Blended from live station reports
Interpolated between live station reports (inverse-distance weighting) — accurate at each dot, a smooth guess in between. No terrain or model background yet, so this cannot show a valley pooling colder than a nearby ridge unless a station happens to sit in it. On the precipitation tab, station dots show a gauge's own last-hour total as a cross-check against the radar estimate underneath — below freezing, gauges read low for real physical reasons and mostly agree with each other while doing it, so a mismatch there is expected, not a fault. On the air quality tab, PM2.5 is regional PurpleAir readings run through the EPA/Barkjohn bias correction — a real improvement over a raw reading, but not yet checked against a real reference monitor near Pelham, so treat it as a good estimate rather than a certified AQI category. Station dots show the raw reading too, as a cross-check, the same way the precipitation tab shows a gauge next to the radar estimate.
Radar motion extrapolation · next 3 hours
Reduces rain-rate error versus a flat persistence baseline (−21% on a tested convective case) but does not yet beat persistence on the rain/no-rain call itself — treat onset times as approximate, not precise.
WRF, 1 km grid, HRRR boundary conditions
One run per cycle, no bias correction, no multi-run skill verification yet — unlike the forecast below, which is calibrated against a year of this station's own history. Shown to see what the model produces, not because it is proven accurate. See wrf/README.md for the one verification run so far.
HRRR at this grid point
5-minute readings
How far off recent forecasts have been, checked against what actually recorded.
Each day's typical miss in degrees — grey is the raw computer model, blue is after correcting it for real measurements. Day-to-day bumps are just harder or easier weather, not the model changing.
The forecast made a day ahead, laid over the temperature that actually occurred — shows whether the timing and shape of the day were right, not just whether the average was close.
Live USGS gauges near Pelham
Flood category (action/minor/moderate/major stage) is only shown where NWS has actually defined one — most small streams here, including Beaver Brook, don't have an official threshold, so they show a level and a trend instead of a risk label. Gauge coordinates are the monitoring station's own location, not necessarily the nearest point on that stream to the house — Beaver Brook's gauge, for instance, is 3.9 km (2.4 mi) away along the brook, not the ~300 yd where it's closest to the property. Gauges outside the map's edges (a few of the 23 below) still show in the table, just not plotted.
Freshness of each ingest source