Historical temperature analysis for cities worldwide (and any custom lat/lon), using free Open-Meteo reanalysis data (daily, from 1940). Produces both the distribution of temperatures and how they change through the years.
🌍 Live site: https://yasoftwaredev.github.io/temperatury/ (auto-rebuilt from
fresh data on each push and yearly). Covers world capitals plus major cities
(~290 locations across 7 regions), picked from the interactive Leaflet map or
a region-grouped dropdown. Available in 6 languages — Polish, English,
German, French, Spanish and Ukrainian (switcher in the header; charts, captions
and the reading guide are all localised). Add a city in config.py, a language
in i18n.py.
A composite dashboard plus a dozen standalone PNGs and two interactive
(Chart.js) widgets in output/:
| View | Question it answers |
|---|---|
| Annual trend | How does the yearly mean change over time? (LOESS curve + robust Theil–Sen trend, °C/decade, Mann–Kendall significance) |
| Warming stripes | The Ed-Hawkins one-stripe-per-year anomaly graphic — blue→red warming at a glance |
| Anomaly bars | How does each year compare to the 1961–1990 norm? (blue = cooler, red = warmer) |
| Seasonal-cycle shift | First-decade vs last-decade monthly curve — which months warmed most (often winters) |
| Growing-season length | Days/year warm enough to grow (thermal season, daily mean ≥ 5 °C), with a trend |
| Hot & freezing days | How many days a year exceed 18 °C or stay below 0 °C, and how is that changing? |
| Diurnal range | Mean daily max − min (day vs night) — a shrinking range is a greenhouse fingerprint (needs daily max/min) |
| Month × year heatmap | Which months/seasons are warming? (discrete 2 °C bands) |
| Monthly anomaly heatmap | How much has each month shifted vs. its 1961–1990 norm? (seasonal cycle removed, robust scale) |
| Day-to-day swings | How often big day-to-day temperature jumps (≥6 °C) happen each year — temperature variability/volatility, with a trend |
| Annual precipitation | Total yearly rainfall (mm) over the years, with a trend (where precipitation data is cached) |
| Monthly min–max range (interactive) | Pick any year and see where its 12 monthly means fall within the full historical envelope |
| Monthly record high/low (interactive) | Each month's all-time record daily high/low, with any selectable year's extremes — "hottest day ever in July?" |
Every view is time-aware: the hot/freezing-day counts track the distribution's tails year by year, and the monthly-anomaly view normalises each month to its own baseline so the warming signal stands out even where the absolute heatmap is dominated by the seasonal cycle.
uv venv .venv
uv pip install --python .venv -r requirements.txt.venv/bin/python main.py # Warszawa, 1940..last full year
.venv/bin/python main.py --location krakow # preset: krakow/gdansk/wroclaw/poznan
.venv/bin/python main.py --all # every preset city, linked by a nav bar
.venv/bin/python main.py --lat 48.85 --lon 2.35 --name Paris
.venv/bin/python main.py --start 1980 --end 2024 --refreshThe published site (--all) builds one page per city with a switcher in the
header; the root index.html shows Warszawa.
Downloaded data is cached under data/; pass --refresh to re-download.
config.py— locations, paths, API constantsdata.py— download + cache daily temperaturesplots.py— the chart functions (LOESS/Theil–Sen/Mann–Kendall stats) + dashboard compositioninteractive.py— the Chart.js monthly-range / records widgets (year picker)report.py— generates the static HTML page (with the city switcher)main.py— CLI entry point and summary printoutassets/appearance.js— the visitor-facing appearance picker. Each reader chooses a look (style: objective / editorial / product / atlas, plus accent, headline face, density, header treatment and light/dark) and it is remembered on their device. Everything is driven bydata-*attributes on<html>; the stylesheets route through tokens andcharts.jsre-colours the canvases on athemechangeevent. A tiny inline bootstrap in each page's<head>applies the saved choice before first paint, so there is no flash.assets/page.src.css,assets/landing.src.css— Tailwind v4 source for the city-page and world-map stylesheets. Edit these, then runtools/build-css.shto regenerate the committedassets/page.css/assets/landing.css(which the build ships verbatim — the site itself needs no Node/Tailwind toolchain). New UI styling should prefer Tailwind utility classes; the CLI compiles only the utilities actually used acrossreport.py,interactive.pyandassets/*.js.
overall mean daily temp : 8.41 °C
warming trend : +0.29 °C / decade
warmest year : 2024 (11.46 °C)
coldest year : 1940 (5.72 °C)
Data source: Open-Meteo historical reanalysis (ERA5). Free for non-commercial use.
The free Open-Meteo quota only lets one machine download a limited number of cities per day, so contributions from several people speed this up a lot. GitHub is the shared source of truth - no coordination or hand-assigned "chunks" needed.
Non-technical contributor? See CONTRIBUTING.md for a step-by-step guide (install Git + Python, then one command). The short version:
git clone --depth 1 https://github.com/YASoftwareDev/temperatury.git
cd temperatury
bash tools/daily-chunk.sh
tools/daily-chunk.sh pulls the latest data, downloads a chunk of whatever
cities are still missing (as much as today's quota allows), then sends the new
data back automatically by the best route available:
- push straight to the repo - if you have write access;
- open a Pull Request from your fork - if you have a GitHub account and the
ghCLI (gh auth login); - otherwise it packages the files and prints exactly what to do (e.g. email the archive).
It creates its Python environment on first run and is safe to run once a day (by
hand or from cron: 30 0 * * * cd /path/to/temperatury && bash tools/daily-chunk.sh).
Because every city's file is identical no matter who downloads it, two people
fetching the same city just no-op after a pull - the missing pool only shrinks.
