Current unsteady multi-field simulation data-sets consist of millions of data-points. To efficiently reduce this enormous amount of information, local statistical complexity was recently introduced as a method that identifies distinctive structures using concepts from information theory. Due to high computational costs this method was
IntroductionDue to the increase in compute power, computer simulations result in inceasingly large data sets. Often, the data is 3-dimensional, time dependent and multivariate. Visualization techniques can help reduce the giant amount of information provided by the data to the important facts. An example for the importance of such techniques is aircraft design. To reduce costs large parts of the development and testing of an aircraft are performed using computer simulations. Thus, flight quality and performance can be tested in many different scenarios. In order to analyze the behavior of the aircraft, developers have to extract relevant features from the simulation and visualize them. Relevant structures are for example the wake of an airplane or vortex breakdown. Understanding the wake of an airplane is important to decide at which intervals airplanes can start and land. Planes that enter the wake of a preceding flight might experience large turbulences that can cause a crash. Similarly if the bend that a delta wing flies is to sharp the vortex that keeps it in the air might break down causing a crash. Finding such structures early in the design process is an important task. In most application domains, scientific data visualization is used to understand both the mean features as well as unusual patterns in the data.A standard technique to detect relevant structures is feature extraction. Post et al. [PVH * 03] provide a detailed examination of techniques that are available in the field of flow visualization. The methods are divided into approaches based on image processing, vector field topology, physical characteristics and selective visualization. For a detailed explanation we refer the interested reader to this publication. Opposed to feature extraction methods that extract single formations, structure-based visualization focuses on a partitioning of the entire domain. Techniques can be classified as cluster-based or integral-line based methods. Salzbrunn et al.[SJWS07] provide a good overview over these approaches. Methods from both fields have in common that they need a priory information about the structures that are to be found. Moreover, most techniques were developed for a special type of field, e.g., scalar or vector valued. Thus, a precise definition of what is to be found is needed that cannot always be given. For example, there exists no general definition of a vortex.Recently Jänicke et al.[JWSK07] adopted local statistical complexity, a concept from information theory, and used the method to automatically detect structures in an unsteady multi-field that deviate from the average behavior in the field. Due to high computational costs, the method was so far...