Our climate and environment are currently experiencing very strong anthropogenic pressures. Earth System models (digital twins) predict serious consequences for the future, but their long-term predictions are still burdened by considerable uncertainties and are often contradictory, especially regarding climate extremes. In addition, it is still almost impossible to predict whether or not the next season, the next year or the next decade will be especially anomalous. This inability has repercussion on civil protection, on expected production of renewable energy and hinders society to adapt to climate change.
Artificial Intelligence can be applied here to improve the capabilities of Earth System models, or to replace them entirely, to predict future frequencies of climate extremes events. For this goal, we need to analyse past extremes, investigate their drivers and understand their long-term trends. Artificial intelligence methods are based on the identification of past patterns of co-variations and require large data sets, ideally comprising many long periods and variables.
For this purpose, we use and create long meteorological reanalysis, spanning the last centuries, using Earth System models and as much available observations and indirect observations (proxy data) as possible. These data sets inform us that the last centuries witnessed not only very strong meteorological extremes (heat waves, floodings, very cold years, wind storms, forest fires, etc. ) and also oceanic extremes (ocean heat waves, coastal upwelling) that also had an economic impact. As today, those extremes appeared to the unprepared society as sudden, unexplained hazards. Our current research aims at providing society with useful predictions of the probabilities of those extremes and climate variations, so that society can adapt in time.