Here is a recent experiment from my personal lab: using DuckDB-WASM, Mosaic and GeoArrow to query a Parquet file directly in the browser without a backend.

The architecture uses Mosaic's global Coordinator to manage state between linked views using SQL predicates. Because DuckDB-WASM returns standard Arrow tables, the WKB output is converted to GeoArrow point vectors before being passed to the GeoArrowScatterplotLayer for rendering. Credit goes to Kyle Barron for his work on the @geoarrow/deck.gl-layers package.

<video src="/earthquake-demo.mp4" controls autoplay muted loop style="max-width: 100%;"></video>

Desktop-only demo with bi-directional brushing: https://linked-views-demo.netlify.app/

(thanks to Henry Rodman for the vrt and vsicurl hints below)

Step 1: Find the right GRIB2 file

Browse NOAA’s HRRR model (limited to CONUS, see below for global options) via the ASDI bucket.

File structure:

s3://noaa-hrrr-bdp-pds/hrrr.YYYYMMDD/conus/hrrr.t{HH}z.wrfsfcf{XX}.grib2

• t{HH}z = model run hour
• f{XX} = forecast hour (f00 is analysis, f01–f48 are 1–48h forecasts)

Example URL for wind at 10m on 2025-05-12 06Z, 4-hour forecast:

https://noaa-hrrr-bdp-pds.s3.amazonaws.com/hrrr.20250512/conus/hrrr.t06z.wrfsfcf04.grib2

Step 2: Understand the wind data

Wind data is split into two bands:

• U = east-west component (positive = eastward)
• V = north-south component (positive = northward)

You'll need both bands for visualization.

Step 3: Create the TiTiler URL

Use vsicurl to let TiTiler access the remote GRIB2 file directly:

https://your-titiler-instance.xyz/cog/preview.png?rescale=-127,128&url=vrt:///vsicurl/https://noaa-hrrr-bdp-pds.s3.amazonaws.com/hrrr.20250512/conus/hrrr.t06z.wrfsfcf04.grib2

Step 4: Extract wind bands

Get the U and V wind components (in this case, bands 10 and 11):

https://your-titiler-instance.xyz/cog/preview.png?rescale=-127,128&url=vrt:///vsicurl/https://noaa-hrrr-bdp-pds.s3.amazonaws.com/hrrr.20250512/conus/hrrr.t06z.wrfsfcf04.grib2?bands=10,11&format=png

This gives you the necessary tiled wind png, ready for visualization. It should look something like this one:

Step 5: Visualize with DeckGL

Pass your complete TiTiler URL to the ParticleLayer from WeatherLayers in Deck.GL:

new ParticleLayer({
  id: 'wind-particles',
  image: 'THE_URL_GOES_HERE',
  imageType: 'VECTOR',
  imageUnscale: [-127, 128],
  bounds: [-134.1214, 21.1222, -60.8912, 52.6287],
  clipBounds: [-134.1214, 21.1222, -60.8912, 52.6287],
  // rest of the config
})

The result should look something like this:

<video src="/wind-viz.mp4" controls autoplay muted loop style="max-width: 100%;"></video>

What about GFS?

I haven’t tested it yet, but the same approach should work for GFS data (NOAA GFS on AWS) with the only difference being the path structure.

It lets you explore the progression of wildfire perimeters over time, overlay wind and terrain data and export animations or videos, all directly in the browser.

• Explore the tool: https://earthdata.nasa.gov/dashboard/tools/fire-event-explorer
• How and why it was built: https://developmentseed.org/blog/2025-06-03-fire-exploration-tool/

<video src="/fire-event-explorer.webm" controls autoplay muted loop style="max-width: 100%;"></video>

Note: VEDA is a set of open-source components and services for easily working with earth observation data, as envisioned by NASA IMPACT.