[Paper Review] Probability Estimates of a 21st Century AMOC Collapse

Emma Smolders, René M. van Westen+1 more|arXiv (Cornell University)|Jun 17, 2024

Climate variability and modelsEnvironmental Science3 citations

TL;DR

This study provides observationally constrained probability estimates for a 21st-century Atlantic Meridional Overturning Circulation (AMOC) collapse using reanalysis data and a CMIP5 climate model. It identifies salinity at the southern Atlantic boundary as the optimal predictor, estimating a 59% probability of collapse before 2050 (mean tipping time: 2050; 10–90% confidence interval: 2037–2064).

ABSTRACT

There is increasing concern that the Atlantic Meridional Overturning Circulation (AMOC) may collapse this century with a disrupting societal impact on large parts of the world. Preliminary estimates of the probability of such an AMOC collapse have so far been based on conceptual models and statistical analyses of proxy data. Here, we provide observationally based estimates of such probabilities from reanalysis data. We first identify optimal observation regions of an AMOC collapse from a recent global climate model simulation. Salinity data near the southern boundary of the Atlantic turn out to be optimal to provide estimates of the time of the AMOC collapse in this model. Based on the reanalysis products, we next determine probability density functions of the AMOC collapse time. The collapse time is estimated between 2037-2064 (10-90% CI) with a mean of 2050 and the probability of an AMOC collapse before the year 2050 is estimated to be 59 +/- 17%.

Motivation & Objective

• To estimate the probability of an AMOC collapse in the 21st century using observational data rather than proxy reconstructions or idealized models.
• To identify the most optimal observation regions and variables for predicting AMOC tipping based on a high-resolution climate model simulation.
• To improve early warning signal reliability by using physically grounded indicators like freshwater transport and restoring rate instead of relying solely on variance and autocorrelation.
• To provide a robust, observationally constrained estimate of AMOC tipping time with quantified uncertainty, addressing limitations in prior statistical and proxy-based approaches.

Proposed method

• Used a CMIP5 Community Earth System Model (CESM) simulation to identify regions and variables most predictive of AMOC collapse under freshwater forcing.
• Identified salinity data near the southern Atlantic boundary (34°S) as the optimal predictor of AMOC tipping time, based on strong correlation with overturning freshwater transport ($F_{\mathrm{ov}}$).
• Applied a restoring rate ($\lambda$) estimator derived from an autoregressive model to quantify system resilience, which is less sensitive to noise than traditional early warning signals (EWS).
• Used change point (CP) analysis with Kendall-tau testing on the restoring rate time series to detect significant shifts indicating approaching tipping points.
• Calibrated the model-based tipping time distribution using reanalysis products (ORAS5, GLORYS12V1, SODA3.15.2) to derive probability density functions of the AMOC collapse time.
• Combined statistical inference from reanalysis data with model-derived optimal observables to produce observationally constrained probability estimates.

Experimental results

Research questions

• RQ1 Which oceanic regions and variables provide the most reliable early warning signals for an impending AMOC collapse?
• RQ2 How can reanalysis data be used to constrain the probability distribution of AMOC tipping time in the 21st century?
• RQ3 Does the classical EWS (variance and autocorrelation) reliably predict AMOC collapse, or are alternative indicators like the restoring rate more robust?
• RQ4 What is the estimated probability of AMOC collapse before 2050, and what is the associated uncertainty range?

Key findings

• Salinity observations near the southern Atlantic boundary (34°S) are identified as the optimal predictor of AMOC collapse, outperforming subpolar gyre SSTs.
• The probability of an AMOC collapse before 2050 is estimated at 59% ± 17%, based on reanalysis data and model-derived optimal observables.
• The mean tipping time for AMOC collapse is estimated at 2050, with a 10–90% confidence interval of 2037–2064.
• The restoring rate ($\lambda$) estimator provides a more robust indicator of system stability than variance or autocorrelation, as it is less sensitive to external noise fluctuations.
• Classical EWS based on subpolar gyre SSTs fail to detect the AMOC collapse in the CESM model, highlighting the need for alternative, physics-based indicators.
• The freshwater transport by the overturning component ($F_{\mathrm{ov}}$) at 34°S shows a minimum just before collapse, confirming its value as a physical early warning signal.