Climate-Informed Multi-Scale Stochastic (CIMSS) Hydrological Modeling: Incorporating Decadal-Scale Variability Using Paleo Data

Tuesday, 16 December 2014
Mark Andrew Thyer, University of Adelaide, Adelaide, Australia, Ben J Henley, University of Melbourne, Parkville, Australia and George A. Kuczera, University of Newcastle, Callaghan, Australia
Incorporating the influence of climate change and long-term climate variability in the estimation of drought risk is a priority for water resource planners. Australia’s highly variable rainfall regime is influenced by ocean-atmosphere climate mechanisms which induce decadal-scale variability in hydrological data. This talk will summarize research on the identification of appropriate models for incorporating decadal scale variability into stochastic hydrological models. These will include autoregressive, hidden Markov models and a Bayesian hierarchical approach which combines paleo information on climate indices and hydrological data into a climate informed multi-time scale stochastic (CIMSS) framework. To characterize long-term variability for the first level of the hierarchy, paleoclimate and instrumental data describing the Interdecadal Pacific Oscillation (IPO) and the Pacific Decadal Oscillation (PDO) are analyzed. A new paleo IPO-PDO time series dating back 440 yr is produced, combining seven IPO-PDO paleo sources using an objective smoothing procedure to fit low-pass filters to individual records. The paleo data analysis indicates that wet/dry IPO-PDO states have a broad range of run lengths, with 90% between 3 and 33 yr and a mean of 15 yr. Model selection techniques were used to determine a suitable stochastic model to simulate these run lengths. For the second level of the hierarchy, a seasonal rainfall model is conditioned on the simulated IPO-PDO state. Application to two high quality rainfall sites close to water supply reservoirs found that mean seasonal rainfall in the IPO-PDO dry state was 15%–28% lower than the wet state. Furthermore, analysis of the impact of the CIMSS framework on drought risk analysis found that short-term drought risks conditional on IPO/PDO state were far higher than the traditional AR(1) model.