Transcending limitations of stationarity and the return period: process-based approach to flood estimation and risk assessment

✍ Scribed by Murugesu Sivapalan; Jos M. Samuel

• Publisher: John Wiley and Sons
• Year: 2009
• Tongue: English
• Weight: 131 KB
• Volume: 23
• Category: Article
• ISSN: 0885-6087
• DOI: 10.1002/hyp.7292

✦ Synopsis

Traditional flood frequency analysis and estimation are underpinned by the critical assumption of stationarity (I.E. Aust., Institution of Engineers Australia, 1987; FEH, 1999). The widely used notion of "return period" is a classical concept arising from assumed invariance of the probability distribution function (pdf) of annual maximum flood peaks over the period of flood records and/or design life of hydraulic structures. These assumptions, i.e. of stationarity, independent and identical pdf of flood peaks, and the concept of the return period, underpin traditional flood frequency analysis and flood risk assessment are no longer appropriate in a fast changing environment, and thus pose serious problems for flood management (Milly et al., 2008). The main causes of non-stationarity are human-induced land use and land cover changes, other human interferences in the water cycle through dams and other flood control structures, and greenhouseinduced climate change. As an example of the effect of climate change, Coulibaly and Dibike (2004) reported, on the basis of three different downscaling approaches of Global Climate Model (GCM) outputs, an overall increasing trend in the frequencies of flood events for the 2020s, 2050s and 2080s as compared to the 1960-2000 period in major tributaries within the Saguenay catchment, Quebec, Canada. Non-stationarity can also arise due to low frequency variations of climate, such as inter-annual (El Nino-Southern Oscillation (ENSO) cycle) and inter-decadal [Interdecadal Pacific Oscillation (IPO) cycle] variations, and shifts of climate of uncertain origins or causes. This is especially relevant when available flood records are relatively short, as demonstrated by Kiem et al. (2003) in the Newcastle region of Australia, and by Samuel and Sivapalan (2008a,b) in the Perth, Newcastle and Darwin regions. There is therefore an urgency for a new flood frequency analysis framework, and the revamping of flood estimation procedures used worldwide to accommodate the non-stationarities arising from such changes or long-term variabilities. Previously purely data-based statistical approaches have been proposed for dealing with non-stationarity: they involved detecting and separating the temporal trends in the observed flood record that may be the root cause of the non-stationarity (Cunderlik and Burn, 2003). However, there are limits to how far purely data-based techniques could be used to extrapolate beyond the observed record. It is generally felt that methods based on improved process understanding can be more effective in assessing the effects of land use and land cover changes, and climate change, on flood frequencies (Cunnane, 1985; Calder, 1993). In this paper, we articulate a new flood frequency analysis framework that advances current methods at least in two respects: (i) it involves a more process-based, derived flood frequency approach capable of incorporating the effects of environmental change and longterm variability, and (ii) it overcomes the limitations of the assumptions of stationarity and of independent and identical pdfs of annual