NivoSense

NivoSense turns Earth observation data into precise water forecasts for the people who decide who gets water.

In Mediterranean and Arid climates, snow is the natural reservoir. Mountains hold winter precipitation as snow and meter it out across spring and summer, the difference, downstream, between a full dam and a dry one. But that buffer is collapsing fast: snow cover duration in the Alps has fallen by −36 days, a loss described as unprecedented in the last six centuries.

A lot of basins depend on this water for consumption, agriculture or energy generation and if the resources are going to be scarce in the future we need to be ready: we cannot farm rain, but we can adapt with time to our water availability. But how?

We fuse Sentinel-2 NDSI + Copernicus ERA5 + Copernicus DEM + wind reanalysis through a LightGBM pipeline to deliver, basin by basin:

🛰️ Current and future snowpack: snow cover at 10 m resolution with different selectable time forecasts.

📈 Inflow trends, short and long-term forecast: estimation of the inflow of every river basin water availability stoplight system (good, alert, critical).

The platform is working and live under: https://storage.googleapis.com/nivosense-web/index.html?

All the progress and commits can be seen in the GitHub repository: https://github.com/mariopeces/NivoSense

We know this is a need for all water resource managers. That is why we aim for EU river basin authorities (110+), irrigation communities (7,000+ only in Spain), and energy companies that control hydro plants (800+). They need to allocate water rights, set reservoir releases, declare drought stages, and commit hydro generation to the grid months ahead. The information is there, but it is not unified into a single platform that can actually save days in the water resource decisions.

¿What is our status? 

• We developed the first prototype in Sierra Nevada mountain range (Spain) and we created a scalable and reproducible workflow that can be applied to any mountain range on earth.
• Validation conversation open with UGR (Universidad de Granada).

Our roadmap now is: 

• Validate now with an institution → scale to all Mediterranean basins → scale to global snow-fed basins (Atlas, Andes, Himalaya, Rockies).
• Regarding the methodology: fine-tuning of the model until getting >85% pixel precission -> add Sentinel-1 data (RADAR) to model snow depth and snoe water equivalent per pixel (SWE) --> fully automate data ingestion and processing.

Our team is a Three-card team layout:

We believe that predicting and modelling the snow water is crutial for mediterranean and arid climates, and the situation will only become worse: Alps is already starting to transition to a mediterranean climate and some countries (Spain, Greece, Italy) are experimenting increase of aridity. These snowpacks and how they behave in the future will act as a crutial element to predict water access and therefore guarantee that no excesses are being done, securing equitable access to everyone that needs it.

• 💧 60–80 % of inflow into Iberian reservoirs originates as mountain snow. Forecasting that snow → water transition is exactly what EU Space data enables and what current tools miss at sub-basin scale.
• 🛰️ Sentinel-2 + Copernicus services are the only data source that can produce a 10 m, multi-decadal, pan-European snow record. EO data is not a layer on top of the product — it is the product.
• End users are public bodies under EU policy (Water Framework Directive basin authorities, EU-funded irrigation). Output is downstream policy support, not a consumer toy.

These were the spatial data used for the project:

Ingestion from CDSE and the Climate Data Store runs at build time, producing forecast COGs and per-basin JSON written to a public GCS bucket. Runtime serves pre-computed tiles, low latency, and respects Copernicus access policies.

APIs, SDKs, frameworks, libraries:

• ⚛️ Frontend: React 18 · Vite 5 · TypeScript · Tailwind 3 · MapLibre GL JS 4 · Recharts.
• 🐍 Backend: Python 3.11 · FastAPI · rio-tiler · Rasterio · NumPy · Pandas · GeoPandas · Shapely.
• 🤖 ML: LightGBM · scikit-learn · xarray · Dask.
• 🛰️ EO tooling: CDSE (Sentinel-2 L2A) · Copernicus CDS Python client (ERA5) · GDAL.
• ☁️ Cloud: Google Cloud (Cloud Run · Cloud Storage · Artifact Registry · Cloud Build, europe-west1) · Docker.
• 📦 Formats: Cloud-Optimised GeoTIFF · GeoJSON (will be PMTiles in the future)· JSON.

Thank you for reading the project! If anyone is curious or wants more information, you can contact the coordinator: [email protected]