Utah Weekly Crash Forecast  ·  Jul 27 – Aug 2, 2026

For today's day-of-year, we pull every crash within ±7 days across 2021–2025 — a 15-day rolling window across five years, roughly 75 days of comparable historical observations.

The baseline is a year-weighted average (recent years weight more: 0.28 (2025), 0.24, 0.20, 0.16, 0.12 (2021)). Each forecast day starts as the baseline scaled by its day-of-week multiplier. Holiday weekends (Memorial Day, Independence Day, Pioneer Day, Labor Day, Thanksgiving) receive an 8–12% bump for recreational travel.

The risk score for each hazard combines relative volume (60%), severity weighting (KSI and pedestrians up-weighted because Vision Zero and HSIP optimize for these outcomes), and a year-over-year trend bonus.

The forecast also uses weather. UDOT tags every crash with the weather at the time of the crash (rain, snow, fog, etc.) and the road surface (wet, snow/ice, dry). For each weather category we compute an empirical day-level rate: how many crashes per day occur when, say, snow is reported, versus a comparable day without snow. That ratio becomes a multiplier — for example, snow days in Utah historically run about 1.4–1.8× the dry-day rate. When the 7-day weather forecast from Open-Meteo calls for snow on a given day, that day's projection is multiplied by the snow multiplier from your own data. When the empirical sample is too small (fewer than 3 weather-days in the window), we fall back to FHWA's published rates from How Do Weather Events Impact Roads? (FHWA-HOP-15-038).

Locations are joined to UDOT's road network by street_id — the same quarter-mile segments used by the PMTiles crash map. Every count on this page ties back to a column in udot_stats.csv, and every address links to the crash map zoomed to that segment.

How hotspots are ranked. Most hazard hotspots are ranked within the same ±7-day seasonal window. But for rare, high-severity categories — KSI (killed or seriously injured), pedestrian, and bicycle crashes — a 15-day window holds too few events to rank locations reliably, so a single seasonal incident could outrank a persistently dangerous corridor. For these categories we instead rank hotspots on the full 2021–2025 record, surfacing structurally dangerous segments rather than seasonal flukes. Ties are broken deterministically so rankings are reproducible. Because a recurring severe or pedestrian/bicycle crash is a meaningful structural signal regardless of how busy a road is, these categories are always presented as definitive hotspots. For higher-volume hazards, segments with too few crashes to rank reliably are instead flagged as a low-confidence signal. All other projections, shares, and risk scores remain window-based.

What "#1 hotspot" vs. "recent activity" means. "#1 hotspot" is a confident label — the top segment has enough crashes to stand out reliably. "Recent activity" is a cautious label, shown only when a category is so sparse that its busiest segment has too few crashes to call a definitive hotspot. A common category like weekday crashes shows "#1 hotspot" precisely because it has plenty of data, while a rare category in a quiet week may show "recent activity" instead.