On the predictability of SSTA indices from CMIP5 decadal experiments

Abstract

Sea surface temperature anomaly climate indices in the tropical Pacific and Indian Oceans are statistically significant predictors of seasonal rainfall in the Indo-Pacific region. On this basis, this study evaluates the predictability of nine such indices, at interannual timescales, from the decadal hindcast experiments of four general circulation models. A Monte Carlo scheme is applied to define the periods of enhanced predictability for the indices. The effect of a recommended drift correction technique and the models' capabilities in simulating two specific El Niño and La Niña events are also examined. The results indicate that the improvement from drift correction is noticeable primarily in the full-field initialized models. Models show skillful predictability at timescales up to maximum a year for most indices, with indices in the tropical Pacific and the Western Indian Ocean having longer predictability horizons than other indices. The multi model ensemble mean shows the highest predictability for the Indian Ocean West Pole Index at 25 months. Models simulate the observed peaks during the El Niño and La Niña events in the Niño 3.4 index with limited success beyond timescales of a year, as expected from the predictability horizons. However, our study of a small number of events and models shows full-field initialized models outperforming anomaly initialized ones in simulating these events at annual timescales.

An Assessment of Drift Correction Alternatives for CMIP5 Decadal Predictions

“…This effect is likely to influence the performance of ICDC to a lesser extent. This is another motivation for considering results obtained by optimizing over the period of skillful predictability (1–12 months; Choudhury et al, ) to be more accurate than those obtained by optimizing over the longer timescale.…”

“…This effect is likely to influence the performance of ICDC to a lesser extent. This is another motivation for considering results obtained by optimizing over the period of skillful predictability (1-12 months; Choudhury et al, 2015) to be more accurate than those obtained by optimizing over the longer timescale. Figure 2 shows the best method of drift correction (as markers on the bars) for the different climate indices (as the abscissa) along with the reduction in RMSE compared with the RMSE from MDC (as the ordinate), for all the models (as the colors).…”

“…Recent attempts to test the skill of these climate prediction systems have reported predictability to vary from few months to few years (Boer et al, ; Choudhury et al, ; Lienert & Doblas‐Reyes, ; Mehta et al, ). This may be because the oceanic or atmospheric state is inherently unpredictable beyond a few seasons depending on the variable and region.…”

“…Further, in almost all cases, the multimodel ensemble outperforms the individual models, and thus, the study conclusively suggests using forecasts based on multimodel averages. We also show some additional benefit to be gained by drift correcting each model/metric using their best correction method prior to model averaging and suggest that the results presented here would help potential users expend time and resources judiciously while dealing with outputs from these experiments.Recent attempts to test the skill of these climate prediction systems have reported predictability to vary from few months to few years Choudhury et al, 2015;Lienert & Doblas-Reyes, 2013;Mehta et al, 2013). This may be because the oceanic or atmospheric state is inherently unpredictable beyond a few seasons depending on the variable and region.…”

Seasonal prediction skill of East Asian summer monsoon in CMIP5 models

“…The main datasets used for comparison in this study include the following: (1) monthly precipitation data from the Global Precipitation Climatology Project (GPCP; Adler et al, 2003); (2) monthly circulation data from the ECMWF Interim reanalysis (ERA-Interim; Dee et al, 2011); and (3) monthly mean SST from the National Oceanic and Atmospheric Administration (NOAA) improved Extended Reconstructed SST version 4 (ERSST v4; Huang et al, 2015). All the model data and the comparison data are remapped onto a common grid of 2.5 • × 2.5 • by bilinear interpolation to reduce the uncertainty induced by different data resolutions.…”