Zarr-Cesium: Browser-Native Visualisation of Terabyte-Scale Multidimensional Scientific Data Without Backend Infrastructure

Environmental sciences increasingly rely on cloud-hosted terabyte-scale multidimensional datasets stored in formats such as Zarr. While cloud-native storage and computation ecosystems have matured rapidly, visualization workflows often still depend on server-side tile generation, raster preprocessing pipelines, and persistent infrastructure. These approaches duplicate data, increase operational complexity, and separate visual products from their underlying scientific datasets. This talk presents Zarr-Cesium, an open-source TypeScript library that enables interactive visualization of multidimensional Zarr datasets directly in the browser using CesiumJS and WebGL2. The software streams chunked array data from object storage and renders scalar fields, volumetric slices, and animated vector fields entirely client-side, without backend services or tile-generation pipelines. The presentation will discuss the software architecture and engineering decisions required to make large-scale browser-native scientific visualization practical. Topics include cloud-native data access, multiscale pyramid selection, CF-compliant metadata interpretation, request cancellation strategies, GPU-based rendering pipelines, and performance optimisation for terabyte-scale datasets. The talk will also reflect on broader Research Software Engineering themes, including reducing infrastructure barriers for scientific dissemination, designing sustainable open-source geospatial software, and enabling reproducible interactive visualisation workflows. Zarr-Cesium is currently used operationally within the National Oceanography Centre’s AtlantiS visualisation tool and has been released as open-source software to support wider adoption across environmental and geospatial sciences.