When we talk about weather forecasts, many questions come to mind. Here are some answers from our Data & Calculations department.
How are weather forecasts developed?
Weather forecasts are the result of simulations carried out using digital models on powerful supercomputers, which can perform up to several billion billion operations per second (1018 flops). These are then raw forecasts, or "Numerical Weather Forecasts" (NWP in French) and NWP "Numerical Weather Production" in English. What we call "numerical models", or "weather models" in everyday language, are in fact complex mathematical equations describing the physicochemical phenomena operating on the surface of the Earth, in the atmosphere or the oceans.
However, to predict future weather, it is essential to know the current weather as well as possible. This is why digital models are initialized from an analysis representing the known and current state of the atmosphere, as accurately as possible. To do this, measuring devices (weather radars, balloons, planes, boats, weather stations and satellites) continuously observe the parameters and collect the information used by the models. This estimate of the future atmospheric state is deduced from a process called data assimilation consisting of correcting a previous forecast (draft) by current observations and this through sequential methods or variational analysis.
What are the limits of these forecasts?
To date, and in the world, only a few countries (United Kingdom, Germany, United States, Canada, France, Russia, Japan, Australia...), consortia (Aladin, Hirlam, Harmonie...) and more recently companies, have their own digital model. The Global Forecast System (GFS) produced by the American National Center for Environmental Prediction (NCEP), the Integrated Forecast System (IFS) produced by the European Center for Medium-Range Weather Forecasting (ECMWF), ARPEGE or AROME produced by Météo-France, ICON calculated by the German DWD, are models that can predict the state of the atmosphere up to 15 days, with resolutions ranging from 1 to 25 km depending on the model.
However, despite the improvement in knowledge and the increased computing power of computers in recent years, no single model is currently able to provide perfect forecasts, even if short- and long-term forecasts have made significant progress over the last 20 years.
Indeed, weather observations remain insufficient and representations of the physical processes of the atmosphere or oceans are imperfect. Since the atmosphere itself behaves chaotically, errors, even tiny ones, made at the start of the numerical simulation increase exponentially over time, and can result in very different forecasts, sometimes even after a few hours.
This is why even the most sophisticated numerical weather forecasts become significantly more efficient when expertise is applied to them, before being distributed to end users.
How are METEO CONSULT forecasts developed?
The use of numerical models makes it possible to meet the growing needs of users all over the world. However, the chaotic nature of the atmosphere is an intrinsic characteristic that eludes us. Each scenario provided by the models therefore represents a possible but always hypothetical state of the atmosphere.
Based on this observation and its conviction based on 35 years of experience, METEO CONSULT favors a probabilistic approach that aims no longer to focus on a single scenario (deterministic approach), but on several scenarios based on different so-called "set" models. This set of scenarios makes it possible to better represent the different possible states of the atmosphere.
We have thus built a computer architecture capable of continuously processing large volumes of data from numerous digital models and observations around the world. Among all the available scenarios, a statistical algorithm, developed by our meteorological experts, builds a METEO CONSULT multi-model scenario. This single scenario is representative of the group containing the most simulations with similar properties (clustering) thus making it possible to isolate the most probable scenario. We associate it with a reliability measuring the degree of confidence that can be given to it, compared to other so-called minority scenarios (less probable, but still possible).
The raw data of this multi-model scenario is then transformed into expert data thanks to the implementation of:
This expertise, unique to METEO CONSULT, allows us to offer our public forecasts with high added value, with optimized reliability, both for so-called sensitive weather (wind, gusts, clouds, rain, humidity, sunshine, etc.) and for other more technical parameters intended for professionals.
How does this expertise translate in concrete terms?
METEO CONSULT has equipped itself with its own digital weather forecasting system, developed by our data engineers and meteorological engineers.
This complete system (atmospheric, terrestrial or oceanic) offers a wide range of forecasts at all time and space scales, aiming to best satisfy our users (from immediate forecasting to seasonal forecasts, with finer resolution in sensitive areas).
Our weather forecasting system is broken down as follows:
How are the different scenarios of other models used?
We offer our users a model comparator with, on the one hand, the forecasts of the METEO CONSULT model and, on the other hand, the forecasts from three widely used and freely accessible global models, ARPEGE*, GFS**, IFS*** and regional models such as AROME* and LaMMA**** which benefit from finer resolutions allowing to better integrate the site effects which are particularly marked near the coasts. We work a lot on the freshness of the "runs" (updates) to provide the very latest information. Other models will later enrich this comparator.
This educational tool allows us to measure both the added value of our expert forecasts compared to those of other models, but also to better understand the potential divergences of scenarios when the weather situation is less reliable.
But the quality of the service offered to our users also depends on our experts who are still very human and have now 35 years of experience in the field. Indeed, they monitor, analyze and describe on our media, or during briefings, the weather situations on a daily basis and provide our users with reliability and security.
*models produced by Météo-France (France) **Global Forecast System, model produced by the NCEP (United States) *** Integrated Forecast System, model produced by the CEP (European Forecasting Center) **** scenario from the WRF atmospheric model, configured by the LaMMA Consortium
