Analysis methods – Artificial Intelligence
Today’s environmental science observation resources are based on increasingly complex measurement systems (multi-instrument aspects, in-situ measurements/active-passive remote sensing, measurement networks and campaigns, technological developments, etc.). As a result, the amount of associated data is increasing rapidly. Analysis methods for the Earth Observation Centre represent all the approaches and techniques used to process, analyze and understand data collected from different observation sources.
Among these methods, Artificial Intelligence (AI) represents a real breakthrough in the field of Earth sciences, particularly with regard to atmospheric and oceanic observations. AI makes it possible to analyze large quantities of data from a variety of sources, including satellites, weather radars, sensor networks and surface weather stations. AI algorithms (machine learning and deep learning) can be used to extract valuable information from this data, identifying patterns, trends and complex relationships.
The development of Machine Learning (ML) applications is based on three essential components:
– Data from which we wish to learn, validate and test an ML model
– ML methods and models, which must be adapted to the data and the objectives to reach. Model development also requires expertise in the data and the application domain.
– A hardware and software infrastructure for processing data and optimizing ML models.
At IPSL, ML methods are developed by researchers from the SAMA community in collaboration with data and application domain experts (https://www.ipsl.fr/en/research/research-themes/statistics-for-analysis-modelling-and-assimilation-sama/).
IT support, both hardware and software, is provided by IPSL’s ESPRI team, in particular with the ESPRI-IA group, which focuses on AI engineering issues (https://www.ipsl.fr/article/espri-ia-oeuvrer-pour-une-communaute/).
The aim of the SAMA (Statistics for Analysis, Modelling and Assimilation) theme is to make better joint use of observations and modelling techniques in order to better represent the climate, geophysical fluids and their constituents and improve their forecasting at IPSL. Since the end of the 2000s, the growth of machine learning has accelerated sharply, generalized by deep learning applications in language processing and computer vision, which have made it possible to carry out tasks that would have been unimaginable few years ago. A major challenge for SAMA is to adapt most of these new tools to environmental science issues.
ESPRI-IA is a working group dedicated to the community of IPSL researchers, engineers and students who use AI as a research tool (in particular Machine Learning and Deep Learning). Its mission consists in helping the community of AI users in the IPSL laboratories and to promote technical and methodological support for AI by sharing experience within this community.
In particular, ESPRI-IA manages and maintains the HAL computing cluster, which provides software libraries and GPUs for the development of Machine Learning models. ESPRI-IA also helps users to use the supercomputers offered by IDRIS for deep-learning when more computing resources are required.
