Assessing effectiveness of critical slowing down in a non-idealised timeseries of the Greenland ice sheet in a paleo-climate model run

Critical slowing down (CSD) is a widely used indicator of resilience loss in systems approaching an abrupt transition. In this study, we analyse a paleo-climate run of the Parallel Ice Sheet Model that shows an abrupt collapse of the Baffin Bay ice shelf, investigating whether CSD signals are present in nearby ice sheet basins. We carefully assess the assumptions underlying CSD in the context of this non-idealised timeseries. We show that the assumptions can be considered sufficiently fulfilled, through the presence of stationary noise, an alternative stable state, a timescale separation between response and forcing and the likely existence of a fold-type bifurcation via the melt-elevation feedback. We find a clear and statistically significant increase in variance and autocorrelation in the North West basin, adjacent to the location of the abrupt ice loss, beginning around -9 ka. This is appropriately before the abrupt change and therefore provides a meaningful forewarning of the subsequent ice loss. Statistical analysis, across a wide range of detrending bandwidths and window lengths, confirms that this signal is not an artefact of parameter choices. Further analysis of the forcing aids us in rejecting the null hypothesis that the indicators reflect inherited forcing variability rather than genuine resilience loss. Our results demonstrate that meaningful early warning signals can be recovered from non-idealised model output, even when assumptions are not perfectly met. We argue that CSD-based methods can provide valuable information when analysing non-idealised model timeseries, but need to be applied with careful assumption testing, parameter sensitivity analysis and statistical rigour.