ArcGIS ParquetLayer with MeteoSwiss Forecasts

This is a small ArcGIS Maps SDK for JavaScript 5.1 prototype that loads MeteoSwiss Open Data from Apache Parquet files directly in the browser.

Open the live app

The app uses MeteoSwiss E4 local forecasts: hourly weather values for about 6,000 locations in Switzerland, including temperature, precipitation, wind, and cloud cover. The data is converted to Parquet with Python, hosted as static files, and displayed with ParquetLayer.

The idea

This project is a practical test of a simple pattern:

Open data snapshots
-> Python download and parsing
-> pyarrow writes Apache Parquet
-> shapely prepares derived geometries
-> ArcGIS Maps SDK for JavaScript loads the files with ParquetLayer

The result is a static forecast explorer that can step through time and switch between raw points, square cells, and hexagons. The same approach can fit open data portals, forecast snapshots, generated model outputs, and read-only layers that are refreshed on a schedule.

Start with these files

• index.html is the Calcite and ArcGIS web-component shell.
• js/app.js is the core ArcGIS ParquetLayer setup.
• js/forecast-extras.js contains the renderers, legend, time controls, hover cards, and popups.
• python/scripts/refresh_sample_data.py refreshes the generated sample data.
• docs/data-generation-workflow.md explains the data parsing and Parquet export.

If you only want the ArcGIS loading pattern, start with js/app.js.

Load Parquet in ArcGIS Maps SDK

The important part is small: create a ParquetLayer, point it at a Parquet URL, and describe the WKB geometry column.

const layer = new ParquetLayer({
  data: new ParquetFilesData({ urls: [parquetUrl] }),
  geometryEncoding: new GeometryEncodingWkb({
    field: "geometry",
    orientation: "counter-clockwise",
  }),
  geometryType: "polygon",
  spatialReference: { wkid: 4326 },
  outFields: ["*"],
});

layer.definitionExpression = "valid_time = '2026-06-28T11:00:00Z'";

In this app, definitionExpression selects one forecast hour at a time. A small .arcgis-timeinfo.json sidecar provides the UI with forecast times, labels, value ranges, source links, and feature counts. The ArcGIS layer itself reads the Parquet file.

Build the same kind of app

1. Start with a dataset that can be represented as points, lines, or polygons.
2. Write an Apache Parquet file with a geometry column encoded as WKB.
3. Add GeoParquet metadata so the geometry column is explicit.
4. Host the Parquet file over HTTP with the rest of the static app.
5. Load it with ParquetLayer and ParquetFilesData.
6. Use attributes such as valid_time, category, model run, or scenario ID in definitionExpression when the app needs filtering.

Data files

• data/meteoswiss_points_all-points_48h.parquet contains direct MeteoSwiss forecast points.
• data/meteoswiss_surface-square-128x72_all-points_48h.parquet contains a derived square-cell visualization.
• data/meteoswiss_surface-hex-128x72_all-points_48h.parquet contains a derived hexagon visualization.

The square and hexagon files are derived visualization layers, not MeteoSwiss gridded forecast products.

Run locally

Serve the repository root over HTTP. Do not open index.html as a file:// URL.

npx vite --host 0.0.0.0 --port 3107

Open http://localhost:3107/.

Vite is only used as a local static server. The published app has no Vite config, package.json, bundling step, or npm dependency tree.

Refresh the data

Point-only refresh:

uv run --project python python/scripts/refresh_sample_data.py --force

Full demo refresh with point, square, and hexagon files:

uv run --project python python/scripts/refresh_sample_data.py --geometry all --force

GitHub Actions runs the full refresh every six hours before deploying GitHub Pages:

uv run --project python python/scripts/refresh_sample_data.py --geometry all --max-age-hours 6

Source data

• MeteoSwiss Open Data catalog
• E4 local forecast documentation
• FSDI STAC collection
• MeteoSwiss Open Data terms

Attribution and license

Project code is released under the MIT License. See LICENSE.

The generated data is derived from MeteoSwiss Open Data, published under CC BY 4.0 according to the MeteoSwiss terms of use. When reproducing or redistributing the meteorological data, cite: Source: MeteoSwiss.

The square and hexagon surfaces also use the Swiss boundary from FSDI/swisstopo for clipping. When publishing those surfaces, cite: Source: MeteoSwiss; © Data: swisstopo.

This app sets those notices with ParquetLayer.copyright, so they appear in the ArcGIS map attribution control.

References

• ArcGIS ParquetLayer sample
• ArcGIS ParquetLayer API
• ArcGIS ParquetFilesData API
• Esri blog: Parquet feature layer beta in ArcGIS Online
• Apache Parquet
• GeoParquet 1.1.0