June 24, 2017
 

Visualizing complex systems

Ruth Falconer and Mark Shovman
Developing optimization techniques to render large, complex data sets and determining the most effective way, in terms of human perception, to display the data represent significant challenges.

Interactive 3D visualization is an increasingly important tool for understanding complex systems. It provides visual analysis, which complements traditional statistical analysis, of the emerging patterns or behaviours of such systems. We have employed visualization techniques to help us understand the link between observed patterns and lower-level system properties in complex systems such as fungal networks, soil systems, gene regulatory networks and urban and rural sustainability.

Visualization can reveal information flow within a complex system, allowing depiction of interactions, transformations and structure over time. Development of visual tools to support information flow presents several challenges. These include ensuring the required computer power to run a computationally intensive model of the complex system while concurrently providing resources to a similarly intensive 3D interactive interface. Another challenge is how to visualize a complex system so that it can be efficiently perceived and comprehended, given the specific capacities of human visual perception.

With respect to the first challenge, and in part stimulated by the computer-games industry, much progress has been made in terms of the amount of data that can be rendered at interactive frame rates. Visualizations divide into two types, including those that use graphical-processing-unit (GPU) hardware acceleration and those that do not. GPU hardware acceleration employs the GPU to perform key functions faster than is possible on a general-purpose CPU. Most commercial visualizations use hardware acceleration, and GPUs are now widely employed for scientific computation, e.g., for lattice Boltzmann methods for fluid dynamics.

In our visual simulations we use hardware acceleration to achieve interactive frame rates while simultaneously running scientific models. Specifically, we have used volume visualization to depict 3D soil obtained by x-ray computed tomography (CT), which consists of an intensity value for each voxel (3D pixel). Volume visualization is used to make 2D projections of 3D data (derived from CT or magnetic-resonance-imaging scans) with the intention of gaining some insight into the structure of the imaged object. Understanding how the microscopic heterogeneity of soil affects microbial and chemical transport is key to developing effective land-management regimes. Volume visualization of soil data, for example, helps to shed light on the features within the data:1 see Figure 1(a). Likewise, spatiotemporal sustainability indicators of urban and rural developments can be displayed using GPUs to exploit novel visualization techniques such as overlays of multivariate sustainability information (blend or weave) onto the 3D physical landscape, based on programming vertex and pixel shaders:2 see Figure 1(b). This allows us to explore the effects of alternative planning options.


(top) Volume visualization of 3D soil obtained by computed tomography. Transparency is mapped to voxel (3D pixel) intensity and used to indicate how spatial structure affects biomass distribution of two fungal colonies (red and green). (bottom) Indicator values overlain on a 3D landscape.


InterSense-900 interactive wand that allows full 3D interaction with submillimetre precision.

Currently, development of most novel visualization techniques does not typically involve objective assessment of their efficiency in terms of human perception and comprehension. Instead, developers rely on insights, general design principles and some heuristic guidelines. We have designed objective tests of visualization techniques based on theories and methods used in research on visual perception and comprehension.3,4 With these tests, we have assessed human performance (speed and accuracy) in common visual analytical tasks using interactive 3D scatter plots depicting artificial data sets. The data sets were constructed to include trends, anomalies and a varied degree of randomness (see Figure 3). The results indicate that 3D scatter plots can be efficient in detecting nonlinear trends but not for revealing singular anomalies. We have also assessed the usefulness of a 3D interactive device, the InterSense-900 wand (see Figure 2), in these tasks. We found that, contrary to our expectations, allowing full 3D view control is actually counterproductive in simple visual analytical tasks, leading to a longer learning curve and slower performance. We plan to continue testing with a different visualization technique, a link chart, commonly used to depict graphs and networks.


3D-scatter-plot visualization of a data set that includes a nonlinear trivariate trend and several anomalies, none of which can be seen in 2D.


Authors

Ruth Falconer
University of Abertay Dundee

Mark Shovman
University of Abertay Dundee


References
  1. A. Kravchenko, R. Falconer, D. Grinev and W. Otten, Fungal colonization in soils of contrasting managements: modelling fungal growth in 3D pore volumes of undisturbed soil samples, Ecolog. Appl.. In press.

  2. J. Isaacs, D. Gilmour, D. Blackwood and R. Falconer, Enhancing urban sustainability using 3D visualisation, J. Urban Design Plan., In press..

  3. M. M. Shovman, A. Szymkowiak, J. L. Bown and K. C. Scott-Brown, Changing the view: towards the theory of visualisation comprehension, 13th Int'l Conf. Inf. Visualis., pp. 135-138, 2009.

  4. M. Shovman, K. Scott-Brown, A. Szymkowiak and J. Bown, Use of ‘pop-out’ paradigm to test graph comprehension in a three-dimensional scatter plot, Percept. 37 ECVP Abstr. Suppl., pp. 79, 2008. http://www.perceptionweb.com/abstract.cgi?id=v080284


 
DOI:  10.2417/2201009.003295