An experimental 20-member pilot dataset providing a seamless ensemble precipitation forecast for the Netherlands. Generated every 5 minutes extending up to 6 hours ahead, combining 5-minute real-time radar accumulations with the Harmonie-Arome ensemble using pySTEPS.
Derived from the high-resolution UWC-West HARMONIE-AROME Cy43 numerical weather prediction model, developed in collaboration with meteorological services of the Netherlands, Iceland, Denmark, and Ireland. Provides 2m Temperature, Wind forecasts at selectable levels (10m to 300m), and Global Solar Radiation at 1-hour frequency in GRIB format.
Mapping & Visualization
Interactive maps and geographical data visual layers are powered by:
Map Library:MapLibre GL JS, an open-source, GPU-accelerated vector map renderer.
Map Style & Basemap: CartoDB Dark Matter base tiles provided by CARTO, built using data from OpenStreetMap contributors.
How it Works
Nimbus is built using a high-performance stack:
Rust Backend: Built with Axum and Tokio, subscribing to the KNMI MQTT broker for instant updates, processing NetCDF, GRIB1, and HDF5 files, and executing parallel statistical reductions with Rayon.
Lossless WebP Packing: Precision float grids are packed into the Red and Green channels of WebP images for highly compressed, 16-bit lossless transmission to the client.
GPU Acceleration: The frontend uses custom WebGL fragment shaders to unpack WebP pixels on-the-fly, render smooth coordinate projections, and execute 60fps GPU particle advection for wind velocity fields.