Neural Network-Based Classification of Atmospheric Circulation Types: A Novel Application of Autoencoders for Time Decomposition of Climate Data

Abstract

Abstract This study presents a novel approach that employs Autoencoders (AE) - a type of artificial neural network - for the temporal decomposition of sea level pressure (SLP) data represented in a T-mode structure where the columns or variables are time series and rows or observations are grid points. The analysis aims to explore the effectiveness of non-linear AE in overcoming potential limitations of linear models like Principal Component Analysis (PCA) in terms of classifying atmospheric circulation types (CTs) within the southern region of Africa. After applying both PCA and AE for the time decomposition of daily SLP data from 1950 to 2022, the resulting spatial composite maps generated by each method were compared to assess consistency. The findings reveal consistency between the spatial maps from the two methods, with 58% of the maps showing congruence matches greater than 0.94. However, when examining the correctly classified dates (i.e., the true positives) with a threshold of 0.8 congruence coefficient between the spatial composite map representing the CT and the dates grouped under the CT, AE outperformed PCA with an average improvement of 29.2%. Hence given the flexibility of AE and capacity to model complex non-linear relationships, this study suggests that AE could be a potent tool for climate data decomposition and CT classification, leading to a more accurate representation of actual circulation variability.