Treasure Island Crime Rate Trends — San Francisco
Treasure Island is a manmade 400-acre island in San Francisco Bay, built for the 1939 Golden Gate International Exposition and later used as a U.S. Navy base until 1997. Today it is a small residential community and event venue under long-term redevelopment, connected to the city by the Bay Bridge.
Treasure Island had a quiet March 2026. One category crossed the anomaly threshold — a below-trend signal in burglary — and a second registered as a zero-event period. With only two signals and no spikes, the month's shape is one of broad, sustained reduction across tracked categories rather than any single dramatic move.
Burglary is the most concrete illustration: 17 incidents in the current 12-month window against 40 in the prior 12, a 57.5% reduction. The multi-year baseline for that category had been running at 40 incidents per year. Aggravated assault, theft from vehicle, and other larceny all show similar year-over-year reductions — 61.1%, 50.0%, and 44.7% respectively — and none crossed the threshold this month, meaning those declines are now structural rather than fresh signals.
Notable signals 1
Burglary
The past 12 months saw 17 incidents — about 58% below the 40 average from prior years.
All categories, last 24 months
Each panel: recent monthly count vs. trailing 12-month context. MoM is the most recent month vs. the one before; 12mo YoY compares the trailing year to the year before that.
What's been quietly true for a year
Spikes get attention. Sustained shifts shape policy. These are multi-quarter patterns where the past 12-month total differs meaningfully from the year before — they often precede the baseline resetting.
No sustained shifts surfaced this month.
What next month likely looks like
Forecasts trained through March 2026, with a likely range we're 95% confident the actual count will fall inside. Categories with too little recent volume — or violent categories at the neighborhood level — show no forecast and are surfaced through signals above instead. See the methodology page for the gating rules.
Aggravated Assault
Too low-volume per neighborhood for a reliable point forecast — see the rare-event and streak-break signals above instead.
Arson
Below the volume threshold for a reliable forecast — too few incidents in recent months to project from.
Burglary
Homicide
Too low-volume per neighborhood for a reliable point forecast — see the rare-event and streak-break signals above instead.
Motor Vehicle Theft
Other Larceny
Robbery
Too low-volume per neighborhood for a reliable point forecast — see the rare-event and streak-break signals above instead.
Sexual Assault
Too low-volume per neighborhood for a reliable point forecast — see the rare-event and streak-break signals above instead.
Theft from Vehicle
Below the volume threshold for a reliable forecast — too few incidents in recent months to project from.
Vandalism
How Treasure Island compares
Peer neighborhoods picked by closest 12-month burglary volume — a pragmatic v1 of peer matching. Demographic / housing-stock peer matching isn't built yet (we deliberately don't ingest income or race data alongside crime). Volume similarity has the right intuition: “neighborhoods experiencing comparable burglary levels.”
Japantown
14 incidents over the past 12 months — 3 below Treasure Island's 17.
Open page →Twin Peaks
13 incidents over the past 12 months — 4 below Treasure Island's 17.
Open page →Portola
22 incidents over the past 12 months — 5 above Treasure Island's 17.
Open page →Recurring local terms (last 12 months)
Top terms in incident descriptions for Treasure Island, excluding generic crime taxonomy. Useful as texture — what kinds of specifics show up here that don't show up elsewhere.
Hour-of-day, day-of-week, and seasonality
Distribution of bucketed incidents in this neighborhood across the full analysis window. Useful for routine context — shopping-strip thefts vs. late-night assaults read very differently when you can see when each typically happens.
How we built this page
Data → Anomalies → Forecast → Page
Incident data is pulled from SFPD's open dataset, mapped to 10 NIBRS-aligned categories, and aggregated to neighborhood × category × month. Anomalies are surfaced using strict thresholds (~p < 0.01). Forecasts are Prophet with low-count gating; violent categories at the neighborhood level skip the forecast and show rare-event / streak signals instead.
Spike rule: 12-mo total > baseline mean + 2.5σ AND ≥ 20 incidents AND 6-mo confirms. Drop rule: 12-mo total < baseline mean − 2.5σ AND baseline mean ≥ 20. Rare event: any incident in the last 90 days, no prior comparable in ≥ 5 years.