A Sequential Autoencoder for Teleconnection Analysis

Abstract

Many aspects of the earth system are known to have preferred patterns of variability, variously known in the atmospheric sciences as modes or teleconnections. Approaches to discovering these patterns have included principal components analysis and empirical orthogonal teleconnection (EOT) analysis. The latter is very effective but is computationally intensive. Here, we present a sequential autoencoder for teleconnection analysis (SATA). Like EOT, it discovers teleconnections sequentially, with subsequent analyses being based on residual series. However, unlike EOT, SATA uses a basic linear autoencoder as the primary tool for analysis. An autoencoder is an unsupervised neural network that learns an efficient neural representation of input data. With SATA, the input is an image time series and the neural representation is a unidimensional time series. SATA then locates the 0.5% of locations with the strongest correlation with the neural representation and averages their temporal vectors to characterize the teleconnection. Evaluation of the procedure showed that it is several orders of magnitude faster than other approaches to EOT, produces teleconnection patterns that are more strongly correlated to well-known teleconnections, and is particularly effective in finding teleconnections with multiple centers of action (such as dipoles).

Assessing Teleconnections-Induced Predictability of Regional Water Cycle on Seasonal to Decadal Timescales Using Machine Learning Approaches

“…Recent work suggests that ML approaches can predict the Nino3.4 index more skillfully than dynamical forecast systems with lead times of more than a year (Yan et al, 2020, He and Eastman 2020, Ham et al 2019, Dijkstra et al 2019, largely using different types of neural networks, for example, deep neural networks, convolution neural networks (CNN), and recurrent neural networks like the convolution long short-term memory. The lack of sufficient observational data was overcome in one study (Ham et al 2019) by using Earth system model (ESM) simulations: a CNN network was trained with global SST and heat content data from historical simulations of 21 CMIP5 models to predict the Nino3.4 index.…”

“…Deep generative networks, like autoencoders-decoders (AE), in combination with explainable AI methods (tools like heat maps), offer a new avenue to nonlinearly identifying variability patterns of tropical Pacific SSTs (and thus an index). Shallow AEs have been shown to replicate spatial and temporal characteristics of PCs (He and Eastman 2020). A more targeted approach to identifying generative networks that elicit statistically significant ENSO teleconnections to remote regions globally-able to cut through the noise of intrinsic atmospheric variability-could enhance our predictive understanding of ENSO's remote impacts.…”

ML-based regionalization of climate variables to forecast seasonal precipitation for water resources management

“…Another extraction algorithm similar to PCA, called independent component analysis (ICA), has been applied to climate studies (Westra and Sharma 2005, Westra et al 2010, Forootan et al 2018. Also, an autoencoder, an unsupervised learning technique belonging to artificial neural network, has recently got the attention in climate studies due to its dimensional reduction property (Saha et al 2016, He andEastman 2020). Since these techniques can extract critical information from globally gridded climate variables to only a few indicators, it is easier to teleconnect the indicators with local seasonal precipitation for forecasting rather than globally grided climate variables.…”

Assessing Tropical Pacific-induced Predictability of Southern California Precipitation Using a Novel Multi-input Multi-output Autoencoder

“…Some studies (e.g. Tang & Hsieh, 2003;He & Eastman, 2020) have also demonstrated the use of autoencoders to effectively identify modes of climate variability, including those related to ENSO.…”