How Weather Forecasting Works: Radar, Satellites, Models, and Stations

The Weather Forecasting Process

Operational forecasting follows a repeating cycle. Each new observation can confirm the previous forecast, expose an error, or reveal a development that began between model runs.

1. Weather instruments collect observations from stations, balloons, aircraft, ships, buoys, radar systems, and satellites.
2. Quality-control systems examine the data for faulty sensors, impossible values, location errors, and transmission problems.
3. Data assimilation combines observations with a recent model forecast to create an estimated state of the atmosphere.
4. Numerical weather prediction models calculate future conditions on a three-dimensional grid.
5. Post-processing systems correct recurring model biases and translate model variables into local forecast products.
6. Meteorologists compare models and observations before issuing forecasts, watches, warnings, and probability information.
7. The cycle begins again as fresh observations arrive.

This process runs many times each day. Short-range forecasts may be updated more often when thunderstorms, tropical cyclones, freezing rain, fog, or other fast-changing hazards are present.

Weather Stations Measure Conditions Near the Surface

Surface weather stations provide direct measurements from the layer of air where people live and where many daily weather effects are felt. Automated stations operate at airports, research sites, farms, roads, ports, islands, and other locations.

Why Station Placement Matters

A thermometer beside sun-heated concrete will not describe the same environment as one above natural ground. Wind readings can also be distorted by buildings, trees, slopes, and nearby structures. Formal observing networks use placement and exposure standards so that measurements from different stations can be compared.

Even a properly installed station describes only its immediate surroundings. A valley may be colder than a nearby hillside during a calm night. Coastal temperatures can differ from inland readings, while a city may remain warmer than the countryside around it. Meteorologists therefore examine both individual reports and the wider pattern.

Manual Observations Still Add Detail

Automated sensors are efficient, but some weather features are easier for a trained observer to identify. Human reports can describe cloud forms, hail size, blowing snow, funnel clouds, storm damage, or changes in visibility that a station may not classify correctly. These reports are especially useful during severe weather.

Weather Balloons Sample the Upper Atmosphere

Surface conditions reveal only the lowest part of a weather system. Much of its motion and energy is found higher in the atmosphere, where jet streams, temperature layers, moisture bands, and upper-level pressure patterns shape future weather.

A weather balloon carries a small instrument package called a radiosonde. As the balloon rises, the radiosonde transmits measurements of pressure, temperature, humidity, and position. Changes in its GPS position allow wind speed and direction to be calculated at different heights.

The resulting vertical profile is called an atmospheric sounding. Forecasters use soundings to examine:

• temperature changes with altitude;
• dry and moist layers;
• wind direction and speed at different levels;
• freezing levels;
• temperature inversions;
• thunderstorm instability;
• wind shear;
• the likelihood of rain, snow, sleet, or freezing rain.

Radiosondes offer precise measurements along one rising path. Their main limitation is spacing: launch locations may be hundreds of kilometers apart, and routine launches occur at set times. Aircraft observations, satellite sounders, ground-based instruments, and model estimates help fill the gaps.

Radar Tracks Precipitation and Motion

Weather radar sends short pulses of radio energy into the atmosphere. When a pulse encounters raindrops, snowflakes, hailstones, insects, birds, or other targets, some energy is scattered back toward the antenna. The radar measures the returned signal and the time required for it to arrive.

The travel time shows how far away the target is. The strength and character of the return provide information about the target. By scanning at several elevation angles, radar builds a three-dimensional sample of precipitation around the site.

Reflectivity

Reflectivity describes the amount of transmitted energy returned to the radar. Stronger returns often indicate larger, more numerous, or more water-coated particles. Weather displays commonly express reflectivity in dBZ, a logarithmic unit.

Reflectivity is not a direct measurement of rainfall reaching the ground. Hail can create an intense return, while dry snow may produce a weaker signal despite reducing visibility. Precipitation can also evaporate below the radar beam before reaching the surface, a process called virga.

Doppler Velocity

Doppler radar measures changes in the phase of returned energy. These changes reveal whether targets are moving toward or away from the radar and how quickly they are moving along the radar beam. This is called radial velocity.

A single radar does not measure the complete wind direction at every sampled point. Motion across the beam is not measured as directly as motion along it. Forecasters interpret velocity patterns to find rotation, convergence, divergence, wind shifts, and other storm-scale motions.

Dual-Polarization Radar

Dual-polarization systems transmit and receive energy in horizontal and vertical orientations. Comparing the two returns helps estimate particle shape, size, consistency, and concentration.

This information can help distinguish rain, snow, melting precipitation, and hail. It may also reveal non-weather targets such as insects or debris lofted by a damaging circulation. The data improve precipitation estimates, but interpretation still depends on the storm environment and radar geometry.

Radar Limitations

• Earth curvature: the radar beam samples higher parts of the atmosphere as distance from the antenna grows.
• Terrain blockage: hills and mountains can partly or fully block the beam.
• Ground clutter: buildings, terrain, wind turbines, and other fixed objects may create echoes.
• Non-weather targets: birds, insects, dust, and ocean waves can appear on radar.
• Beam overshooting: distant shallow precipitation may remain below the beam.
• Attenuation: some radar wavelengths lose energy while passing through heavy precipitation.
• Bright band: melting snowflakes can produce an enhanced return that may exaggerate rainfall estimates.

Radar mainly shows what precipitation is doing now. A forecast of where a storm will move next requires motion analysis, atmospheric observations, model guidance, and meteorological judgment.

Satellites Observe Weather Across Large Areas

Weather satellites measure electromagnetic radiation reflected or emitted by Earth, clouds, oceans, and the atmosphere. They do not simply take ordinary photographs. Their instruments observe several wavelength bands, including bands that human eyes cannot see.

Geostationary Satellites

A geostationary satellite orbits above the equator at the same angular rate that Earth rotates. From the ground, it appears to remain over the same longitude. This position allows frequent observation of the same broad region.

Geostationary imagery is useful for monitoring developing thunderstorms, cloud motion, tropical cyclones, fog, volcanic ash, smoke, dust, and atmospheric moisture. Rapid scanning can show changes over intervals of only a few minutes, depending on the satellite and selected scan mode.

Polar-Orbiting Satellites

Polar-orbiting satellites travel much closer to Earth and pass near both poles while the planet rotates beneath them. Successive passes cover different strips of the surface, producing broad global coverage.

Their instruments can measure temperature and moisture through layers of the atmosphere, sea-surface temperature, snow and ice, clouds, vegetation, and other environmental conditions. These observations are valuable over oceans and remote land areas where surface stations are sparse.

What Satellite Channels Show

Cloud-top temperature can help estimate cloud height because the upper atmosphere is usually colder than the lower atmosphere. A very cold top may indicate a tall cloud, but temperature alone does not prove that severe weather is occurring. Forecasters compare satellite data with radar, lightning observations, surface reports, and the surrounding environment.

Other Observations Fill the Gaps

Modern forecasting uses far more than stations, balloons, radar, and satellites. The atmosphere crosses national borders and ocean basins, so observing systems must cover many environments.

Aircraft

Commercial and research aircraft can report wind and temperature during flight. Some systems also provide moisture or turbulence information. Aircraft data are especially useful near busy flight routes and during ascent and descent, when they create partial vertical profiles near airports.

Ships and Ocean Buoys

Ships and moored or drifting buoys measure conditions over the ocean. Reports may include pressure, wind, air temperature, sea-surface temperature, wave height, and wave period. These observations support marine forecasts and help models describe weather systems before they reach land.

Ground-Based Remote Sensors

Wind profilers, lightning networks, ceilometers, radiometers, and satellite-navigation receivers can provide additional atmospheric information. Each instrument observes a different property or layer. The value comes from combining measurements that complement one another.

Data Assimilation Creates the Starting Point

A model needs an estimate of atmospheric conditions at the start of its forecast. Observations alone cannot provide a complete three-dimensional map because they are unevenly spaced, collected at different times, and affected by measurement error.

Data assimilation blends recent observations with a short model forecast called the background or first guess. The system considers observation uncertainty, model behavior, location, time, and relationships among atmospheric variables. The product is an analysis: the model’s best estimate of the current atmosphere on its grid.

This step explains why an observation does not simply replace the value in one nearby grid cell. A pressure report, satellite radiance, or aircraft wind may influence a broader area and several related variables. Quality control is applied before and during assimilation so that a faulty instrument does not distort the analysis.

Small errors remain. Some regions have fewer observations, some features are too small to resolve, and every instrument has limits. These starting errors matter because the atmosphere is a chaotic system. Over time, slightly different initial states can produce noticeably different forecasts.

Numerical Weather Models Calculate Future Conditions

A numerical weather prediction model divides the atmosphere into a three-dimensional grid. It applies mathematical equations describing motion, pressure, temperature, moisture, radiation, and energy. The model advances forward through a series of time steps, calculating how the simulated atmosphere changes.

Global and Regional Models

Global models cover the whole planet. They are needed because weather systems interact across long distances and air entering a region may have crossed an ocean or continent.

Regional models cover a smaller area and can often use finer grid spacing. They may describe coastlines, terrain, thunderstorms, and local wind patterns in greater detail, but they still depend on information from a global model along their boundaries.

Well-known operational systems include the U.S. Global Forecast System, the European Centre’s Integrated Forecasting System, Germany’s ICON system, and the UK Met Office Unified Model. Their output differs because each system uses its own grid, data-assimilation methods, physical approximations, update schedule, and computing design.

Resolution and Grid Spacing

Grid spacing is the distance between calculation points in the model. Finer spacing can describe smaller features, but it requires much more computing work. Vertical resolution also matters because clouds, inversions, jet streams, and precipitation processes develop through many atmospheric layers.

A feature smaller than the effective model resolution may be smoothed, misplaced, or absent. Fine grid spacing does not guarantee a better forecast if the starting analysis is poor or if the model handles a physical process badly.

Parameterization

Many atmospheric processes occur at scales too small or too complex for a model to calculate directly at every grid point. Parameterization estimates their combined effects.

Depending on the model, parameterized processes may include:

• cloud formation and precipitation;
• turbulent mixing near the surface;
• heat and moisture exchange with soil, plants, snow, and water;
• solar and infrared radiation;
• small-scale convection;
• drag caused by terrain that the grid cannot describe directly.

Different treatments of these processes are one reason two models can start with similar conditions and still produce different temperatures, cloud cover, rainfall, or storm tracks.

Deterministic and Ensemble Forecasts

A deterministic forecast is one model simulation produced from one estimated starting state. It gives a single predicted sequence, such as a low-pressure center crossing a certain location at a certain time.

An ensemble forecast contains many related simulations. Members may begin with slightly different initial conditions, use varied model settings, or both. The spread among the members shows how sensitive the outcome is to uncertainty.

Ensembles support probability forecasts for rain, snow, temperature thresholds, wind speeds, tropical cyclone tracks, and other events. They also help meteorologists avoid placing too much weight on one visually convincing model map.

How Raw Model Output Becomes a Local Forecast

Model grid values are not automatically the final public forecast. A grid cell may cover mountains, valleys, water, countryside, and urban land. The predicted average may not match any one location inside that cell.

Statistical Post-Processing

Post-processing systems compare current model output with past model errors and local observations. They can adjust temperature, precipitation probability, wind, cloud cover, and other variables. Some systems combine several models, giving different weights according to recent or historical performance.

Local Knowledge

Meteorologists account for effects that a model may describe poorly. These include sea breezes, lake-effect snow, cold-air pooling in valleys, downslope warming, mountain precipitation, urban heat, fog-prone terrain, and coastal wind shifts.

They also monitor whether the current atmosphere is following the expected path. If a front moves faster than predicted or morning clouds persist longer than expected, the next several hours may require adjustment.

Forecast Communication

A finished forecast must state more than a number. Timing, location, confidence, and possible impact affect how useful it is. “Rain likely during the afternoon commute” gives a different decision signal from a daily rain icon with no timing.

Forecast precision should match forecast confidence. When model solutions disagree, a range or probability often communicates the evidence better than an exact value.

Nowcasting and Longer-Range Forecasting

Forecast methods change with the time range.

Nowcasting emphasizes what is happening and how observed features are moving. It is well suited to precipitation already visible on radar, though storm growth and decay can make simple motion-based projections fail.

Longer-range forecasting relies more heavily on ensembles and broad atmospheric patterns. A seasonal outlook does not predict the weather for each day. It describes the probability that an average condition over a period will fall above, near, or below a reference range.

Why Forecasts Change

A changing forecast does not always mean the earlier forecast was careless. New observations may show that the atmosphere developed differently from the earlier estimate. Later model runs then begin from a better-informed starting point.

Forecasts can change because:

• a storm develops between observation sites;
• the measured track or speed differs from the predicted track;
• new satellite or aircraft data improve the starting analysis;
• ensemble members shift toward another scenario;
• small temperature errors alter whether precipitation falls as rain, snow, or ice;
• cloud cover, soil moisture, snow cover, or sea temperature differs from the model estimate;
• local terrain effects become clearer as the event approaches.

Updates often narrow uncertainty as an event gets closer, but not every situation becomes easier. Thunderstorms can remain hard to place even a few hours ahead because their development depends on small boundaries, pockets of instability, and local wind interactions.

How to Read a Forecast More Accurately

Check the Valid Time

A forecast is tied to a period. “Tonight,” “by morning,” and “during the afternoon” refer to different windows. Model maps also carry valid times that may differ from the time the model run began.

Read Precipitation Probability Correctly

A probability of precipitation expresses the chance that measurable precipitation will occur at a specified location during the stated period. It does not directly state how long the rain will last, how hard it will fall, or what percentage of the day will be wet.

Separate Probability from Amount

A high chance of rain can produce only a small amount. A lower-probability thunderstorm may produce heavy rain where it forms. Probability and expected accumulation answer different questions.

Use Ranges for Temperature and Wind

Local temperature can vary with elevation, cloud cover, vegetation, urban surfaces, and distance from water. Wind forecasts also differ between sheltered locations, open ground, ridges, bridges, and coastlines. Gusts should be read separately from sustained wind.

Do Not Treat One Model as the Forecast

Individual model images are simulations, not official forecasts. A single snowfall total or storm track can change sharply between runs. Ensemble trends, recent observations, and forecasts prepared by meteorological services provide better context.

Watch for Hazard Messages

Routine forecast icons cannot carry every risk. Watches, warnings, advisories, and impact statements provide information about timing, affected areas, expected severity, and protective action. Their names and definitions vary by country.

Forecast Accuracy Has Physical Limits

Forecast skill generally decreases with lead time. Observation gaps, measurement error, unresolved processes, and model approximations all contribute. Atmospheric chaos then allows small starting differences to grow.

Large weather systems are often predictable farther ahead than the exact location of a thunderstorm or a narrow band of heavy snow. Temperature may be forecast well while cloud cover is missed. A storm track may be accurate even when rainfall totals are not. Forecast quality therefore depends on the variable, location, season, weather pattern, and time range being judged.

The best forecasts are updated estimates built from observation, calculation, comparison, and human interpretation. Stations describe the surface, balloons sample the air above, radar follows nearby precipitation, satellites provide wide-area coverage, and models connect those measurements through time. Each update brings the forecast into closer contact with the atmosphere as it actually develops.

References

• NOAA National Weather Service – Observation Equipment (Explains surface, upper-air, marine, aircraft, radar, and satellite observation systems used in forecasting.)
• NOAA National Weather Service – Weather Radar (Describes Doppler radar operation and its use in observing precipitation.)
• NOAA NESDIS – How Reliable Are Weather Forecasts? (Explains satellite observations, weather models, forecast lead time, and changing reliability.)