Climate Variability and Weather Noise

There is an active research debate regarding how weather and climate interact – this is particular noted in terms of understanding ENSO predictability and the influence of “weather noise”. This weather noise view argues that the irregularity of ENSO and ultimately the loss of predictability are largely driven by stochastic forcing. If the stochastic forcing or noise is of primary importance then it is possible that the details (e.g., space-time structure and state dependence) of the noise are also important – that is weather and climate interactions. The overarching goal of this dissertation research is to develop a framework to study the role, if any, that noise plays in sustaining/modifying/modulating ENSO and tropical Pacific climate variability and predictability. The methodology used here follows two different approaches: (a) assessing the impact of the noise in a non-phenomenological manner recognizing that it can occur on all temporal and spatial scales and (b) assuming the noise is associated with a specific phenomena with clear constrains on its location and spatial-temporal structure. A noise reduction technique, namely the interactive ensemble (IE) approach is adopted to reduce non-phenomenological noise at the air-sea interface due to internal atmospheric dynamics in a state-of-the-art coupled general circulation model (CGCM), namely the Community Climate System Model (CCSM3). To study the impact of weather noise and resolution in the context of a CGCM, two IE experiments are performed at different resolutions. Atmospheric resolution is an important issue since the noise statistics will depend on the spatial scales resolved. A simple formulation to extract atmospheric internal variability is presented. The results are compared to their respective control cases where internal atmospheric variability is left unchanged. The non-phenomenological noise reduction has a major impact on the coupled simulation and the magnitude of this effect strongly depends on the horizontal resolution of the atmospheric component model. Specifically, applying the noise reduction technique reduces the overall climate variability more effectively at higher resolution. This suggests that “weather noise” is more important in sustaining climate variability as resolution increases. ENSO statistics, dynamics, and phase asymmetry are all modified by the noise reduction, in particular ENSO becomes more regular with less phase asymmetry when noise is reduced. All these effects are more marked for the higher resolution case. In contrast, ENSO frequency is unchanged by the reduction in the weather noise, but its phase-locking to the annual cycle is strongly dependent on noise and resolution. At low resolution the noise structure is similar to the signal, whereas the spatial structure of the noise deviates from the spatial structure of the signal as resolution increases. It is also suggested that event-to-event differences are largely driven by atmospheric noise as opposed to chaotic dynamics within the context of the large-scale coupled system, suggesting that there is a well-defined “canonical” event. The next step is to study the importance of phenomenological noise forcing of the climate system. Here, westerly wind bursts or events (WWBs or WWEs) are taken as example of phenomenological noise forcing of the tropical Pacific Ocean. The impact of parameterized WWBs on ENSO variability in CCSM3 and CCSM4 is analyzed. To study the impact of WWBs three experiments are performed. In the first experiment, the model is integrated for several hundred years with no prescribed WWBs events (i.e., the control). In the second case, state-independent WWBs events are introduced. In other words, the occurrence, location, duration, and scale of the WWBs are determined (within bounds) randomly. For the third case, the WWBs are introduced but as multiplicative noise or state-dependent forcing, modulated by SST anomalies. State-dependent case produced larger ENSO. There is very little difference between the control and the state-independent WWB simulations suggesting that the deterministic component of the burst is responsible for reshaping the ENSO events. There is a shift towards a more self sustained mechanism as the experiments progress from the control to the state dependent WWBs. Overall, the parameterized WWBs have the capability to modify the ENSO regime in the CGCM, demonstrating the importance of sub-seasonal variability on interannual time scales. This study also investigates the effect of parameterized WWBs on the diversity of ENSO warm events, namely eastern Pacific (EP) and central Pacific (CP) ENSO in CCSM3 and CCSM4. It is found that parameterized WWBs tend to enhance EP variability more relatively to CP variability. This enhancement in the case of state-dependent WWBs forcing is due to an increase in the so-called thermocline feedback as opposed to the so-called zonal advective feedback. Lastly, we test whether phenomenological stochastic forcing of the form of WWBs impacts ENSO predictability. An ensemble ENSO prediction experiment is presented in which CCSM3 control and CCSM3 with state-dependent WWBs parameterization are used as both truth and as predictor systems. The inclusion of WWBs does not improve nor degrades ENSO predictability if the truth lacks WWBs activity. ENSO predictability increases substantially if a forecast system that produces WWBs activity is used to predict a truth that includes these wind events. It is also found that the so-called forecast spring prediction barrier (SPB) is partially caused by the lack of WWBs representation in the forecast system. The argument for the SPB is that the coupled system is more susceptible to noise forcing in spring. The signal-to-noise ratio (SNR) is larger with WWBs, so it must be that the increase in the signal due to state-dependent WWBs is more important that the added noise. These results were further validated with the more recent version of this model, namely CCSM4. It turned out that CCSM4, at least with the coarse resolution used here, has a much-reduced seasonality in the SNR and therefore reduced seasonality in the forecast skill of SSTA. To further test these results, CCSM3 with and without WWBs parameterization were used to make real predictions of observed tropical Pacific SSTA. It was demonstrated that predictability skill are enhanced when WWBs were included and these improvements were mostly over the low SNR season. These results were further validated by a case study of a warm event with considerable WWBs activity, mimicking the strong 1997-98 event. It was found that the presence of WWBs in the prediction system enhances the forecast ensemble spread, leading to a more reliable probabilistic forecast. But most importantly, the number of ensemble members depicting the correct “truth” increases considerably. This is best observed for those forecasts progressing through the SPB.