Managing Storm Uncertainties for a Safe and Efficient Air Transport System: weather forecasting using artificial intelligence (PID2021-122323OB-C33)
Universidad Carlos III de Madrid (Spain)
About
Within the general objective of the coordinated project, the objective of this subproject is to leverage the improvements in NWP prediction along with machine learning techniques in order to improve thunderstorm prediction at both the short time horizon (12 hours), typical of nowcasting, and longer timescales compatible with the forecast horizons of NWPs, up to 36 hours. Using historical forecasts, we aim to train a machine learning model to predict images of storms captured by different sensors, including satellites, radar, meteorological stations, and lightning detectors. Our hypothesis is that there is a correlation between NWP parameters and convective weather phenomena. By exploiting advanced neural network architectures, we aim to identify correlations that may have been missed by classical meteorological analysis.
In particular, we will:
- Exploit the spatial characteristics of the data when predicting convective events by developing different convolutional neural network (CNN) architectures based on image segmentation (CNN-based encoder-decoder architectures, U-nets, PSPnet). The objective is to deliver predictions at the continental scale for both for the short-term (approx. 12 hours) and long-term (approx. 2436 hours) look-ahead horizons, to feed both short-term and long-term trajectory prediction and optimization blocks.
- Exploit the temporal characteristics of the data by developing long short-term memory (LSTM) architectures for the development of convection prediction models. The objective is to deliver predictions at the national scale for both for the short-term (approx. 12 hours) and long-term (approx. 2436 hours) look-ahead horizons, to feed both short-term and long-term trajectory prediction and optimization blocks.
- Increase the temporal and spatial resolution of the data to better capture the physical phenomenology of the thunderstorm development, for which we will rely on General Adversarial Networks (GANs) and Physics Informed Neural Networks (PINNs) architectures. The objective is to deliver predictions at the airport scale focusing only on the short-term (approx. 12 hours) look-ahead horizon to feed the short-term trajectory prediction and optimization blocks.
- Implement a showcase demo of a Service-Oriented Weather Product that the different aviation stakeholders (namely airlines, air navigation service providers, Eurocontrol as the network manager) could access.
RESEARCH TEAM
Manuel Soler Arnedo
masolera@ing.uc3m.es
WORK TEAM
Ricardo Vinuesa
rvinuesa@mech.kth.se
Alejandro Cervantes Rovira Alejandro.cervantesrovira@unir.net
Álvaro Moreno Soto amsoto@ing.uc3m.es
Javier García-Heras Carretero
gcarrete@ing.uc3m.es
Abolfazl Simorgh asimorgh@pa.uc3m.es
Aniel Jardines ajardine@ing.uc3m.es
Eduardo Andrés Enderiz eandres@ing.uc3m.es
EXPECTED SCIENTIFIC IMPACT
This multidisciplinary project brings together the following branches of knowledge: meteorology (ensemble weather forecasting), aeronautics (aircraft trajectory analysis), and mathematics (uncertainty, artificial intelligence, optimization). In this subproject, the focus is on the application of artificial intelligence to meteorological prediction with an application to the aviation domain. This ensures an interdisciplinary impact of the results.
The dissemination of scientific discoveries will be one of the inherent activities of StormATS. We will regularly attend International conferences and publish in top-ranked JCR journals to maximize the impact of our research. We will also foster the internationalization through participation in European networks (continuing a series of activities started within the last 7 years, note that StormATS is a follow-up of MetATS and OptMet projects). The different SESAR programs have funded knowledge networks where uncertainty and meteorology have been present as thematic challenges. SESAR 1 funded ComplexWorld, in which Damián RIVAS (US) was the scientific coordinator, and SESAR 2 funded Engage, in which Damián Rivas (US) and Manuel Soler (UC3M) participated as research leaders of the thematic challenge related to weather services and ATM. These networks had dedicated workshops, tracks in conferences, and funded different activities. It is expected that the oncoming SESAR 3 (Starting in Jan. 22) will also fund new knowledge networks.
Ultimately, the goal is to transfer the attained know-how to both society and the industry. Our strategy is to carry out innovation projects to deliver solutions to the real world with companies interested in the results of the project. This strategy goes in two directions:
- We expect to build up transnational consortiums, which include both academia and industry, for European calls within SESAR 3 or Horizon Europe, thus consolidating a scientific and technical network of world-recognized researchers in the field of weather uncertainty and aviation, as we have done previously through the Clean Sky project POTRA, and H2020 SESAR projects TBO-Met, PSA-Met, FMPMet, ISOBAR, START, FlyATM4E, and ALARM. Indded, the disruptions caused by severe weather in air traffic management is addressed in the cluster 4 `Digital, Industry and Space’ from the Horizon Europe Strategic plan 2021-2024. This cluster covers a full section on driving the Artificial Intelligence revolution with a special focus on the synergies between civil, defense and space industries. Similarly, Artificial Intelligence (together with climate change and drones) is one of the main pilars of SESAR 3.
- UC3M team is working towards the creation of a spin-off company, for which the weather service demo we intend to develop within subporject 3 will be a pilar. The team has gone already through the incubator program CRECE – https://www.uc3m.es/investigacionapoyopdi/CR3CE – and is working on a business plan together with UC3MS entrepreneurship service. This spin-off company will leverage the results of Aniel Jardines PhD thesis and the activities conducted under the umbrella of ISOBAR project, in which UC3M is participating together with CRIDA/ENAIRE, Aemet, MeteoFrance, and Eurocontrol, among others. All of them have shown interest in preliminary results obtained using artificial intelligence to predict weather, which we intend to extend in this project.
Storm-ATS 