A Global-Scale Analysis of Hydrologic Extremes using Hidden Climate Indices

Hydrologic extremes (floods and intense precipitations) are among Earth’s most common natural hazards and cause considerable loss of life and economic damage. Describing their space-time variability in relation to climate is hence important for scientific and operational purposes. This presentation describes the use of an innovative probabilistic framework to jointly analyze global datasets of floods and extreme precipitations. This framework is based on the idea that the temporal variability of the data is induced by hidden climate indices that are unknown and therefore have to be estimated directly from the data. This is to be contrasted with the usual approach using predefined standard climate indices such as ENSO or NAO for this purpose. In statistical terms, a two-level hierarchical model is used. The first level jointly describes floods and intense precipitations, with hidden climate indices treated as latent variables. The second level describes the temporal variability of the hidden climate indices (including trend and persistence components), and the spatial variability of their effects.

This model is applied to station-based datasets describing seasonal maxima of streamflow and precipitation at the global scale, corresponding to more than 3,000 stations over a 100-year period (1916-2015). Several hidden climate indices governing the joint temporal variability of streamflow and precipitation data are identified. They affect floods and intense precipitations over large (continental) spatial scales and in a highly structured way. Overall these hidden climate indices do not present noticeable trend or persistence components, suggesting that they represent mostly interannual modes of variability. By contrast, when the same model is applied to precipitation data only, the estimated hidden climate indices are affected by stronger and mostly upward trends: this confirms that increasing intense precipitations do not identically translate into increasing floods, as highlighted by the latest IPCC report. Finally, we demonstrate that hidden climate indices can be predicted to some degree from atmospheric variables such as pressure, wind, temperature etc. This allows reconstructing the probability of occurrence of hydrologic extremes in the distant past using long reanalyses such as 20CR.