Mapping Conceptual Spatial Models into GIS

(NOTE: a revised version of this paper has appeared as Kemp, K. K. (1997) "Integrating traditional spatial models of the environment with GIS". In Proceedings of 1997 ACSM/ASPRS Annual Convention and Exposition, Technical Papers Volume 5: Auto-Carto 13, Seattle, WA. American Society of Photogrammetry and Remote Sensing and American Congress on Surveying and Mapping. pp. 23-32.)
 
The following is a preliminary summary of my conclusions from discussions held in Canberra during October. It is prepared simply as a means of organizing all the ideas generated in the discussions. The ultimate insightful review is pending...

Preliminary premises

1. The conceptual spatial models used by environmental modelers differ in significant ways from the spatial data models provided in current GI systems. Simple mappings between these models are not currently possible which means that modelers usually must modify their models in order to use GIS. This is sometimes difficult and may result in incorrect use of available data and misinterpretation of model results.

 
2. The objects of study, traditional sampling designs and modeling techniques used by individual environmental science disciplines lead to discipline specific conceptual spatial models. These conceptual spatial models vary significantly between disciplines. Thus, there are significant differences in how different sciences discretize space, sample spatially distributed phenomena and extrapolate from their discrete samples to the phenomena being studied.

 
3. However, it is possible to deconstruct these differences such that the fundamental common characteristics of conceptual spatial models can be identified and measured.

 
4. These characteristics can be used to develop interoperable interfaces, data models or other elements of GI systems which will enable environmental modelers to use them more efficiently.

• Describe and characterize conceptual spatial models used by environmental modelers for data collection and for model development.

 
• Using this information, devise direct mappings between these conceptual spatial models and digital spatial models.

 
• Using these mappings, outline some improvements to the design of new interfaces, data models and/or other GIS components in order to improve the efficiency with which environmental modelers can use GIS and to assist in the development of interoperable components for GIS.

 

Conclusions about the premises

1. The conceptual spatial models used by environmental modelers differ in significant ways from the spatial data models provided in current GI systems.

 
• [The term "conceptual spatial model" refers to the analog models used to constrain or inform data collection activities and/or used during the conceptualization of process models.]
• By definition, environmental models are environmentally determined. Thus since many environmental phenomena are fields, environmental models are fundamentally continuous. Hence, environmental modelers generally have a continuous view of the world and most seem to find this compatible with the cellgrid or pointgrid data models. Therefore, there are no fundamental differences between the disciplines.

 
2. The objects of study, traditional sampling designs and modeling techniques used by individual environmental science disciplines lead to discipline specific conceptual spatial models.

 
• If environmental modelers generally do perceive their phenomena as continuous or see their phenomena as being environmentally determined, then it seems that their conceptual spatial models do not vary significantly.
• However, objects of study do vary from superimposed continuously varying phenomena (e.g. climatology) to objects embedded in continuous matrices (e.g. mining geology) to independent objects (e.g. entomology). However, environmental determinism is a fundamental principle in the prediction of the occurrences of many of the phenomena and so they can all be seen to exist within a continuous matrix or at least on a continuous probability surface.
• In some sciences, traditional data collection and representation techniques have relied on the discretization of space and of phenomena, particularly in soils, geology and vegetation mapping. In these cases, data collection requires experts who can interpret all the environmental clues, some of them unspecified and unmeasureable, and make conclusions about the distribution of classes of the phenomenon being mapped. In such cases, the data which is ultimately recorded (i.e. mapped) is not the fundamental observed phenomena, but an inferred classification.
• However, environmental modelers in these same fields are now working with models based on field (i.e. continuous) variables. Classes can be extracted as needed for any set of criteria and/or by using various statistical techniques.
• Thus, significant conceptual differences may not be within the different sciences, but between the scientists and the managers.
• At the management end of modeling applications, continuous results are too difficult to integrate conceptually, particularly when there are several environmental gradients involved. Thus classifications of results of continuous analysis are needed.

 
• Themes for further consideration:
• What role does classification play in science? Is the need to classify during data collection now unnecessary given current computing power and massive digital storage?
• What role does expert knowledge play? Can modelers replace or simulate the expert knowledge of the field scientists? (No, of course not. Models are simplifications of reality, some independent variables cannot yet be modeled or predicted.)
• Is it best if the classification systems for management applications are based on repeatable, mathematical methods, rather than on expert knowledge? Does anyone care if the process is a black box?
 
3. However, it is possible to deconstruct these differences such that the fundamental common characteristics of conceptual spatial models can be identified and measured
• There are no fundamental differences. Continuous fields, in some cases with embedded objects, may provide the unifying theme.
• However, it is possible to conceive of a continuous environment composed of homogeneous discrete units such as watersheds. The scale of the process determines if this is possible since in many regional scale models, processes below the watershed level are insignificant. This permits an assumption of homogeneity even within a continuous context.
 
• Themes to consider
• The role of scale and the need for classification may provide some useful points of departure for fine tuning the idea that continuous fields can form the fundamental conceptual model for environmental modeling.
• Can we identify fundamental and derivative environmental phenomena? For example, fundamental ones may be terrain, geology, solar radiation; derivative ones are soil, vegetation, fauna, rainfall, radiation on the surface…
• Burrough and Frank (1995) have itemized several dimensions which could form a basis for characterizing spatial models. A revised and expanded version of these dimensions is included in the section "Enumerating the characteristics of conceptual spatial models" below.
 
4. These characteristics can be used to develop interoperable interfaces, data models or other elements of GI systems which will enable environmental modelers to use them more efficiently.
• Interoperability can be based on a common conceptual reality. Objects and phenomena should be conceptualized within their physical environment and their attributes and relationships expressed in ways which allow interfaces to translate these generic qualities into system specific values. This means that reality would form the central interface between different environmental models and databases. All data passed through the interface would be returned to the appropriate variation of a continuous model and then redefined as required for specific software.
• Is it possible to design a software product which would assist environmental scientists and managers to itemize the critical elements of the environment and to understand and express the relevant spatial and aspatial components. Such a product might construct objects (in an OO sense) ready for computation.
• Need to review current literature to enumerate the critical characteristics across many different environmental science disciplines.
• Look for evidence of a universal continuum theory.
• Evaluate different data collection philosophies. Does each collect from an continuum? Does the scale of sampling match the process scale? How is count data handled? Is it embedded in continuity?
• Look for truly spatial models. What distinguishes them? Can these factors be itemized and used to determine the difference between spatial and aspatial models? Look for aspatial models disguised as spatial ones. Look for "despatializing" techniques.

Issues

1. Is continuity, possibly with embedded objects, "the" conceptual model for environmental modelers?
• By definition, environmental models are environmentally determined. Thus since most environmental phenomena are fields (is this true?), environmental models are fundamentally continuous. (How do with historical events fit into this continuous framework?) Given this, can there be only continuous surfaces and continuous volumes with objects embedded in or on these continua?
• Is geology fundamentally different? Do they need a conceptual temporal model as well to explain the intersecting lithology? In geology they have objects within continuous matrices, but also the continuous bodies are discontinuous at boundaries. However, rock is continuous, so there is something everywhere in the study area. Also, the incorporation of a temporal conceptual model might allow several different continuities to be combined for complete understanding of the historical development of the rock or landforms.
 
2. However, the need for classification remains.
• What is the role of classification in sciences with continuous views of space? Must we eventually classify in order to understand?
• Classification allows many different factors to be summarized and understood conceptually, though not necessarily analytically. Classification is needed in when there are too many unspecified independent variables and experts are relied upon to assess the complex environment and account for these unspecified variables. Thus, classification is still necessary in some data collection situations, in visualization and, possibly, in management applications of model output.
• However, these roles for classification are now somewhat diminished since many analytical and data collection tasks can now be handled using continuous frames of reference.
• Classification is a critical aspect of remote sensing applications. In fact, this may be the reason RS has not met widespread acceptance. Most RS uses classify it rather than make it a fundamental source data layer.
• What do boundaries mean? It depends on whether the lines are drawn during landscape description or during modeling. Is it the boundaries or the interiors which are more important? Some ecological niches exist in transition zones (= boundaries?) rather than in the interiors. The ability to generate alternate, appropriate sets of boundaries is possible with continuous models.
• There may be one fundamental landscape unit -- the watershed. It does seem to bound a number of fundamental environmental variables. Note that while ridges may be geologically determined and the terrain on either side of ridges are likely to have different soils, vegetation and microclimates, the other set of distinct lines in terrain - rivers - do not correlate well with abrupt changes in environmental variables.
• Pattern requires classification. Can it be evaluated within a continuous context?
 
3. Do managers need different spatial models?
• Given the ability to understand environmentally determined phenomena using continuous models and the need to classify to understand, the conceptual differences lie not between environmental science disciplines but between scientists and managers. Is there a difference between modeling for prediction versus modeling for description and/or management? Do managers need a more discrete (i.e. classified) view of space or do we simply need to educate managers to work with data in forms other than classified maps?
 
4. Scale
• Classification is a scale issue. Scale determines whether space can be divided into objects yet still treated as continuous phenomena. Continuous phenomena may be expressed by discrete units if the scale of the process is larger than the units.
 
5. Expert knowledge plays a major role in the understanding and modeling of environmental systems.
• How is expert knowledge incorporated into models of processes? What role does it play in classification and data collection? Can we be explicit about the incorporation of expert knowledge in data collection and modeling activities?
 
6. Conceptual temporal models also need to be addressed.
• Which sciences assume change and which are static?
• Historical and episodic events affect the environment but these cannot be represented or modeled well. This is also a scale issue.
• What about objects and continuity in time? Can space be substituted for time or vice versa (e.g. succession demonstrated by going up an elevation gradient or astrophysical location equating with time)?
 
7. Can models be usefully classified as either spatial or aspatial?
• Is there a significant difference? Are aspatial models just spatial models at regional scales in which spatial heterogeneity is not relevant at large process scales?
• How do aspatial models and aspatial data incorporate space? How is space despatialized for aspatial modeling and data collection? How do aspatial models represent changes which have an impact over space?

Enumerating the characteristics of conceptual spatial models

1. Describe the objects of study
• What are they?
• How big are they? Which of Zubin's object types are they (small object, large object, scene or territory)?
• Are they discrete objects or continuous fields? (B&F)
• Are they capable of precise definition or are they somewhat vague? (B&F)
• Can they be precisely located? Does their location change over time?
• Are they static or dynamic? (B&F) Do their characteristics change over time?
• What is their spatial extent and their spatial relationship to other objects or fields?
• Are objects topologically related?
• Do the entities interact and change the character of other entities of study?
• Do they exist at a single scale or can they be defined at various scales? (B&F)
• Are they internally consistent or is there variation within a single entity?
2. Describe the study area
• How big is it?
• What are its component parts?
• How well are these parts differentiated?
• Do they overlap and/or interact spatially?
• How is this overlap visualized and/or represented?
3. Describe the process being studied
• How do the component parts interact?
• How does space affect the process?
• What scale does the process operate at? (How can this be quantified?)
4. Describe the data
• What is measured?
• Can the phenomena be measured by complete enumeration or must they be sampled? (B&F)
• What is the sampling strategy?
• How well is the spatial distribution of the phenomena captured?
• How well is the spatial variability captured? How is it accounted for?
• Are sample sites representative or significant?
• How precise are the sampling locations? Are they modified in the field?
• How are the phenomena measured?
• Which measurement scale is used - nominal, etc., discrete or continuous?
• Which measurement units are used?
• What is the accuracy and precision of these measurements?
• How is the field data, collected, recorded, stored and analyzed?
• How is the data stored in the database related to the geographic reality?
• Are the representations generalized or detailed? (B&F)
• Are the locations precise or general?
• Is the spatial extent of the phenomena recorded or left for interpolation?
• Is spatial variation accounted for in the data storage?
• What does each number in the computer represent? What is it a measurement of?
• How does the dimensionality of entities in reality compare to the dimensionality of their representation?
• What are the major sources of error in the data? How is this accounted for this?
5. Describe the process model
• Explain how the components of the landscape/environment are represented in the model.
• What parts are represented directly? Which indirectly?
• Does the model incorporate spatial effects? Are these implicit in the data or explicit in the model's functions?
• How well does the model represent reality? Whatís missing? How is this accounted for?
• How well does the data required by the model represent the components of the landscape that are being modeled?
• What are the major sources of error in the model? How is this accounted for?
• What kind of results are expected from the model?

What next?

Acknowledgments

My thanks to everyone who gave me some of their time and many of their ideas. I hope I will have a chance to continue these discussions in the future. Also, I would like to thank CSIRO Division of Information Technology, CSIRO Division of Water Resources and the ANU Department of Geography who provided support for my travel to Canberra, and, of course, the NCGIA who allowed me to escape the Santa Barbara office for 5 weeks.