Item

Abstract

Observational datasets are increasingly used to detect critical slowing down (CSD) as a measure of stability or resilience changes in key Earth system components. However, most observational datasets have inherent, often non-stationary missing-data-distributions, as well as biases and uncertainties, all of which can influence the CSD analysis. In particular, sea-surface temperature (SST) and salinity-based indices have been used to detect CSD for a possible collapse of the Atlantic Meridional Overturning Circulation (AMOC). Here we present an in-depth uncertainty analysis of AMOC CSD based on SST and salinity fingerprints. We first use uncertainties provided with the HadSST4 and HadCRUT5 SST datasets to generate uncertainty ensembles and estimate the uncertainty of SST-based AMOC fingerprints. We then construct stringent and conservative significance tests on the CSD indicators in the EN4.2.2, HadISST1 and HadCRUT5 datasets. We use surrogate testing that incorporates the influence of data processing steps and non-stationary uncertainties can have on CSD. We find that the properties of the observational datasets could in theory cause false indication of CSD, but that in the cases we examine, CSD indicators in the Atlantic are still present and significant. Our results highlight the importance of taking into account the properties of the specific observational dataset used when calculating higher-order statistics.