Optimal localized observations for advancing beyond the ENSO predictability barrier

The existing 20-member ensemble of 50 yr ECHAM5/MPI-OM simulations provides a reasonably realisti Monte Carlo sample of the El Niño–Southern Oscillation (ENSO). Localized observations of sea surface temperature (SST), zonal wind speed and thermocline depth are assimilated in the ensemble using sequential importance sampling to adjust the weight of ensemble members. We determine optimal observation locations, for which assimilation yields the minimal ensemble spread. Efficient bservation locations for SST lie in the ENSO pattern, with the optimum located in the eastern and western Pacific for minimizing uncertainty in the NINO3 and NINO4 index, respectively. After the assimilation of the observations, we investigate how the weighted ensemble performs as a nine-month probabilistic forecast of the ENSO. Here, we focus on the spring predictability barrier with observation in the January– March (March–May) period and assess the remaining predictive power in June (August) for NINO3 (NINO4). For the ECHAM5/MPI-OM ensemble, this yields that SST observations around 110 W and 140 W provide the best predictive skill for the NINO3 and NINO4 index, respectively. Forecasts can be improved by additionally measuring the thermocline depth at 150 W.