Sampling variability and the changing ENSO–monsoon relationship

The impact of sampling variability on the correlation between all-India rainfall (AIR) and the El Niño-Southern Oscillation is investigated in a large ensemble of seasonal climate simulations made using the European Centre for Medium-Range Weather Forecasting Ensemble Prediction System at T319 (64 km). The analyzed runs consist of 51 ensemble members initialized each May 1 for the period 1980–2011 and integrated for 7 months. 10,000 pairs of 32-year timeseries of June–September (JJAS) mean AIR and NINO3 indices are created from this database by randomly drawing one of the 51 ensemble members for each year. The correlation between each pair of AIR and NINO3 series is then calculated, generating a distribution of AIR–NINO3 correlation values. The model is reinitialized with observations each May 1 and thus all members are drawn from the same background state by construction and any differences in correlation are attributable to sampling variability. The spread in the calculated correlation values and the differences between 32-year segments are sufficient to explain the observed variations in AIR–NINO3 correlation since the beginning of the 1900s, including the sharp decrease in correlation strength since the late 1970s. Sampling variability thus represents a strong null hypothesis for the observed changes and one that cannot be rejected at the 95 % level based on our simulations.