Contribution of tropical instability waves to ENSO irregularity

Abstract

Tropical instability waves (TIWs) are a major source of internally-generated oceanic variability in the equatorial Pacific Ocean. These non-linear phenomena play an important role in the sea surface temperature (SST) budget in a region critical for low-frequency modes of variability such as the El Niño–Southern Oscillation (ENSO). However, the direct contribution of TIW-driven stochastic variability to ENSO has received little attention. Here, we investigate the influence of TIWs on ENSO using a $1/4^\circ$ ocean model coupled to a simple atmosphere. The use of a simple atmosphere removes complex intrinsic atmospheric variability while allowing the dominant mode of air−sea coupling to be represented as a statistical relationship between SST and wind stress anomalies. Using this hybrid coupled model, we perform a suite of coupled ensemble forecast experiments initiated with wind bursts in the western Pacific, where individual ensemble members differ only due to internal oceanic variability. We find that TIWs can induce a spread in the forecast amplitude of the Niño 3 SST anomaly 6-months after a given sequence of WWBs of approximately $\pm \,45\%$ the size of the ensemble mean anomaly. Further, when various estimates of stochastic atmospheric forcing are added, oceanic internal variability is found to contribute between about $20\%$ and $70\%$ of the ensemble forecast spread, with the remainder attributable to the atmospheric variability. While the oceanic contribution to ENSO stochastic forcing requires further quantification beyond the idealized approach used here, our results nevertheless suggest that TIWs may impact ENSO irregularity and predictability. This has implications for ENSO representation in low-resolution coupled models.

Appendix: The atmospheric boundary layer model

As discussed in Sect. 2, we use an Atmospheric Boundary Layer Model (ABLM) to freely determine the air temperature $T_{air}$ and air humidity $q_{air}$. Our implementation is based on the cheapAML model of Deremble et al. (2013), following earlier work by Seager et al. (1995). The model solves single layer advection-diffusion equations for $T_{air}$ and $q_{air}$,

$$\begin{aligned} \frac{\partial T_{air}}{\partial t}&= -\nabla _h\cdot \left( \varvec{U} T_{air}-\kappa \nabla _h T_{air}\right) + \frac{1}{\rho _a C_p h}\left( F^+ -F^-\right) - \frac{1}{r_T}\left( T_{air} - T_b\right) , \end{aligned}$$

(4)

$$\begin{aligned} \frac{\partial q_{air}}{\partial t}&= -\nabla _h\cdot \left( \varvec{U} q_{air}-\kappa \nabla _h q_{air}\right) + \frac{1}{h}\left( F_Q^+ -F_Q^-\right) - \frac{1}{r_T}\left( q_{air} - q_b\right) , \end{aligned}$$

(5)

where $\varvec{U}$ is the prescribed 10m wind field, $\kappa$ is an isotropic horizontal diffusivity, $\rho _a$ is the density of air, $C_p$ is the heat capacity of air, h is the spatially variable depth of the atmospheric boundary layer, $r_T$ is a restoring time-scale that is non-zero only over land (where it takes the value 0.1 days) and $T_b$ and $q_b$ are background restoring fields for air temperature and humidity.

As discussed in Deremble et al. (2013), the imbalance of heat loss from the top of the boundary layer, $F^+$, and heat gain from the ocean $F^-$ are parameterized using long-wave radiative fluxes and the air–sea sensible heat flux (solar radiation and the latent heat flux both pass through the boundary layer at first order). Heat is lost via long-wave radiation from the top of the boundary layer using an average lapse rate of $0.0098\,^\circ$C m$^{-1}$. The upper and lower fluxes of moisture, $F_Q^+$ and $F_Q^-$ are represented by evaporation and entrainment at the top of the boundary layer. The advecting wind-velocities $\varvec{U}$, the boundary and over-land air temperature and air humidity and the spatially variable boundary layer depth h are taken from the ERA Interim 1980–2014 July–December average discussed above. All air–sea fluxes are determined using the ROMS bulk flux routines, based on Fairall et al. (1996). Due to the constant wind speeds and lack of storm systems, we use a large diffusivity of $\kappa =5 \times 10^5$ m$^2$s$^{-1}$.

In regions with high SST, the air temperature determined by the ABLM has a tendency to warm too much due to the absence of convection. This excessive warming in convective regions was also noted by Deremble et al. (2013), but they did not suggest a solution other than restoring. In order to avoid this unphysical warming we include a simple threshold on the surface air temperature, chosen as $28\,^\circ$C. This crudely models the effects of convection, which above this threshold mixes the air column vertically until the surface air temperature is once again below the threshold, returning the system to marginal stability. The presence of a threshold SST of around 27–28 °C above which convection occurs is well supported in the literature (e.g. Graham and Barnett 1987; Johnson and Xie 2010). Wind convergence also plays an important role in modulating convection (Graham and Barnett 1987). However, as we have a temporally constant wind field and do not resolve any synoptic scale variability we do not include a parameterization for this effect.

Our implementation of the ABLM includes several tuning parameters, such as the effective height of upwards long-wave radiation out of the boundary layer, the convective air temperature threshold and the constant of proportionality $\alpha$ relating the entrainment of humidity at the top of the boundary layer to the surface fluxes. The best parameter set was found to be $\alpha =0.3$ [compared to the value of 0.25 used by Deremble et al. (2013)], a $28\,\,^\circ$C threshold and long-wave radiation from the top of the boundary layer. The remaining biases include a tendency to be too warm and wet in the warm and wet regions and too cool and dry in the cool regions [as also noted by Deremble et al. (2013) in a fixed SST experiment]. This bias is likely due to the absence of low-cloud feedbacks.