Uncertainty introduced by flood frequency analysis in projections for changes in flood magnitudes under a future climate in Norway

Abstract

Study region: The study considers 115 unregulated catchments in Norway, with areas from 6 to 15,449 km2 and flood generation regimes ranging from snowmelt-dominated to ‘mixed’ (snow-melt and rainfall) to fully rainfall-driven. Study focus: Bias-corrected EURO-CORDEX RCM output for RCP 8.5 is used to generate an ensemble of 500 hydrological simulations for assessing changes in flood magnitudes under a future climate. Flood estimates are based on three extreme value distributions (EVDs), Gumbel, Generalised Extreme Value and Generalised Pareto, with confidence intervals calculated using parametric bootstrapping, and uncertainty introduced into the ensemble by the flood estimation is evaluated using variance decomposition. Changes in EVD parameters under future conditions are also assessed. New hydrologic insight for the region: There are large differences in projected changes between catchments, with median estimates ranging from −48 % to +99 % for the 200 year flood. Flood magnitudes in all catchments with rainfall-dominant or mixed flood regimes are expected to increase. EVDs with a shape parameter (GEV and GPD) indicate larger increases in higher flood quantiles than the Gumbel distribution (e.g. by 5–8 percentage points for the 200 year flood). Flood frequency estimation contributes 30–52 % of the total ensemble range in individual catchments. Location and scale parameters generally increase in catchments with increasing flood magnitudes, and some catchments with mixed flood regimes also exhibit increases in the shape parameter under future conditions.