SF

Toronto Shelter Forecast

Seasonality lab for shelter demand, unmet callers, and occupancy pressure
-series span
MSTLdecomposition
7 / 365seasonal cycles

Seasonality console

Fold demand onto the rhythms the forecast has to survive.

Choose a series, then compare weekday behavior, the annual cycle, long-run trend, residual volatility, and year-over-year shape. This view is for explaining when pressure is structural, seasonal, or noisy.

Series

Day-of-week profile

Mean value by weekday; useful for operational staffing rhythm.

weekly

Month-of-year profile

The 12-point annual cycle; watch for winter peaks and shoulder-season breaks.

annual

MSTL decomposition

Long trend, weather-linked cold overlay, and residual noise on one inspection surface.

trend + residual

Year-over-year overlay

Each calendar year on a Jan-Dec axis to expose repeated seasonal shape.

comparison
Method. MSTL from statsmodels with weekly and annual periods. Seasonal strength = 1 - var(residual) / var(residual + seasonal). Values describe signal structure; they are not causal estimates.