Hyperlocal verification-driven forecasting · observations, model skill, and the garden record
Forecast discussion yard AFD
Next 72 hours — temperature PropertyCast vs models
Next 72 hours — precipitation blend
Chance of precipitation
Amount per hour
Aurora planner SWPC Kp × cloud cover
Frost odds — next 3 nights from verified error history · cover when odds beat each crop's cost/loss
Not reporting yet
Work windows next 72 h · yard wind at plant height
Watering call ET0 deficit × ensemble rain odds
Tomato disease TOMCAST
Chill hours 32–45°F hours · Utah units · Oct 1 – Apr 30
Insect timing degree days
Soil temperature learned from your probe
Garden season to date
Growing degree days since Apr 1
Your yard's frost dates 40+ yr reanalysis × your offset
Harvest & bloom ETAs learned from your journal
Not reporting yet
Model skill 30-day · 6–24 h lead
Which PropertyCast you see v2 must beat v1 · Diebold-Mariano, 30 d
Reliability diagrams forecast % vs how often it happened · on the diagonal = honest odds
Probability scorecard are the odds honest?
Station health neighbor consensus QC
Monthly verification month to date
Data sources feed health at publish time
Wind shelter by direction forecast wind minus your station, by forecast direction · 6–24 h
Not reporting yet
Observed — last 7 days temp / dewpoint
What PropertyCast doesn't cover read before relying on it
Not a warning service. For severe storms, blizzards, floods and fire weather, follow the National Weather Service. PropertyCast has no radar nowcasting and issues no watches or warnings.
One backyard. These are forecasts for a single property in the Red River Valley. Its exact location is withheld on purpose, so conditions a few blocks away can differ.
Temperatures are at sensor height. Frost can form on the ground and on low leaves while the sensor still reads above 32°F, especially on calm, clear nights. Frost odds and cover calls are tuned to the sensor, so err toward covering tender plants.
Wind is sheltered by design. The station sits below standard 10 m height, behind a tree line, so it reads lower than the airport. Work windows use this yard wind; high-wind alerts use open-air model gusts, because trees and roofs feel the wind above the canopy.
Snow and winter precipitation are undercounted. The station's rain gauge isn't heated, so snowfall totals and frozen precipitation are unreliable until it melts.
Scores take time. A 24-hour forecast can only be graded 24 hours later. Corrections switch on per model, lead time and weather type only after enough verified cases, and rare situations fall back to broader corrections.
Gaps in the record. The station reported from Sep 2020 to Feb 2024 and again from Oct 2026; in between, the nearby airport (KFAR) stood in. Climatology, frost dates and chill comparisons use only real station data.
Neighbors are personal stations. The neighbor comparison uses nearby home weather stations of varying siting and quality; outliers are dropped before the median, but it is a consistency check, not a reference standard.