Dykes & Mountain : Data Viz II

Figure 1

Time Geography approach. A two-dimensional model of space is used with an additional (third) independent continuous dimension employed to represent time. The key concept is the trajectory of mobile individuals over (2D) space and through time; static phenomena (e.g buildings, transport networks) occupy the same location in space through time. Here three stages are shown for an individual moving from one static phenomenon to another via a network link. A traditional planimetric representation (with the link as a grey line and the individual a black cross) is contrasted with the trajectory through time (the black line).

Figure 2

figure 02

Graphical output from the Location Trends Explorer. A two-year log of 80,000 points collected by a single individual is shown shaded by time. The map (left) covers most of Europe and some of North Africa. The height of the bars on the time view (right) indicates the frequency at which points were logged. An interactive selection tool is being used to focus in on a particular period (see figure 3).

Figure 3

figure 03

Graphical output from the Location Trends Explorer. A subset covering the Sinai peninsular is shown shaded by speed; this is revealed to be slower in general for node points where much time is spent and faster for the links between them. In the time view, the temporal subset selected in figure 2 is shown. The focus tool has been used to select a particular period of activity for which additional detail is evident. The map is re-scaled to show the selected 'time-space' (the spatial range defined by the user defined temporal limits). Additional focusing using the map can select 'space-times' as shown by the lasso tool created with the cursor. When such a space is selected the time view is updated appropriately.

Figure 4

figure 04a
a) Weekly spatio-temporal log.

figure 04a
b) Weekly spatio-temporal log with exogenous information.

In these two figures lines are used to connect successive points. The shapes and patterns displayed in figure 4a allow particular episodes and behaviours to be identified. The background map showing the extent of land and sea adds context and helps explain the patterns. More sophisticated geographic information allows us to make more advanced inferences.

Figure 5

figure 05a
a) Spotlight showing weekly spatio-temporal log.

figure 05b
b) Spotlight with superimposed exogenous information.

Use of the spotlight metaphor. Varying colour lightness to show density of occupation of space can be useful to 'throw light upon' additional exogenous data in areas that are frequently visited when graphically exploring time-space data sets. Here the full week's activity is represented by colour lightness (figure 5a). We are able to use colour saturation and hue to show contextual spatial information that help us interpret the time space data. The data can be split into various temporal periods (e.g. days of the week, hours of the day, daytime/nighttime) and sequenced to throw light on relevant locations in a suitable order. Interactive or animated representations reveal dynamic changes in the focus of activity and the relevant contextual spatial data is revealed to aid interpretation - hence the spotlight metaphor. A drawback of the technique is the requirement of large amounts of computing power to create and manipulate the raster structures used. EDA with high levels of interaction and quick response is less easy to achieve with such bulky continuous data structures than the discrete points used in LTE. The surfaces reported here were calculated independently in a non-interactive environment before being loaded into more interactive software. This is a stopgap approach and a more thoroughly integrated solution is required so that density surfaces can be incorporated into the suite of real-time visualization tools offered by EDA software such as LTE and spotlights derived from them.
Crown Copyright Ordnance Survey. An EDINA Digimap/JISC supplied service.

Figure 6

figure 06a
a) Surface feature network.

figure 06b
b) Surface feature network and magnitude (spotlight).

Surface feature networks and spotlights derived from northern section of weekly log. Feature categories are symbolised with colour hue : peaks in red, pits in gray, ridges in yellow, channels in blue, passes or saddles in green. The densities (surface value) use colour lightness in 'spotlight' fashion, highlighting the densest areas. The width of the border at the edge of the surface reveals the scale of kernel used to calculate the network of features.

Figure 7

figure 07a a) 5 cell window (500m).	figure 07b b) 7 cell window (700m).
figure 07c c) 9 cell window (900m).	figure 07d d) 15 cell window (1500m).

Surface feature networks and spotlights generated from northern section of weekly log. Deriving networks at a variety of spatial scales reveals different structural components in the data. Calculating the networks at a range of spatial scales results in the identification of features that relate to events at different spatio-temporal scales in this instance. The 500m kernel reveals peaks at stopping points on a series of recreational walks recorded in the log. The peaks identified in the 900m kernel highlight areas of repeat activity such as 'home', quays and favourite locations for lunch. The channels define 'no go areas' that are not navigable by boat due to shoals, or foot due to vegetation and access restrictions imposed following an outbreak of disease. The surface network derived from the 1500m kernel identifies the locations at which most time was spent : 'home' (the centre of the peak) and the main tracks along the island. This network of four nodes and three vertices describes the activity extremely succinctly.

Figure 8

figure 08a
a) Position 1.

figure 08b
b) Position 2.

figure 08c
c) Position 3.

Geocentric parallel plots - Each parallel plot (top right frame) shows four variables relating to occupancy levels of households in enumeration districts (EDs) in an area of Leicester, UK. The vertical axes are scaled and, from left to right, relate to the percentages of households in each unit with : less than 0.5 persons per room (ppr); 0.5 - 1.0 ppr ; 1.0 - 1.5 ppr ; at least 1.5 ppr. Each of the three figures uses shading to relate the map polygons to corresponding lines on the parallel plot. Colour lightness represents the distance from a single geographic point of interest to each of the EDs in each case. It varies from highly visible black for close EDs to faint/non visible white for distant cases. The location of interest is represented by the red target symbol on the map in each figure. For a route loaded directly from a GPS receiver, spatial/statistical trends can be observed as the target point moves through the scene. An early location along the route (figure 8a) shows an area with relatively low levels of overcrowding. As the GPS track is followed the pattern changes and the local EDs, shaded in darker grays on the map and in the parallel plot, display greater variation and more overcrowding (see figure 8b). At the end of the track the geocentric shading of the multivariate parallel plot (figure 8c) reveals a very different neighbourhood. In each figure the panoramic images located close to the target are displayed to provide local qualitative information to the analyst. The plots can be used to show variation amongst independent as well as dependent variables.

Paper	Seeking Structure in Records of Spatio-Temporal Behaviour : Visualization Issues, Efforts and Applications
Authors	Jason A. Dykes, David M. Mountain
Title	Editor / Publisher Download
Notes	This page contains links to the figures used in the following paper : Dykes, J. A. and Mountain, D. M. (2003) Seeking structure in records of spatio-temporal behaviour: visualization issues, efforts and applications, Computational Statistics and Data Analysis, 43(Data Visualization II Special Edition), 581-603.