Extraction and predictability of coherent intraseasonal signals in infrared brightness temperature data

This work studies the spatiotemporal structure and regime predictability of large-scale intraseasonal oscillations (ISOs) of tropical convection in satellite observations of infrared brightness temperature (). Using nonlinear Laplacian spectral analysis (NLSA), a data analysis technique designed to extract intrinsic timescales of dynamical systems, the field over the tropical belt and the years 1983-2006 (sampled every 3 h at resolution) is decomposed into spatiotemporal modes spanning interannual to diurnal timescales. A key advantage of NLSA is that it requires no preprocessing such as bandpass filtering or seasonal partitioning of the input data, enabling simultaneous recovery of the dominant ISOs and other patterns influenced by or influencing ISOs. In particular, the eastward-propagating Madden-Julian oscillation (MJO) and the poleward-propagating boreal summer intraseasonal oscillation (BSISO) naturally emerge as distinct families of modes exhibiting non-Gaussian statistics and strong intermittency. A bimodal ISO index constructed via NLSA is found to have significantly higher discriminating power than what is possible via linear methods. Besides MJO and BSISO, the NLSA spectrum contains a multiscale hierarchy of modes, including the annual cycle and its harmonics, ENSO, and modulated diurnal modes. These modes are used as predictors to quantify regime predictability of the MJO amplitude in data through a cluster-based framework. It is found that the most predictable MJO regimes occur before the active-MJO season (November-December), when ENSO has a strong influence on the future statistical behavior of MJO activity. In forecasts initialized during the active-MJO period (February), both ENSO and the current state of MJO are significant predictors, but the predictive information provided by the large-scale convective regimes in is found to be smaller than in the early-season forecasts.