Enhancing the performance and reliability of national weather warning systems by use of deep learning techniques applied on radar, satellite and ground meteorological observations

Project facts

Project promoter:
Babes-Bolyai University of Cluj-Napoca(RO)
Project Number:
RO-RESEARCH-0026
Status:
Completed
Final project cost:
€1,070,685
Donor Project Partners:
Norwegian Meteorological Institute(NO)
Other Project Partners
National Administration of Meteorology RA(RO)
Programme:

Description

The number and intensity of severe weather events is increasing, leading to loss of goods, property, and human lives. In 2017, a storm caused 5 deaths in the Romanian city of Timisoara, triggering the creation of a nation-wide emergency alerting system. Improved weather forecasting, especially for severe weather events expected less than 6 hours in the future, also known as nowcasting, is expected to help mitigate the outcome of such events.
Issuing accurate weather warnings is difficult for meteorologists, as they must consider changes in the speed and direction of wind at different heights, air temperature and pressure, cloud cover, as well as the effect of terrain features and climate - all changing from one hour to the next. Decisions must be made quickly and broadcast to people in the affected area in time to take necessary precautions.
Our project uses artificial intelligence to analyze present and past meteorological data gathered by satellites, weather radars and ground stations. Powerful computers will allow our programs to compare current conditions to past ones and inform meteorologists when and where dangerous weather events are expected faster than they could otherwise determine. This will allow meteorologists to give people more precise and earlier warnings.
The project team is comprised of researchers from the Babes-Bolyai University, who will contribute their expertise in artificial intelligence; meteorology experts from the Romanian National Meteorological Administration will provide the data and interpretation for Romania, while their colleagues at the Norwegian Meteorological Institute will do the same for Norway. Once ready, the jointly developed system will be extensively tested by on-duty meteorologists, who will evaluate its precision and speed in the nowcasting process. Population in both countries will benefit by having more time to prepare and, with a reduced risk of false alarms will have more confidence in meteorological alerts.

Summary of project results

The WeaMyL project contributed to enhancing the performance and accuracy of forecasting severe weather over the next 6 hours, also known as nowcasting. This was achieved by analyzing past weather data using machine learning algorithms. Severe weather events can lead to damage or loss of property, goods, and human lives. Since their number and intensity is increasing in many regions of the world, nowcasting is of broad and current interest for meteorologists as well as the public at large. Issuing accurate nowcasting warnings is challenging for meteorologists, as they must consider various factors including changes in wind speed and direction at different altitudes, air temperature, pressure and the effect of local terrain features and climate. Thus, by employing deep learning models for extracting accurate and meaningful insights from large amounts of weather data we obtained high precision in predicting the occurrence and the areas affected by severe meteorological phenomena. The project created a deep learning-driven platform for early and accurate forecast of severe phenomena that was integrated in the national nowcasting warning systems from Romania and Norway.  Thus, the WeaMyL project contributed to improving decision-making in the event of severe weather.

The main result of the project is the WeaMyL software platform. The platform comprises a forecasting component for early forecast of meteorological phenomena and the Annotated Atlas of Meteorological Observations, a curated database of meteorological events and data annotated to facilitate information retrieval and data analysis. WeaMyL’s Forecasting Platform generates short term predictions for meteorological products such as wind speed at different elevations, precipitation or humidity, thus facilitating early and accurate forecast of severe phenomena. The Annotated Atlas is useful for research purposes and for supporting operational meteorologists in short-term weather forecasting using statistical and comparative analyses of past meteorological conditions. The WeaMyL system, which has been deployed at the Romanian National Meteorological Administration and the Norwegian Meteorological Institute, benefits from enhanced deep learning models that are able to accurately predict meteorological product values up to 2 hours in the future and from an Annotated Atlas providing intelligent information retrieval. WeaMyL integrates deep learning methods for precise nowcasting and Big Data approaches for managing large amounts of meteorological data. The Forecasting Platform provides real-time and near real-time meteorological data for the embedded deep learning model, further used to generate short term predictions for meteorological products.

The main result of the project is WeaMyL, a new open-source software platform with a more general and complex target than existing solutions for weather nowcasting. The direct beneficiaries of WeaMyL are National Meteorological Institutes, with the public at large being the most important indirect beneficiary. The integration of the WeaMyL platform with the software systems of the National Meteorological Services from Romania (NMA) and Norway (MET) allows meteorologists to issue more precise and timely severe weather alerts. From a meteorological standpoint, both partner countries will benefit from WeaMyL, as the platform is intended to be used by meteorologists to analyze the weather for the upcoming 5-6 minutes up to 2 hours ahead. Considering the complex nature and spatio-temporal evolution of convective storms, the predictions provided by the deep learning-driven Forecasting Platform offer key information to both NMA and MET forecasters in the process of monitoring and nowcasting severe storms. The second component of WeaMyL, the Annotated Atlas of Meteorological Observations, is deployed at both NMA and MET and will be useful for research purposes and to support meteorologists and forecasters to compare the current weather situation with past weather events when considering issuing weather warnings and forecasts.

Summary of bilateral results

The WeaMyL project significantly contributed to one of the main objectives of the NO Grants programme on strengthening bilateral relations between Romania and Norway. WeaMyL involved a team of researchers with various backgrounds and an international partnership with the prestigious Norwegian Meteorological Institute. The main benefit of the partnership has been in the opportunity of the joint team members to work in an interdisciplinary consortium with expertise from three domains: intelligent systems, software development and engineering, and the operational and research meteorology. An additional benefit has been the collaboration within the WeaMyL consortium, which contributed to improving the know-how of the joint team members in all the project topics and produced high-quality scientific results. The bilateral partnerships established within the WeaMyL project also had a significant impact on partner institutions.Technical challenges were mainly related to the integration of the WeaMyL main components (Forecasting Platform and Annotated Atlas) in the existing hardware and software infrastructure of the Romanian National Meteorological Administration. Due to the large volume of meteorological data and different data formats used by the two meteorological partners, practical challenges arose regarding the manipulation and integration of radar and satellite data in the training steps of the machine learning models. These challenges were successfully addressed by the hard work of all Consortium members and through proper communication and collaboration.

Information on the projects funded by the EEA and Norway Grants is provided by the Programme and Fund Operators in the Beneficiary States, who are responsible for the completeness and accuracy of this information.