The Guardian ran a data-visualisation roundup on 9 April and gave Melbourne Seasons a shout out. We are a sole trader working out of Fitzroy North, so this is not something that happens often.
The quote was short:
How many seasons does Melbourne have? As an unwilling Melbourne resident I'd say it has two, cold and colder. But software developer [...] has built a very cool app that suggests there are actually seven. The methodology page on this app is fascinating, even using cosine similarity to match season names with weather data.
Most of Melbourne Seasons is a Python pipeline. It fetches 75 years of daily weather data from Open-Meteo, falls back to NASA POWER when rate limits hit, runs kernel changepoint detection across 16 climate variables, and matches each detected season against a library of archetypes (Furnace, Deluge, Gale, Frost, and a few more).
Cosine similarity lives at the matching step. Each archetype is a vector of expected values across the 16 variables. Each detected season is another vector. The closest cosine match wins the name. It is a small corner of the project, and we loved that the Guardian zoomed in on it.
We would like to make more hyper-local data projects. Open data is great, and Melbourne has a lot of it.