From NetCDF workarounds to fast AI weather visualizations with Earthmover
Team Beyond Weather
Published on: 2026/05/12
3 min
Why visualisations are so important
At Beyond Weather, we build tailored long-range weather forecasts for clients who need to understand weather risk beyond the standard forecast horizon. But producing a forecast is only part of the work. To be useful in day-to-day decision-making, the data also needs to be easy to explore, compare and interpret.
We believe weather-critical industry need to have fast and clear answers to their weather needs. They should not have to work with raw files or wait for heavy visualizations to load. They need a clear, responsive interface that helps them understand what the model is showing, where signals are emerging and how conditions may develop over time.
For us, this made visualization a core part of the product experience. The question was not whether we had valuable forecast data. The question was how to make that data available to clients in a way that felt fast, intuitive and reliable.
The challenge: visualizing scientific data for clients
In the early stages of Beyond Weather, our team worked directly with NetCDF files on a local computer. NetCDF is a powerful format for scientific weather and climate data, but it is not designed for smooth, client-facing web visualization.
To make forecasts visible in an online interface, we relied on workarounds: converting data into formats such as GeoTIFFs, PNGs or vector layers and then serving those through a web application. These approaches helped us move forward, but they added complexity and slowed down the user experience. In some setups, loading data in the web interface could take seconds. Furthermore, the added conversions created complexity for us in the backend. The raw forecast data and the end-user were simply too far apart.
For clients using forecasts to assess weather-driven volatility, that matters. A visualization layer should help users explore signals, compare scenarios and understand spatial patterns quickly. It should not become a bottleneck between the forecast and the decision it is meant to support.
Initial browser visualization of Beyond Weather weather model output.
Earthmover as the infrastructure layer
Earthmover provided the missing infrastructure layer. Through Earthmover’s Flux API, Beyond Weather could serve forecast data as fast, on-demand map tiles directly from cloud-native scientific data workflows. Because our team was already working with Zarr, Xarray and Google Cloud Storage, the move towards Earthmover’s ecosystem was a practical step rather than a full rebuild.
Flux made it possible to turn model output into browser-based visualizations without forcing the data through slow intermediate formats. Alongside Flux, we adopted Icechunk, Earthmover’s open-source tensor storage formatstorage engine, creating a stronger foundation for operational AI model training and data delivery.
For Beyond Weather, this was not just a technical improvement. It allowed us to focus engineering capacity where it creates the most value: improving long-range forecast skill, fine-tuning AI-models and translating weather signals into decision support for sectors exposed to weather risk.
Faster visualization, scalable delivery
The implementation moved quickly. With one developer, working closely with the Earthmover team, Beyond Weather implemented the new visualization setup in one week. Forecast data that previously depended on slow and imperfect workarounds could now be rendered directly in the browser.
Loading times improved from seconds to milliseconds after optimization. A complex feature supporting reduced Gaussian grids, the native grid format for ECMWF’s AIFS model, was also resolved within the same implementation window.
The result is a more scalable route for delivering AI weather forecasts to clients and partners. This was further strengthened by the launch of Earthmover’s Data Marketplace, where Beyond Weather became a launch partner.
Operational Beyond Weather forecasting interface powered by Earthmover infrastructure.
Explore what long-range weather signals can mean for your decisions
Long-range forecasting will always involve uncertainty. The value lies in identifying where meaningful signals emerge, understanding what they imply for a specific market or operation, and making them available early enough to support action.
Get in touch with Beyond Weather to explore how AI-based long-range forecasts can support planning, trading and risk decisions in your organisation.
Team Beyond Weather
Stay in the loop
Subscribe to our newsletter and get updates on the state of AI-enabled S2S forecasting.
