Data-Driven Compute Overlays for Interactive Geographic Simulation and Visualization

Patrick Komon TU Wien, Gerald Kimmersdorfer TU Wien, Adam Celarek TU Wien, Manuela Waldner TU Wien · 2025

Read Data-Driven Compute Overlays for Interactive Geographic Simulation and Visualization by Patrick Komon TU Wien, Gerald Kimmersdorfer TU Wien, Adam Celarek TU Wien, Manuela Waldner TU Wien (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-23364.txt.

Introduction

Patrick Komon and co-authors put live simulations on top of 3D maps, in a 2025 paper. Their overlays are computed on the graphics card in several steps from mixed data sources, then draped over native or web maps. The showcase is snow: users can twist the settings of snow cover and avalanche models and watch the terrain answer at once.

Large avalanche runs finish in milliseconds to seconds, orders of magnitude faster than a standard Python pipeline. That speed changes the tool from a report generator into a sandbox. Forecasters and planners could test what-if winters directly on the slopes they manage.

Catalog record

CREATORPatrick Komon TU Wien, Gerald Kimmersdorfer TU Wien, Adam Celarek TU Wien, Manuela Waldner TU Wien
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-23364

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Data-Driven Compute Overlays for Interactive Geographic Simulation and Visualization Patrick Komon TU Wien, Gerald Kimmersdorfer TU Wien, Adam Celarek TU Wien, Manuela Waldner TU Wien Abstract We present interactive data-driven compute overlays for native and web-based 3D geographic map applications based on WebGPU. Our data-driven overlays are generated in a multi-step compute workflow from multiple data sources on the GPU. We demonstrate their potential by showing results from snow cover and avalanche simulations, where simulation parameters can be adjusted interactively and results are visualized instantly. Benchmarks show that our approach can compute large-scale avalanche simulations in milliseconds to seconds, depending on the size of the terrain and the simulation parameters, which is multiple orders of magnitude faster than a state-of-the-art Python implementation. \onlineid 0 \vgtccategory Research \vgtcinsertpkg \teaser Simulation of the Madlein avalanche (Tyrol, 1984) from its recorded release areas (highlighted in blue), computed and visualized in a few milliseconds. The color encodes the velocity. Introduction keywords 3D geographic visualization, geographic simulation, WebGPU. Data-driven overlays for geographic maps showing, for instance, steepness of the terrain, precipitation, or points of interest, are typically computed offline. This limits the applicability to static data. To overlay 2D or 3D maps with visualizations of dynamic data or results of interactively steerable simulations, the overlays either need to be limited to a very small area or require remote access to a powerful server. In this work, we explore the use of WebGPU to compute data-driven overlays for online maps dynamically with the support of compute shaders. We present a workflow for running simulations and instantly generating overlays for 3D maps and demonstrate how our prototype weBIGeo can speed up conventional workflows by orders of magnitude. In summary, our contributions are as follows: 1. An API to specify compute workflows for instant generation of data-driven multi-resolution overlays for 3D maps. 2. The first open-source 3D geographic simulation and visualization tool weBIGeo for interactive native and web-based applications with compute shader support, 3. Two use cases – snow cover and avalanche simulation ( \cref fig:teaser) – demonstrating the potential of data-driven overlays. Our source code is available on \href https://github.com/weBIGeo/webigeoGitHub. An online demo showcasing the avalanche simulation use case can be accessed here: \url https://webigeo.alpinemaps.org/ (WebGPU-capable browser required, for example Google Chrome on Windows). 1 Background: (Web-Based) GeoVis Figure 1: A node graph generated from the workflow specification for the avalanche use case ( \cref sec:avalanches) This workflow generates a multi-resolution Avalanche Overlay Texture (right) from terrain model tiles in a given Region AABB (left). Common geographic visualization techniques are glyphs placed on top of maps and color overlays [ 12 ] . The focus of our work lies on overlays. Here, we review three principle paradigms to data-driven map overlays particularly for the web: Map tiles are the foundation for all popular online map applications, where map clients request pre-rasterized map images depending on the currently shown zoom level and visible field of view [ 13 ] from a dedicated server – also known as “slippy maps” [ 8 ] . Data-driven overlays can be similarly pre-rendered into offline tiles and overlaid onto a base map [ 6 , 15 ] . A well-known open-source library to build layered maps is OpenLayers [ 16 ] . Offline tiles are useful when the data to be overlaid itself is big and static. The disadvantage is that the server has to re-compute the tiles whenever the underlying data changes. Users have no or only very limited possibilities to interact with the data. An alternative approach is to render the final images to be viewed by the client directly on the server. The client only forwards users’ interaction with the map (e.g., panning and zooming) and with the data (e.g., filtering) to the server and finally shows the image transmitted by the server. The application is therefore largely independent from the client’s hardware resources. In the geographic context, examples for such a server-side analyis and rendering approach are HORA 3D for interactive flooding simulations [ 14 ] and applications built on top of ParaView [ 2 ] . The disadvantage is that the server needs to be powerful enough to serve all requests at the same time. In addition, interactive applications require a lot of bandwidth for smooth interaction, such as flying over the 3D terrain in \cref fig:teaser. The most flexible approach is client-side analysis and rendering : the data to be analyzed remains on the client, and the client handles all interactivity, simulation, and rendering itself. The major disadvantage, clearly, is that the client needs to be powerful enough – in terms of CPU, GPU, and memory – to handle the computation of the overlays and all user interactivity. Client-side overlays can be computed pixel-wise for every frame in the fragment shader based on the raster tiles used for rendering the 3D map. Examples are risk maps computed from local terrain properties [ 7 ] or difference maps of flood simulation scenarios [ 9 ] . There are some major shortcomings to this approach: First, overlays need to be recomputed whenever user navigation requires a new map tile to be loaded, even if the underlying data does not change. Second, the resolution of the overlay is bound to the resolution of the underlying map tile. Third, pixel-wise analysis is not sufficient for more complex scenarios like avalanche simulation (see \cref sec:avalanches). It has been shown that some more complex geospatial analyses can be significantly sped up by GPGPU, using, for instance OpenCL [ 4 ] or CUDA [ 22 ] . Vollmer et al. [ 20 ] compute levels-of-detail of pre-computed static overlay textures on the GPU so that the user can interactively change aggregation parameters. In contrast, we propose a workflow and present an implementation to interactively compute overlay textures directly from the raw data. Up until recently, graphics APIs for the web were limited to vertex and fragment shaders. With the new web-based graphics and compute GPU API WebGPU , it is now possible to employ compute shaders in the browser. WebGPU is increasingly generating interest for interactive web-based visualization of large data. In particular, it has been used for interactive isosurface extraction [ 18 ] , graph drawing [ 5 ] , and aggregate 2D GeoVis [ 11 ] . WebGPU is generally seen as a performant successor of WebGL in the context of WebGIS applications [ 19 ] and is appreciated for its code portability between web, desktop, and mobile map applications [ 1 ] . We extend these works by contributing a workflow, a software architecture, and performance benchmarks for data-driven compute overlays for 3D maps. To the best of our knowledge, weBIGeo is thereby the first online map with interactively computed data-driven overlays. 2 Data-Driven Compute Overlays weBIGeo exposes a simple and flexible API to construct compute workflows of varying complexity. Compute workflows are specified as a sequence of compute nodes, similarly to dataflow models for visualization provenance [ 17 ] or for steering simulation ensembles [ 21 ] . Each compute node has at least one input and one output, as illustrated in the node graph in \cref fig:compute-pipeline. Raw input data can come in various forms – from single

Read the complete text