---
title: DMI Aarhus Collector
emoji: "🌤️"
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.47.2
python_version: "3.10"
suggested_hardware: cpu-basic
startup_duration_timeout: 15m
app_file: app.py
license: apache-2.0
---

# DMI Aarhus Collector

Operational Hugging Face Space for Aarhus weather ingestion, prediction generation and snapshot publishing. Maintained by `Ciroc0`.

## What this Space does

- fetches Aarhus forecasts from Open-Meteo's DMI HARMONIE-backed endpoint
- fetches Aarhus observations from Open-Meteo Archive
- builds and updates `training_matrix.parquet`
- loads active model bundles from `Ciroc0/dmi-aarhus-weather-data`
- generates 48-hour future predictions
- verifies old predictions against actual observations
- publishes `frontend_snapshot.json` for the public frontend

## Scheduler

| Time | Job | Notes |
| --- | --- | --- |
| `00:35`, `03:35`, `06:35`, `09:35`, `12:35`, `15:35`, `18:35`, `21:35` | `generate_predictions()` | Writes fresh future predictions |
| Every hour `:12` | `verify_predictions()` | Adds actuals to expired predictions |
| Daily `05:45` | `update_daily()` | Refreshes training data for the recent 7-day window |

Startup also runs a catch-up thread if datasets are missing or stale.

## Inputs and outputs

### Input

- `Ciroc0/dmi-aarhus-weather-data`

### Output

- `training_matrix.parquet` to `Ciroc0/dmi-aarhus-weather-data`
- `predictions_latest.parquet` to `Ciroc0/dmi-aarhus-predictions`
- `frontend_snapshot.json` to `Ciroc0/dmi-aarhus-predictions`

## Current prediction targets

- `temperature`
- `wind_speed`
- `wind_gust`
- `rain_event`
- `rain_amount`

## Notes

- The current code still supports legacy fallback files such as `data.parquet` and `predictions.parquet`.
- `frontend_snapshot.json` is the primary integration contract for the Vercel frontend.

## License

Source code in this Space is released under `Apache-2.0`. Preserve attribution to `Ciroc0` and keep applicable notice/license files when redistributing the code.