The automation runs daily at 21:00 and consists of three main Python scripts:

• Purpose: Fetch forecast data from:
  • OpenWeatherMap
  • WeatherAPI
• Output:
  • JSON summary → docs/{city}_forecast.json
  • PNG comparison chart → docs/{city}_comparison.png
• Includes:
  • Average temperature & rain probability
  • High/low temperatures
  • Comparison between both APIs

• Purpose:

  • Reads each city's forecast JSON
  • Scrapes forecast data from public sources:
    • YR.no
    • Meteoblue
    • Météo France (for Paris) or Météo Belgique (for Brussels)
  • Uses GPT-4 (OpenAI API) to:
    • Compare forecasts only between 09:00–21:00
    • Assess confidence in API predictions
    • Add a human-readable, natural-language gpt_comment
    • Classify alignment as: full, partial, or divergent
    • Comment intelligently avoids vague phrases like “most forecasts agree” unless exceptions are made explicit

• Output:

  • Appends gpt_comment and alignment to each city's forecast JSON
  • Comments are short, optionally playful, and context-aware (e.g., “cozy Sunday”, “good start to the week”)

• Purpose:
  • Reads enriched forecast JSON + PNG chart
  • Sends a clean and structured Pushover notification per city
  • Includes:
    • Summary (from JSON)
    • GPT insight comment
    • Attached PNG chart
    • Title includes alignment emoji (✅, ⚠️, ❌)
• Multi-user support: You can add multiple recipients by comma-separating user keys in the PUSHOVER_USER_KEY secret

• Trigger: Daily at 21:00 CET (19:00 UTC)
• Uses main.yml workflow and GH_PAT secret to auto-commit outputs (not the default GITHUB_TOKEN)

• OpenWeatherMap
• WeatherAPI
• OpenAI GPT-4 API

• HTML scraping of YR.no, Meteoblue, Météo France, and Météo Belgique using BeautifulSoup
• Rate-limiting and errors are handled

• All outputs saved to docs/ folder
• Publicly accessible via GitHub Pages for chart access if needed

• Uses Pushover for direct push alerts
• Image, text, and summaries are sent with rich formatting + emojis
• One message per city, combining summary + GPT insight + PNG chart

• GPT comment is parsed to detect forecast agreement level:
  • alignment = "full" → forecasts agree
  • alignment = "partial" → minor differences or slight outlier
  • alignment = "divergent" → major mismatch (e.g. rain vs no rain)

Forecast validation includes scraped data from trusted meteorological services, chosen for their reputation and accuracy:

• YR.no 🇳🇴 – From the Norwegian Meteorological Institute, widely praised for scientific rigor across Europe.
• Météo France / Météo Belgique 🇫🇷 🇧🇪 – Official government agencies, known for high-resolution local accuracy.
• Meteoblue 🌍 – Swiss-based ensemble forecaster, known for clarity and consistency across Europe.

These sources form a "ground truth" to assess API forecasts.

• Temperature lines:
  • 🔴 OpenWeatherMap
  • 🟠 WeatherAPI
  • ⚫ Average temp (thin dotted line)
• Rain lines (dashed bars):
  • 🔵 OWM
  • 🔹 WeatherAPI
• Grey Band: Fills the gap between the OWM and WeatherAPI temperature forecasts to show uncertainty/disagreement — clearly visible but not intrusive
• Heat Bands:
  • Warm zone (24–29.9°C): light pink (mistyrose, alpha=0.12)
  • Hot zone (30°C+): darker red (lightcoral, alpha=0.2)
  • Deliberate alpha choices ensure the consensus band remains clearly visible even in hot or warm conditions
• Labels at 12:00 & 18:00
• Title format: Paris Tomorrow – Day Forecast

• ✔ Daily GitHub Action is live
• ✔ JSON & PNG generated for Paris and Brussels
• ✔ ChatGPT analysis integrated
• ✔ Alignment detection working
• ✔ Pushover notification fully automated 🎉

scripts/
├── weather_notify.py         # Fetch APIs + generate chart
├── compare_and_analyze.py    # Scrapes + GPT + alignment tagging
├── send_to_pushover.py       # Push alert sender (final step)

docs/
├── paris_forecast.json       # JSON summary
├── paris_comparison.png      # Chart with annotated lines
├── brussels_forecast.json
├── brussels_comparison.png

• Daily at 21:00 CET

Add these under GitHub → Settings → Actions → Secrets:

GH_PAT=your_personal_token
OPENWEATHER_API_KEY=your_openweather_key
WEATHERAPI_API_KEY=your_weatherapi_key
OPENAI_API_KEY=your_openai_key
PUSHOVER_API_TOKEN=your_app_token
PUSHOVER_USER_KEY=your_user_key[,your_boyfriends_key]

You can add multiple user keys by comma-separating them. No spaces.

1. Create a file weather.env:

PUSHOVER_USER_KEY=your_user_key
PUSHOVER_API_TOKEN=your_pushover_token
OPENWEATHER_API_KEY=your_openweather_key
WEATHERAPI_API_KEY=your_weatherapi_key
OPENAI_API_KEY=your_openai_key

2. Install dependencies:

pip install -r requirements.txt

3. Run manually:

python scripts/weather_notify.py
python scripts/compare_and_analyze.py
python scripts/send_to_pushover.py

• Expand to more cities (e.g., Amsterdam, Berlin)
• Weekend trends or weekly summaries
• Archive via GitHub Pages
• UI dashboard for review

Pushover Message:

Paris: ☀️ Avg 27.1°C (0°C range), 0% rain
High 31°C / Low 22°C

🤖 GPT: Forecasts align well — looks like a warm, dry day. Perfect for a stroll by the Seine.

📊 Chart attached.

Q: Can I add someone else to receive alerts?
A: Yes, just update your secret:

PUSHOVER_USER_KEY=your_key,partner_key

Pushover will send to both recipients at once.

Built for curiosity, consistency, and comparing clouds ☁️