SNAPSHOT AND COMPOSITE CHANGE DATA
MARK G. SCOTT
research paper submitted in partial fulfillment of the requirements for the
MASTER OF SCIENCE in GEOGRAPHY Portland State University 2001
THE STATE OF CHANGE DETECTION IN GIS
Change detection with maps developed long before the development of spatial databases. Manual map overlay, traditionally used to compare source material, was done with much difficulty. The contributions of GIS to change detection include map registration, re-projection, and scaling functions. GIS provide an ideal environment to perform change detection using methods that have been developed to collect, organize, and evaluate spatial data.
Change studies often rely on historic map sources used to inventory potential changes to mapped features including:
| - change of map feature
identity over time
- change of map feature location over time
- change of map feature shape over time
- change of map feature size over time
- change of map feature extent over time
Time and temporal attributes are overlooked in the modern development of spatial database oriented GIS software. Improvements must involve the development of conceptual models of temporality for spatial data. The study of time has been in effect the study of change within the database.
Demonstrating change with GIS is functionally dependant on the database design and how it integrates time data in a database with spatial data. Change detection usually does not involve interpolation. Temporal states are discrete and change detection merely measures the difference between two of them. There are three models that have traditionally stood out because they integrate time data with spatial databases in a way many others have followed.
- The Snapshot Model: time information is stored as separate but temporally homogenous spatial data (Armstrong, 1988). This model is useful for a continuous interpolation of temporal data but any assessment of the past is complicated by the fact that past conditions have been influenced in ways not detectable with snapshot data alone.
- Base map with overlay: A base map with overlays uses the method where a new data are generated for each database used in a sequence of overlays. This method creates an environment where one can analyze the whole data set. The base layer with overlay approach is superior to the time slice composite because it represents the boundaries of both states and versions (Langran and Chrisman 1988).
- The Time-Space Composite: time integrated into the topology of spatial objects (Langran and Chrisman, 1988). This model is useful for discrete event driven and temporal representation.
Change with map data encounters interesting nuances, such as how much change to location and shape can occur before it is not useful to recognize the thing as the same feature? Does change detection look for changes in location, identity, and shape? What about changes in other attributes? What about changes in the topology of spatial relationships even if the location and shape remain constant? How best can these data be queried?
In this study, snapshot and composite data models have been selected for further exploration because they provide the most contrast between the three methods. Although there are many examples of base maps with overlay, these are not discussed in order to demonstrate more thoroughly applications that involve thematic data suitable for snapshot and composite modeling.
In change studies, the snapshot model is the simplest method to apply because change detection is obtained by overlapping temporally separate versions of one theme in 2-D space. Snapshot data are used as sampled states containing temporally interpolated versions of points, line segments, and area objects that can be attributed and queried. The important technical contribution from GIS to the snapshot involves the isolation of discrete areas representative of change between the two snapshot moments. GIS can readily manipulate the display and overlay of the source data using the same extent and scale.
Change detection using a sequence of snapshots adds implied temporality demonstrating changes to map features including pattern, location, and size. One important behavior of snapshot data is that areas affected by change often exist at the fringes of adjacent areas that remain stationary. Vrana (1998) points out that fringes are in the process of becoming something else. What that something else is depends on the behavior of the phenomenon. What we know about it depends on how we model the data.
Armstrong (1988), and Langran and Chrisman (1988) observe a redundancy in the snapshot model that becomes apparent when considering how the snapshot works. In this case, every version of the snapshot contains a redundancy. The temporal snapshot uses historical time not real time. Temporal joins tell us coincident historic and valid time but cannot tell us how geographic features are bounded by events.
With the display of snapshot data, what is not on top is buried beneath other layers. Changes in time are implied through overlapping areas and therefore the existence of a feature often depends on being on the top. Engenhofer (1998, 4) pointed out in these cases, "existence is based on visibility and non-existence is based on invisibility".
One distinction between snapshot and time space composite methods involve how time is stored. The snapshot method does not integrate time directly into the database. Snapshots are time independent, meaning that each snapshot only retains the date of the original source. Because snapshots lack the topologic data structure required to store temporal information, snapshots have no means of integrating an object's state amongst temporally adjacent states. The snapshot data model treats objects contained within each snapshot as a homogeneous state independent of all other states.
2.2 Integrating Time with Spatial Data
The function of time in a spatial database is to provide information used to query spatial objects at some point along a timeline. The temporality of spatial data is determined by how time is modeled. A GIS database can either ignore or assume temporal information determining the capacity to query where and when. "Spatial database structures provide a means for describing the metrical, topological, and attribute characteristics of objects," (Armstrong 1988, 1). Spatio-temporal databases included with the composite method involve more complex ways of handling time data. Armstrong (1988) described several design principles that applications can be built on.
- Static Database: the simplest form contains only the most recent information. Legacy data are discarded along with any possible way to make a connection to temporal information.
- Static Rollback: past states are stored making it possible to rollback database time. However, these data use transaction time, not related to the actual time the event occurred.
- Temporal Database: combines transaction time and valid time. The utility to store temporal information with this method is two-fold because transaction time relays the sample times and valid time, when the event actually happened. Temporal databases are better suited in situations where a natural lag effect exists in the data, one ideally that has the potential to go un-noticed.
- Historical Database: A database where temporal individuality is preserved amongst events. This method of data storage uses a more accurate or "valid time" (Armstrong 1988), adding specificity to the placement of events on a timeline. Vrana (1998) points out that many historical data are inherently present as legacy data in land information system applications and are often and unfortunately ignored.
Temporal databases using a time-space composite architecture described by Langran and Chrisman (1988) and Yaun (1995) can respond to spatio-temporal queries because time data are integrated with space data. "Spatial considerations are made to account for geometrical transformation, forming, or dissolving of aggregates, and temporal considerations are made to account for the denotation of the event" (Armstrong 1988, 2).
The space-time composite is designed to manage and analyze spatio-temporal data using both temporal and historic database designs. The snapshot database is limited to static and static rollback designs. Composite data contain all of the spatial and temporal relationships between attributes and features with rules to maintain the integrity of the data (Yuan 1995). Feature identification is maintained throughout a temporal evolution of geometrical changes and topological relationships between temporal neighbors. Uitermark (1998) described space-time data as reusing updated information that propagates between topographic databases with different spatial and temporal resolution.
Fragmentation and Aggregation
Compiling temporally versioned spatial data into a composite can demonstrate how or if time influences the topology of spatial data. "Time is a phenomena that can be perceived only by its effects" (Langran 1992, 1). Space must be divided into time-space objects by time. Composite data work because individual and groups of individual objects are associated with a greater assemblage of aggregates.
Fragments are spatial objects in the most discrete and unique version. Fragments are created from the accumulation of geometric changes caused by overlaying individual layers together to form new ones. Worboys (1995) describes complex spatial objects undergoing discrete change when overlaying map versions. Chrisman (1992) also observed composite data decomposing into increasingly smaller pieces.
The second noticeable change to the topology found in composite data involves the topology of aggregates created by temporal neighbors. Every mapped feature in composite data are an assemblage of everything that feature ever was or will become. Conceptually, moving forward in a composite, spatial data decompose into smaller pieces. Moving backward, reversing time requires that fragmented spatial data be re-assembled into an aggregate. The continuity of identity involved with aggregation is made possible by the presence of time dependant attributes in the database. Attributes re-assemble fragments forming aggregates or disassemble aggregates into fragments.
2.3 Measuring Time, Space, and Theme
Sinton (1978) breaks down mapped thematic data and describes it as containing three components, theme, location, and time. Each feature contained within a map can be described in terms of all three aspects. Sinton established a general set of rules that define analysis involving those aspects of data. Measuring one aspect, time, theme, or location: requires that variation within a second component is systematically controlled, and the third component is fixed, or in a sense, ignored.
Time is normally fixed in map data because maps are versioned with the date the data were collected. Static maps of individual feature objects measure the location of these objects, given a systematic control of how to define them categorically. Time is fixed and theme is "controlled": location is being measured. If time is fixed and location is "controlled", theme can be measured. Such an example is a raster representation of continuous or discontinuous surfaces, where cell locations are systematically "controlled" and the nature of the thematic phenomena is measured. Fixing the location and controlling the time is useful for measuring spatial changes to the theme over a known period of time: essential to temporal modeling. Examples of Sinton's analytical framework for measuring time, theme, and location for a variety of source themes are presented in Table 2.
|U.S. Census Data||Time||Location||Theme|
Table 2 Representation of geographic data in various formats (Sinton 1978).
Expanding on Sinton's framework, Yuan (1995) identifies six spatial and temporal situations that apply to the study of change. The simplest example involves short-lived events that do not change the spatial properties of the feature, but cause a change to the features attribute identity. These types of change are associated with the static rollback database design. The use of static geographic boundaries such as snapshot data, themes can be classified by their attributes.
Yuan's temporal framework includes complex considerations for dynamic data that lends itself to the space composite design and use of temporal and historic database design. Time and location can be controlled by a spatial topology defined independently by attributes in the database. The location of independent events that move, shrink, and expand are measured because the attributes can be fixed while controlling time. Examples presented in Table 3 demonstrate measurements of theme, time, and location for both static (Type I - III) and dynamic types of map features (Type IV Ð VI).
|Change Type||Description of scenario||Fixed||Controlled||
|I. Fixed||Duration of event or attribute||location||attribute||time|
|II. Category||For a given point in time certain phenomenon may change its characteristics from site to site.||time||attribute||location|
|III. Static||For a given period of time where attributes may change from site to site through time.||time||locations||attributes|
|IV. Trans||For a given event where its characteristics or processes may change at sites through time.||attributes||locations||time|
|V. Mutation||For a given area where attributes may change site to site and from time to time||location||time||attributes|
|VI. Movement||For a given event where its location may change from time to time||attributes||time||locations|
Table 3 Spatial and temporal scenarios (Yuan 1995)
The thematic component of data refers to attributes about objects that are not locational or temporal. The theme has two significant contributions to change study: first by controlling the content of the database and second, by determining the spatial configuration. Change detection can be modeled as individual snapshots, base map with overlay, or composite. Due to the spatial configuration introduced by theme, one approach may be more suitable than another. Two common examples of mapped thematic data that are examined in the present study are vegetation (forest cover) and surface water (hydrography).
To demonstrate the snapshot method, Chapter 3 is devoted to the use of snapshots in the evaluation of surface water in Portland Oregon between 1880 and 2000. Data were collected for the years 1880, 1909, 1922, 1933, 1945, 1950, 1976, 1983, 1990, and 2000. The base snapshots are evaluated in three ways. First, by overlapping the separate extents of surface water as a series of separate time slices. Second, temporally joining the significant historic events as ancillary information demonstrating the way these historic events have shaped the present surface waters. Third, quantifying the historic information from the snapshot database.
Chapter 4 is devoted to applying a composite data model to land cover changes in the Gray's River Watershed between 1942 and 1996. The Gray's River composite data demonstrates time as an entity of land cover classification. Because of the way the composite model handles time objects, vegetation data exist in multiple states at multiple sample dates. The Gray's River change data conceptually works with an assemblage of time versioned attributes within a vector spatial topology.
The composite data are evaluated by using classification schemes to quantify land cover changes. The most important question concerning vegetation, in addition to "where" or "when" is "what". How did the vegetation exist in previous states? And how did it change from one state to the next? These questions can be answered by how events are modeled within the database.
There are differences in the level of information modeled in the approaches presented in Chapter 3 and 4. Surface water as snapshots in Chapter 3 must rely on a temporal information temporally joined, while Gray's River land cover in Chapter 4 exists as composite integrating temporal data directly in the database. Chapters 3 and 4 discuss data sources and how these data are assembled using the principles of snapshot and composite data.