Forecast PointsWill it be rain or snow, and how much?

Type and amount are one question with two parts. A point forecast that answers them separately hides the fact that the wettest outcomes are usually the warmest ones. Here are nine ways to show both parts together, ordered from plain counting to model-based statistics.

Demo storm · synthetic 51-member ensembleWashington, D.C. · next 48 hAll charts live from one dataset
MILE1
BasicCounting + conditional ranges

The Point Card

Answers “what kind, and if so how much?” in one glance. Each rung is the chance of that outcome, and the line under it is what to expect if it happens.

MILE2
BasicFrequency format

Fifty-One Forecasts

Answers “how many forecasts say snow, and how much do they say?” Every tile is one ensemble member. People read counts out of a group more accurately than percentages.

Rain tiles show inches of rain. Changeover and snow tiles show inches of snow on the ground at the end. Hover or tap a tile for the member’s full totals.

MILE3
BasicMosaic plot

The Mosaic

Answers the whole joint distribution in one rectangle. Row height is the chance of each type; segment width is the chance of each amount given that type. Every block’s area is its joint probability.

Rain row in inches of rain; changeover and snow rows in inches of snow. Darker means more. Hover a block for its joint and conditional odds.

MILE4
BasicExceedance curves + joint table

Odds of At Least

Answers “what are the odds of at least 4 inches?” Drag a threshold. The table shows the joint odds of both rain and snow past a threshold, next to what you would wrongly get by multiplying them.

Both at once: P(rain ≥ row and snow ≥ column)

Big number: share of members with both. Small “vs” number: the product of the two separate odds, which is what you get if rain and snow amounts were unrelated. They are not.

MILE5
IntermediateType probability by hour + accumulation fan

When It Changes

Answers “when does rain turn to snow, and how fast does it pile up?” The top bars are the chance of each type every 3 hours. The fan below is the spread of the running total.

Running total of

Fan bands: middle 50% of members (dark) and middle 80% (light). The line is the median.

MILE6
AdvancedBivariate kernel density

The Joint Cloud

Answers “why can’t I just multiply the odds?” Each dot is a member placed by total water and the share that falls as snow. The shaded cloud is the smoothed joint density; the dashed curves are equal snowfall.

MILE7
AdvancedCompositional data

The Triangle

Answers “is this a changeover or an ice storm?” Each member’s water is split three ways: rain, mixed (sleet and freezing rain), snow. Corners are 100% of one type; dot size is total water.

MILE8
AdvancedHurdle mixture model

Ridges

Answers “what’s the full range of amounts for each type?” A two-part model: first the chance of each type (the bar), then a fitted distribution for the amount given that type (the ridge). The dry outcome is the spike at zero.

Ridges are gamma fits to the member amounts (ticks underneath). Shaded band is the 10th–90th percentile; the bold line is the median.

MILE9
AdvancedScenario clustering

Storylines

Answers “what are the ways this storm could go?” Members are grouped by machine into a few storylines, each with its odds, its totals and its hour-by-hour type.

Strip: most common type in each 3-hour block for that storyline, Sat 7 AM to Mon 7 AM. Lighter means fewer members agree.

Open the Storylines drill-down → hour-by-hour members, temperature, timing, drivers, cluster confidence and run-to-run odds.

EXIT⇢
Recommendation

Where Each One Lives

Following the network rule: if a chart needs a meteorologist to read it, it belongs on Forecast Eagle.

OptionReads inShows timingShows dependenceHome
1 · Point Card3 secondsNoPartlyShip first. Header of every point page and the widget.
2 · 51 Forecasts10 secondsNoYesTap-to-expand under the card; strong social graphic.
3 · Mosaic15 secondsNoYesPoints detail view. The full table, drawn to scale.
4 · At Least15 secondsNoYes (table)Points page for decisions; joint table for Weather Trader.
5 · When It Changes10 secondsYesNoShip second. Under the card; feeds Forecast City Today.
6 · Joint Cloud1 minuteNoBestForecast Eagle; explains why the joint matters.
7 · Triangle1 minuteNoYesForecast Eagle for ice-versus-snow events.
8 · Ridges30 secondsNoYesEagle; the fitted model behind it powers miles 1–4.
9 · Storylines15 secondsYesYesShip third. Plain-language scenarios people remember.

Suggested stack for a Forecast Points page: the Point Card on top, When It Changes beneath it, Storylines as the third panel, with Fifty-One Forecasts behind a “see every forecast” tap. The hurdle model from Ridges is the engine that should feed all of them once it is calibrated against observations.