Interpreting the reproduction of climate natural variability and extreme events using probabilistic neural network architectures

This project aims to use artificial intelligence methods to reproduce various characteristics of the Earth climate system. More specifically, we know that the climate system is subject to natural variability, a physical phenomenon that dictates that multiple future conditions are possible within atmospheric physical constraints. Modelling this phenomenon is currently very costly with large models that solve the physical equations of the atmosphere. Having neural network models that can accurately model natural variability would be a huge benefit to the climate science community. We plan to use recent neural network architectures that have promising properties for generating probabilistic outputs that can be related to natural variability. Gaining a better understanding of this system will have applications for studying regional scale climate extremes with rapid production of plausible climate conditions.

Faculty Supervisor:

Julie Carreau

Student:

Partner:

Ouranos Inc

Discipline:

Earth science

Sector:

Accommodation and food services; Agriculture; Professional, scientific and technical services; Public administration

University:

Polytechnique Montréal

Program: