Variational Autoencoders-based Detection of Extremes in Plant Productivity in an Earth System Model

TL;DR

This study applies a variational autoencoder (VAE) to detect extreme events in gross primary productivity (GPP) in CESM2 simulations across four AR6 CONUS regions and three historical-to-future periods under SSP5-8.5, benchmarking against singular spectral analysis (SSA). The VAE reconstructs GPP from 12-month sequences with a latent space of dimension 5, using a reconstruction loss plus KL regularization, and anomalies are flagged via percentile-based thresholds; SSA serves as a non-linear baseline. Results show strong spatial and temporal concordance between VAE and SSA in identifying extremes, though VAEs yield higher thresholds and better capture non-linear dependencies, with projected intensification of extremes by 2050-80, especially in CNA and WNA regions. The work demonstrates the viability of deep-learning-based anomaly detection for carbon-cycle extremes, offering computational efficiency and potential for real-time monitoring and improved understanding of carbon-climate feedbacks.

Abstract

Climate anomalies significantly impact terrestrial carbon cycle dynamics, necessitating robust methods for detecting and analyzing anomalous behavior in plant productivity. This study presents a novel application of variational autoencoders (VAE) for identifying extreme events in gross primary productivity (GPP) from Community Earth System Model version 2 simulations across four AR6 regions in the Continental United States. We compare VAE-based anomaly detection with traditional singular spectral analysis (SSA) methods across three time periods: 1850-80, 1950-80, and 2050-80 under the SSP585 scenario. The VAE architecture employs three dense layers and a latent space with an input sequence length of 12 months, trained on a normalized GPP time series to reconstruct the GPP and identifying anomalies based on reconstruction errors. Extreme events are defined using 5th percentile thresholds applied to both VAE and SSA anomalies. Results demonstrate strong regional agreement between VAE and SSA methods in spatial patterns of extreme event frequencies, despite VAE producing higher threshold values (179-756 GgC for VAE vs. 100-784 GgC for SSA across regions and periods). Both methods reveal increasing magnitudes and frequencies of negative carbon cycle extremes toward 2050-80, particularly in Western and Central North America. The VAE approach shows comparable performance to established SSA techniques, while offering computational advantages and enhanced capability for capturing non-linear temporal dependencies in carbon cycle variability. Unlike SSA, the VAE method does not require one to define the periodicity of the signals in the data; it discovers them from the data.

Figures (11)

• Figure 1: US AR6 regions reference map.
• Figure 2: Architecture of variational autoencoder designed for detecting GPP extremes
• Figure 3: Comparison between original (blue line) and VAE-based reconstructed GPP time series (orange line) for the period 1850-1880 in four CONUS AR6 regions.
• Figure 4: Convergence of model training and validation loss over epochs for the four study regions.
• Figure 5: Spatial distribution of total negative extremes in GPP during 1850--1880, for VAE (top) and SSA (bottom) for each of the AR6 regions analyzed separately. A separate color schemes are used for each region to highlight the gradients.