Benchmarking prediction skill in binary El Niño forecasts

Abstract

Reliable El Niño Southern Oscillation (ENSO) prediction at seasonal-to-interannual lead times would be critical for different stakeholders to conduct suitable management. In recent years, new methods combining climate network analysis with El Niño prediction claim that they can predict El Niño up to 1 year in advance by overcoming the spring barrier problem (SPB). Usually this kind of method develops an index representing the relationship between different nodes in El Niño related basins, and the index crossing a certain threshold is taken as the warning of an El Niño event in the next few months. How well the prediction performs should be measured in order to estimate any improvements. However, the amount of El Niño recordings in the available data is limited, therefore it is difficult to validate whether these methods are truly predictive or their success is merely a result of chance. We propose a benchmarking method by surrogate data for a quantitative forecast validation for small data sets. We apply this method to a naïve prediction of El Niño events based on the Oscillation Niño Index (ONI) time series, where we build a data-based prediction scheme using the index series itself as input. In order to assess the network-based El Niño prediction method, we reproduce two different climate network-based forecasts and apply our method to compare the prediction skill of all these. Our benchmark shows that using the ONI itself as input to the forecast does not work for moderate lead times, while at least one of the two climate network-based methods has predictive skill well above chance at lead times of about one year.

Physics-based 18-month lead forecast of Indian Summer Monsoon Rainfall

“…This becomes a 'boon' rather than a 'bane'. This concept has been successfully used in extending lead of useful prediction of ENSO 19,27 as well as in extending the skill of useful prediction of East Asian monsoon 18 . The strength of our physics-based hybrid model is the ability of the model to successfully predict all extremes of ISMR (droughts/ oods) within the independent veri cation period of 31 years (1980 to 2011) for a mediumlead of 5-month and at a long-lead of 18-month forecasts.…”