To get the most out of weather forecasts, it is important for us to be able to obtain reliable localised forecasts for the places where we undertake our professional or personal activities.
An advanced system known as “data4web” has been producing such localised forecasting data for several years, and is responsible for the forecasts that are displayed on the MeteoSwiss app and website. The graphics and numeric values are generated by seamlessly combining all available numeric models along with the best post-processing data sources (see the PostprocVeri project) and the Nowcasting system.
In order to convert model grids with varying resolutions into a detailed localised forecast, data4web uses various sophisticated interpolation and correction techniques that enable local effects to be taken into account. Furthermore, data4web succinctly summarises the complex probabilistic information of ensemble models by calculating statistics such as quantiles.
The data4web data are updated every 10 minutes. Every day, data4web produces a total of around 1 billion data sets for 6,000 locations in Switzerland.
The first challenge faced by data4web is to produce weather data from the start of the day in question for up to 10 days into the future without interruption, for each of the points for which a forecast is available. In order to cover the entire range of parameters as accurately as possible, data4web combines the highest-performing forecasting systems available to it.
The sources used in the production chain are as follows:
For certain parameters where data from statistical post-processing are available and create added value, such as for wind and temperature, data4web uses these to improve the quality of the forecasts. Data4web thus draws on the following complementary data:
The role of data4web is therefore to combine these different sources optimally to ensure that they are coherent, and, in certain cases, to improve the quality of the data in a fully automated manner.
Even the high-resolution models do not allow for an exact reproduction of the subtleties of spatial structures established as a result of the complex topography of the Alps. In order to forecast local weather conditions as accurately and as detailed as possible, the results obtained from models with a range of resolutions (between 1 km and 18 km) need to be corrected and harmonised. The use of interpolation and downscaling techniques makes it possible to transform a parameter calculated at the resolution of the model grid to a more appropriate resolution for the user.
The aim of data4web is to provide local forecasts that reflect the weather conditions as perceived by a user in a precise location.
To do this, data4web deploys several approaches, for example by integrating very high-resolution (50 m) topographic indices, but also by spatially incorporating certain parameters such as cloud cover (since the user can only see the sky as it appears right above their head).
To meet users’ needs as closely as possible, more than 40 parameters are calculated at each site, and the most important of these are available on the MeteoSwiss app and website. Certain parameters consist of hourly values relating to simple weather variables, such as temperature, wind speed or precipitation, while others represent combined variables in the form of graphics, or even aggregated variables relating to specific time periods ranging from 1 to 24 hours.
Since the most recent forecasts are the most accurate, one of the principles of data4web is to utilise the most recent updates available from the forecasting models (with an update frequency every 10 minutes for INCA, every 3 hours for COSMO-1E, every 6 hours for COSMO-2E, and every 12 hours for the global model IFS-ENS). Each time a new update becomes available for a model, the data4web computation chain is triggered and the data updated.
Data4web thus produces around 1 billion data sets a day in total.
Recent developments have also made new probability-based parameters available, such as quantiles or threshold overrun probabilities. These parameters will enable the range of graphical representations on the website and app to be improved in the near future. Last but not least, data4web now integrates forecasts from a statistical machine-learning based point post-processing for all wind parameters, with their demonstrated increase in quality. The next important step will be to progressively extend such statistically improved forecasts to more weather variables.
While there will always be an element of uncertainty in the field of meteorological forecasting, one thing is certain: interest in local and personalised forecasts will continue to rise in the future. MeteoSwiss is therefore committed to responding to this growing need by improving the quality and reliability of its services as much as possible.