Why weather forecasts disagree, and how to use them

Different models and different update times

Modern forecasts come from numerical weather prediction. A model takes observations from weather stations, buoys, ships, aircraft, balloons and satellites, builds a picture of the atmosphere at a starting time, and uses the equations of physics to step that picture forward. Several national and international weather centres run their own models. Some cover the whole globe; others cover a limited region in finer detail. They differ in how they are built and in how they use the observations, so they produce somewhat different answers from the same weather.

Models are run on a fixed schedule, commonly every six or twelve hours for global models and more often for some regional ones, and the results take a few hours to become available. Two apps may therefore be showing runs made many hours apart even if they use the same model. Presentation adds further differences, such as hourly values in one source and three-hour averages in another.

Forecasters also run a model many times with slightly altered starting conditions, a set known as an ensemble. If the runs stay close together the forecast is robust. If they spread apart, small uncertainties are growing and the outcome is genuinely less predictable.

Grid cells and what they cannot see

A model divides the atmosphere into a three-dimensional grid. The horizontal spacing ranges from a kilometre or two in the finest regional models to ten kilometres or more (several miles) in global ones. Each cell holds a single value for wind, temperature and so on, representing an average across the whole cell.

Inside the model the land is simplified to match. Hills are smoothed, and the coastline follows the edges of the grid rather than the real shore. Anything smaller than a few grid cells, such as a headland, a narrow channel, a steep valley, a small island or a harbour, is absent or blurred. The figure an app shows for an exact point is worked out from the nearest cells, and for a spot on the coast those cells may be treated partly or wholly as land, where friction makes the modelled wind lighter than it is over the water.

Local effects no forecast shows well

Funnelling. Wind squeezed between islands, through a gap in hills or along a narrow channel or valley speeds up and tends to follow the line of the gap.

Headlands. Wind accelerates around the ends of headlands and capes, where currents are often strong too.

The lee of land. High ground gives shelter downwind, but the air there is often turbulent, with sudden gusts from changing directions.

Sea and land breezes. Daytime heating of the land draws in an onshore breeze that can strengthen, weaken or reverse the general wind near the coast.

Showers and thunderstorms. These bring abrupt gusts and wind shifts. A model may correctly predict showers in an area without placing any individual one.

These effects are consistent from one day to the next in a given wind direction, which is why local knowledge is so valuable. The article on onshore, offshore and cross-shore wind explains the coastal effects in more detail.

Forecast and observation are different things

An observation is a measurement made by an instrument at a known place and time. A forecast is a prediction. The distinction is easily lost, because the "current conditions" shown by many apps and websites are model output for the present hour rather than a measurement.

Real observations from a nearby weather station or buoy are a useful check. If the model said 10 knots for this hour and the station is recording 18, the forecast for the next few hours deserves suspicion. Bear in mind where the instrument is: a sheltered inland airfield and an exposed breakwater will report very different winds on the same day.

CatchWindow takes its forecast data from weather models through Open-Meteo and from public agencies, so the same limits apply to the figures on its location pages: they describe the general conditions around a point, not the precise wind in a particular cove.

Why time matters

Forecasts are most reliable for the next day or two and become less so with each day further ahead. Beyond several days, the general pattern, such as settled or unsettled, windy or light, is usually more trustworthy than any hourly figure. Timing is a frequent source of error: a front or wind shift can be forecast correctly and still arrive a few hours early or late.

How to use this

Compare two or three independent sources. If they agree, plan with more confidence. If they differ, plan for the less favourable one.

Prefer the official marine forecast for decisions. It is produced by the national weather service for your waters, comes with warnings, and often reflects a forecaster's judgement across several models.

Look at trends. Check whether each new update brings a change earlier or later, stronger or weaker, and whether the outlook is improving or deteriorating.

Compare the forecast for the present hour with a real observation nearby.

Look out of the window. The sky, the flags and the surface of the water are the final observation, and they outrank every app. If what you see is worse than what was forecast, act on what you see.