Science Week 2016: Ensemble Prediction Systems - The Rising Capability of Numerical Weather Prediction
15 November 2016
Numerical weather prediction (or NWP), the process of predicting future weather patterns by using a computer model of the atmosphere, has advanced remarkably in the last 100 years. NWP can trace its origins to Louis Fry Richardson's (1881-1953) dreams of a “forecast factory”, where tens of thousands of human computers would calculate the future weather as quickly as it was changing. By the second half of the twentieth century, computer models based upon the primitive equations of atmospheric dynamics started to become operational. Current models are capable of detailing complex physical processes and features such as valleys and mountains.
Despite major advances, problems still remain. Firstly, the number of available weather observations may be quite large, but certain areas still lack the spatial resolution needed to accurately capture all meteorological phenomena. Ideally, one would need an observation every couple of metres over land and sea to have a complete description of the initial atmospheric state at any one time. Secondly, in order to model the atmosphere's complex physical processes, assumptions about these processes must made. Making these unavoidable assumptions can lead to errors in the model meaning uncertainty can creep into the forecast. Thirdly, and perhaps the most inescapable issue when it comes to forecasting, is the inherent chaotic nature of the atmosphere. Even the slightest difference in the initial state of the forecasting model can lead to completely different forecasts.
The main tool for tackling these problems is what is known as an Ensemble Prediction System (or an EPS). These systems aim to take the different sources of uncertainty into account by using multiple forecasts with slightly different initial conditions. Using a multitude of forecasts instead of just one single deterministic vision of the future atmosphere increases the chance of correctly forecasting an event and allows the issues outlined above to be taken into account.
Met Éireann currently use two different Ensemble Prediction Systems; the EPS of the European Centre for Medium-Range Weather Forecasts (or ECMWF, of which Ireland is a member state) and the Grand Limited Area Model Ensemble Prediction System (or GLAMEPS). The ECMWF's ensemble uses 51 different forecasts (or members), predicting future global atmospheric states out to 15 days and beyond. The ECMWF is constantly upgrading its ensemble and recently announced that the horizontal resolution will double to 18km allowing for an even better representation of global weather patterns. GLAMEPS was developed jointly by the HIRLAM and ALADIN consortia to focus on high-impact weather over Europe such as storms. Compared to the ECMWF EPS, GLAMEPS is more short-range focused, with a forecast length of 54 hours. This 52 member ensemble does however have a finer horizontal resolution running at approximately 8km.
Sample output from 28 members of the GLAMEPS ensemble showing the % chance of rainfall over 1 mm between 00 UTC Tuesday November 15th and 00 UTC Wednesday November 16th 2016.
For further information contact Alan Hally (alan dot hally AT met dot ie).