Language
Country/Area
No result found
The secret of global weather forecasting: How do you use the most powerful supercomputers to calculate future weather?
With the advancement of technology, the accuracy of weather forecasts has made breakthrough progress. Numerical Weather Prediction (NWP) uses mathematical models to describe the atmosphere and oceans to predict future weather based on current weather conditions. Although the earliest attempts date back to the 1920s, it was not until the advent of computer simulations in the 1950s that numerical weather forecasting was able to produce realistic results.
Several forecast models are run around the world, from global to regional, using current meteorological observations from radiosondes, weather satellites and other observing systems as input.
Meteorologists use this data to initialize models, then apply basic equations of atmospheric fluid dynamics and thermodynamics to predict weather over the next few days. Although current supercomputer performance continues to increase, the forecast accuracy of numerical meteorological models is still limited to a range of about six days. Factors that affect forecast accuracy include the density and quality of the observational data used as forecast input, as well as imperfections in the model itself.
Even with more powerful supercomputers, the forecasting skill of numerical forecasting models is limited to a range of about six days.
To improve forecast accuracy, meteorologists have developed post-processing techniques such as model output statistics (MOS) to improve error handling in numerical forecasts. These techniques help meteorologists mitigate the effects of chaotic behavior, extending forecast accuracy to many areas, especially the prediction of tropical cyclone tracks and air quality.
History of Numerical Weather Forecasting
The history of numerical weather prediction dates back to the 1920s, when meteorologist Lewis Fry Richardson attempted to create atmospheric forecasts using tedious hand calculations. It was not until 1950 that the widespread use of computers significantly reduced the calculation time for predictions. That year, the ENIAC computer was used for the first time to produce weather forecasts based on simplified equations, marking a pioneering period in numerical forecasting.
By 1954, Carl-Gustav Rossby's team at the Swedish Meteorological and Hydrological Institute had used the same model to successfully generate the first practical weather forecasts. By 1955, numerical weather forecasting in the United States began to operate under the Joint Numerical Weather Prediction Unit (JNWPU), marking the United States' active involvement in numerical weather forecasting.
In 1956, Norman Phillips developed the first successful climate model capable of realistically depicting the monthly and seasonal patterns of the troposphere.
As computer power has increased, the size of initial data sets has also increased, and new atmospheric models have been developed to take full advantage of these computing resources. These advances have enabled meteorologists to more accurately predict climate change and its impacts, although challenges remain. For example, the models still do not perform well for processes that occur in narrow areas, such as wildfires.
Initialization and calculation process
In numerical weather forecasting, initialization is the process of inputting observational data into the model to generate the initial state. The main inputs come from observations from national weather services, including radiosondes launched from weather balloons and weather satellites. This data is processed and converted into usable values for the model's mathematical algorithms, which are then used to predict future weather.
Observational data are collected in a variety of ways, including from weather balloons that rise into the stratosphere and from weather satellites.
Besides the initialization process, processing these observations requires significant computing power. Modern weather models rely on a series of mathematical equations to predict future weather conditions. Most of these equations are nonlinear partial differential equations and therefore cannot be solved exactly, and numerical methods are often used to obtain approximate solutions. Furthermore, different models use different solution methods, which may include finite difference methods or spectral methods.
Post-processing and integrated prediction
Even after processing, numerical predictions are never perfect, so model output statistics (MOS) have been developed to correct the predictions. These statistical models are adjusted based on the three-dimensional fields generated by numerical models, surface observations and climate conditions at specific locations. They can correct for lleol effects and model biases, making predictions more accurate.
Since the 1990s, ensemble forecasts have been widely used to quantify the uncertainty of forecasts, helping meteorologists assess forecast confidence and extend the validity period of forecasts.
This approach assesses uncertainty by analysing multiple predictions, either from different physical parameterisations of the same model or from different initial conditions. This not only improves the accuracy of weather forecasts, but also promotes more in-depth research on the impact of climate change.
While our predictive capabilities are improving with the advancement of technology, many challenges remain. In the future, can we find a better balance between predictive accuracy and a changing climate?
