• 14 Dec 18
  • Posted by admin
  • 0 Comments

Uncertainty and climate predictions in BINGO

Increasing levels of atmospheric CO2 from anthropogenic sources are causing global warming. The global warming signal in response to higher CO2 levels manifests itself as a long-term, gradual trend towards higher temperatures across the globe. Alongside this long-term trend, the climate system has its own natural variability, ranging on temporal scales from annual, to decadal, to multi-decadal. The magnitude of this “internal variability” can be particularly strong at regional scales and can act to either damp or amplify the long-term climate change signal.

The field of decadal predictions seeks to predict the near-future climate, up to 10 years in advance, by predicting/modelling this internal variability of the climate system. This is achieved by initialising an earth system model with observed states of the climate which are known to slowly evolve and exert control on climate – for example ocean temperatures, salinity and sea-ice cover. The “long-term memory” of such climate-system components is key to any skill which decadal predictions may have. The field of decadal predictions is still relatively new, but to date skill has been demonstrated in predicting, amongst other things, seasonal mean temperatures and cyclone frequencies.

The BINGO project utilises decadal predictions from the MiKlip project. Due to the inherent uncertainty in estimating initial conditions, a 10-member ensemble of decadal predictions is created by initialising the earth system model separately with 10 different sets of initial conditions. The sensitivity of the predictions to perturbations in the initial state help to indicate the levels of confidence in the predictions, i.e. predictions highly sensitive to initial perturbations will exhibit large intra-ensemble differences and hence lower levels of certainty. The ability of decadal predictions to provide skilful predictions of extreme events remains an open question.

For prediction of localized extremes such as intense thunderstorms, skilful prediction is particularly challenging and has not yet been demonstrated. This is because such events are so rare and small-scale in nature that any increased risk cannot hope to be captured in a 10-member, 10-year ensemble. The occurrence of such extremes at a given location is also influenced by factors whose variance may not be discernible at the decadal time-scale, or may not even be predictable at all beyond short lead times. For larger-scale extremes, such as continental-scale heat waves or droughts, there are stronger grounds to suggest that such events can potentially be predicted. This, of course, is predicated on the extreme in question having been caused, or at least influenced, by components of the earth system which vary on the decadal scale – which is not necessarily the case for all large-scale extremes.