A unified account of Earth's climate variability: dominant modes as eigenmodes of one stochastic operator
Abstract. The Madden–Julian Oscillation, the El Niño–Southern Oscillation, the Indian Ocean Dipole, the Atlantic zonal mode and the annular modes are conventionally explained by separate theories with separate parameters. We show that they can emerge from one minimal description. The climate state is modelled as four fields on the sphere, an atmospheric streamfunction, a column moisture anomaly, an ocean mixed-layer temperature and an upper-ocean heat content, whose joint evolution is written as a single linear stochastic equation. The operator contains rotation, dissipation, air–sea and moisture coupling, and the geography of the continents, and no mode-specific physics. Its eigenvalue spectrum separates into three families without any tuning of individual modes: damped atmospheric patterns with e-folding times of about a week, an eastward-propagating equatorial moisture mode with a period of 47 days, and basin-confined coupled oscillators with periods of years. Integrating the system under stochastic forcing reproduces the observed spectral bands, with peaks at 4.29 yr for Niño 3.4, 1.54 yr for the Dipole Mode Index, 1.54 yr for the Atlantic index and 69.1 days for the moisture index, together with red backgrounds for the annular indices. The same periods obtained from the resolvent of the autonomous operator, which requires no integration and carries no sampling error, are 4.06 yr, 1.54 yr and 1.54 yr. The interannual periods are not fitted: they follow from the zonal width of each ocean basin, with a fitted exponent of 1.00 on the resolvent peaks and 1.07 on the eigenmode periods, and they reproduce the observed ordering and approximate spacing of the Pacific, Indian and Atlantic modes. Four natural forcing terms are then added, followed by an anthropogenic one. Annual modulation of the coupling produces seasonal phase locking of approximately the observed magnitude. Stochastic forcing produces the red spectra of the extratropical modes without any oscillatory mechanism. An imposed volcanic cooling produces a damped oscillatory Pacific response with a rebound near two years and a memoryless extratropical response. Slow modulation of the dissipation gives the mode variances a common low frequency component. Anthropogenic forcing is treated last and separates into two channels: an additive radiative term, which by superposition cannot alter any anomaly statistic, and a parametric dependence of the operator on the background state, which can. Parameterising the second through five physical channels gives a shortening of the interannual periods governed by the increase in equatorial wave speed alone, so that the fractional period change equals minus the fractional change in wave speed, and an amplitude response that is the residue of four opposing contributions of comparable magnitude. The second result predicts that projections of ENSO amplitude should be small and of uncertain sign, which is what the data show.
The authors developed an idealized, linear, global model of low-frequency climate variability that, they claim, offers insights into the dynamics of the observed internal and forced climate modes. This may be so, but, in my opinion, the model is insufficiently described in the present version of the paper, while the implied correspondence between the simulated and observed variability is highly subjective, all of which asks for a major revision before this paper becomes suitable for publication.
In particular, I’d suggest the authors include the formulas for the actual, complete set of equations for all model components, as well as the exact formulations of the way these linearized components are coupled and forced (before projecting these equations onto the spherical harmonics), including the justification for the choices of the final model parameters. Apparently, the authors felt that all these equations are well known and the verbal description of the model is sufficient but I disagree.
It looks like the equations are:
(1) For psi: barotropic vorticity equation on a sphere with topography linearized about the mean state with a prescribed uniform zonal flow - there is an unspecified vorticity source associated with the precipitation and latent heat release (?);
(2) For q: Moisture advection by the barotropic flow, removing access moisture by condensation through an unspecified parameterization - this needs to be spelled out in the explicit formulation. What’s the climatological distribution of specific humidity, also uniform? How does one decide whether the specific humidity exceeds saturation value to result in condensation?
(3, 4) The ocean component is most cryptic: it is “represented by the recharge structure of Jin (1997), generalised from a two-variable system to a field” - whatever this is supposed to mean (is it that each spherical harmonic of T, h represents a recharge oscillator with the same parameters?). It is also coupled, in the tropics, “to the atmosphere … through surface wind stress and convective heating.” There is no physical description of how this coupling is set up and how the coupling parameters are derived. What are the actual physical equations that govern the evolution of the oceanic thermocline in this model (see Jin 1997 part 2 for an example of what I mean by the physical equations)? Is it only valid in tropics? (Why then global claim?)
(5) Forcing: How exactly is Eq. (4) used to derive the covariance structure of the noise forcing for the final system? Some discussion of the stochastic forcing appears in section 5.1 (first sentence), well after the simulation results under this stochastic forcing are discussed (which is not a very logical way of presenting things). It looks like noise specification is quite arbitrary and ad hoc. The way the anthropogenic forcing is included also seems to be quite heuristic. Adding seasonal dependence to model damping parameters leads to the seasonal cycle of model variance - something that is hardly unexpected.
The authors’ punchline is that their model produces tropical spectra in Fig. 4 with broad spectral peaks mimicking the time scales of the observed tropical phenomena (although I must say the spatial patterns of these modes are not particularly consistent with the observed patterns!). They go on to show that these simulated time scales are sensitive to land-ocean mask configuration, which may not be that surprising given the chosen structure of the model.
All in all, in my view, the authors set up a toy model of the global atmosphere coupled to a toy tropical model representing the ocean component of the climate system and its coupling to the overlying atmosphere, yet on a global grid with realistic land-sea contrast. This model has a bunch of tropical modes mimicking some of the observed climate modes in terms of the periods and, less so, spatial patterns, yet without realistic extratropical teleconnections (cf. Branstator 1985, for example). The modal structure is shown to strongly depend on the land-sea mask, which is an interesting point.