Radar nowcast when will it rain, when will it stop, how sure is it
National radar (NOAA MRMS) is sampled every ~2 minutes and sliced to a 5 km box over the plateau. mswx tracks the precipitation field's motion and projects it forward 60 minutes. Every new scan re-predicts when rain starts and stops at the house pixel — agreement across recent scans, newest weighted heaviest, is the confidence.
Last 60 min observed → next 60 min nowcast (house pixel, mm/hr)
Recent scans agree on timing?
Next 24 hours — chance of rain (NWS raw vs blended vs blended+radar)
Recent radar observations (latest 15)
How this works
MRMS (Multi-Radar Multi-Sensor) is NOAA's national radar mosaic — a per-cell instantaneous precipitation rate (mm/hr) blended from every weather radar and rain gauge in the country, updated every ~2 minutes at 1 km resolution. KHTX (Hytop, AL) is the dominant radar over Monte Sano at ~50 mi NE.
The system continuously pulls the freshest scan, slices a 5×5 km box around the plateau, and records the rate at the house pixel plus the box mean, max, and 90th percentile. The headline — raining or not, how hard, when it starts and stops — reads the house pixel; the box max is only the "heavier cell nearby" signal, since a hot cell 2 km away is not rain on the plateau.
Motion comes from per-pixel optical flow (pysteps Lucas-Kanade) across recent scans, so different cells can move different ways. Nowcast advects the current field forward and samples whatever passes over the house pixel — i.e. "each storm keeps moving the way it's moving." Solid for organized rain at 0–60 min, shakier for isolated pop-up cells.
Confidence is an ensemble measure: every 2-minute scan produces a fresh prediction of when rain starts (or stops). When recent scans cluster tightly on the same time, confidence is high; when they scatter, it's low. Newer scans count more (weights halve every ~3.5 min), so when a storm changes its mind the headline follows the fresh scans instead of the stale ones. It is not a single model's self-reported probability — it's measured stability of the prediction over time.
Chance of rain compares three series: raw NWS gridpoint POP, the multi-source blend (POP_BLEND), and the same blend bent toward this radar signal inside the 60-minute horizon (POP_BLEND_RADAR). The main site now uses the radar-bent value.