×
AI Weather Nowcasting With StormScope-US · From Satellite Observations to Interactive Forecasts
Temporary demo pipeline for SC26 Chicago. It goes offline after the event.
Live · --:--
Forecast
--:--
Forecast initialized --:-- · GOES-19 ABI
Init
01 · AI Weather Nowcasting

From Satellite Observations to AI Forecasts

StormScope-US uses six GOES-19 observations to predict twelve future satellite frames at 10-minute intervals, covering the next two hours.

WHY NOWCASTING
  • NWP takes hours to run
  • Nowcasting starts from what satellites see in real time:
    • Advection-based: moves a frozen cloud field
    • Generative: clouds also form and dissolve
WHY IT MATTERS
Solar power rampsSevere stormsOutdoor events
02 · Earth Observations

Observing the Atmosphere from Space

GOES-19 (GOES-East) observes the Americas from geostationary orbit, approximately 36,000 km above the equator at 75.2° W.

Its Advanced Baseline Imager measures 16 spectral channels and scans the contiguous US every 5 minutes.

Two Satellite Composites
16spectral channels
5 minCONUS scan cadence
3 kmmodel grid over CONUS
03 · THE AI MODEL

StormScope-US predicts the next two hours of clouds

A generative diffusion model from NVIDIA, distilled for speed (Flash). It forecasts eight GOES channels directly in image space on a 3 km grid, learning how clouds move and grow instead of solving the physics.

INPUT · 6 OBSERVED FRAMES
−50−40−30−20−100
↓
StormScope-US Flash on GPUFew-step distilled sampler · bf16
↓
+10 minThe new frame is fed back in. 12 rollouts reach +120 min.
  • Conditioned on topography
  • Ensemble-of-experts denoising across noise levels
  • NVIDIA L40S GPU · ~3.5 s per 10-minute step
04 · THE PIPELINE

Earth2Studio wires data, model and storage together

NVIDIA's open-source Python package for AI weather and climate. You can swap data sources, models and outputs without rewriting the pipeline.

  1. 1
    FetchLatest 6 GOES-19 CONUS frames from NOAA Open Data on AWS
  2. 2
    PredictStormScope-US rolls forward 12 × 10 min on the GPU
  3. 3
    WriteUpload forecasts to Arraylake while GPU runs
  4. 4
    ServeMap tiles served via Earthmover Compute
05 · DATA FLOW

From satellite to screen in 5 minutes

  1. GOES-19Scans the contiguous US from orbit
  2. NOAA Open Data on AWSProcesses and publishes the scan
    ~4 min
  3. DownloadEarth2Studio fetches the latest frames
    ~2 s
  4. Inference on NVIDIA GPU saving to ArraylakePredicts 2 hours ahead
    ~40 s
  5. Earthmover ComputeRenders map tiles straight from the data
    ~2 s
  6. Your browserLoads all 13 frames and plays the loop
    ~12 s
NOAANVIDIAEarthmover
06 · Earthmover Arraylake + Icechunk

Compute and Upload Together. Publish Once.

Forecast data is uploaded to Earthmover Arraylake while StormScope-US generates subsequent predictions on GPUs.

Icechunk atomically publishes each complete forecast as a versioned snapshot.

Overlapped Uploads · Atomic Publication · Versioned History
Overlapped PublicationSuccessive GPU prediction steps overlap delayed uploads. A completion barrier waits for every upload, then one atomic Icechunk commit exposes the complete forecast snapshot to readers. Widths illustrate overlap and are not measured durations. Execution Time · Schematic GPU I/O +10 +10 +20 +20 … … +120 +120 All writes finish One Atomic Icechunk Commit
07 · THE DATA LAKE

Every forecast run is a commit

Each 10-minute step uploads to Arraylake while the GPU computes the next one, and a single versioned Icechunk commit publishes all 12 at once. Icechunk atomically publishes each complete forecast as a versioned snapshot.

Recent published forecasts
Latest Forecast Publications—
TransactionalIcechunk: open-source, cloud-native Zarr storage
Time travelOpen any past forecast by its snapshot ID
Browse this dataset in Arraylake →
08 · YOUR TURN

Take the forecast with you

Raw tilesLatest run, 10-min forecast, served via Earthmover Compute
Animated viewerEvery run, every lead time
Live SQL…
How cloudy over the venue, next 2 h · PostgreSQL wire protocol
←/→ Browse · Space Pause
1 / 8