Physics captured by data-based methods in El Ni\~no prediction

TL;DR

This paper investigates how data-driven methods, specifically CNNs, capture physical processes in El Niño prediction by analyzing their performance on distorted physics simulations, revealing strengths and limitations in representing key feedback mechanisms.

Contribution

It demonstrates the ability of CNNs to correct certain physical distortions in El Niño models and identifies specific feedback processes that are challenging for machine learning methods to represent.

Findings

CNNs can correct for ocean adjustment process distortions

CNNs struggle with distortions in upwelling feedback strength

Data-based methods reveal insights into physical process representation

Abstract

On average once every four years, the Tropical Pacific warms considerably during events called El Ni\~no, leading to weather disruptions over many regions on Earth. Recent machine-learning approaches to El Ni\~no prediction, in particular Convolutional Neural Networks (CNNs), have shown a surprisingly high skill at relatively long lead times. In an attempt to understand this high skill, we here use data from distorted physics simulations with an intermediate complexity El Ni\~no model to determine what aspects of El Ni\~no physics are represented in a specific CNN-based classification method. We find that the CNN can adequately correct for distortions in the ocean adjustment processes, but that the machine-learning method has far more trouble to deal with distortions in upwelling feedback strength.