Skip to main content
U.S. flag

An official website of the United States government

Publication Date
17 July 2023

Assessing Tropical Pacific-induced Predictability of Southern California Precipitation Using Machine Learning

Print / PDF
Powerpoint Slide
Image
|
Science

A novel Multi-Input Multi-Output AutoEncoder (MIMO-AE) machine learning network is designed to capture the non-linear relationship of Southern California precipitation with Tropical Pacific Ocean sea surface temperatures, which is projected onto an index (MIMO-AE index). MIMO-AE network is trained on a historical simulation with the DOE’s Energy Exascale Earth System Model (E3SMv1) and observational data and its skill of predicting sub-seasonal to seasonal Southern California precipitation is evaluated.

Impact

MIMO-AE provides significantly enhanced predictability of Southern California precipitation for a lead-time of up to four months as compared to El Niño Southern Oscillation (ENSO) indices, like Niño 3.4 index and ENSO Longitudinal Index. MIMO-AE learned SST anomaly patterns associated with Southern California precipitation strongly influence processes that drive precipitation over the region, allowing MIMO-AE to provide enhanced predictive skill. The study demonstrates that machine learning approaches could significantly improve the predictability of regional precipitation on sub-seasonal to seasonal time scales.

Summary

Traditional El Niño Southern Oscillation indices, like the Niño 3.4 index, although well-predicted themselves, fail to offer reliable sub-seasonal to seasonal predictions of Western US precipitation. A novel Multi-Input Multi-Output Autoencoder-decoder (MIMO-AE) machine learning network is constructed to capture the non-linear relationship between Southern California precipitation and Tropical Pacific Ocean sea surface temperatures. The MIMO-AE is trained on both monthly Tropical Pacific SST anomalies and Southern California precipitation anomalies simultaneously. The co-variability of the two fields in the MIMO-AE shared nonlinear latent space can be condensed into an index, termed the MIMO-AE index. A transfer learning approach is used to train a MIMO-AE on the combined dataset of 100 years of output from a historical simulation with DOE’s Energy Exascale Earth Systems Model (E3SMv1) and a segment of observational data. Long Short-Term Memory networks are used to assess sub-seasonal to seasonal predictability of Southern California precipitation from the MIMO-AE index. MIMO-AE provides enhanced predictability of SC-PRECIP for a lead-time of up to four months as compared to Niño 3.4 index and the El Niño Southern Oscillation Longitudinal Index. This is likely because MIMO-AE learned SST anomaly patterns associated with Southern California precipitation strongly influence processes that drive precipitation over the region, lending predictive skill.

Point of Contact
Salil Mahajan
Institution(s)
Oak Ridge National Laboratory
Funding Program Area(s)
Additional Resources:
ALCC (ASCR Leadership Computing Challenge)
NERSC (National Energy Research Scientific Computing Center)
Publication