Back To Schedule
Monday, October 14 • 3:45pm - 4:00pm
Data-Driven Super-Parametrization Using Deep Learning: Climate Models that Do Not Affect Our Climate

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!


Certain physical processes that play key roles in the weather-climate system occur at such small spatial and fast time scales that resolving them can be very expensive. These subgrid-scale processes (denoted by variable Y), are parameterized using semi-empirical schemes as a function of the large-scale-slow variables (X) that are explicitly solved. Multi-scale numerical models dubbed super-parameterization (SP), improved simulation of climate variabilities and extremes, but is computationally prohibitive for many applications. Here, we show recurrent neural networks (RNNs) can be used for data-driven super-parameterization (DDSP): To solve for X numerically at low resolution, and emulate Y at higher numerical resolutions using RNNs. With a multi-scale Lorenz 96 chaotic system as the testbed and examining both predicted short-term trajectory (weather forecasting) and reproducing long-term statistics (climate simulation) we show that DDSP outperforms state-of-the-art and can achieve the accuracy of SP at a much lower computational cost.


Ashesh Chattopadhyay

Presenter, Rice University

Adam Subel

Rice University

Pedram Hassanzadeh

Rice University

Krishna Palem

Rice University

Monday October 14, 2019 3:45pm - 4:00pm CDT
BRC 280

Attendees (3)