A research paper submitted in partial fulfillment of the requirements for the degree of
MASTER OF SCIENCE in GEOGRAPHY Portland State University 2001



This chapter explores change analysis in a GIS application created to measure, quantify, and analyze the spatial variation of landcover over time. It examines how historic landcover data are assembled and attributed as a space-time composite database. This chapter looks at some examples of querying the composite database to determine how forest landcover age data can be used to demonstrate change. The chapter concludes with a discussion evaluating the data model.

In contrast to snapshot data, composite data integrates the event data as part of the spatio-temporal model. The temporal chronology of events present in the database separate past, present, and future object states. In the database these states form a temporal topology, a condition that exists in composite data where a single state at one point in time contains references to multiple states at adjacent points in time. This characteristic can be used to better understand the chronology of changes to both land cover and topology by considering states as bounded by temporal events.

The diagram on Figure 6 explains how states are partitioned by events. The state recorded in the database occurs at the point in time when the measurement takes place. This point of measurement is where there is evidence to confirm that a state exists in a certain form of existence. How long the state exists is determined by the way the database stores time. In Figure 6 events are separated from the point of measurement by lag time, where there is a difference in time between the event and where the event creates a change of state.

Taking into consideration the interaction between events, states, and lag time, the composite data model becomes an appropriate choice to use with forest landcover because of the way the forest changes state in time. If we assume that the state of the forest is determined by age, then there are a number of definable parallels between the data model and forest cover.

The information contained within the database relies on the structure of the composite data model. Each land cover state is bounded by events that establish relative beginning and ending point of a land cover type. We assume that the events that bound a state in forest data are known as 'clear-cuts'. Clear-cuts can easily be identified on source material because they do not contain trees, but are surrounded and delineated by trees. When a forest is clearcut, all the trees are cut about the same time and all the young trees that grow in place will be similar in age. These changes to the forest state occur repetitively across the landscape in contiguous and non-contiguous, irregular shaped patterns. Knowing when and to what extent these events occurred allows us to determine the state of the forest at some point in time.


4.1 Historic Map Sources in the Gray's River

Landcover data obtained from map and air photo sources (Table 6) include the years: 1942, 1953, 1964, 1976, 1983, 1990, and 1996. Data for 7 time slices came from various sources listed on Table 6. These sources include hundreds of Washington Department of Natural Resources (WDNR) low elevation aerial photographs including stereo-pairs. No satellite imagery was used as a data source.

Source Scale Source Source Type Digitized Features Source Date
1:12,000 Washington Department of Natural Resources orthophoto quadrangles clear cut 1996
1:24,000 Soil Conservation Service, U.S. Department of Agriculture controlled mosaic clear cut


1:12,000 Washington Department of Natural Resources low elevation stereo pairs clear cut 1964-1990
1:24,000 War Department, Corps of Engineers, U.S.ARMY controlled mosaic, photo map old-growth, clear cut 1942

Table 6 Data Sources.


4.2 Converting Map Sources into GIS

Converting aerial photography into GIS proved to be a formidable endeavor. The critical part about interpreting event data needed for the model to work involved identifying and delineating these occurrences in the cleanest format possible. Procedurally, hundreds of 1:12,000 un-rectified photos were interpreted and their contents merged onto 3 rectified air-photo bases used for digitizing. Since each mosaic contained the delineations of multiple air photos, the digitizing process was procedurally more efficient than digitizing each photo separately.

The digitizing resulted in the creation of three composite data layers. The three layers were then combined in a spatial overlay that spatially joins both topology and attributes. In the composite database each of the digitized objects identity must be maintained throughout the temporal evolution of topological relationships obtained through the overlay of separate time slices (Yaun 1995).

The overlay produces a single composite database needed for modeling states of forest cover. Included in the post overlay database are two attributes used to uniquely identify objects, the first is a system generated identification number [forest_id] the second a code [code]. Further development of the [code] in the database is required to make the attributes a useful source of information.


4.3 Data Development

The attribute information contained in the post overlay database resides in the cryptic attribute [code] produced from interpreting forest states from air-photos. The key procedure to data development is extracting the attributes contained within the [code] to other fields within the database. Table 7 provides examples of how the 3-digit coding scheme relates to text information contained in other fields. The first digit place in [code] defines an event as taking place between a beginning and ending point. The second digit defines beginning and ending points for a second event.

[CODE] First digit First Attribute Event Description [CODE] Second digit Second Attribute Multiple Event Descriptions
100 cut between 1942-1953    
200 cut between 1953-1964    
300 cut between 1964-1976 310 cut in 1942 and again between 1964-1976
400 cut between 1976-1983 410 first cut between 1964-1976, second cut between 1976-1983
420 first cut between 1953-1964, second cut between 1976-1983
430 Power Lines, first cut between 1964-1976, second cut between 1976-1983
500 cut between 1983-1990 510 first cut between 1942-1953, second cut between 1983-1990

first cut between 1964-1976, second cut between 1983-1990

530 first cut between 1953-1964, second cut between 1983-1990
600 cut between 1990-1996 610 first cut between 1942-1953, second cut between 1990-1996
620 first cut between 1953-1964, second cut between 1990-1996
630 first cut between 1964-1976, second cut between 1990-1996

Table 7 Internal structure of the coding scheme.


The database in a static rollback form stores the occurrence of events and the state of the forest at any point in time as attributes. Because forest age is dependent on time, the attribute describing the state of the forest depends on when it is measured and the database rolls back to the date for when the event is recorded. Age as an attribute value is computed as accruing from an event. For example, forest age immediately following a clearcut event is equal to zero in the database. The forest age measured from some time after the event is equal to zero + the time elapsed after the event. The database does not make this calculation automatically.

The database models forest states from the past to the present by information stored as attributes. Events trigger changes to the state to the forest in the database by changing the value of the forest age attributes. These changes created by events basically establish a chronology to the cycle of forest growth: as young trees grow old, they are replaced by old trees, and so on.

In the database, land cover attributes make special considerations for time data by storing age information. Land cover attributes provide detail about the forest state and describe object states at different points in time. For example, the attribute text "first cut between 1953-1964, second cut between 1983-1990" describes the state of the forest as being conditional to one event occurring between 1953-1964 and another event between 1983-1990. The text does not apply or have significance to other events occurring at any other time period. The temporality of the attributes contained in the database become clearer with query expressions designed to evaluate forest changes.


4.4 Query Development

The temporal and land cover information content of the database is accessible through fields contained within the database. These fields contain attributes that return specific details about both historic events and forest states for different spatio-temporal objects. For example, the database fields contain text and classification schemes including nominal, ordinal, and relative typologies. By including alternative classification schema with fields, the database includes quantifiable land cover classifications for pertinent land cover states. Information contained in Table 8 is specific to age-based forest landcover and designed to quantify forest clear-cutting events.

Field Name Description
Code Forest Landcover Classification System Code
Fourtytwo Forest Landcover age in 1942
Fiftythree Forest Landcover age in 1953
Sixtyfour Forest Landcover age in 1964
Seventysix Forest Landcover age in 1976
Eightythree Forest Landcover age in 1983
Ninety Forest Landcover age in 1990
Ninetysix Forest Landcover age in 1996
Yearuse The year first cut (date of the first event to set age = 0)
Standtype Stand generation (first, second, or third) growth in 1996
Stand_age Stand age in 1996
Period_cut Various dates when first, second, and third cutting occurred
Acres Area / 43,560

Table 8 Fields contained in the composite data attribute table.


To demonstrate the usefulness of the database, we present spatial queries that identify and return states of forest cover at different times. For example: determining the acreage of old-growth forest in 1976. To access this information requires the use of the attributes contained in the field [1976]. The field contains forest age values that range from 0 to 999. Selecting the old-growth component means that the query expression identifies and separates old-growth polygons having an attribute = '999' from polygons containing other values. The Arc-plot expression "reselect [1976] = 999" returns the selection of polygons representing the extent of old-growth in 1976.

Identifying and selecting polygons at different points in time between 1942 and 1996 in the database require the use of the fields [1942, 1953, 1964, 1976, 1983, 1990, and 1996]. For example, the same kind of database selection used to identify and return the extent of old-growth polygons in 1976 can be applied to 1983, the extent of old-growth land cover 7 years later. These fields make it easy to select and compare the extent of each forest land cover age group contained within the database at various points in time.

Another example of the information obtainable in the database involves a query designed to identify the state of forest age in 1996. To do this, the field [stand_age] is classified by age. Because the values contained in [stand_age] range from '0' to '999', a forest classification scheme partitions age into a user-defined number categories displays the extent of forest land cover age relative to the most recent event. In this example, the '0' value identifies the most recently affected by clearcut events and the '999' value identifies those polygons that have never been affected by such an event.

Another example identifies and returns the total number of events that have occurred. To do this, the database returns text contained in the field [period_cut]. The text contains descriptions of beginning and ending points for each clear-cut event occurring at any point between 1942 and 1996. One example of this text obtained for a location in the upper section of the watershed, "first cut between 1942 and 1953, second cut between 1990 and 1996". This information can be used to identify polygons where up to three events have occurred


4.5 Evaluating the data model

These data were constructed to quantify changes to forest states by having the capacity to identify features by location, date, and age. The database in its present form relies on the principles involved with a static rollback design. According to Armstrong (1988) each event occurs at a database transaction time when it actually did not. Therefore a lag exists in the database between the event and when the event changed the state of the forest.

In the present form of the database, the text "first cut between 1942 and 1953, second cut between 1990 and 1996" indicates the occurrence of two events. Since the database relies on cartographic time, there is no way to tell how far an actual event is away from how it exists in the map evidence. There is no way to tell exactly when an event occurred, just an indication that it occurred between 2 points of time. Error present in cartographic time is not constant because the samples used to obtain map evidence are not equal. Events vary from their actual occurrence as little as six years to as much as 12 years.

The temporal database is adaptable to what Armstrong (1988) describes as a historic database design. Making the transition from static rollback to historic only affects the attribute table because the topologic relationships are the same for both database designs. The main difference between designs is that the historic database relies on a valid time attribute and the temporal design does not. If the database is converted from static rollback to historic or temporal, then the time lag associated with cartographic time goes away because valid time becomes part of the database. For example, the present static rollback database returns the text, "first cut between 1964 and 1976, second cut between 1990 and 1996". In the historic or temporal database this text would read, "first cut July 13, 1966, second cut September 9, 1996". Incorporating valid time in the database allows the database to return specific information about when an event occurred rather than settling for cartographic time and the use of dependant variables where events are always waiting for the next map version to occur. Unfortunately, the valid time information required for the historical database is not included with the source mapped evidence.

One way to improve on the existing database is to use a generalized classification scheme. The error introduced by not knowing exactly when an event occurred is reduced if the measurement can include the highest number of sample points possible. In this case, stretching out the classification scheme by reducing the number of possible categories reduces the likelihood of misclassification because each group contains more data. For example, forest age classifications use the Washington Forest Practice Rules Board. The classification identifies trees less than 50 years of age into two groups, 1-29 and 30-49 years old. Because the classification uses 2 age groups rather than a higher number, the result is a reasonably accurate classification of 1-50 year old trees.

At some future point the database must be updated. Composite data requires changing the topology, adding a lot of baggage to accommodate new events. Adding new events may require recalculating significant portions of the attribute table. The integration of new events can be done with the use of spatial overlay processing. New events are integrated into the database at the most recent endpoint of the timeline.

<-- Chapter 3
Return to Menu
Chapter 5 -->