Where reading KMA forecasts covers how to interpret figures like rain probability and sky condition, this page covers how those figures get made in the first place. From observation to supercomputer calculations to a forecaster's final call, here's the scientific process behind a weather forecast.
A forecast starts with knowing exactly what state the atmosphere is in right now. A dense network of ground stations (ASOS/AWS) measures temperature, wind, and precipitation in real time; weather balloons (radiosondes) sample the upper atmosphere; and geostationary weather satellites watch cloud and moisture movement from space. Add a nationwide radar network and ocean buoys, and you get a dense observation net capturing the current state of both atmosphere and ocean.
Using these observations as "initial conditions," a numerical weather prediction (NWP) model divides the atmosphere into a fine grid and applies fluid-dynamics and thermodynamics equations to calculate the atmosphere's state hours or days ahead. Since 2020, Korea's KMA has run its own numerical model, KIM — the world's ninth independently developed model — on its supercomputers "Maru" and "Guru." This amount of computation is far beyond what a personal computer could handle, requiring a dedicated supercomputer at the National Center for Meteorological Supercomputer.
Even with the same model, initial observations inevitably carry small errors. "Ensemble forecasting" runs dozens of simulations at once, each with these errors varied slightly. The share of simulations that predict rain becomes the rain probability. In other words, a 70% rain probability isn't a forecaster's hunch — it's closer to a statistical result where 70% of dozens of simulations predicted rain.
Numerical model output (guidance) isn't published as-is. A forecaster compares results across multiple models and adds experience from what actually happened under similar pressure patterns in the past to finalize the forecast. This step is also where local features that models can miss — mountain terrain, coastal effects — get corrected for.
The atmosphere is chaotic — even a tiny error in the initial conditions gets amplified sharply over time, often called the "butterfly effect." Because of this, no matter how precise the observations are, uncertainty grows the further out the forecast reaches, so tomorrow's and the day after's short-term forecasts are far more accurate than one a week out. To narrow this gap, KMA is targeting a next-generation supercomputer by 2026 that raises resolution from 3km to about 1km.
sanCheck takes this forecast and observation data straight from KMA and combines it with feels-like temperature, UV, and fine dust to answer "is it okay to go out right now." Knowing how a forecast is made helps explain why the same 30% rain probability might mean bringing an umbrella one day and not the next.