Data Time

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

EXPLORE MULTIVARIABLE SPATIO-TEMPORAL DATA WITH THE TIME WAVE

CASE STUDY ON METEOROLOGICAL DATA

Xia Li a,b, *, Menno-Jan Kraak a

a

ITC- International Institute for GeoInformation Science and Earth Observation

PO Box 6, 7500 AA Enschede, the ***********-@***.**

b

College of Earth Science And Resources, Chang an University

Commission II

KEY WORDS: Time wave, Temporal exploration, Time space

ABSTRACT:

Traditionally the GIScience community is well able to deal with the locational and attribute component of spatio-temporal data.

However, the methods and techniques to deal with the data s temporal component are less developed. This paper introduces a

conceptual framework that combines user tasks, available temporal data and visualization theories to discuss temporal visualization.

Two limitations of existing method are improved by introducing the time wave environment which is a close combination of

temporal graphic representation and temporal interactive tools, and operates in so-called time space. A case study based on

meteorological data illustrates the approach.

precipitation per month, season or year? How is it linked to

1. INTRODUCTION

location space (e.g. maps) and attributes space (e.g. diagrams).

Many of the most important challenges our society is facing

today, such as global climate change, economic development In the wider context of GIScience research on temporal data

and infectious diseases depend on spatio-temporal data to detect analysis and modelling (Goralwalla, Ozsu et al. 1998) and

and analyze changes as well as trends to support problem temporal visualization are receiving increased attention

solving. Especially the temporal component of the data should (Mackinlay, Robertson et al. 1991; Harrison 1994; Allen 1995;

be studied carefully to understand the changes and their impacts. Brown 1998; Harris, Hetzler et al. 2000). This includes papers

Current data collection techniques offer a wide variety of that offer an overview over existing visualization methods

thematic data in many different spatial and temporal resolutions. (Andrienko, Andrienko et al. 2003; Aigner, Miksch et al. 2007).

From a temporal perspective, earth observation techniques Most of these focus on applications in the information

provide data with temporal resolutions varying from weeks, visualization field or on discussions of specific aspects of

days, hours, to even minutes. The challenge faced is how to temporal visualization such as static and dynamic

process, manage, and use these continuous streams of data to representations (Muller 2003), linear time (Silva and Catarci

support problem solving and decision making. The application 2000) and the visualization process (Chi 1998). Others take a

of graphic representations in a dynamic and interactive more overall approach. Aigner et al (Aigner, Miksch et al.)

geovisualization environment is part of the solution. discussed temporal visualization based on a time-oriented

framework. Andrienko (Andrienko, Andrienko et al.) focused

Geovisualization integrates approaches from disciplines on spatio-temporal exploration and considers the time

including cartography with those from scientific visualization, visualization from both data and user tasks. However, compared

image analysis, information visualization, exploratory data with location space, and attribute space, time space hasn t been

analysis and GIScience (Dykes, MacEachren et al. 2005). The studied very well.

graphic representations, mostly maps, are used to stimulate

(visual) thinking about geospatial patterns, relationships and In this paper, a proposed solution is to take a closer look at the

trends. This is strengthened by looking at the data in a number integration of several visualization theories accepted in

of alternative ways. Playing with the data using these multiple GIScience and information visualization, and see how to extend

representations without constraints (traditions) will trigger the these with a specific temporal component. This is done by

mind of the users, and lead to an increase in their knowledge. analyzing user tasks to structure the temporal data. Hereby the

This does improve our understanding of how to use limitations of existing temporal visualization method will be

visualization to get a better insight into spatial data, but not discussed and a specific graphic representation will be

necessarily in temporal data. This is partly due to the fact that suggested. To structure time space the suggested graphic

most methods and techniques used to solve the geo-problems representation should allow the user to represent the temporal

are from either location or attribute perspective. This paper will data, to display different views on time (e.g. linear or cyclic),

look at the geo-problems from a temporal angle, from what is and last but not least should allow for interaction. For this the

called time space (Li and Kraak 2008). Time space is a time wave was introduced (Li and Kraak 2008). In the next

visualization space to represent time and answer the temporal section, a commonly accepted visual problem solving approach

questions; for example, what is the temporal distribution of is discussed with specific attention for the nature of the data, the

* Corresponding author.

575

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

well (day, season). The granularity of time is defined by the

user task and the visualization environment. After this

different units in which time is expressed, and can be used to

background information a temporal conceptual framework for

define the scale of time. These granularities inherit from both

visualization is presented, followed by a case study that

linear and cyclic characteristics (hours, seasons).

demonstrates the alternative approach.

From the temporal data aspect, the views on time such as linear

or cyclic, continuous or discrete, single scale or multiple scale,

2. VISUALIZATION THEORY AND TIME SPACE

should be considered and addressed in time space properly. In

2.1 Visualization theory relation to the task each view might have different requirements

for temporal visualization. For example, the question how did

A common approach to support problem solving with the city expand? could be answered with discrete multiple

visualization is shown in the scheme in Figure 1. A set of tasks, views showing a set of snapshots of the city s extent.

translated into questions is executed in an appropriate Alternatively an animation could be used to answer the same

visualization environment with suitable data to solve the question in continues view. The dynamic nature of the

problem at hand. Three keywords are of importance: user tasks, animation could also attract attention to other aspects of the

data framework and visualization framework. The visualization expansion. Based on the character of the data, a linear or cyclic

framework includes the graphic representation and the representation might by appropriate. Both could express either

functional tools to play with the graphics. continuous or discrete time. Although there might be a

straightforward solution in the selection process of the visual

representation, it could be useful to try other visualizations as

well, because alternative view might reveal patterns or aspects

of the data that could remain hidden in the straightforward

solution.

2.2.2 Temporal user tasks and temporal visualization

Even though the selection of a graphic representation is based

upon the nature of the data, the user task plays an important role

as well, since at the end, the graphic representation has to

answer questions. For instance, the question When did

countries get their independence? might require different

representations depending on the focus of the question. A

timeline with names and years along will do if the temporal

distribution is the focus. If one is also interested in the spatial

Figure 1. Visual problem solving: The relation between user distribution then a world map with labels indicating the year of

tasks, a data framework and a visualization framework independence could the most suitable representation. If both are

of interest, a space-time-cube (H gerstrand 1967) could be a

Relevant questions are: How to abstract questions based on the possible solution. When one would also be interested in the

data? How to address the data in a visualization environment? season of independence the graphic representation should be

How to decide on the right graphic representation(s) and the able to handle both linear and cyclic time and the time wave

required functions to answer the questions? Most cartographic might be a solution.

and geovisualization theories are studied by looking at the

interrelationship of these tightly coupled aspects, such as Bertin Shneiderman s visual information seeking mantra is widely

(1983), Macrachren (1995), Peuquet (2002) and Andrienko accepted in the visualization field and can be applied in time

(2006). space. First, the temporal graphic representation is used to

locate the time and show the temporal overview, and then

2.2 Time space temporal interactive tools are required to carry out temporal

zooming and temporal filtering options to get the details-on-

Time space includes temporal graphic representations and demand. This process is often an interactive and iterative

temporal interactive tools. Here the requirements of time space process, which is supported by a close combination of

will be discussed from temporal data and temporal user tasks representation and interactive tools in time space. However, this

perspectives. close combination has not been realized in temporal

visualization. There are many temporal graphic representations

2.2.1 Temporal data and temporal visualization which could display a temporal pattern of data, such as

Compared with attribute and location, time is considered ThemeRiver (Havre, Hetzler et al. 2000), Stacked bar chat

special. Attribute and location can change over time, but the (Harris 1999), People garden (Xiong and Donath 1999),

reverse is not possible. The notion of time is difficult to grasp. MultiCombs (Tominski, Abello et al. 2003) and etc. To assist

This has led to many different views on time. Examples answering the questions, temporal interactive tools should allow

relevant in GISciense are found in Frank(1994), Goralwalla, one to identify single or multiple points in time, identify points

Ozsu et al. (1998) Peuquet (2002) and Aogner, Miksch et al. at periodic intervals, or define intervals of certain length in both

(2007) . It can be said time passes continuously, but it is the linear and the cyclic format. Some have experience with this

observed and measured in discrete points or intervals. This approach, like Koike et al. (1997) with TimeSlider and Edsall et

results in a view of continuous time versus discrete time. Time al (1997) experimented with a time wheel query device in the

can also be considered as relative (e.g. last week) or absolute TEMPEST system. Above temporal visualization methods are

(e.g. May 27th). The continuous nature of time makes it linear either a temporal representation or a temporal interactive tool

(before, after) but it obviously holds cyclic characteristics as only.

576

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

3. TEMPORAL VISUALIZATION AND TIME WAVE:

AS A NEW APPROACH IN TIME SPACE

Most temporal representation are either controlled by a timeline

(linear time) (Mackinlay, Robertson et al. 1991; Plaisant,

Milash et al. 1996; Silva and Catarci 2000; Wijk 2002) or a

time wheel (cyclic time) (Mackinlay, Robertson et al. 1994;

Carlis and Konstan 1998; Harris 1999; Daassi, Fauvet et al.

2002). However, this does not always result in satisfying

solutions because many phenomena have both linear and cyclic

characteristics. The time wave (Figure 2; Li and Kraak 2008) is

one potential solution for representing both the linear and cyclic

nature of data. It is a combination of the timeline and the time

wheel and offers an alternative view by its own. With the

timeline and the time wheel it is difficult to show multiple time Figure 3. The interactive options to move from time space into

scales. However, the time wave can show multiple time scales location and attribute space supported by CMV principle

by nesting different waves with different wavelengths and

amplitudes based on their temporal scale. 4. CASE STUDY: TIME SPACE AND

METEOROLOGICAL DATA

To see how to create a proper time graph based on the analysis

of temporal data, user tasks, a case study is discussed in this

section. The data set contains observations of nineteen

meteorological stations in Beijing, China. For each station,

known at location (x, y, z), the temperature and the dominant

Figure 2. The time wave as a construct of the time line and the land use information are given. The temperature was measured

time wheel combining the linear and cyclic nature of time (Li at minute interval over the months in July 2007 and July 2008.

and Kraak 2008) A data overview is shown in Figure 4. The location data is point

data in context of geographical units (districts in Beijing city);

Time wave not only could show the linear and cyclic time the attribute data is quantitative (temperature) and qualitative

temporal pattern of data as a temporal representation, but also (land use); the temporal data has linear characteristics

carry out zooming and filter as an interactive tools. For (measurements over a month) and multi-scale cyclic

temporal interactive tools, the time wave can do both linear and characteristics (July for two year and days and nights for 24

cyclic operation as well. In addition, the interactive time wave hour periods).

also support one to represent some locational and attribute

characteristics in relation with the phenomenon s temporal

distribution which can be explored.

Furthermore, the time wave allows users to interact with the

temporal reference or temporal representation. Supported by

Coordinated Multiple View (CMV) (Roberts 2005; Roberts

2008) principle, one could identify or select any time instant or

interval at various scale, or select a temporal pattern, to show

the corresponding information in the location space or attribute

space. Figure 3 shows an interpretation of time space with the

time wave combining the representation and manipulation

functions to carry out the interactive and iterative exploration

process. The figure also demonstrates the link between time

space, location space and attribute space. With a temporal

question, one starts in time space and depending on the nature Figure 4. Case study data: the characteristics of the

of the question, the answer can be given in time space itself or meteorological stations in the Beijing area.

one jumps out of time space into location or attribute space.

For instance, the question related to year of independence of the The user task is to find out when the temperature reaches its

countries is an example where one moves to location space daily maximum value at the different meteorological stations. Is

when one needs to see spatial patterns next to temporal patterns. the pattern related to the longitude, latitude and elevation of

In answering complex question one might have to jump from each meteorological station? What are the other impact factors,

space to space in an extensive iterative process to identify and such as land use? Is there any difference between those two

compare temporal, spatial and attributer patterns. years? The temporal questions, such as when, how often, in

what order, synchronization and trends, always have a relation

to other data components. For instance, from object perspective

(station name, station ID), location perspective (height;

east/west; north/south), attribute perspective (low, high, or

average temperature; land use category like urban, vegetation,

water, or other).

577

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

located on the wave at the time when the highest temperature

was observed. It is obvious that a wave like this will only offer

Based on figure 4, figure 5 gives a schematic overview of the

and overview. A first glance at the data shows that the highest

coordinated multiple view environment in which time space

temperature is reached a few hours after 12 noon (the top of the

plays a key role. The time wave is selected as graphic

wave). However, both at the beginning and the end of the

representation since it can handle the temporal data

month an anomaly in the temporal pattern can be observed.

requirements of the case study: linear and cyclic multi scale

These two anomalies have been further explored in the Figure.

time. The wave length could be months, weeks, days or any

In both cases one can first zoom in the wave to see more details.

other time unit one would need. It also allows interaction and is

At the beginning of the month (the left of the Figure) two

directly linked to location and attribute space which is needed if

stations are rather late in reaching their maximum temperature

one has to switch spaces while executing the user tasks. In

and at the end of the month (the right of the Figure) there is a

Figure 5 this is represented by the in and out arrows. Location

single day on which all stations are late in reaching their

space show the map with the position of the meteorological

maximum temperature. In the first situation it makes sense to

stations and attribute space shows a scatterplot of the station s

get more details on those two particular stations, and switch to

temperature versus land use. Figure 6 gives a snapshot of the

location space to see if their location is special. The map

actual software. The time wave is a plug-in for the Udig open

doesn t reveal anything special, since stations with similar

source GIS. The time wave can also include a visual

geographic condition can be found. For the second situation it

representation of the data s attribute and location characteristics.

was decided to switch to attribute space and have a closer look

In Figure 5 symbols related to the meteorological stations, the

at the temperature ranges of all stations at that particular day

temperatures observed, as well as land use could be plotted on

and the days before and after. It can be observed that the

the wave. Figures 7 and 8 give examples related to the case

maximum temperature for that particular day was also

study.

considerably lower than the days before and after. The available

data do not give a good explanation. Other meteorological

parameters, for example cloud cover, may be introduced to give

an explanation. The experts will have to continue exploring the

data and perhaps obtain additional information in order to be

able to explain the anomalies.

Figure 5. The working environment of the time wave, with

options to visually represent characteristics from location and

attribute space on the wave

Figure 7. Working with the time wave: identify and explain

temporal patterns. At the top the overview of July 2007

showing all stations at the moment these observe the highest

temperature. In the middle zooming in on two clear anomalies

in the monthly temporal pattern. At the lower left a map in

location space and the lower right a temperature diagram in

attribute space.

Figure 8 shows another useful function of the time wave. The

top wave is the normal time wave, here showing again the

moment that the daily temperature reaches its maximum at each

meteorological station for three days. It is possible to create a

set of parallel waves (Figure 8b). For each meteorological

station a separate wave is created to obtain a better view on the

Figure 6. Snapshot of the time space/time wave environment as patterns. This approach has been derived from the parallel

implemented in the open source GIS uDig coordinate plot. The wave in Figure 8c shows the data sorted

based on the values of July 1st. This result in having the station

Figure 7 gives two examples of how time space works in with the earlier highest temperature in the lower wave. Keeping

practice. In the upper section of the Figure an overview of the this sort order the following days show a different pattern. It

month July 2007 is given. The wave length represents a day. means the order of each stations reach its maximum

For each day and each meteorological station, a symbol is temperature varies over different day. It is possible to sort the

578

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

solving approach. The nature of the temporal data and the user

data based on any other available variables. For instance, one

task at hand are the driving forces to select a suitable graphic

could sort based on the height of the stations (Figure 8d), or sort

representation. In time space the graphic representation focuses

from east to west or north to south. After sorting, one might

on the temporal aspect of the data. Here the time wave is

observe patterns that will stimulate one to jump to location or

introduced. It not only combines linear and cyclic time, but also

attribute space, to look for more detail, to switch back to

combines temporal data representations and interaction, and

overview mode or even to retrieve more data. Working with the

allows a limited representation of attribute and location data.

time wave to explore temporal patterns is clearly an iterative

The exploratory activities are guided by the visual information

process. In the next section the specific advantages and possible

seeking mantra of Shneiderman. Depending on the user task,

limits of the approach will be discussed.

one can switch from one space to the other, and into any of the

information seeking modes (overview, zoom / filter, and details

on demand).

A case study based on the data observed at meteorological

station in Beijing has demonstrated the time space framework

and the capabilities of the time wave. Temporal patterns and

distribution are studied from time space switching to location

space (maps) and attribute space (diagrams) when required. The

time wave is a good example of how an alternative view on the

data might reveal patterns not always obvious from traditional

graphic representations. In overview mode it is a very suitable

visualization to provide an impression of the nature in data,

being linear or cyclic. In this mode the wave easily reveals

anomalies as well, as for instance in figure 7. This effect can be

strengthened by moving the horizontal line (x-axis) vertically

through the wave.

It is not argued that the time wave is the only visual

representation in time space. Graphics based on the time line or

time wheel only, would be preferred in certain situations. This

very much depends on the nature of the data and the task at

hand. However, it is claimed that the optimal solution space is

time space interactively linked with location and attribute space.

This allows the user to flexibly tackle the problem from many

different perspectives.

Future work will deal with a detailed validation of the time

wave in other case studies and its relation to graphic

representation in location and attribute space. Known data sets

will be used to see if and how the time wave might discover

known patterns and possibly reveal new temporal patterns as

well, based on tasks executed by users. In relation to the task

space (see figure 3) further work is ongoing in which existing

temporal visualization methods are be analyzed on their

strength and weakness in the context of the actual temporal

problem of users. If successful this could lead to a kind of

advisory system that assist the user to select a suitable graphic

representation in time space dedicated to the user task one is

dealing with.

ACKNOWLEDGEMENTS

Figure 8. Working with the time wave: comparing temporal

This research Supported by the Programme of Introducing

patterns. a) an overview of the first three days of July 2008

Talents of Discipline to Universities, 111 project No:

showing when the maximum temperature was observed for each

B08039

station; b) temporal stretch by creating a set of parallel waves to

un-clutter the pattern as seen in the upper wave; c) sorted wave

based on the temperature values of July 1, 2008; d) sorted by

REFERENCE

station s elevation (highest elevation in lowest parallel wave).

Aigner, W., S. Miksch, et al., 2007. Visualization Time-

5. CONCLUSIONS AND DISCUSSIONS

oriented Data--- A Systematic View. Computers and

Graphics, 31, pp.401-409.

In this paper an alternative approach to work with spatio-

Allen, R. B., 1995. Interactive Timelines as Information System

temporal data has been presented based on a visual problem

Interfaces. Symposium Libraties, Japan.

579

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. 38, Part II

Mackinlay, J. D., G. G. Robertson, et al., 1991. The Perspective

Andrienko, G. and N. Andrienko, 2006. Exploratory Analysis of

Wall: Detail and Context Smoothly Integrated. ACM

Spatiall and Temporal Data. Springer, Berlin,

CHI'91, New York.

Germany.

Muller, W., Hetdrun Schumann, 2003. Visualization Methods

Andrienko, G., N. Andrienko, et al., 2003. Interactive maps for

for Time-dependent data-Overview. Winter

visual exploration of grid and vector geodata. ISPRS

simulation conference 2003, New Orleans USA.

Journal of Photogrammetry and Remote Sensing,

Peuquet, D. J., 2002. Representations of Space and Time.

57(5-6), pp.380-389.

Guilford, New York.

Andrienko, N., G. Andrienko, et al., 2003. Exploratory spatio-

Plaisant, C., B. Milash, et al., 1996. LifeLines: visualizing

temporal visualization: an analytical review. Journal

personal histories. Conference on Human Factors in

of Visual Languages & Computing, 14(6), pp.503-541.

Computing Systems Vancouver, Canada ACM

Bertin, J., 1983. Semiology of graphics : diagrams, networks,

Roberts, J. C., 2005. Exploratory Visualization with Multiple

maps. University of Wisconsin Press, Madison.

Linked Views. Exploring Geovisualization. J. Dykes,

Brown, I. M., 1998. A 3D Interface for Visualization of Serial

M. A.M. and M.J.Kraak. London, Elsevier. 8: pp.159-

Periodic Data. ACM GIS'98, Washington DC USA.

180.

Carlis, J. V. and J. A. Konstan, 1998. Interactive Visualization

Roberts, J. C., 2008. Coordinated Multiple Views for

of Serial Periodic Data. ACM UIST'98, San Francisco.

Exploratory GeoVisualization. Geographic

Chi, E. H., et al., 1998. Visualizing the Evolution of Web

Visualization: Concepts, Tools and Applications. M.

Ecologies. ACM CHI'98, CA USA.

Dodge, M. Mcderby and M. Turner. Chichester, John

Daassi, C., M.-C. Fauvet, et al., 2002. Multiple Visual

Wiley & Sons Inc. 3: pp.25-48.

Representation of Temporal Data. Proceedings of the

Silva, S. F. and T. Catarci, 2000. Visualization of Linear Time-

13th International Conference on Database and

oriented data: a survey. Web Information Systems

Expert Systems Applications Springer-Verlag.

Engineering, 2000.

Dykes, J., A. M. MacEachren, et al., Eds. 2005. Exploring

Tominski, C., J. Abello, et al., 2003. Interactive Poster: Axes-

Geovisualization. Elsevier Science & Technology

Based Visualizations for Time Series Data. Poster

Books

Compendium of IEEE Symposium on Information

Edsall, R., M.-J. Kraak, et al., 1997. Assessing the

Visualization, IEEE.

Effectiveness of Temporal Legends in Environmental

Wijk, J. v., 2002. Image based flow visualization. SIGGRAPH,

Visualization GIS/LIS '97 Annual Conference and

San Antonio, USA.

Exposition Cincinnati, Ohio

Xiong, R. and J. Donath, 1999. PeopleGarden: Creating Data

Frank, A., 1994. Different Types of "Times" in GIS. GIS and

Portraits for Users. 12th Annual ACM Symposium on

Computational Science Perspectives.

User Interface Software and Technology (UIST '99),

Goralwalla, I. A., M. T. Ozsu, et al., 1998. An Object-Oriented

NC; USA.

Framework for Temporal Data Models. Temporal

Database-Research and Practice. O. Etzion, S.

Jajodia and S. Sripada. Berlin Heidenberg, Springer. 1:

pp.1-35.

H gerstrand, T., 1967. Spatial Process. University of Chicago,

Chicago.

Harris, R. L., 1999. Information graphics: a comprehensive

illustrated reference Oxford University Press US New

York

Harris, S., B. Hetzler, et al., 2000. ThemeRiver: Visualization

Theme Changes Over Time. IEEE Symposium on

Information Visualization, USA, IEEE Computer

Society.

Harrison, B. L., R.Owen, R.M.Baecker, 1994. Timelines: An

Interactive System for the Collection of Visualization

of Temporal Data. Graphics Interface' 1994,

Canadian.

Havre, S., B. Hetzler, et al., 2000. ThemeRiver: Visualizing

theme changes over time. IEEE Symposium on

Information Visualization, Los Alamitos, USA, IEEE

Computer Society.

Koike, Y., A. Sugiura, et al., 1997. TimeSlider: An Interface to

Specify Time Point Proceedings of UIST '97.

Li, X. and M.-J. Kraak, 2008. The Time Wave. A New Method

of Visual Exploration of Geo-data in Time space.

The Cartographic Journal, 45(3), pp.1-9.

MacEachren, A. M., 1995. How Maps Work: Representation,

Visualization, and Design. Guilford press, New York,

USA.

Mackinlay, J., G. Robertson, et al., 1994. Developing calendar

visualizers for the information visualizer.

Proceedings of the 1990 ACM conference on

Computer-supported cooperative work, Los Angeles,.

580

Contact this candidate