IEEE COMPUTING IN SCIENCE AND ENGINEERING (CISE), SPECIAL ISSUE 2021 1 In-situ visualization of natural hazards with Galaxy and Material Point Method Greg Abram, Andrew Solis, Yong Liang, and Krishna Kumar Abstract—Visualizing regional-scale landslides is the key to conveying the threat of natural hazards to stakeholders and policymakers. Traditional visualization techniques are restricted to post-processing a limited subset of simulation data and are not scalable to rendering exascale models with billions of particles. In-situ visualization is a technique of rendering simulation data in real-time, i.e., rendering visuals in tandem while the simulation is running. In this study, we develop a scalable N:M interface ar- chitecture to visualize regional-scale landslides. We demonstrate the scalability of the architecture by simulating the long runout of the 2014 Oso landslide using the Material Point Method coupled with the Galaxy ray tracing engine rendering 4.2 million material points as spheres. In-situ visualization has an amortized runtime increase of 2% compared to non-visualized simulations. The developed approach can achieve in-situ visualization of regional- scale landslides with billions of particles with minimal impact on the simulation process. Index Terms—In-situ visualization, Material Point Method, TACC Galaxy, Ray tracing. I. OVERVIEW AND MOTIVATION L ANDSLIDE runouts are regional-scale events that can bury whole towns (e.g., 2018 Southern California mud- flows that followed a series of wildfires, 2014 Oso landslide caused 43 fatalities in the US) or devastate entire regions (e.g., 2016 Kaikoura, New Zealand earthquake recorded 100,000 landslides within a 12,000 sq. km area). The frequency of these regional-scale landslides is increasing with devastating earthquakes, extreme precipitation, and wildfires caused by climate change. Even in these risk-prone communities, where the residents are aware of the threat posed by the landslides, the residents repeatedly show indifference to these threats and consequently fail to act. The human capacity to deny danger endangers lives and property [1]. In 2020, twenty-two extreme events cost the United States $95 billion in damages. The underlying problem is the uncertainty of the actual event and a lack of understanding of its potential impacts. How to communicate the reality of potential risk to a diverse group of individuals, who must understand and support a complex set of actions that reduce the risk for the whole community? Visualization is the key to communicating scientific results effectively to engineering decision-makers and the public. Given the limited cognitive capacity in humans, visualizing the complex threats as images enhances the brain’s capacity to per- ceive opportunities and make decisions. Visual representation Department of Civil, Architectural and Environmental Engineering, Univer- sity of Texas at Austin, TX, 78712 USA e-mail: (krishnak@utexas.edu). Texas Advanced Computing Center, University of Texas at Austin, TX,. University of California at Berkeley, CA,. Manuscript received September 3rd, 2021. of complex information reduces the cognitive load on human information processing and increases human “absorptive ca- pacity” for problem-solving [2]. However, these visualizations must be physically sound to be perceived as realistic and ac- cepted as valid. Visualization is also context and user-specific, i.e., different users may be interested in different aspects of the experimental or simulation data sets. Hence, the same data set may be represented in several forms – perhaps at varying levels of detail, emphasizing or deemphasizing different regions and features, and employing different visualization techniques to best present the information. Despite the landslide risks, most regional-scale landslide hazard analyses do not consider downstream impacts and the aerial extent of debris-flow runouts [3]. Traditional numerical methods such as the Finite Element Method (FEM) primarily focuses on the onset of failure but provides limited information on the post-failure runout mechanisms due to mesh distortions at large deformations [4]. Modeling the impacts of the poten- tial runout of large landslides is possible with new tools such as Material Point Method (MPM) [5]. MPM is a mesh-free method that discretizes the domain as a collection of material points moving on a background grid, and Newton’s laws of motion determine their deformations. In this study, we employ the CB-Geo MPM code (https://github.com/cb-geo/mpm), and a typical MPM computation cycle is shown in fig. 1. For more information on MPM and the code implementation, refer to [5]. MPM results of the landslide runout can be translated into compelling visualizations to inform decision-makers of potential hazards. A regional-scale landslide simulation of a kilometer-scale runout with MPM requires billions of material points and mil- lions of cells in the background grid. Scaling applications to simulate a billion particles require exascale high-performance computing (HPC) architectures with efficient Message-Passing Interface (MPI). The CB-Geo MPM code exploits a hy- brid MPI+X approach to achieve regional-scale simulation of landslides. In addition, as the landslide runout progresses, the workload distributed across the compute nodes becomes imbalanced. We adopt a distributed graph-partitioning tech- nique to load balance both the background mesh and particle distribution dynamically. Data retrieval and visualization in High-Performance Com- puting applications have long been a bottleneck. Current techniques for visualizing exascale simulations involve writing a temporal slice of a subset of the data to disk, often at a much coarser resolution than the original data, which leads to a significant portion of information being disregarded and potentially lost, followed by post-processing. In practice, this arXiv:2109.02754v1 [physics.geo-ph] 6 Sep 2021