Next-Generation Geospatial Visualization: From Traditional Meshes to Gaussian Splats and NeRFs

Recent advances in remote sensing and computer vision have reshaped how geospatial data is captured, visualized, and analyzed. Drones equipped with high-resolution cameras and sensors enable rapid, detailed surveys from multiple angles, providing near-real-time insights into complex environments. While photogrammetry-based workflows yield accurate 3D meshes, they often demand substantial computational power and lengthy processing times.

Emerging techniques such as Gaussian splatting and Neural Radiance Fields (NeRFs) offer compelling alternatives to traditional mesh-based methods. Gaussian splatting represents 3D scenes as point-based “splats” with mathematical distributions, enabling faster, photorealistic rendering. NeRFs employ neural networks to generate volumetric reconstructions from sparse image inputs, capturing intricate lighting and geometry with minimal manual intervention. Together, these methods reduce post-processing complexity and enhance visual fidelity.

In this session, we demonstrate novel applications of Gaussian splats and NeRFs within ArcGIS and discuss how these approaches can integrate with familiar mesh-based workflows. We also explore ways to extend existing 3D streaming standards, such as OGC’s Indexed 3D Scene Layers (I3S), to incorporate these emerging data types. Finally, we showcase real-world examples demonstrating how blending Gaussian splats and conventional meshes enables richly detailed, interactive visualizations at multiple scales. This convergence promises more efficient collaboration, cost-effective workflows, and deeper insights into rapidly evolving built and natural environments.