Weather forecasting

Science and technology that predicts future atmospheric states using observations, numerical models, and increasingly AI

Weather forecasting

Overview

Weather forecasting is the activity of analyzing physical changes in the state of the atmosphere with observational data and numerical models to inform in advance the weather of a specific region and time. At its core is numerical weather prediction (NWP), which solves equations of fluid dynamics and thermodynamics on supercomputers, and recently AI-based forecast models are rapidly being put into practical use. A forecast consists of the stages observation → data assimilation → model integration → post-processing (bias correction) → delivery.

Main content

Historical background

The origins of forecasting lie in the empirical rules and proverbs of ancient agrarian societies (e.g., "If swallows fly low, it will rain"). Modern forecasting began when Le Verrier in France produced weather charts, prompted by the sinking of a fleet in a Black Sea storm during the Crimean War in 1854. In 1922 Richardson proposed the concept of numerical weather prediction, and in 1950 the first numerical weather forecast calculation was performed on the ENIAC in the United States. After the launch of the TIROS-1 satellite in 1960, satellite observations were introduced, and in the 1990s ensemble forecasting became widespread.

Observation system

Automatic weather stations (AWS) on the ground, radiosondes (twice a day, 00 and 12 UTC), aircraft, ships, and buoys, Doppler weather radar, geostationary and polar-orbiting weather satellites, GNSS atmospheric precipitable water, lidar, and ceilometers monitor the state of the atmosphere in three dimensions. Observations are exchanged worldwide in real time, and the WMO's WIS/GTS system handles this.

Numerical weather prediction models

The momentum equation, mass conservation, thermodynamic equation, equation of state, and water vapor equation are numerically integrated on a grid. Cloud, radiation, turbulence, and land surface processes are approximated by parameterization, and higher grid resolution better simulates local phenomena. Representative models include ECMWF's IFS, the U.S. GFS, the U.K. UM, Korea's KIM, and the regional model WRF. Data assimilation (3D/4D-Var, ensemble Kalman filter) for accurately creating the initial field determines half of forecast performance.

Ensemble forecasting and probabilistic forecasting

Dozens of ensemble members are integrated to reflect uncertainty in initial conditions and the model, and their distribution is used to present precipitation probability and temperature ranges. Because of chaos theory (Lorenz's butterfly effect), the useful limit of deterministic forecasting is about two weeks, and ensembles complement that limit with probabilistic information.

Types of forecasts

Forecasts are divided into nowcasting (within 6 hours), short-range (3 days), medium-range (10 days), long-range (1 month), seasonal forecasts, and climate outlooks. Tailored forecasts by sector (impact-based forecasting) are also expanding in agriculture, aviation, shipping, energy, construction, and disaster safety.

Accuracy and limitations

Precipitation intensity, localized heavy rain, mountainous regions, and convective precipitation remain difficult to predict. Statistical post-processing (MOS, bias correction) and multimodel ensemble techniques are used to reduce model bias.

Delivery and use

Public meteorological agencies including the Korea Meteorological Administration and private weather service providers deliver forecasts via apps, broadcasts, disaster text messages, and APIs. Forecasts go beyond simple information provision to become a basis for decision-making in social and economic systems such as disaster prevention, logistics, agriculture, power supply, and insurance.

Latest trends

Since 2023, AI-based forecasting has been changing the game. DeepMind's GraphCast, Huawei's Pangu-Weather, FourCastNet, FuXi, GenCast, and others have shown similar or better accuracy than conventional numerical models at hundreds to thousands of times faster speeds. ECMWF integrated AIFS (AI Integrated Forecasting System) into its operational system in 2025, and the Korea Meteorological Administration is also pursuing pilot operation of AI forecast models and advancement of hourly nowcasting of precipitation. Meanwhile, as climate change increases the frequency of extreme weather such as heat waves, heavy rain, and typhoons, the importance of probabilistic forecasting, impact-based forecasting, and risk communication is growing. Data assimilation using nontraditional observations such as satellites, drones, and private data, and kilometer-scale ultra-high-resolution ensembles are also major trends.