Dynamical attribution of oceanic prediction uncertainty in the North Atlantic: application to the design of optimal monitoring systems

Abstract

In this study, the relation between two approaches to assess the ocean predictability on interannual to decadal time scales is investigated. The first pragmatic approach consists of sampling the initial condition uncertainty and assess the predictability through the divergence of this ensemble in time. The second approach is provided by a theoretical framework to determine error growth by estimating optimal linear growing modes. In this paper, it is shown that under the assumption of linearized dynamics and normal distributions of the uncertainty, the exact quantitative spread of ensemble can be determined from the theoretical framework. This spread is at least an order of magnitude less expensive to compute than the approximate solution given by the pragmatic approach. This result is applied to a state-of-the-art Ocean General Circulation Model to assess the predictability in the North Atlantic of four typical oceanic metrics: the strength of the Atlantic Meridional Overturning Circulation (AMOC), the intensity of its heat transport, the two-dimensional spatially-averaged Sea Surface Temperature (SST) over the North Atlantic, and the three-dimensional spatially-averaged temperature in the North Atlantic. For all tested metrics, except for SST, ∼ 75% of the total uncertainty on interannual time scales can be attributed to oceanic initial condition uncertainty rather than atmospheric stochastic forcing. The theoretical method also provide the sensitivity pattern to the initial condition uncertainty, allowing for targeted measurements to improve the skill of the prediction. It is suggested that a relatively small fleet of several autonomous underwater vehicles can reduce the uncertainty in AMOC strength prediction by 70% for 1-5 years lead times.

Spatiotemporal Patterns of Chaos in the Atlantic Overturning Circulation

“…The intrinsic power spectral density decreases monotonically at lower frequencies whereas the forced mode dominates at long time scales. This suggests that in the future generation of climate models with eddy-resolving ocean models, projections of future changes in the North Atlantic overturning would be somewhat limited at interannual time scales but might benefit from better predictive skills at decadal and longer time scales, consistent with the theoretical investigation of Sévellec et al (2018). Although the dynamical origin of this large-scale intrinsic mode of AMOC variability remains to be investigated in depth, the physical processes responsible for their emergence could involve spatiotemporal inverse cascade of kinetic energy (Arbic et al, 2012(Arbic et al, , 2014.…”

The active and passive roles of the ocean in generating basin-scale heat content variability

“…In contrast to the established techniques of using a passive tracer (e.g., Banks & Gregory, 2006;Xie & Vallis, 2012;Marshall et al, 2015;Garuba & Klinger, 2016 or a slab ocean model (e.g., Dommenget & Latif, 2002;Dommenget, 2010;Clement et al, 2015;Wang & Dommenget, 2016) to investigate the role of the ocean, we have utilised a novel adjoint-based approach (Sévellec et al, 2018). The use of an adjoint model has uniquely allowed us to causally attribute heat content variance to di↵erent variables, times, and locations at the surface, by projecting onto surface sensitivity fields a realistic stochastic representation of atmospheric fluxes diagnosed from a coupled climate model.…”

“…Unlike a conventional model, which integrates anomalies forward in time, an adjoint model describes the sensitivity of a metric of interest (here heat content) to past changes (here stochastic atmospheric forcing), establishing causes, rather than e↵ects (Errico, 1997). This has been leveraged to attribute the sources of temporal ocean variability in response to historical atmospheric forcing (Pillar et al, 2016;Smith & Heimbach, 2019) and establish the evolution of oceanic variance in response to representative stochastic atmospheric forcing (Sévellec et al, 2018).…”

Issues in the theoretical foundations of climate science

“…Enriching the characterisation of climate states can also be undertaken with the help of climate models. For example, climate models are already being used to characterise how distributions of climatic quantities on interannual and decadal timescales depend on underlying physical processes (Daron and Stainforth 2013, Hawkins et al 2016, and Sévellec et al 2017.…”