- Acronym: DeepAtmo
- Referencia: PID2023-151817OA-I00
- PAIDI Code: RNM119
- Funding Entity: Ministry of Science and Innovation
- Period: 01/09/2024 - 31/12/2027
- PI: Ana del Águila Pérez
- Budget: 115000 €
- Researchers: Inmaculada Alados Arboledas (Universidad de Málaga); Inmaculada Foyo Moreno (Universidad de Granada); Pablo Ortiz Amezcua (Universidad de Granada); Sol Fernández Carvelo (Universidad de Granada); Gregori de Arruda Moreira (Education and Technology of São Paulo, IFSP, and researcher at Center of Lasers and Applications, Brasil); Alexander Haefele (MeteoSwiss, Switzerland)
Abstract
The DeepAtmo project is designed to address urgent environmental challenges and aims to improve the understanding of atmospheric phenomena, specifically aerosols and clouds, which are critical in climate change research and environmental monitoring. This project will focus on making significant contributions to Atmospheric Sciences research in the field of lidar-based remote sensing techniques, applying Artificial Intelligence (AI) methodologies. Accurate Predominant Atmospheric Component (PAC) identification and Aerosol Particle Type (APT) classification of lidar/ceilometer measurements can be achieved using Deep Learning (DL) techniques. This approach is crucial for improving climate change predictions by addressing the problem of uncertainty in climate models, as highlighted in the Intergovernmental Panel on Climate Change (IPCC) reports. The project highlights the need for accurate identification of CAPs and APTs, as they profoundly influence weather patterns and climate dynamics. Traditional methods, while useful, face complexity in handling large volumes of data, with DL being a promising solution. The novelty of DeepAtmo lies in its innovative approach to both APT identification and APT classification from remotely sensed measurements. Instead of traditional signal processing and inversion, the project treats lidar-derived time series as images. DeepAtmo will significantly improve atmospheric composition analysis and aerosol characterization, creating a robust DL application framework for 2D lidar technique products. This innovative approach will employ segmentation models and Convolutional Neural Networks (CNNs), combined with learning optimizations such as transfer learning or self-supervised learning. This fusion of atmospheric research with AI will be scalable from simple to more complex instruments based on lidar techniques. Thus, the algorithm will be extensible to large lidar-based networks, such as E-PROFILE, which has more than 400 ceilometers across Europe. The products generated by the network are assimilated by the European Centre for Medium-Range Weather Forecasts (ECMWF) and will therefore reduce the uncertainties of numerical climate prediction models. DeepAtmo will be developed in the Atmospheric Physics Group (GFAT) of the University of Granada at the Andalusian Institute for Earth System Research (IISTA-CEAMA), which has extensive experience in lidar techniques. This project opens the opportunity to initiate a new line of research within the GFAT activities applying AI technology to the challenges of atmospheric science. In particular, the principal investigator, Dr. Ana del Aguila, has extensive international experience in the field of remote sensing and the application of AI techniques. The research team is complemented by Dr. Inmaculada Foyo, expert in aerosol-cloud interactions and radiation, and Dr. Inmaculada Alados, expert in atmospheric radiation using neural networks. In addition, the team is composed of international and national collaborators with extensive experience in remote sensing based on lidar techniques and the project manager of the European network of ceilometers E-PROFILE.