Uncertainty Visualization of 2D Morse Complex Ensembles Using Statistical Summary Maps

Tushar M. Athawale, Dan Maljovec, Lin Yan, Chris R. Johnson, Valerio Pascucci, Bei WangIEEE TVCG 2019 · OpenReview

TL;DR

This paper introduces "statistical summary maps" to visualize uncertainty in 2D Morse complexes derived from noisy scalar field data. By analyzing ensembles of these complexes, the authors propose two new visualization techniques, the Probabilistic Map and the Survival Map, to capture variations in local extrema and gradient flows, offering a more robust understanding of topological structures in the presence of data uncertainty.

What Problem Does This Paper Address?

Scientific visualizations often rely on topological descriptors like Morse complexes to understand scalar field data. However, real-world data is rarely perfect; it's often corrupted by noise from acquisition or processing. This noise can significantly alter the computed Morse complex, making it difficult to discern the true underlying structure. When dealing with ensembles of simulations or measurements (where each member has slight variations), visualizing the uncertainty in these topological structures becomes crucial. The challenge is to represent not just the average structure, but also how it varies across the ensemble, which current visualization methods struggle to do effectively for Morse complexes.

Methods

The core contribution of this paper is the introduction of two novel visualization techniques:

  • Probabilistic Map: This map quantifies the positional uncertainty of local extrema (maxima and minima) across an ensemble of Morse complexes. It leverages "mandatory local maxima" (identified through a specific noise model) as anchors. For each point in the domain, it computes a probability distribution indicating which local maximum its gradient flow is most likely to lead to. This is visualized using color blending, where regions with consistent flow to a single maximum are shown in a solid color, and uncertain regions show a mix of colors. Interactive queries allow users to explore these probability distributions at specific points.
  • Survival Map: This technique focuses on the uncertainty in local gradient flows. It uses a process called "persistence simplification," where topological features (like saddle-maximum pairs) are iteratively removed based on their significance (persistence). For each ensemble member, a "survival measure" is computed for each point, indicating how often its gradient flow direction remains unchanged during simplification. The Survival Map is then the average of these survival measures across the ensemble, visualized as a heatmap where yellow indicates high survival (consistent flow) and blue indicates low survival (variable flow).

The overall pipeline involves computing Morse complexes for each ensemble member, identifying mandatory local maxima, clustering them, and then generating the Probabilistic and Survival Maps.

Datasets

The authors evaluate their approach on:

  • Synthetic Datasets:
    • Ackley Dataset: A standard benchmark function used to demonstrate the core concepts of the Probabilistic and Survival Maps.
  • Mixture of Gaussians: Used to validate the noise model and its relationship with mandatory local maxima.
  • Himmelblau's Function: A multimodal function used to showcase the methods on a more complex landscape with multiple local maxima of similar heights.
  • Real-World Datasets:
    • Kármán Vortex Street Dataset: A fluid dynamics simulation dataset used to analyze flow velocity uncertainty.
  • Weather Dataset: A climate dataset involving vector fields, used to demonstrate the application to a different type of scientific data.

Key Insights and Ideas

  • Mandatory Local Maxima as Anchors: The concept of "mandatory local maxima" provides a robust way to identify and track critical points across noisy ensemble members, forming the basis for the Probabilistic Map.
  • Persistence Simplification for Flow Uncertainty: Using persistence simplification as a tool to probe gradient flow stability is a clever way to quantify how "reliably" a point's flow direction is maintained across different ensemble members.
  • Statistical Summary Maps as a New Abstraction: The Probabilistic Map and Survival Map offer a new paradigm for visualizing topological uncertainty, moving beyond simply showing an average structure to explicitly representing variations and confidence.
  • Probabilistic Map for Boundary Uncertainty: The Probabilistic Map not only shows uncertainty in where critical points end up but also provides "expected boundaries" that are more robust to noise than boundaries derived from a single mean field.

Key Results

The paper demonstrates the utility of the Probabilistic Map and Survival Map across various synthetic and real-world datasets.

  • Probabilistic Map: Effectively visualizes regions of certainty (where gradient flows consistently lead to the same local maximum) and uncertainty. Interactive queries allow for detailed exploration of these uncertain regions, revealing the mix of possible flow destinations. For the Ackley dataset, the Probabilistic Map's expected boundaries were shown to be more accurate than those from the mean field's Morse complex, especially under non-parametric noise.
  • Survival Map: Successfully highlights areas where gradient flows are stable or unstable across the ensemble. For the Ackley dataset, it correctly identified high peaks (yellow regions) and low peaks (blue regions) consistent with the ground truth. For the Kármán Vortex Street dataset, it provided insights into the uncertainty of vortical structures.
  • Qualitative Improvements: The visualizations provide intuitive insights into the structural variations and positional uncertainties of Morse complexes, which are not apparent from standard visualizations of mean fields or individual ensemble members.

Limitations and Future Work

  • Noise Model Dependence: The current noise model assumes a specific relationship between the noise level and the persistence of topological features. While the paper discusses heuristics for violating this model, a more generalized noise model would be beneficial.
  • 2D Focus: The current work is limited to 2D scalar fields. Extending the methods to higher dimensions (3D and beyond) is a significant challenge, especially in developing effective visual mappings for higher-dimensional uncertainties.
  • Computational Cost: While not explicitly detailed, computing ensembles of Morse complexes and then deriving summary maps can be computationally intensive.
  • Future Work: The authors plan to generalize their noise model, extend the approach to higher dimensions, and explore new visual mappings for higher-dimensional uncertainties.

Summary updated

Abstract

Morse complexes are gradient-based topological descriptors with close connections to Morse theory. They are widely applicable in scientific visualization as they serve as important abstractions for gaining insights into the topology of scalar fields. Data uncertainty inherent to scalar fields due to randomness in their acquisition and processing, however, limits our understanding of Morse complexes as structural abstractions. We, therefore, explore uncertainty visualization of an ensemble of 2D Morse complexes that arises from scalar fields coupled with data uncertainty. We propose several statistical summary maps as new entities for quantifying structural variations and visualizing positional uncertainties of Morse complexes in ensembles. Specifically, we introduce three types of statistical summary maps – the probabilistic map , the significance map , and the survival map – to characterize the uncertain behaviors of gradient flows. We demonstrate the utility of our proposed approach using wind, flow, and ocean eddy simulation datasets.
BibTeX
@article{athawale2019uncertainty,
  author = {Tushar M. Athawale and Dan Maljovec and Lin Yan and Chris R. Johnson and Valerio Pascucci and Bei Wang},
  title = {{Uncertainty Visualization of 2D Morse Complex Ensembles Using Statistical Summary Maps}},
  journal = {IEEE Trans. Vis. Comput. Graph.},
  year = {2019},
  url = {https://openreview.net/forum?id=1ZaXCic5Ul},
}
Figure