GIS/LIS (1994), p860-869, copyright GIS/LIS
A wildfire GIS needs fire and related data organized in an efficient way in order to provide a good support for wildfire studies and operations. Wildfire consists of both spatial and temporal characteristics and they need to be addressed in a wildfire GIS. Conventional GISs lack for a sufficient support for managing temporal information in order to facilitate the development of a wildfire GIS. This study briefly reviews three main space-time models proposed in GIS literature and discusses their problems with respect to modeling wildfire information. This study proposes a three domain model for a better support of organizing and managing wildfire information in a GIS. The three domains model is built based on four conceptual wildfire models identified by knowledge acquisition techniques. They are models of locational snapshots, fire entities, entity snapshots, and fire mosaics. The three domains model can support analytical and modeling needs for all of the four conceptual models and may be developed as a theoretical framework for modeling geographic information in GISs. Applications and future work of the three domains model to the other fields are discussed.
The need for incorporating temporal information into a GIS has been emphasized (for a detailed review, see Langran, 1992). Most of proposed GIS space-time models can, however, only store static historical information in a discrete time structure. This is similar to conventional GIS spatial data models, which are not good at representing continuous space.
Conventional GIS spatial data models use a set of layers to represent distributions of geographic themes. Geographic data are encoded according to locations. Reality is described by location-fixed cells in a raster framework, or by location-fixed points, lines, or polygons in a vector structure. As such, GIS spatial data models can be only used to represent spatial distribution, spatial pattern, and spatial arrangement.
[End Page 860]
Three main approaches have been proposed to the incorporation of temporal information into conventional GIS spatial data models. First, independent time-stamped layers are added to a GIS database if any changes occurred (Armstrong, 1988, Figure 1). For example, a data layer represents burned areas in 1988 in California and another layer represents burned areas in 1989 for the same region. This approach can support spatial query for a particular year but cannot or is not efficient to facilitate temporal query, such as "where has been burned more than 5 times in the past 10 years?" The second approach extends the first one by overlapping all time-stamped layers to a space-time composites layer (Langran and Chrisman, 1988, Figure 2) so that every space-time composite has a unique history compared to the changes of its spatially connected neighbors. For example, a burning history of a space-time composite will be different from all its spatially adjacent space-time composites. In this case, GIS data configuration needs to be reconstructed whenever new data are introduced because space-time composites will be changed by the overlay of the new spatial data and the old space-time composites. The third approach extends the space-time composite model to a three-dimensional spatiotemporal object model (Worboys, 1990, Figure 3). This model needs no reconfigurations for the introduction of new temporal information but will repeat unchanged information. This model also has limitation on storing dynamic information such as fire spread and motion (for a detailed review, see Yuan 1994b).
[End Page 864]
Wildfire, as well as other natural disasters, requires an information system which can support both spatial and temporal query, analysis, and modeling. A fundamental problem is to store spatiotemporal data in an efficient form in terms of data storage and operation. This study uses wildfire as an example and proposes a model, the three domains model, to fulfill both of the requirements. The next section will discuss the spatial and temporal information needed for building a wildfire information system. Then the following section will present the three domains model and elaborate its design of data storage and operation. The final sections will examine the potentials and applications of the three domain model.
[End Page 862]
In order to design a GIS data model which is able to support spatiotemporal reasoning about wildfire, Yuan (1994a) conducted a series of wildfire interviews and surveyed fifty-five wildfire literature's in terms of data requirements on spatial and temporal scales and resolution. Four major conceptual models of wildfire were found. They are named locational snapshots, fire entities, entity snapshots, and fire mosaics (Figure 4).
The locational snapshot model is similar to Armstrong's snapshot model. However, the locational snapshot model emphasizes that spatial units and locations are pre-determined. Reality is described by individual spatial units. Studies of fire forecasting use the model the most since most of data used in these studies are derived from remote sensing and cells (pixels) are pre-determined by remote sensors. Data for fire forecasting are updated on a regular basis and all updates are performed on a regional basis. The model of fire entities is used to describe fire spread and motion according to individual fires' spatial positions through time because studies of fire growth and spread analyze each fire by its burns at a point in time. Fire spread or motion is described by its changes of burns in positions through time. The model of entity snapshots differs from the model of locational snapshots in that descriptions are done according to a focused entity, such as a fire, instead of locations. Two major fire characteristics stressed in fire effects are fire intensity and the depth of a burn in each burns. Studies of fire effects emphasize the comparison of post- and pre-fire conditions and interpret their difference by the incidence of a fire or a series of fires. Fire effects are described by its pre- and post-fire characteristics in burns. However, whether areas were burned at the same time is not the focused in this model. Its emphasis is placed on the homogeneity of burns and vegetative or ecological settings. Studies of fire history delineate a study area into a set of fire mosaics to show the distribution of the most recent burns in the area. The model of fire mosaics describes reality by a set of spatially exhaustive polygons, each of which is different from its neighbors in terms of a certain attribute, such as the date of the last burn.
The four conceptual models form a fire information cycle. Information output from one model can be useful to another model. In addition, the models of locational snapshots and fire mosaics, in fact, describe reality from a locational point of view; that is spatial units are determined before the descriptions of fire or environmental characteristics. This kind of description is a location-centered approach to reasoning about "where, then when and then what " In contrast, the
[End Page 863]
models of fire entities and entity snapshots describe reality from an entity point of view so that they conceptualize reality according to an entity-centered thinking of "what, then when and then where." Therefore, the organization of fire data in a GIS will be either linking temporal and fire characteristics to spatial units to model the data from a locational perspective, or linking spatial and temporal characteristics to fire entities, such as the 1988 Yellowstone fire, to model the data from an entity perspective.
[End Page 864]
The three domains model is built based on the location-centered and entity-centered approaches to describing reality (Figure 5). The model has three domains: semantics, time, and space. In a wildfire information system, the semantic domain consists of wildfire's concrete or abstract concepts of aspatial and atemporal properties, such as names of individual fire events, fire intensity, fire types, or forest stands. The temporal domain consists of temporal objects of points and lines, which represent instance time and time intervals respectively. The temporal domain describes existence of semantic objects and spatial objects, and further supports temporal analysis and reasoning, such as fire frequency or fire cycles. The spatial domain is composed of spatial objects of points, lines, polygons, cells, and volumes. Each of them represents zero-, one-, two-, or three- dimensional spatial unit. Each domain can have its own database management system (DBMS) for data storage, maintenance, and operation.
The three domains model supports both the location-centered and entity-centered approaches to describing reality. The models of locational snapshots and fire mosaics conceptualize reality in a fashion of linking spatial objects, through temporal objects, and then to semantic objects. Descriptions about temporal and semantic objects are modeled as attributes of locations in such an approach. On the other hand, the models of fire entities and entity snapshots describe reality by mappings from semantic objects, through temporal objects, and then to spatial objects. As such, temporal and spatial properties are modeled as attributes of semantic objects. Further, the three domains model shows the duality of individual versus field views of reality in geographic data modeling.
[End Page 865]
The location-centered approach is used in conventional GIS layer-view models and forms the central idea of the snapshots, space-time composites, and ST-object models as reviewed in the introductory section in this paper. Because reality is conceptualized as layers of themes, snapshots, and composites, this approach can only be used to model locational characteristics but is good for fire danger rating and fire management. Fire danger rating is based on information on weather, vegetation, and topography. Cartographical modeling is used to weight all the three factors at a location (a cell) to compute for its fire danger index (for example, see Van Wagner, 1989). Therefore, fire danger rating needs data to be organized on a locational basis for easy computation. Also in simulation or prediction of fire behavior, additional computation on estimated fire's rate of spread for each location is performed on a locational basis (Kessell, 1990; Zack and Minnich, 1991). In fire management, an area is delineated into several management units. Each unit is set up a plan for management operations, such as prescribed burning cycles, research or protection of endangered species, and ranking of fire protection for example, the fire management plans set in the South Florida Water Management District (K. Horn personal comm.). These spatial units are pre-defined and usually will not change. Therefore, data organized in a location basis according to these management units are most suitable for fire management.
The entity-centered approach is somewhat similar to the concepts of object-oriented modeling but does not emphasize the concept of encapsulation. However, both of the approaches describe reality by modeling individual concepts or entities in the real world. Such entities include forest stands, fire burns, and fire crews. The entity-centered approach describes temporal and spatial properties as attributes of entities. Therefore, it is able to model an entity's motion or spread as changes in locations through time, and also it can describe changes of an entity's characteristics through time. The former is useful for studies of fire behavior and for fire fighting; the latter, for studies of fire effects and comparison of ecosystems' responses to wildfires. In firefighting, important information is what is to a burning fire, such as fire's rate of spread, fire intensity, percentage of containment, and techniques for fire control. As a result, a fire information system will enhance its capabilities in supporting firefighting and studies of fire behavior by organizing data according to fire entities in a semantic domain, and then describing temporal and spatial characteristics of these fire entities in temporal and spatial domains. As such, the three domains model links a fire entity in a semantic domain, through correspondent temporal object(s) in a temporal domain to correspondent spatial object(s) in a spatial domain. By this way, the three domains model can also facilitate studies of fire effects. In these studies, a burn is chosen for analysis then
[End Page 866]
time of interest is identified, and then spatial properties and other characteristics, such as resprouting potential and soil nutrient status, are described (Moreno and Oechel, 1991). The three domains model stores burns in a semantic domain. Once a burn is selected for study, correspondent temporal and then spatial objects are retrieved for analysis. A feedback appears from the spatial domain to the semantic domain if other semantic properties are of interest, for example, vegetation survivorship and fuel reduction (Romme and Despain, 1989).
The three domains model is based on implications of semantics, time, and space from the four conceptual models of wildfire. However, its applications are not limited to the modeling of wildfire information because the three domains model is a highly abstract framework for the conceptualization of reality.
For example, in the GIS modeling of a stand of Douglas firs, a semantic domain may include all general characteristics about Douglas firs, such as its life cycle, its ecological relations with other species, growing seasons, and fire cycle. As such, these semantic concepts can be applied to all other Douglas fir stands. A temporal domain may record the dates of field survey and other activities. A spatial domain may store spatial objects to show its location. However, the stand may be surveyed multiple times and may appear in more than one place. This idea can be expressed by linking the same semantic object representing a general concept about Douglas firs to multiple temporal objects and multiple spatial objects. As a consequence, a retrieval of the semantic object, the stand of Douglas firs, will return information on all surveying dates and all distributed patches.
Another example will be water pollution. The source of pollution may be a point, a line, or an area but the result is more likely to be an area or a volume. The characteristics of pollutants may be described in a semantic domain, such as hazardous levels, chemical contents, reaction time, and rate of dispersion. Again, these general properties can also be applied to other pollution events resulted from the release of this pollutants. However for this event, a temporal object may be linked to a point in its spatial domain to represent that the event starts with a point source. Another temporal object representing a later point in time or an interval may be linked to an area in the spatial domain to indicate that the pollution has dispersed to an extent. Another temporal object may be linked to a 3D spatial object to show that these pollutants have reached ground water at that time. In this case, the same semantic object, which represents an event of water pollution, has multiple linkages to temporal objects to indicate when measures or estimates have
[End Page 867]
been done and the event of water pollution exists during that period. Also these temporal objects have multiple linkages to spatial objects. Therefore, the dispersion of these pollutants is described by a sequence of spatial objects through time.
Additional examples for the applications of the three domains model include recording landuse changes, soil distributions, and hurricane sweep. The three domains model allows many to many mappings. A semantic object can be linked to multiple temporal objects to show its transactions, a temporal object can be linked to multiple semantic objects to show coexistence of multiple entities. Also the temporal object can be linked to multiple spatial objects to express the concept that the semantic object (a geographic entity or event) appears in multiple places in one time, such as more than one burns can occur simultaneously. A spatial object can be also linked to multiple temporal objects since some spatial properties may not change over time.
Four wildfire conceptual models are identified for various wildfire studies and operations. They are locational snapshots, fire entities, entity snapshots, and fire mosaics. The four conceptual models can form a wildfire information cycle and complete information flows from one model to another. Further analysis shows that the four conceptual models encounter two ways of describing reality: location-centered and entity-centered approaches. A three domain model of semantics, time, and space domains is built to support the needs for the analysis of wildfire from both views by directional linkages among the three domains. The entity-centered approach is supported by linkages from semantic objects, through temporal objects, to spatial objects; whereas, the location-centered approach is supported by linkages from spatial objects through temporal objects to semantic objects.
Boundary problems are often the main issue not only in modeling temporality in a GIS but also in identifying geographic objects. Many geographic phenomena have fuzzy boundaries. For example, there may not have a clear cut between burned and un-burned areas but a transition zone. The transition zone may be a function of other environmental factors, which boundaries may also appear to be fuzzy. Changes in boundaries usually require reconstructing and reorganizing a spatial database. Therefore, identifiers assigned to spatial objects may not be able to maintain in a temporal GIS, which may cause problems in keeping track of the path or the history of a geographic entity. This is also a
[End Page 868]
problem for temporal objects since they consist of geometrical properties and therefore they have boundaries. Further study is required to the design of a system manager which can maintain the persistence of spatial and temporal identifiers.
Most of GIS studies emphasize modeling the spatiality of geographic phenomena. Therefore, conventional GIS data models are designed in a way that reality is described according to locations. However, comparably less research on geographic semantics. As interests increased in applying concepts from the object oriented paradigm in computer programming and database management to GIS data modeling, a better understanding of defining a geographic entity or feature becomes important. More work is needed to the study of geographic semantics and analysis of their semantic relationships. This paper only proposes a conceptual design for the three domains model. Further work is required to test its representational completeness regarding to wildfire and other environmental problems, such as insect infestation and oil spill.
The author would like to thank Dr. David Mark, Dr. Hugh Calkins, and Dr. Paul Densham for their valuable suggestions.
Armstrong, M. P. 1988, Temporality in spatial databases. Proceedings of GIS/LIS'88, volume 2, pages: 880-889.
Kessell, S.R. 1990, An Australian geographical information and modeling system for natural area management. International Journal of Geographical Information Systems, 4(3):333-362.
Langran G. and Chrisman, N.R. 1988, A framework for temporal geographic information systems. Cartographica, 25(3):l-14.
Langran G. 1992, Time in Geographic Information Systems. Taylor & Francis, New York. Moreno, J.M. and Oechel, W.C. 1991. Fire intensity effects on germination of shrubs and herbs in southem California chaparral. Ecology, 72(6): 1193-2004.
Romme, W.H. and Despain, D.G. 1989. Historical perspective on the Yellowstone fires of 1988. Bioscience, 39:695-699.
Worboys, M.F. 1992, Object-oriented models of spatiotemporal information. Proceedings of GIS/LIS'92, volume 2, pages 825-834.
Yuan, M. 1994a, Acquiring experts' knowledge to build wildfire representation in GISs. Proceedings of ESRI User Conference'94 (published on CDs).
Yuan, M. 1994b, Representation of Wildfire in Geographic Information Systems. Ph.D thesis, State University of New York at Buffalo.
Zack, J.A. and Minnich, R.A. 1991, Integration of geographic information systems with a diagnostic wind field model for fire management. Forest Science, 37(2):560-573.
[End Page 869]