THE METEOROLOGICAL RISK:
A RISK THAT CAN BE
FORECASTED
![]()
As a young research engineer with Météo-France at the beginning of the 1970's, the author of this article was working on how to forecast the drifts of hydrocarbon slicks. The Torrey Canyon affair was still on everyone's mind and people were seeking to develop tools to manage such crises. It was in this context that he received a request from an official in charge of maritime safety asking him to develop a model for which the input would be the dispersion coefficients of the type of petroleum involved, and the output would be not only the place where the slick would beach and the volume beached, but also the tonnage of detergent required and the length of floating barrages to be used! This only goes to show the heavy symbolic weight that models carry for certain decision-makers, who consider them to be some kind of a black box that eliminates doubt. Needless to say, neither the knowledge nor the means of calculation available at the time enabled the author to answer the request.
It is only recently that Météo-France has been able to avail itself of a powerful operational model for oil slick drifts. However, the recent Erika oil-spill has served to remind us that, even with such a tool, we are still very far being able to answer all the questions asked in 1970 and we may indeed wonder whether we will ever be able to answer them. Consequently, we may usefully reconsider the whole notion of modelisation. This is an essential precondition for managing risks that have a meteorological origin.
From this point of view, 1999 was without a doubt one of the "great" years of the 20 th century. France was not spared its share of disastrous weather. From the deadly avalanches in January and February to the exceptional storms at the end of December, with the catastrophic floods in the south of France in November followed by the snowfalls that hit the Rhône valley thrown in for good measure, we were not spared anything. In short, our compatriots, who are generally used to a very pleasant climate, were sharply made aware of how vulnerable our society is when confronted with phenomena whose scale is out of the ordinary. Hence the questions, even the polemics as to the responsibility of the various players, among which the national weather service, Météo-France, inasmuch as its forecasts constitute a key link in the risk management chain.
Let us state from the outset that this national service has no reason to be embarrassed concerning its performance during these various episodes, and indeed no one seriously challenged it, once the first moments of impatience had passed. Nevertheless, it can be useful to identify the weatherman's responsibilities and possibilities when such events take place.
Natural phenomenon, human disaster
Meteorological phenomena are natural phenomena, over which no one has any control.
However they at least have an advantage over earthquakes, which is that they can today be forecasted more often than not. From this point of view, the weather services play the role of a watchman, who warns and who pulls the alarm bells, not only for the authorities in charge of public safety measures, but also for the general population. Except for cases of blatant failure, to attack the weather service after a disaster is to return to the age-old tradition of ancient despots: kill the bearer of bad news as a scapegoat.
A natural phenomenon only becomes a disaster with respect to its direct or indirect consequences on human activities. A good example would be the famous El Niño, which is often presented as the great climatic troublemaker, the source of all the present meteorological evils on the surface of the globe. Yet if the first Spanish colonisers of Peru gave it the name of the infant Jesus, a positive sign, it was because it appeared to them beneficial, bringing the saving rains for Andean farming. Its demonisation only came later, when anchovy fishing and guano collection became the leading local industries. El Niño makes anchovies vanish and then the seabirds disappear, thus generating an economic disaster. Of course, much later, we discovered that these variations of the temperature of the equatorial Pacific Ocean are intimately related with the global climate, but that's another story.
The fact that a natural phenomenon is considered to be a disaster or not is therefore based on how much the area under consideration is occupied by man and his activities and on how the population and the authorities react to the forecasts and their consequences. From this point of view, the management of risks related to the vagaries of the weather is something that concerns a whole chain of responsibilities, the first link of which is the weather service.
The model at the heart of the process
Meteorological phenomena can be forecasted, but within certain limits of which we need to be aware. This is the notion of predictability that any risk manager absolutely needs to know, for this is what makes it possible to assess the uncertainties that remain once you think you have foreseen everything.
It has been now 150 years since the first meteorological services were organised in France, Great Britain, the Scandinavian countries and elsewhere.
Right from the outset, their concern was to manage the risks of an activity that was in those days entirely dependent on the vagaries of the wind, i.e. sailing, in particular the navy.
Gradually they were extended to the protection of people and property. Then, in the 20th century, aerial navigation took over, and with it came economic preoccupations. Starting from almost nothing, scientifically and technically speaking, meteorologists little by little developed a complex system of observation and forecasting which has made the atmosphere the most closely monitored natural environment. One hundred and fifty years of practice continually confronted with reality, supported by active scientific research and benefiting from advances in technology, has made it possible to attain an understanding of the mechanisms of atmospheric phenomena and to predict their evolution with a precision that is satisfactory for the users.
This is a remarkable demonstration of the effectiveness of synergy between operational applications and scientific research.
Today we can distinguish four main types of weather forecasting:
a) Short-term forecasting, which bears on the evolution of the phenomena between 12 and 72 hours. This is the oldest type of forecasting. It also goes by the name of general forecasting or synoptic forecasting. One may rightly consider it to be the historic core business of meteorologists. Done "by hand" until the 1970's, it is today almost entirely digitalised and is used as a support for important safety and economic decisions. We shall come back to this later.b) Medium-term forecasting is interested in the period comprised between 4 and 10 days. It has been truly operational since the 1970's, when sufficiently powerful computer tools first started to compute the necessary models. It was this need for powerful tools that lead the European meteorological services to found the European Centre for Medium-Term Meteorological Forecasts (ECMTMF), which has become a centre of excellence in its field and daily produces this kind of product. It is a tool to serve planning and early warning needs.
c) Immediate or very short-term forecasting is interested in everything that happens between 0 and 12 hours, with a preference for what happens in the first few hours. This is an activity that is currently developing in line with the improvements affecting information systems. This is because such forecasting needs high performance tools, not only for modelling the phenomena but also for hour-by-hour monitoring, when it's not minute-by-minute. The data coming from automatic stations, radars and satellites need to be given to the forecaster in real time via infographic systems with integrated model outputs. This is a tool for managing operations directly subjected to weather hazards. It is at the very core of the projects being developed following the storms of December 1999.
d) Long-term forecasting is interested in the other end of the spectrum, beyond 10 days and up to a month, even a season. Here it is a question of providing trends for the general characteristics of the weather for the period under consideration. This is also a new activity, which is still very much at the research stage. Its operational possibilities are recent, thanks particularly to advances in the knowledge about the coupling mechanisms between the ocean and the atmosphere that has been accumulated over the last two decades of the 20 th century. This is where we find El Niño and its distant influences, but that is another story. There are only a few meteorological services that are currently publishing such forecasts. Already they appear as important tools for planning.
Whatever the length of the forecast, the meteorologist's work rests on three fundamentals: observing, modelling, predicting. At the heart of the process we therefore have the model, i.e. a set of physical hypotheses representing, to the best of our knowledge, the phenomenon to be predicted. The Norwegian front-theory, defined at the beginning of the 20 th century, the archetype of the physical model that can be processed by hand, was the tool for short-term forecasting until the advent of the computer. Its success was such that the notions of fronts and disturbances have passed into everyday language and are still used to interpret the results produced by modern digital models.
Today, the hypotheses of the models are translated into equations, which are then converted into IT programmes via calculation algorithms.
Contrary to what common language might lead one to think, we are not dealing here with mathematical models. Only physics is involved, through fluid mechanics and thermodynamics. This type of model is the tool par excellence for short and medium-term forecasts. They are now very complex tools that give a good representation of the atmosphere under normal conditions, and are only caught out on a few occasions, as at the end of 1999! Now that the dynamic is well mastered, the current operational models are distinguished one from another essentially by the way they treat physical phenomena (radiation, water cycle, convection, turbulence, etc.) And this is what matters.
At the time of the storms of December 1999, Météo-France's Arpège model was one of the few to correctly simulate the phenomenon. The physical pattern of this model, that had been modified and improved following a forecast failure in December 1998, is undoubtedly the reason for this good performance. Arpège nevertheless underestimated the scale of the deepening of the depressions and the force of the winds, which means that we are not yet fully familiar with the mechanism that allows storms to intensify over the earth. In short, there is still need to work at improving the modelisation of the atmosphere.
From this point of view exceptional phenomena prove to be very stimulating, but the sources of error are not only to be found in the models.
Observation, analysis and predictability
A model is nothing if it is not fed relevant data at the initial moment of the forecast. The second tool of the meteorologist is therefore the overall, complex and costly observation system, associating in situ measurements with the latest advances in tele-detection from space. Though the adequacy of the model to the phenomenon being studied is fundamental for the quality of the forecasts, the adequacy of the observation system not only to the phenomenon but also to the model is just as important, something that tends to be forgotten by some people. Actually, the result of a forecast depends closely on the initial condition fed into the model. The slightest error is amplified as time passes and sometimes the system can develop aberrant configurations. The atmosphere, being a complex environment, is subject to deterministic chaos - a notion developed by E. N. Lorenz - which limits forecasting capacity, because it will certainly never be possible to have an exact representation of the initial state. This also defines predictability as being the limit beyond which it will never be possible to forecast the weather using deterministic modelisation techniques.
Meteorological services have established a composite observation system, the Global Weather Watch (GWW), which includes earth, sea and air-based measuring stations, as well as observation satellites. From these are taken measurements of physical parameters (temperature, humidity, wind, etc.) but also information on so-called perceived weather (visibility, cloud cover, precipitation, etc.). This system, developed to serve synoptic forecasting is now used to feed the models. Whereas it provides good coverage for the developed countries of the northern hemisphere, it is not very dense over the oceans and the southern hemisphere. Properly maintaining this system and increasing its performance is at the heart of the concerns of the Worldwide Meteorological Organisation (WMO) and its member organisations. This is how it has been possible to identify one of the sources of error in the forecasts of the December 1999 storms as being a lack of data at the right moment and the right place in the vicinity of Newfoundland.
Between the observation and the model there is a fundamental operation that takes place: analysis and assimilation. It consists of making the information drawn from the observations comply with the characteristics of the model being used. This analysis was for a long time a manual operation: the forecaster drew his isobar charts, and placed his fronts by applying the Norwegian theory. Today it is automatic, based on the amendment of a previous forecast by the data effectively observed and using a specific piece of software adapted to the forecast model. This operation results in the rejection of data that do not conform to the model. Such data is indeed more often than not erroneous but in certain cases, it is because they were outside the limits envisaged by the analysis software! This is what happened with the storms of December 1999: the relevant data was eliminated because it was surprising for the model and the observers! It was something "never seen before"!!
In any case, it will never be possible to give an exact representation of the initial state of a forecast, and errors amplify with the time factor, thanks to deterministic chaos. Now if this sets a limit to the predictability of weather, estimated to be some fifteen days, we nevertheless benefit from these evolutions for quantifying the validity of medium-term forecasts, by set forecasting. This is based on the comparison of a sample of forecasts, generally around fifty, done from the disturbances of the analysis used for operational forecasting. If this set of forecasts is consistent, the confidence index is high, if there is a dispersion of the results, the confidence index is low. This is the direction of information provided by Météo-France in its 4 to 7 day forecasts. In Europe it is the ECMTMF that provides its members with the necessary information to carry out such forecasts.
Forecasting, informing and raising awareness
The techniques described above have allowed a constant improvement in forecast scores. The claimed quality of 5-day forecasts is now equal to that of 1-day forecasts twenty years ago. But this is based on criteria, which however scientifically relevant they may be, are nonetheless esoteric for the user of the forecasts.
Try to explain to someone that the quadratic discrepancy of the geopotential error at 500 hPa has never been smaller, when he's just been through a misty day after the forecast spoke of a clear sky The fact is that between these results and the information provided there remains an important stage: adapting the information provided by the model to the weather perceived by the user.
The term "perceived weather" covers the most important fact for the one who is at the end of the chain: visibility for the aircraft pilot or the car driver, wind gusts for the crane operator, good weather for the tourist, rain or hail for the farmer, and so on. If certain elements of information such as highest or lowest temperatures can be automated by what is called statistical adaptation of model output, others such as cloud cover, precipitations, wind gusts, etc. still depend on the forecaster's expertise and the technical means that allow him to comprehend the exact state of the atmosphere. This forecasting of perceived weather, though important for short and medium term forecasting, is at the heart of the problem of immediate or very short-term forecasting. Hence its difficulty, for the slightest error of estimation will in this case be immediately sanctioned.
This is where the weather forecast user comes in. He needs to be able to evaluate what is the perceived weather for his activities, to know its exact consequences and then to formulate his needs. In short, the user needs to be aware of the meteorological risk that he faces. Certain communities that are directly subjected to the vagaries of the weather - sailors, flyers, farmers, civil engineering contractors, etc. - have since a long time acquired this meteorological consciousness and are therefore able to formulate precise requests for managing their activities.
On the other hand, in the area of civil security, in the widest sense of the term, where it is not only the risk managers who are concerned but also and above all the general public, things are somewhat more difficult. The success of prevention requires the raising of the awareness of the users. Even though the weather forecaster should contribute to such awareness raising, his role is first and foremost to inform on the basis of precise questions enabling an improvement in the readability of this information. This is one of the chief lessons drawn from the meteorological events of 1999 by Météo-France and its supervising authorities.
A sailor knows what to expect from a force 4 or 10 Beaufort. The West Indian inhabitant knows the difference between a tropical depression and a hurricane and knows what to do as the successive stages of the hurricane alert are announced. The same goes for an American in the Middle West during the tornado season. But who in our usually temperate climes knows the difference between 100 mm and 400 mm precipitations, or between a 110 kph wind and a 150 kph gust?
On the face of it, no one, apart from a few specialists who are directly concerned.
Hence the idea of introducing a risk scale, associating the meteorological phenomenon with its potential consequences. The idea was put forward after the floods that hit the Aude département in Southwest France, along the lines of what has been done for avalanche risks.
Work is under way in conjunction with the local Direction Départementale de la Sécurité Civile (DDSC - Civil Security Office at the level of the département). The next step will be to associate this risk scale with the measures to be taken by the authorities and the information to be made available to the public. Improving the management of meteorological risk must also include this type of approach.
By way of conclusion
The meteorological risk is predictable. But the limits of predictability teach us that there will always be phenomena that slip through the net of the meteorological services, even though we are actively working to reduce their number by advances in observation, modelisation and forecasting. To continually bear this in mind can only help to reinforce the confidence level between the meteorological services, the authorities responsible for managing the risk and the general public. Last year's events showed the importance of developing a meteorological risk consciousness among the authorities and the citizens of a country known for the clemency of its climate. A consciousness of prevention, when planning for the development of urban and rural environment and the activities that go with it, and a consciousness of protection when the phenomenon hits, are very closely linked. One hopes that the exceptional events of 1999 will make it possible to give our country this meteorological risk awareness and encourage the authorities to organise accordingly.
François Gérard,
Head of the Oceanography Department
Direction Générale - Météo France
.
Translation by Andrew WILES
© Institut Européen de Cindyniques -Lettre n° - 31 - July 2000