Chapter 10. Reading a weather forecast

Three lines are enough to give tomorrow's weather, and practically nobody knows what it took to write them. This chapter goes back up the chain the other way: from the three lines to the models, from the models to the observations, and from the observations to the previous chapter, which explained how we measure.

There is a practical reason for this detour. A reader who knows how a forecast is made also knows what it can and cannot say, and so uses it better and expects fewer miracles of it. A forecast is not an announcement: it is the result of a calculation, made from an imperfect picture of the atmosphere, and corrected by people whose trade it is.

The chapter ends with the warning map, which is not a forecast but a signalling of danger. The difference between the two is important, and it is the last thing this volume teaches about everyday weather.

Météo-France writes it in a sentence that could serve as the title of this chapter: "observation is the basis of any weather forecast". To forecast the weather, one must first know precisely the state of the atmosphere at the moment the calculation begins.

That knowledge comes from several sources which complete one another. From space, Météo-France states that dozens of satellites observe the Earth, its oceans, its ice, its land and its atmosphere. On the ground, radars follow precipitation, as chapter 4 described. Everywhere, automatic stations continuously record the quantities of chapter 9, under the standard conditions of chapter 9. The chain is unbroken: without a white screen at 1.5 metres, no good observation; without good observation, no good forecast.

Out of all those observations, the calculation must draw a single thing: a picture of the atmosphere at the instant it begins. That picture has a name, the initial state, and it is the starting point of everything that follows.

Remember what that picture is, and above all what it is not. It is a reconstruction, made from the observations available, and it is necessarily approximate: the atmosphere is not observed at every point of the globe, it is observed where instruments exist. We therefore never know the atmosphere perfectly, we approach it. The rest of the chapter shows what that approximation costs.

A weather model is a mathematical representation of the atmosphere used to simulate how it will develop. It is given the state of the atmosphere at one instant, and it works out, step by step, the likely state of the following hours.

Météo-France has developed two complementary numerical weather prediction models, named ARPEGE and AROME. Complementary means that they do not do the same job: one covers a large territory with a wider mesh, the other covers a more limited territory with a finer mesh, which lets it represent small-scale phenomena.

These calculations call for machines out of the ordinary. Météo-France stated, at the date on which this volume checked its sources, that its two supercomputers, named Belenos and Taranis, are about 50,000 times more powerful than a personal computer. A figure of that kind ages: the number of machines, their names and their power change as they are replaced, and it is the date of verification of the volume, in its colophon, that says when it speaks from. That power serves three purposes: handling ever more numerous observations, calculating how the meteorological parameters develop, and carrying out the research needed to improve the models.

The last link is not a machine. Human expertise remains indispensable in order to analyse complex results and turn them into concrete information. A model produces fields of numbers; a forecast says whether a coat is needed. Between the two, there is a trade.

Here is the explanation that is most often missing, and it follows directly from what has gone before. The initial state given to the model is a reconstruction, so it is not reality. It carries very small differences, which cannot be avoided: between two stations, an estimate has to be made; over the oceans, measurements are rarer; some quantities are measured less well than others.

Those very small differences do not stay small. At each step of the calculation, they spread and grow. After a few hours they are still negligible: that is why a forecast at twenty-four hours is excellent. After several days, they have grown to the point of changing the result: that is why a forecast at seven days gives a tendency and not a detail.

Remember the practical consequence, which is the only one that counts for a reader. A distant forecast remains useful, but for other questions: it correctly says whether it will be warmer or colder than the normal, drier or wetter, whereas it cannot say whether it will rain on Tuesday between two and four in the afternoon. To read a forecast is first of all to know which question it can carry.

And that also explains why a forecast changes from one day to the next. Each new calculation starts from more recent observations, so from a better initial state. A forecast that corrects itself is not a forecast that was wrong: it is a forecast that has received better data.

A separate scheme exists alongside forecasts, and it does not answer the same question. The one described here is the FRENCH scheme, and it is the example this volume takes: every country has its own, with its own colours, its own thresholds and its own divisions. What carries over everywhere is the IDEA, namely that a public service signals a danger separately from its forecast; the names and the colours, for their part, are to be read from the service of one's own country. Météo-France gives the purpose of its own: "the Météo-France Vigilance scheme aims to draw everyone's attention to the potential dangers of a weather situation".

The difference is clear. A forecast says what is going to happen. A vigilance says how dangerous the expected situation may be. There can therefore be a forecast of rain with no vigilance, and a vigilance without the event being rare: what is signalled is the danger, not the unusual.

The scheme follows several events, namely heatwave, thunderstorms, rain and flooding, waves and coastal flooding, wind, storms, avalanches, snow and ice, and severe cold. Each department is coloured according to the situation, and the map is updated at least twice a day, at six o'clock and at four in the afternoon. It covers the current day and the following one.

The division by department is not an administrative detail: it was chosen to make it easy for people to find themselves on the map. Everyone knows which department they are in, which is not true of a climate zone or a catchment.