Daily Rain Markets

Live analytics for the Kalshi & Polymarket “Where will it rain today?” markets
About this dashboard
This dashboard covers the 24 stations in Kalshi's and Polymarket's “Where will it rain today?” markets — use the Exchange filter above the station list to show a single platform's stations. Each market resolves on its station's official NWS CLI report — trace counts as NO (measurable means > 0.005″). The market day runs midnight–midnight local standard time (1:00 AM – 12:59:59 AM on the local clock during DST; Phoenix reads midnight–midnight year-round).
🔒 You're viewing the free demo market. Pro unlocks all 24 rain markets with storm tracking. Demo: Tampa (KTPA). Subscribe to Pro Sign in

— 

Loading…

Radar & storm approach

light → extreme
Latest
How to read this map & futurecast
How to read this: colors show radar reflectivity — teal/blue is very light rain or drizzle (often trace — remember trace resolves NO), greens are light-to-steady rain (usually measurable), yellow/orange heavier showers, red/purple intense storms. Dashed rings sit at 25 / 50 / 100 miles from the station (orange bullseye). Press Play to loop the last hour of radar — the direction and distance echoes travel per loop tells you the storm motion. Click any echo on the map to get its distance, bearing, and estimated arrival time at the station across typical storm speeds. Auto-refreshes every 5 minutes. Futurecast overlays the HRRR model's simulated radar up to 12 hours ahead — treat it as guidance: it's good at organized systems (lines, fronts) and only directional for scattered pop-up storms, where the pattern is right but individual cells aren't. Inside the next hour, trust the real radar and cell tracks over the model.

The market day

Everything here runs on the official climate day: midnight-to-midnight in the station's local standard time, year-round — the exact window the NWS CLI report summarizes and daily rain markets settle on. During daylight-saving time that's 1:00am–1:00am on the local clock, by design.

Observed rain & resolution

The green bars and the Obs total come from the station's own gauge (METAR/one-minute data), accumulated within the market day. Markets resolve on the NWS CLI report: measurable rain means more than 0.005", and a trace counts as NO — a trace hour shows as a small triangle on the baseline, never a bar. The row locks to YES the moment measurable rain is observed, and settles on the final CLI (next-morning report supersedes the same-evening preliminary).

The EST % column

The estimated chance of measurable rain over the remaining hours of the market day, built from the official NWS 12-hour forecast probabilities — the "70% chance today / 40% tonight" headline numbers from the forecast office. Hourly rain chances are strongly correlated (one weather system spans many hours), and the NWS has already priced that correlation into each 12-hour period's number — so we combine only the two or three periods that overlap the remaining window. Twelve remaining hours of 30% hourly chances reads as roughly the period's own ~30%, not the ~99% that naive hour-by-hour multiplication would claim, and a day whose best period is 55% can't display 90%. Locks to 100% once measurable rain is observed and to 0% after a final NO report.

The hourly chart

  • Blue bars — NWS Rain Chance (PoP): the official NWS hourly probability of measurable precipitation, straight from the forecast office.
  • Orange line — Model consensus: for each hour, the share of forecast models (each model's latest run) whose own forecast puts measurable rain in that hour. Ten of fourteen models painting rain at 4pm = 71%. This is a vote count among independent simulations, not a calibrated probability — HRRR's hourly and extended cycles share one vote so the same physics never votes twice. For each hour, every model votes with its newest run that existed when the hour began — so already-elapsed hours keep the full fleet's opinion as it stood going into that hour, rather than losing every model whose latest run starts later in the day. Hover shows the exact vote (e.g. "8 of 13 models say rain"); as a guard, the line only draws where at least 3 models cover the hour.
  • Thin colored lines — individual models: each model's forecast rain amount per hour, in inches on the right axis. Models are physics simulations that forecast an amount of rain per grid cell, not a probability — a line above zero means that simulation drew rain in that hour. Like the consensus, each hour shows the model's freshest forecast as of that hour, so a line can stitch together successive runs across the day. The five most skillful models at this station are shown by default; click any legend entry to show or hide a model. Small amounts deserve skepticism: a model painting 0.02" over a ~10 km grid cell is sketching drizzle somewhere nearby, not promising rain at the gauge.
  • Green bars — Observed: gauge-measured rain per hour; the ground truth everything else is trying to predict.

Model skill rankings

The skill table under the chart grades each model's daily rain call at this station over the last 90 days, using what the model said the day before (24-hour lead) — the view that matters if you position ahead of the market day. A model "said rain" when its forecast day-total exceeded measurable; verification is against the CLI (trace = dry).

  • CSI — the overall skill score (hits ÷ (hits + misses + false alarms)). It ignores easy dry-day calls entirely, so it can't be inflated in dry climates. Higher is better; it's the ranking column.
  • Rain calls verified — of the days the model said rain, the share where it rained. The "can I trust a rain call?" number.
  • Rainy days caught — of the days it actually rained, the share the model called. The "will it warn me?" number. A model can look perfect on verified calls by rarely speaking — these two columns keep each other honest.

What "the model said rain" means — three different rules, disclosed per model: amount models (GFS, ECMWF, ICON…) forecast inches, not probabilities — their hourly amounts are summed over the market day and any measurable total (0.01"+) counts as a rain call. That's a deliberately low bar: one damp hour on one grid cell triggers it, which is why aggressive models rack up false alarms. Rate models (HRRR/RRFS, marked ᴿ) say rain when any hour shows measurable rain. NBM-PoP (marked ᴾ) is a probability — the National Blend guidance underlying the official NWS forecast — and is counted as saying rain only when some 6-hour period reaches 50% ("more likely than not"), a deliberately high bar that makes it a conservative caller. A day of persistent 20–30% chances is graded as NBM-PoP saying dry.

The honest consequence: because the bars differ, "rain calls verified" and "rainy days caught" describe each model on its own terms — an amount model's 55% verified and NBM-PoP's 85% verified are answers to differently-shaped questions. CSI is the metric designed to survive this: each model is scored on its own calls, hits against misses and false alarms, so the ranking stays fair even though the calling styles differ. Full grades, longer windows, lead-time comparisons and the all-station aggregate live on the Model Accuracy dashboard.

Live signals

Radar rain rate is a multi-radar (MRMS) estimate over the station itself, updated every ~2 minutes — it often shows rain starting minutes before the gauge accumulates its first 0.01", but it is an early signal, not truth. Nearby stations show surface rain at surrounding airports and personal stations. Markets resolve only on the station gauge/CLI, never on radar.

Market quotes

Kalshi (K) and Polymarket (P) prices are pulled roughly once a minute and shown for context — compare them against the EST % and the consensus to spot disagreement between the market and the forecast machinery.