Forecasting
Weather & Meteorology Fundamentals
Course 1 · Chapter 7 · Forecasting
A modern forecast is the end of a long chain: millions of measurements, a computer simulation of the atmosphere, many repeated runs to estimate the uncertainty, and a human forecaster who turns the output into something useful. This chapter follows that chain, and ends with how to read a forecast sensibly. It builds on the idea in Weather & Meteorology Fundamentals 1 that a forecast is possible, but limited. Dates and figures are given from general knowledge and flagged where they should be checked.
Observing the Atmosphere
A forecast can be no better than its starting point, so the first job is to measure the present state of the atmosphere as completely as possible.
| Source | What it measures |
|---|---|
| Surface stations | Temperature, pressure, humidity, wind and rainfall at the ground, including automatic stations in remote places |
| Weather balloons | Radiosondes carried aloft twice a day from sites around the world, reporting temperature, humidity, pressure and wind up through the atmosphere (check the schedule) |
| Ships and buoys | Conditions over the oceans, which cover most of the planet |
| Aircraft | Wind and temperature at cruising height, sent automatically by airliners (check) |
| Weather radar | Where rain, snow and hail are falling, and how it is moving |
| Satellites | Cloud, temperature and moisture profiles, sea temperature and winds over the oceans |
Numerical Weather Prediction
The atmosphere obeys the laws of physics: the movement of fluids, the conservation of energy and mass, and the behaviour of gases and water. Numerical weather prediction divides the atmosphere into a three-dimensional grid of boxes and steps the equations forward in time, box by box, from the measured starting state. The idea was set out by the Norwegian Vilhelm Bjerknes in 1904 and attempted by the British mathematician Lewis Fry Richardson, who calculated one forecast by hand during the First World War, with poor results; he imagined a huge hall of human calculators doing it in real time (check the names, the dates and the details).
The first successful computer forecasts came around 1950, and the method has improved steadily since as computers have grown faster and observations more plentiful (check the date). Roughly, forecasts today are as good at five days ahead as they once were at two or three, though this rule of thumb varies by place and season (check).
Chaos, Ensembles and Uncertainty
As Weather & Meteorology Fundamentals 1 explained, tiny errors in the starting state grow until the forecast loses all value, which limits useful detail to roughly one or two weeks (check the figure). To measure the uncertainty, forecasters run the model many times from slightly different starting states, a set called an ensemble. If the runs agree, confidence is high; if they scatter, the forecast is uncertain and the true outcome could be any of several.
- Probability of precipitation: a forecast of a 40 per cent chance of rain is usually a statement about the chance of rain at a given place in the period, and it does not describe how much of the area gets wet (check the definition used by your own forecaster).
- Confidence: some forecasters give a statement of how sure they are, based on the spread of the ensemble.
- Persistence and climatology: the simplest forecasts, that tomorrow will be like today, or like the average for the date, are the benchmark any real forecast has to beat.
The Human Forecaster
The models are not the end of the matter. A forecaster compares several models, spots where each is known to go wrong, interprets local effects such as hills and coasts, and decides on the wording and the warnings. National services such as the Met Office in the United Kingdom issue warnings graded by likelihood and impact (check the current system), and a good forecast communicates risk and not only the most likely outcome.
Reading a Forecast
- Look at the date range. Detail is trustworthy for a day or two, the broad pattern for several days, and little beyond that.
- Read probabilities as probabilities. A 20 per cent chance of rain means that it will often stay dry, but not always.
- Check the source. A national service and a model-based app may disagree; neither is certain.
- Watch the trend. If successive forecasts keep changing, the situation is uncertain; if they stay the same, confidence is higher.
- Take warnings seriously. A warning is about impact, so look at the area, the time and the level.
Exercises
Name four sources of observations used to start a forecast, and say what the main difference is between a geostationary and a polar-orbiting satellite.
Explain in your own words what numerical weather prediction does, and why a model with a coarse grid cannot represent a single thunderstorm.
What is an ensemble forecast, and what does it tell you if the runs all give much the same answer, compared with when they scatter widely?
Quick Reference
- Observations: surface stations, balloons, ships, aircraft, radar, satellites
- Geostationary satellite: fixed over one spot, continuous view; polar-orbiting: lower, scans strips
- Numerical prediction: physics equations stepped forward on a three-dimensional grid
- Parametrisation: simplified rules for features smaller than the grid
- Ensemble: many runs from slightly different starts; the spread shows the uncertainty
- Reading a forecast: trust near-term detail, treat probabilities as probabilities, watch the trend
Next chapter: the capstone, Weather, Climate and Change, which joins everything together and looks at how to read claims about the climate.