Home
This Title All WIREs
WIREs RSS Feed
How to cite this WIREs title:
WIREs Clim Change
Impact Factor: 7.057

Preserving the coupled atmosphere–ocean feedback in initializations of decadal climate predictions

Full article on Wiley Online Library:   HTML PDF

Can't access this content? Tell your librarian.

Abstract On interannual to decadal time scales, memory in the Earth's climate system resides to a large extent in the slowly varying heat content of the ocean, which responds to fast atmospheric variability and in turn sets the frame for large‐scale atmospheric circulation patterns. This large‐scale coupled atmosphere–ocean feedback is generally well represented in today's Earth system models. This may fundamentally change when data assimilation is used to bring such models close to an observed state to initialize interannual to decadal climate predictions. Here, we review how the large‐scale coupled atmosphere–ocean feedback is preserved in common approaches to construct such initial conditions, with the focus on the initialized ocean state. In a set of decadal prediction experiments, ranging from an initialization of atmospheric variability only to full‐field nudging of both atmosphere and ocean, we evaluate the variability and predictability of the Atlantic meridional overturning circulation, of the Atlantic multidecadal variability and North Atlantic subpolar gyre sea surface temperatures. We argue that the quality of initial conditions for decadal predictions should not purely be assessed by their closeness to observations, but also by the closeness of their respective predictions to observations. This prediction quality may depend on the representation of the simulated large‐scale atmosphere–ocean feedback. This article is categorized under: Climate Models and Modeling > Knowledge Generation with Models
The probability distribution of Atlantic meridional overturning circulation (AMOC) between 24 and 28 °N (a,c) and of the Atlantic multidecadal variability (AMV) (b,d) for the time period 1961–2013. (a) The AMOC distribution during assimilation depends on the assimilation method. Atmospheric nudging only (nudgAf, brown) leads to a narrow distribution similar to the historical ensemble (dashed green), all oceanic assimilations lead to a much wider distribution more similar to ORAS4 reanalysis data (dashed red). Both anomaly nudging (nudgOa, black) and full‐value nudging (nudgOf, purple) resemble the double‐peak distribution from reanalysis data, whereas oceanic ensemble Kalman filter (EnKF) assimilation (enkfOf, blue) results in a one‐peak distribution. (c) In lead year 5, the shapes of the distributions become more similar to the historical ensemble in all hindcasts with some offset in the mean, the largest difference to the historical ensemble remains with the EnKF system. (b) The distribution of the AMV during assimilation only partly depends on the assimilation method (b), with the some differences due to oceanic EnKF assimilation and oceanic full‐value nudging. (d) In lead year 5, the AMV distributions are similar in all simulations. Please note that thick lines represent the cumulated distributions over all 10‐ensemble members, thin lines distributions for each of the 10‐ensemble members, respectively
[ Normal View | Magnified View ]
(a) In the time series of the simulated ensemble mean Atlantic meridional overturning circulation (AMOC) at 26.5°N during assimilation, interannual variability is dominant when only the atmosphere is nudged (nudgAf, brown). Multidecadal variability is introduced by oceanic assimilations. Here, both anomaly nudging (nudgOa, black) and full‐value nudging (nudgOf, purple) lead to a similar AMOC evolution than in the ORAS4 reanalysis data (dashed red), even down to the abnormally low AMOC in 1991. Oceanic ensemble Kalman filter (EnKF) assimilation (enkfOf, blue) results in an AMOC evolution different to any of other simulations before the 2000s. However, for the period after 2004, the observations from RAPID‐MOCHA (solid red, Smeed et al., ) are, except for a mean offset, met by all assimilations. The impact of assimilation decreases over the lead years. (b) In the lead year 5, time series of the simulated ensemble mean AMOC at 26.5°N, the variability is not much different any more from the uninitialized historical ensemble (dashed green)
[ Normal View | Magnified View ]
The pattern correlation of the initialized Atlantic meridional overturning circulation (AMOC) cell (mean and year‐to‐year variability) with that of the uninitialized historical ensemble (Figures and , respectively) is a measure of similarity between the initialized and uninitialized simulations. (a) The pattern correlation for the AMOC mean cells depends on the assimilation method, changes over lead time are small and the differences are maintained over all lead times. (b) The pattern correlation for the year‐to‐year variability also depends on the assimilation method, but it changes over lead time. The two systems applying oceanic nudging have a lower correlation with the uninitialized system during assimilation than the other two systems, but not as low as the ORAS4 reanalysis. Both nudging systems show an increasing correlation and therefore similarity with the uninitialized system after assimilation with some saturation after lead year 5. The impact of the oceanic ensemble Kalman filter (EnKF) assimilation lasts longer, the correlation increases only after lead year 3 as does the uncertainty, and does not reach a saturation before lead year 10. For the system with atmospheric nudging only, correlation is high during assimilation and the first five lead years. It slightly drops off toward lead year 10 with increasing uncertainty
[ Normal View | Magnified View ]
The year‐to‐year variability (1961–2013) of the Atlantic meridional overturning circulation (AMOC) cell in terms of the ensemble mean of the standard deviation of yearly mean values depends on the assimilation method (left column), but this dependence persists only in the first lead years and is hardly visible in lead year 5 (right column). The atmospheric assimilation alone (nudgAf, b, c) accounts only for a minor part in changes when compared to the historical ensemble (upper panel). The major part of these changes comes from the oceanic assimilation. Here, both anomaly nudging (nudgOa, f) and full‐value nudging (nudgOf, h) to ORAS4 reanalysis data results in a transfer of the variability of the re‐analysis (lower panel) to the Max Planck Institute Earth System Model (low resolution, MPI‐ESM‐LR). Oceanic ensemble Kalman filter (EnKF) assimilation of observed T and S profiles (enkfOf, d) shows a similar strong variability. In all hindcast simulations, the variability becomes similar to that from the uninitialized historical ensemble from lead year 5 onward. The lead year dependency of the pattern correlation of the variability of the initialized AMOC cell with that of the uninitialized AMOC cell is shown in Figure
[ Normal View | Magnified View ]
Prediction skill for sea surface temperature in terms of correlation with HadISST near observational data (Rayner et al., ) in the period 1961–2013 is already high in the uninitialized historical ensemble for large parts of the North Atlantic (a), which are dominated by a long‐term trend. Prediction skill is low in areas without a dominant trend: the subpolar gyre and the North Atlantic current. In these areas, initialized hindcast shows partly improved skill in lead years 2–5 depending on the assimilation method: absolute correlation on the left, differences in correlation to the historical ensemble on the right. Atmospheric nudging alone (nudgAf, b,c) improves the lead years 2–5 prediction skill for the North Atlantic current region. The oceanic ensemble Kalman filter (EnKF) (enkfOf, d,e) additionally improves the western part of the subpolar gyre, whereas anomaly nudging to reanalysis data (nudgOa, f,g) results in no additional improvements and full‐value nudging to reanalysis data (nudgOf, h,i) even counteracts the positive impact of atmospheric nudging and leads to deterioration of correlations in the southern part of the subpolar gyre
[ Normal View | Magnified View ]
The time mean (1961–2013) Atlantic meridional overturning circulation (AMOC) cell depends on the assimilation method (left column, see also Smith, Eade, & Pohlmann, ). The hindcasts inherit the general pattern of their AMOC mean cell from the assimilation for lead year all lead years (lead year 5 shown in the right column). The oceanic assimilation method determines how close the mean cell is to either the uninitialized historical ensemble (upper panel) or the ORAS4 reanalysis data (lower panel) used during assimilation. With atmospheric nudging only (nudgAf), the mean cell is very close to the historical ensemble with a distinct clockwise circulation above 3,000 m depth and a anticlockwise circulation below (b,c). A similar mean cell is simulated when additionally the ocean is anomaly nudged (nudgOa) toward reanalysis data (f,g). The ensemble Kalman filter (EnKF) assimilation of full‐value observed oceanic profiles (enkfOf) changes the mean cell significantly (d), these changes also persist in lead year 5 (e). The full‐value nudging toward reanalysis data (nudgOf) effectively imprints large‐ and small‐scale features of the mean cell of ORAS4 onto Max Planck Institute Earth System Model (low resolution, MPI‐ESM‐LR) during assimilation (h) and the large‐scale features still persist in lead year 5 (i). The lead year dependency of the pattern correlation of the initialized AMOC mean cells with the uninitialized AMOC mean cell is shown in Figure
[ Normal View | Magnified View ]
(a) The prediction skill of Atlantic meridional overturning circulation (AMOC) at 26.5°N in terms of correlation with RAPID‐MOCHA observations (Smeed et al., ) over the period 2004–2013 is lead year dependent. Atmospheric nudging alone (nudgAf, brown) and the oceanic ensemble Kalman filter (EnKF) assimilation (enkfOf, blue) result in significantly higher prediction skill, that is, shaded areas do not overlap, than the uninitialized historical ensemble (dashed green) for lead years 1 to 5. Nudging to reanalysis data, either anomaly (nudgOa, black) or full‐value (nudgOf, purple), results in a high correlation during assimilation but in low correlations in lead years 1 and 2, due to an initialization shock (Kröger et al., ). The prediction horizon for AMOC at 26.5°N with the Max Planck Institute Earth System Model (low resolution, MPI‐ESM‐LR) is at five lead years. After this time, skill does not significantly differ between the simulations. (b) The prediction skill of Atlantic multidecadal variability (AMV) in terms of correlation with HadISST (Rayner et al., ) is high for all initialized systems in lead year 1 but differs considerably from lead year 2 to 5. Here, the EnKF system outperforms all other systems while the system with oceanic full‐value nudging is not better than the uninitialized historical ensemble. After lead year 5, the EnKF system looses its good prediction skill, and none of the initialized systems significantly outperforms the uninitialized historical ensemble. The prediction horizon for AMV is at lead year 5, similar to that for AMOC. Please note that the thick line represents the skill of the ensemble mean, the shaded areas indicate the spread in skill based on 95% of the bootstrapped ensemble mean
[ Normal View | Magnified View ]

Browse by Topic

Climate Models and Modeling > Knowledge Generation with Models

Access to this WIREs title is by subscription only.

Recommend to Your
Librarian Now!

The latest WIREs articles in your inbox

Sign Up for Article Alerts