Flood Frequency Analyses with Metastatistical Extreme Value Distribution applied to Simulated River Flows

Abstract:

In flood frequency analysis (FFA), sufficient data are essential to obtain reliable estimates particularly to extrapolate the return periods of floods beyond the gauged records. In practice, short record lengths and poorly gauged river basins are encountered in many parts of the world, both limiting FFA accuracy. Proposed FFA approaches for improving hydrological prediction in ungauged or poorly gauged watersheds can be based on simulated flows from distributed hydrological models forced with atmospheric reanalysis datasets, which have the capability to incorporate a variety of spatially varying land characteristics and hydro-climatic conditions. However, hydrological model simulations generally exhibit important biases in the estimation of flood extremes. As traditional FFA relies on extreme value statistical distribution methods applied to extremes, quantile estimates usually exhibit large sensitivity on peak discharge errors in extreme events and ignore abundant information associated with the entire peak flow record. To address this issue, the Metastatistical Extreme Value Distribution (MEVD), a novel methodology that relaxes some of the assumptions of extreme value theory, is applied here on simulated daily streamflow time series. We deployed a raster-based distributed hydrological model, Coupled Routing and Excess Storage (CREST), in the Connecticut River Basin, and forced it by atmospheric reanalysis precipitation. In a recent study it was shown that although the model exhibited significant skill in prediction of the river flows (efficiency coefficient 0.65-0.8), simulations suffered from peak discharge errors. We compare MEVD with a variety of traditional statistical methods (GEV, GPD, LP3) and apply them to 40 years of simulated flows, to evaluate their accuracy against extreme value analysis derived from observed streamflows at selected USGS stations. This analysis will shed light on the role of MEVD in alleviating some of the hydrological model uncertainty relative to the traditional methods due to its lower dependence on the CREST’s peak flow simulation error. Doing so we expect to demonstrate uses of hydrologic model-simulated flows in hydrologic extreme value analysis of ungauged basins, and expands the practicality of MEVD method for regional flood frequency estimation.