Extreme rainfall, temperature and wind - weather extremes as a result of climate change

Climate Extreme Events and their impacts are getting a lot of attention lately, because their occurrence, severity and spatial coverage are increasing and will likely increase further toward mid and end of century. Many countries are experimenting significant impact of those climate extremes, in what many call a Climate Crisis.

It becomes more and more important to better assess the change of characteristics of climate extremes., according to users and society needs. However, it is not straightforward to correctly assess and quantify uncertainties as well as to find and characterize climate extremes in all available and relevant climate simulations. This is mainly due to the very large number of climate simulations along with very significant associated data volumes. It is unfortunate to limit the number of climate simulations used in a climate change assessment study, only because of those technical and time constraints, as we should use all available information.

The challenge is to provide very efficient tools for end users assessing climate change impacts, in order for them to explore what-if scenarios.  

A novel approach and methodology is being developed within the interTwin context to detect and characterize the changes in climate extreme events using Artificial Intelligence. This will be provided to users as a Digital Twin Application. The underlying method is based on machine learning (ML) techniques and can process large climate simulation ensembles much faster than traditional statistical methods, and also use efficient hardware architecture like GPUs.

These ML techniques have the potential to better assess uncertainties associated with the various projected IPCC (Intergovernmental Panel on Climate Change) scenarios and climate assessments. A generic method using Convolutional Variational Auto Encoders (CVAE) is being developed, and its performance will be assessed by comparing the results to a traditional analysis method. First results are very promising and are in line with expected results.