'Analyses of space-time processes require not only effective tools for multidimensional data, but also detailed spatial and temporal information' (Miller, 1997).
Eric Miller's statement is a crucial observation for any analysis in the field of temporal phenomena. Today we are confronted with huge masses of spatial data which are in one or another way temporally referenced.
Miller implicitly mentions three parts: the data, the handling of them and their analysis. The first aspect of data acquisition is probably further developed than the remaining two. Airborne and satellite remote sensing techniques produce large datasets with a variable temporal resolution often used as input sources for GI-databases. In wildlife research large collections of data are being collected by means of radiotelemetry and especially satellite telemetry. Also in the field of animal behavior research spatial movements of animals are being recorded automatically at intervals of seconds or at even shorter intervals (e.g., Buma et al., 1998).
Some of the methods developed in this work require a high sampling intensity. If the sampling intensity is too low, spurious results may be produced. TT-plots and temporal RDFs require data sets that do not have larger gaps without any information. The present implementation within the TUPF prototype uses interpolation techniques between two observations. This has the benefit of relatively homogeneous plots. If larger gaps occurred in the data, they should be indicated as missing information in a commercial software or other production system, an effect not taken into account in the prototype application developed here.
The advances made in storing and querying such data are quite large, but still require major efforts until they become available in productive off-the-shelf systems (see section 3.4.3). What has not been recognized yet is the fact that by extending GIS with a temporal domain, new analytical methods need to be developed. Today's extensions in this context are mainly concentrating on queries (e.g., Bagg and Ryan, 1997). The discrepancy between the very high level questions asked about spatial change e.g. in global change issues and the available methods for answering such questions is still enormous.
One of the goals set up at the beginning of this work was to develop methods which allow for a better recognition and definition of biologically meaningful phases compared to the methods currently being used. All approaches established in chapters 4-7 are achieving this goal. Especially when used in combination, they provide efficient means to define such phases which need to be analyzed separately.
The conceptional change from a space centered view of the data to a time centered view is completely new in both the biological and the GIS literature body. The TT-plots are a first method performing this shift and thus require some time to become familiar with it. Once familiarized with them, they provide powerful instruments to analyse temporal changes in moving point objects by means of EDA. As this is a completely new approach, it is difficult to compare them with other methods. The only comparable methods are the ones provided by Openshaw and Perrée (1996) discussed earlier and the analytical plots used in sleep research (Borbely et al., 1998,1981), although the latter is used strictly with non-spatial data. In these plots the activity (active, not active) of a monitored person or animal is plotted with the x-axis representing the daytime and the y-axis representing the calendar date. The activity is indicated in black or white at the specific location to recognize temporal patterns within the data.
One of the biggest advantages of TT-plots over traditional approaches is the speed at which changes in the data can be located. The plots are computed within seconds, and make use of our most powerful way of pattern recognition, the visual perception. As an example the definition of homogeneous phases, which normally takes one or two days, can be performed within minutes (pers.comm. Prof. Dr. U. Reyer, University of Zurich).
Even though the form chosen in this work for representing the TT-plots was based on a color scheme, the reader needs to keep in mind that TT-plots consist of a matrix of calculated values such as distances or angle, which can be used for further analysis. They are not to be confounded with visual techniques such as a DTM hill-shading or the like. The calculated values are directly usable and available for further analytical steps or the calculation of derivatives. It may well be the case that other techniques are developed to visualize the data or make their interpretation easier.
The basic concept behind TT-plots is not limited to the examples presented in chapters 5 and 6. It opens up a large new field for extensions and adoptions. A wide variety of parameters and measurements can be used in the calculations of TT-plots. It is also possible that the concept could be adapted to other non-spatial problems where regularities in datasets are of interest.
The interpretation of TT-plots needs some training just like any other analysis method. The idea of creating automatic interpretation mechanisms and programs may be an approach in the future. The design of TT-plots was an open one to depict various kinds of regularities in the data. Using fixed search algorithms in computer programs for this task of finding regularities might contrast this basic idea. Maybe the application of a modified version of Openshaw's STACs (Openshaw, 1991) would allow for an appropriate machine learning process for new occurring regularities.
In chapter 4 a switch from manual, text-oriented selection of data to a graphical selection was made. In combination with effective methods to select for different temporal aspects such as the ones related to the sun or moon allows for a very efficient handling of such aspects in biological analysis. The neglect of these important factors in most previous studies concerned with wildlife animals expresses the need that such instruments for calculating and handling biologically important influences on the behaviors of animals should be available within the most capable analytical framework for such data, the GIS. Today such calculations can only be performed if professional help from astronomers or highly skilled computer technicians with knowledge in Fortran programming. This situation needs to be improved. As it becomes more and more clear that GIS will become the standard platform for analyzing data from wildlife research studies, it seems to be reasonable to stimulate and concentrate efforts for such extensions in GIS.
Bagg and Ryan (1997) recently reported on an application on historic land ownership, creating a temporal model within the Illustra database management system 8.1 to store and query the data. They use a similar approach as used in the TDF concept to retrieve data. Due to the application in historic changes of land ownership, it was constrained to a single linear time aspect. They also implemented different temporal rules of temporal topology (e.g. before, during, overlap, etc.), but did not include mechanisms for dealing with relative time. Time was only considered as a selection criteria and not as a separate domain for analytical purposes as in the present work.
In section 2.2 the need for 'time systems' comparable to coordinate systems in the case of space was expressed. Whenever time is recorded, the information about its reference is needed, although it is almost always forgotten to document it. This includes aspects as for example the system used (UTC, time zone, Julian Date, etc.) the actual time zone, the use of daylight saving time and others. It is an often encountered source of error that researchers forget to adjust their data for the daylight saving time in calculations. These 'time systems' are by far not as complex as the coordinate systems, but complex enough to prevent the casual GIS user to transform between them for calculations like sunset or moon rise. Especially in cases where data are exchanged it is necessary to know exactly what the time values mean.
TDF and Time-Plots are very efficient with respect to computing resources. Temporal RDFs require a lot of geometrical calculations and thus need several minutes to a few hours to be calculated, depending on the temporal resolution and the length of the observation period. In the present implementation of the RDFs all the calculations are performed using vector operations. Implementing the RDF in a grid-based environment would presumably enhance the calculating speed to a large extent. Nevertheless the computing time for temporal RDFs (and of course for the basic RDFs, too) depends strongly on the geometrical complexity of the data as well as the temporal and spatial resolutions required for the analysis in question. Considering the development of computing power over the next five or ten years, these issues probably become neglectable.
A problem that is present throughout the analysis of spatial data in wildlife research and spatial information in general is the problem of scale. Problems of generalization show up when implementing TT-plots or TRDFs. The approach used here to address these issues was a pragmatic one. Today most data collected in wildlife research are collected independently from GIS. The scale and accuracy of the data are determined by the field researcher. In the case of visualization data such as TT-plots, the possibility to zoom into the data allows the user to browse over different scales of the data, provided the user does stay within the limits imposed by the data themselves. The TT-plots are intended for interactive usa, where this is possible. For the creation of visualizations the limits imposed by the hardware (screen resolution) are being used.
In the introduction to this thesis I provided some of the general objectives pursued in this work. I hope I was able to provide a better understanding for time and temporal aspects. The goal of developing improved and more accurate analysis methods for temporal aspects in wildlife research data within geographical information systems was achieved, even though it is clear that this can only be the start of a large field of research. I think it became evident that in the context of temporal geographical information systems not only sophisticated data models to represent temporal aspects are needed, but also new and improved analytical methods need to be developed to adequately analyze the data. This work concentrated on the single data type of point objects. Having seen the interesting topics involved in exploring time in such data, I assume that further developments for other data types such as lines, polygons or fields can be even more exciting, but also more complex. I think the aim to develop analytical instruments to perceive and define biologically 'meaningful' phases in a wildlife researcher's data was fully achieved.
As stated in the introduction, even though most examples and statements were made primarily with a focus on wildlife research, the methodology is of course applicable to any context concerned about the spatial movements of point objects.
A lot of research is being performed with the aim of identifying a very specific thing. The goal in this work was to open up new ways of thinking and analyzing, as my colleage Emmanuel Schmitt once stated:
Questions should be answered in a way that more possibilities emerge. (E.Schmitt, personal communication)
After all, it might be a good advice to keep in mind that not everything we see necessarily needs to be true. As Couclelis (1996) states in her article about 'Geographic Illusion Systems' (GIS?)
... that GIS through its images, and what can be done with them, creates beliefs (mythologies, some would say) and molds habits of mind in thinking about the world unlike any that would exist without it.