Using new and emerging ideas about the growth of singular vectors ('optimal perturbations') in dynamical systems, the dynamics of error growth and predictability in an intermediate coupled model of the El Nino Southern Oscillation (ENSO) is investigated. A mechanism is identified for error growth associated with penetrative-convection anomalies in the atmosphere. Conditions for error growth via this mechanism are most favourable in the central Pacific where sea surface temperatures (SSTs) are relatively warm, and where changes in SST are moderately sensitive to vertical movements of the main oceanic thermocline. The singular vectors of the coupled system were computed using the tangent-linear coupled model and its adjoint. The singular-value spectrum was found to be dominated by one singular vector at all times of the year. The potential for error growth in the coupled model, measured in terms of the growth of energy of the dominant singular vector, is found to vary seasonally, being greatest during the boreal spring. These seasonal variations are associated with the seasonal cycle in SST. During boreal spring and early summer, the SST in the central Pacific is at its maximum, at which time conditions are most favourable for error growth. Springtime is also the time of the 'predictability barrier' for ENSO. The potential for error growth is also influenced by the ENSO cycle itself. The results suggest that error growth will be enhanced during the onset of El Nino and suppressed during the onset of La Nina, which indicates that El Nino may be less predictable than La Nina.
