The types of coupling discussed in the previous section really only represent technical solutions as to how GIS and an environmental simulation model can share the same data rather than being integrated in terms of achieving compatible views of the world. In other words, this type of coupling has not necessarily led to an improvement in the scientific foundation of either GIS or environmental modeling (Grayson et al., 1993) even though it is generally recognized that there have been tangible benefits for both GIS users and modelers. The initial conceptualizations are also a rather simplistic view of the software/database environments that actually occur on projects with the need to work using spreadsheet, database, GIS software, statistical package, word processor, graphics package, RS image processing package, CAD, and environmental simulation model, not necessarily all simultaneously, but certainly at different stages of a project.
There may also be more than one of each—specialized databases for specific types of data and perhaps several simulation models, one for each specific process; maybe even more than one GIS software package. The main difficulty in moving toward a more scientifically rigorous approach to the integrated use of GIS and environmental simulation models is their differing data models (Livingstone and Raper, 1994; Bennett, 1997a; Hellweger and Maidment, 1999; Aspinall and Pearson, 2000; van Niel and Lees, 2000; Bian, 2007). As discussed in Chapter 2, data models are abstractions of reality designed to capture the important and relevant features that will be required to solve a particular set of problems. In the case of GIS, the data model focuses on creating a digital representation of geographical space, the objects contained therein, and their spatial relations.
The emphasis is on location, form, dimension, and topology. Environmental simulation models, on the other hand, are predominantly concerned with spatial processes, their states, and throughput of quantities. One is a static representation, the other is concerned with dynamics. This means that their data models will be quite different and result in database structures (for the purpose of data manipulation) and databases (for the purposes of storage and retrieval) that are also quite different. Consider the examples given in Figure 7.2 and Figure 7.3 for a drainage basin. In both Figure 7.2(a) and Figure 7.2(b), the outer catchment boundary (watershed) and the streams within are the same, but the subdivision into subregions is quite different. Figure 7.2(a) gives the traditional geographical view of subcatchments as being the contributing area of overland flow to stream confluences (identified as 1 to 3). From a modeler’s perspective, these subcatchments may not be homogeneous hydrological response units nor may the confluence itself represent a typical stream reach of relatively stable known properties (e.g., cross-sectional area, wetted perimeter) with which to model flow or from which to collect flow data.
The modeler’s view of the physical basin might thus resemble Figure 7.2(b). The geographical view in Figure 7.2(a) would perhaps result in a GIS representation of two data layers, one containing the streams as a network of lines and the other giving subcatchments as polygons (Figure 7.3(a)). The hydrologic simulation model, on the other hand, would require the elements of the drainage basin to be stored as subbasins, reaches, and junctions, as shown conceptually in Figure 7.3(b). In such an arrangement, the spatial dimension is only implicit with the data structure optimized for simulating how inputs (rainfall) are transformed into outputs (flow at the basin outlet) via their passage through the system.
Again, let us consider an ecological example. Figure 7.4(a) is a typical geographical view of vegetation mapped as succession communities. Each community is mapped with hard, nonoverlapping boundaries to make a polygon representation in GIS possible. While an ecologist might well be interested in the dynamics of the plant succession and its present state, there might not be recognition of distinct communities and they would rarely have abrupt boundaries. Instead, interest might well focus on species response to environmental gradients as in Figure 7.4(b). Furthermore, ecological simulation tends to assume homogeneous landscapes in modeling population dynamics (e.g., the panda–bamboo interaction model of Chapter 5). Within environmental simulation modeling, there is a wide range of data models because each one will need to reflect the numerical methods used to solve the particular process model(s) being simulated.
There is another important way in which the data models of GIS and many environmental simulators differ, and that is the way in which they view flow or motion. For a modeler, there are two views to choose from, the Lagrangian view, which is dominant in GIS, and the Eulerian view, which is dominant in environmental simulation models (Maidment, 1993a; Sui and Maggio, 1999). Euler and Lagrange were both eighteenth-century mathematicians. A Lagrangian model of flow focuses on the object that is moving, such as tracking a car as it moves though the countryside. A Eulerian model of flow focuses on a fixed portion of space through or across which some motion takes place. This would be much like standing on the side of a road watching cars cross your field of view.
A common analytical function in vector GIS is finding the shortest path (distance, time, or cost) between a starting point on a network and one or more destinations. Another, in raster GIS is tracing flow paths across a DEM. These are very much a Lagrangian view. But many simulation models of physical processes (as we saw in Chapter 5) are concerned with states and the changes in states for a specific length, area, or volume of bounded space. So, for example, a model might calculate changes in quantity over time for a specific area consequent on the rates of ingress and leakage across the boundaries of that area. This is a Eulerian view. While tessellations and networks in GIS are suitable spatial arrangements of bounded space for a Eulerian view of flows, there is a marked absence of corresponding functionality. Of course, there are some environmental simulation models that employ a Lagrangian approach, but, in general, there remains a marked dichotomy in the way flows are modeled in GIS and environmental simulations.