and acknowledged as a key goal in the University Consortium of GIS’s (UCGIS) research agenda (McMaster and Usery, 2005). Yuan et al. (2005, p. 132) posit that ‘As the conceptual core of a geographic information system, geographic representations determine what information is available for communication, exploration and analysis. Hence, research in extensions to geographic representations is critical to advancing geographic information science’. In order to investigate change in space and time, the theme of this book, we need to be able to explicitly represent change as it occurs. Existing theories and data models for simulating processes focus on representing the state of the represented system at a moment in time. The future pattern of global temperature from a global climate change model or the distribution of humans in an agent-based simulation of disease spread, for example, only provides information about the status of the attributes of the system at each step of the simulation, attributes such as temperature or agent health at a particular location. Information about the processes defined in the model is typically not expressed or represented in any form. In utilising a process-oriented data model, we gain the advantage of being able to query, analyse and visualise processes. This chapter presents a new process-oriented data model called nen, which can be used to represent process information. The application of the nen data model to process modelling offers a set of modelling results that is complementary to those of traditional models. Its novelty is the provision of a new epistemological window on the modelled results, allowing for new process-oriented queries and analysis. The data model is applied to a small watershed modelling test case, which provides initial scope for simulating geographic processes with the new data model. In what ____________________________________________________________________________________ Dynamic and Mobile GIS: Investigating Changes in Space and Time. Edited by Jane Drummond, Roland Billen, Elsa João and David Forrest. © 2006 Taylor & Francis © 2007 by Taylor & Francis Group, LLC 78 Dynamic and Mobile GIS: Investigating Changes in Space and Time follows, Section 5.2 describes current approaches to theorising and representing processes in GIScience, forming a framework for discussion of the new data model. Section 5.3 presents an alternative approach, describing the new data model, which is then applied with a prototype implementation of a watershed runoff model in Section 5.4. The results of the nen-based approach are then discussed in Section 5.5, followed by consideration of validation of models and results of this method in Section 5.6. Section 5.7 concludes the chapter. 5.2 Process theories and models Current research into dynamic phenomena in GIScience has focused on the representation of object states at each moment of time and over time. This is built upon long-standing theories defining the entities that populate or compose space and time. What is meant by object, are those things that we typically identify and categorize as existing at an instant of time, such as trees, mountains, barrier islands and political boundaries. These are the things dominating metaphysics (Hartshorne, 1998; Rosenthal, 1999), as well as GIScience ontologies (for example, Casati et al., 1998; Fonseca and Egenhofer, 1999; Smith and Mark, 1998; Thomasson, 2001). Spatiotemporal research in GIScience has consequently focused on the dynamics of these entities, i.e. connecting the states of these entities over time (e.g. Tryfona and Pfoser, 2001), or exploring the relationships between objects and the processes that modify them (Bittner and Smith, 2003; Tomai and Kavouras, 2004). As a consequence of the focus on static objects, data models for dynamic phenomena centre on state changes of objects. For example, a global climate change model, while containing process information in the model structure, does not represent or store this information for analysis; rather, the states of the climate system are stored at each instant of time. There is no data object that represents a geographic process that changes over space and time (Yuan et al., 2005). This results in a loss of information about the modelled process, which cannot accurately be regained by interpolating between time slices. For example, in global climate modelling virtually the same future state of increased temperature can be modelled as a result of two very different changes to the model, an increase in solar luminosity or an increase in CO2. It is not immediately obvious which process or processes, such as heat transport or a change in cloud optical depth, caused these results. The static roots of GIS are found in its cartographic origins, which have formed the intellectual framework for much of GIScience research (Kuhn, 2001; Yuan et al., 2005). Kuhn (2001) notes a number of other reasons for such object orientation in geographic and other information systems, including: ⑨ an emphasis on attributes and relationships rather than process and change, ⑨ the weakness of logic-based formal languages in dealing with operations and their semantics, ⑨ and a presumed priority of objects in human (spatial) cognition. © 2007 by Taylor & Francis Group, LLC 5. nen, A Process-oriented Data Model 79 5.3 An alternative process data model An alternative data model for the representation of processes is presented in this chapter, which provides advantages in querying, analysis and exploration of process descriptions under computer simulation conditions - or in silico. The data model is referred to as a nen because its simplest and most abstract graph representation is a node-edge-node triple (Figure 5.1). This simple point process representation, which was used for the watershed prototype described in Section 5.4, can be extended to larger spatial entities, as might be represented by a polygon (Figure 5.2). (x1, y1) (x2, y2) Figure 5.1. Process representation for point feature. A more comprehensive representation is in form of a tuple: (x1, y1, x2, y2, t, st, {a1, a2,...}, {r1, r2, …}). The spatial location of the process is identified by x1, y1, x2, y2, which expresses the spatial extent of the process. The temporal location of the process is defined by t, where a process is represented on a single layer of spatial information rather than lost between time slices. The st represents the spatiotemporal granularity of the process, which may be a function of the amount of energy that initiates the process. For example, given some threshold breaking push, the spatiotemporal granularity expresses how far and over what time period the process will operate in response to that push. The set {a1, a2, ...} defines the attributes of the process. The set {r1, r2, …} defines the rules of the process that govern its dynamics and interaction with other processes. For example, a set of rules for modelling the process of sediment transport in the longshore may define the spatiotemporal extent of an instance of that process as 5m/hour, depending on various relationships it holds between other processes operating in the nearshore. © 2007 by Taylor & Francis Group, LLC 80 Dynamic and Mobile GIS: Investigating Changes in Space and Time Figure 5.2. Process representation for area feature. This data model provides a new epistemological window on geographic processes. Simulating processes with a process data model allows us to ask questions that are not directly answerable with current object-centred formulations, which focus on the states of a system that result from the operation of processes. Our new data model allows us to ask questions such as: ⑨ Where is a process operating at a particular instant of time? ⑨ How has the process changed over time? ⑨ What process(es) caused another process to occur? The answers are not inferred (or interpolated) but are explicitly stored as part of running the process model. How the rules of the process affect the spatial dynamics of the process may therefore also be better explored. 5.4 Watershed modelling application The theory of taking process as a representational primitive has been prototyped with a watershed model within a simulation environment called Flux. 5.4.1 Simulator Flux is written in Java and inherits and extends a number of basic operating classes from the RePast (Recursive Porous Agent Simulation Toolkit) library, which is an open source agent-based modelling environment created by Social Science Research Computing at the University of Chicago1. RePast is primarily used for its display and scheduling classes, and also has the advantage of containing Java classes for importing GIS raster data (ESRI ASCII raster files). Flux contains a set of interfaces and default classes that define the basic structure of the process model, including methods that must be implemented by an inheriting domain model. The 1 http://repast.sourceforge.net/ © 2007 by Taylor & Francis Group, LLC 5. nen, A Process-oriented Data Model 81 objective was to maximise generic functionality within the Flux classes, thereby minimising the code to be developed within the domain model. The output of a simulated model is stored in text files, which can then be queried with a query tool that was developed as part of the initial steps towards process analysis. For a full description of the simulator, see Reitsma and Albrecht (forthcoming). Figure 5.3 presents a sample simulation using the Flux simulator. Each nen, represented by a node-edge-node tuple (as depicted in Figure 5.1), indicates an instance of groundwater flow. The raster backdrop is a digital elevation model of a small sub-watershed, where lighter shades represent higher elevation. At each time step, groundwater flows towards the North-Western corner of the sub-watershed. Figure 5.3. Sample simulation. 5.4.2 Model and simulation For the purposes of testing the methodology a simple watershed model was simulated. The model included the following restricted set of processes: Hortonian overland flow, groundwater flow, infiltration, percolation, saturation excess runoff and surface ponding. The data used to define the parameters for the simulation are taken from the Reynolds Creek Experimental Watershed (RCEW), which is a high- © 2007 by Taylor & Francis Group, LLC ... - tailieumienphi.vn