Rain arrives when a forecast, a radar loop, or your own eyes show cooling air reaching saturation and growing droplets heavy enough to fall. For a specific time and place, the honest answer usually comes from a high-resolution forecast for the next 48 hours, radar for the next 90 minutes, and climate statistics for anything further out. Those three tools answer different questions, and mixing them up is the most common reason people feel misled by weather apps.

Below, the mechanisms behind rain, the numbers that govern it, the cases where forecasts fail, and the practical thresholds you can act on.

How rain actually forms

Rain begins with rising air. As a parcel of air climbs, pressure drops and it expands and cools at roughly 5.4°F per 1,000 feet while unsaturated, a rate called the dry adiabatic lapse rate. Cooler air holds less water vapor, so relative humidity climbs toward 100 percent, the point of saturation. That is why almost every rain-producing process on Earth involves lifting: fronts, mountains, surface heating, or convergence of winds.

Saturation alone does not make rain. Cloud droplets are about 20 micrometers across and fall at roughly 1 centimeter per second, so weak updrafts keep them aloft almost indefinitely. To reach the ground, droplets must grow to roughly 1 millimeter, about 50 times wider, at which point they fall at several meters per second. Two processes do this. In warm clouds, larger droplets collide and merge with smaller ones, a process called collision-coalescence. In cold clouds, ice crystals grow by vapor deposition and then aggregate into snowflakes that melt into raindrops on the way down. A single raindrop can contain the water of around a million cloud droplets.

The counterintuitive part: most clouds never rain. A fair-weather cumulus may hold hundreds of tons of water and still evaporate before anything reaches the surface, because the droplets stay too small and the updraft too weak. Rain is the exception, not the default.

Why forecasts beat folklore

Modern forecasts rest on numerical weather prediction. Supercomputers solve the equations of motion, thermodynamics, and moisture transport on a three-dimensional grid of the atmosphere, starting from millions of observations from satellites, radiosondes, aircraft, and surface stations. The governing equations are deterministic, but the atmosphere is chaotic: two starting states that differ by a rounding error diverge exponentially, which is the core reason forecast skill decays with lead time.

The practical consequence is a skill horizon. Day-one forecasts of whether it will rain are right roughly 80 to 90 percent of the time in mid-latitudes; by day five, precipitation timing and placement get noticeably vaguer; beyond about ten days, a specific rain forecast is closer to climatology than prediction. Ensemble forecasting handles this honestly: instead of one answer, models run dozens of slightly perturbed simulations, and the share that produce rain becomes your probability of precipitation. A 40 percent chance of rain does not mean it will rain for 40 percent of the day. It means that in this situation, rain occurred at your location in roughly 40 of 100 comparable historical or ensemble cases.

The time course of a rain event

Rain rarely starts and stops without structure. A typical frontal system moves through recognizable stages, and knowing them tells you when to expect the first drop and when it ends.

Stage What is happening What you observe
12–36 hours before Front approaches; pressure falls; high clouds thicken Cirrus gives way to a milky overcast
6–12 hours before Clouds lower and thicken; humidity rises Ceiling drops; wind backs or shifts
1–3 hours before Saturated air; drizzle or first light rain Radar echoes intensify; smell of rain
Onset Main rain band arrives Rain rate climbs from trace to steady
Peak Strongest lift and deepest moisture Heaviest rain; lowest visibility
1–3 hours after peak Front passes; wind shifts; cold advection Rain tapers; temperature drops
Post-event Drier air; clearing Broken clouds; cooler, clearer air

That rhythm holds for large-scale systems. Summer convective storms follow a different clock: they build in 30 to 60 minutes, dump rain for 15 to 45 minutes, and collapse, often leaving sun within the hour.

The 90-minute answer: radar and nowcasting

For "will it rain on me in the next hour," radar beats any model. Doppler radar measures reflectivity, which scales with the number and size of precipitation particles, and radial velocity, which reveals wind and rotation. A reflectivity of 20 dBZ is usually light drizzle or snow; 30 to 40 dBZ is moderate rain; above 50 dBZ is heavy rain or hail. Nowcasting algorithms track echo cells frame by frame and extrapolate their motion, which works well for 0 to 90 minutes but degrades as storms grow, split, or die. The professional caveat: radar extrapolation cannot predict initiation. A storm that does not exist yet cannot be tracked.

Where rain forecasts go wrong

Every forecast has known failure modes, and each has a mechanism.

Convective initiation: models struggle to predict exactly where and when afternoon thunderstorms first fire, because the trigger depends on small-scale boundaries and land-surface heating. Rain-shadow and terrain effects: mountains wring moisture from air, leaving sharp gradients. A forecast for a valley town can be badly wrong 10 miles upwind. Microphysics uncertainty: whether a cloud rains depends on ice nucleation and droplet spectra that models approximate crudely. Winter mixed-phase precipitation: the rain-snow line often sits within a few degrees, so a 1°F error can flip the forecast. Urban heat islands: cities run warmer and can shift rainfall downwind, a real but hard-to-resolve effect. Long lead times: beyond about seven days, precipitation forecasts carry low confidence even when temperature trends remain useful.

The most consequential misunderstanding is treating a daily percentage as a schedule. A "60 percent chance of rain" almost never means rain all day, and a "20 percent" day routinely produces a brief downpour somewhere in your county.

Checking a forecast without being fooled

To turn forecasts into decisions, compare the right tool to the horizon you care about and read the probability honestly.

Horizon Best tool Typical accuracy What it cannot tell you
0–90 minutes Radar nowcast High for existing cells Storm initiation
2–24 hours High-resolution model (HRRR-scale) High for timing within a few hours Exact street-level location
1–3 days Global and regional models, ensembles Moderate to high for events Fine timing
4–7 days Ensembles Event likely, timing loose Hour of onset
8–14 days Ensembles and climate outlooks Directional only Specific days
Weeks to months Seasonal climate outlooks Probabilistic shifts Any particular storm

Reconciling thresholds explicitly: treat a 30 percent chance as a reason to carry a jacket but not cancel plans; at 60 percent or higher, plan for rain; at 80 percent, expect it. Timing matters more than probability for commuting, so check radar 60 to 90 minutes before leaving. For outdoor events, use the three-hour window rule: if the forecast shows rain within three hours and the event is weather-sensitive, move it.

Practical steps that hold up

Several actions follow directly from the mechanisms above.

Check the hourly forecast, not the daily icon. The daily summary hides onset timing. Look at radar 60 to 90 minutes before departure; that window beats any model. Read probability of precipitation as a frequency, not a duration. For anything beyond 48 hours, plan around the range, not the point forecast. Use ensemble spread or a forecast confidence indicator when it is offered. Irrigate and schedule outdoor work by soil moisture and the 24-hour forecast, not the ten-day outlook.

What no forecast can deliver

The hard limit is chaos. Forecast skill is not a technology problem alone; it reflects an exponential loss of information about the atmosphere's initial state. Even with perfect observations, the amplification of tiny errors caps useful predictability at roughly two weeks for weather and hours to days for the fine timing of individual showers. Local microclimates add another layer: a single hillside, lake, or urban block can differ from the grid cell a model resolves. And any forecast is probabilistic by nature, so a correct 30 percent forecast still verifies most often as no rain at your location.

The practical upshot: trust the next 90 minutes from radar, the next two days from models, and everything beyond as a probability statement. That is not a weakness of the science; it is an honest description of a chaotic system.