weather-monitor

One dashboard that answers a single question — "is today calm, or not?" — by blending weather, geomagnetic activity, earthquakes/wildfires, news volume, and market volatility into one integral index. Live demo → (neutral placeholder city, no history)

Русская версия: README.ru.md

What it does

Every day (GitHub Actions, once a day) it pulls from ten free/keyless APIs, computes one weighted activity index, renders a 4-tab static site, and — if anything crossed a threshold — pings Telegram:

• Weather — OpenWeatherMap (current + forecast)
• Space weather — NOAA SWPC (Kp/Ap index, solar wind, aurora probability) + Schumann resonance (best-effort, honestly flagged when stale)
• Natural events — NASA EONET (storms/wildfires/ice) + USGS (significant earthquakes)
• "Digital noise" — publication frequency across BBC / Google News / Al Jazeera / RIA Novosti RSS, as an indirect activity signal
• Markets — daily % change on 5 baseline tickers (S&P 500, VIX, MOEX, USD/RUB, EUR/USD) via the Yahoo Finance chart API

Design is inspired by sungeo.net — earth-tone palette, ring gauge, status badges.

The Sri Lanka case

A second, independent index tracks hazard risk for a specific coastal point — the same EONET/USGS feeds, geo-filtered by radius (earthquakes within 2500 km and no older than a configurable age cutoff, storms within 1500 km), with its own weighted scoring (compute_sri_lanka_index in composite.py). It started as personal risk-monitoring for a place the author cares about and stayed in as a worked example of the trap of a stale-but-technically-in-window event silently pinning an index for weeks (see the age-cutoff fix, SRI_LANKA_QUAKE_MAX_AGE_DAYS, documented in the code). Always shown in the Telegram notification alongside live sea/air temperature and wind for the coast, regardless of whether a trigger fired — see Automation below.

The scoring/filtering logic isn't tied to Sri Lanka specifically — swapping the coordinates and radii in sources.py repoints it at any coastal location.

Run it locally

pip install -r requirements.txt
export OPENWEATHERMAP_API_KEY=...   # openweathermap.org/api, free tier
python build.py "London,GB"          # or any other city — no args builds the same placeholder demo published on Pages

Open the generated index.html in a browser.

Data sources (all keyless except OpenWeatherMap)

Source	What it provides
OpenWeatherMap	current weather + forecast
NOAA SWPC	Kp/Ap index, solar wind, aurora probability
Schumann Resonance Live	Schumann resonance (unofficial source, honestly flagged when stale)
NASA EONET	open natural-event feed (storms feed the coastal-hazard index)
USGS	significant earthquakes, filtered to a recent window (feeds the coastal-hazard index)
Open-Meteo / Open-Meteo Marine	air temperature, wind, sea temperature for the coastal point
BBC / Google News / Al Jazeera / RIA Novosti (RSS)	"digital noise" — publication frequency as an indirect indicator
Yahoo Finance chart API	S&P 500, VIX, MOEX, USD/RUB, EUR/USD — daily % change

How the automation works (.github/workflows/daily-check.yml)

flowchart TD
    subgraph SRC["Data sources (all keyless except OpenWeatherMap)"]
        OWM["OpenWeatherMap"]
        NOAA["NOAA SWPC<br/>Kp/Ap, solar wind, aurora"]
        SCHU["Schumann Resonance Live"]
        EONET["NASA EONET<br/>storms/wildfires/ice"]
        USGS["USGS<br/>earthquakes"]
        OM["Open-Meteo / Marine<br/>coastal air+sea+wind"]
        RSS["BBC / Google News /<br/>Al Jazeera / RIA (RSS)"]
        YF["Yahoo Finance chart API"]
    end

    subgraph GHA["GitHub Actions — daily-check.yml (cron, once a day)"]
        SRCPY["sources.py<br/>collect()"]
        MOON["moon.py<br/>phase, no API"]
        COMP["composite.py<br/>integral index +<br/>coastal-hazard index"]
        BUILD["build.py"]
        RENDER["templates.py + pages.py + charts.py<br/>→ 4 HTML files"]
        SUMMARY[("data/latest_summary.json<br/>data/history.jsonl")]
        NOTIFY["notify.py<br/>evaluate() trigger conditions"]

        SRCPY --> COMP
        COMP --> BUILD
        MOON --> BUILD
        BUILD --> RENDER
        BUILD --> SUMMARY
        SUMMARY --> NOTIFY
    end

    PRIVREPO[("Private history repo<br/>(not this repo)")]
    TG["Telegram Bot API<br/>sendMessage"]
    CHAT(["Your Telegram chat"])
    DISCARD["discarded — never committed<br/>(.gitignore *.html)"]

    OWM --> SRCPY
    NOAA --> SRCPY
    SCHU --> SRCPY
    EONET --> SRCPY
    USGS --> SRCPY
    OM --> SRCPY
    RSS --> SRCPY
    YF --> SRCPY

    PRIVREPO -- "history_sync.py pull<br/>(before build)" --> SRCPY
    SUMMARY -- "history_sync.py push<br/>(after build)" --> PRIVREPO

    RENDER -.-> DISCARD

    NOTIFY -- "NOTIFY_MODE=always,<br/>or a trigger fired" --> TG
    TG --> CHAT

Trigger conditions notify.py checks in NOTIFY_MODE=condition (any one fires the alert; in the default always mode it sends every run regardless): integral index ≠ "Calm" · market volatility or news-headline volume >1.5× its historical average (only once enough history has accumulated) · coastal-hazard index ≠ "Calm". The coastal-hazard block is always included in the message body, whether or not it was the thing that fired.

This repository is public and holds only code — neither the real city nor the run history is committed here (see .gitignore and the docstrings in build.py/history_sync.py): a public git history growing on a schedule would quietly leak geolocation and usage patterns. So:

1. History (data/history.jsonl) is synced with a separate private repo via history_sync.py (pull before build, push after) — never written to this public repo.
2. Generated HTML pages are never committed either — each run's output is throwaway, not a persistent site. (The demo/ folder is the one deliberate exception — see below.)
3. Alert condition (notify.py): by default (NOTIFY_MODE=always, the settled mode) a Telegram notification goes out on every run regardless of trigger — the coastal-hazard block is always visible in it. If NOTIFY_MODE=condition is explicitly set, it only sends when a trigger actually fired: the integral index ≠ "Calm", OR market volatility / news volume is notably (>1.5×) above its historical average (once enough history has accumulated), OR the coastal-hazard index ≠ "Calm" (a ~M5.75+ earthquake within 2500 km in the last few days, or an active storm in the basin).

Repository secrets (Settings → Secrets and variables → Actions)

Secret	Purpose
OPENWEATHERMAP_API_KEY	weather API key
WEATHER_CITY	e.g. London,GB — never stored in code
HISTORY_PUSH_TOKEN	GitHub PAT with write access to the private history repo
TELEGRAM_MOTION_BOT_TOKEN	Telegram bot token for alerts
TELEGRAM_MOTION_CHAT_ID	where to send the alert
NOTIFY_MODE	optional; always (default if unset) or condition — see above

The demo/ folder and GitHub Pages

demo/ is a one-time, committed build for a neutral placeholder city (London) — the one deliberate exception to "nothing generated gets committed," precisely because it carries no real location or history. .github/workflows/deploy-pages.yml publishes it to GitHub Pages on every push to demo/; it's a separate workflow from daily-check.yml and never touches the real automation. To refresh the demo after a UI change: run python build.py (no args → the default placeholder city), copy the four HTML files into demo/, commit.

Files

• sources.py — fetches data from every source
• moon.py — moon phase, pure calculation, no API
• composite.py — the integral activity index (normalization + weighted blend) + the separate coastal-hazard index
• charts.py — SVG line/bar/ring charts
• templates.py — shared page shell (nav, CSS)
• pages.py — each tab's body
• build.py — entry point: collects data, computes the index, writes history + summary, renders 4 HTML files
• history_sync.py — pull/push data/history.jsonl to/from the private history repo
• notify.py — checks the alert condition, sends to Telegram
• .github/workflows/daily-check.yml — schedule + steps
• .github/workflows/deploy-pages.yml — publishes demo/ to GitHub Pages

Contributing

See CONTRIBUTING.md. Issues and PRs welcome.

License

MIT