Influences on flood frequency distributions in Irish river catchments
- School of Civil, Structural and Environmental Engineering, University College Dublin, Dublin, Ireland
Abstract. This study explores influences on flood frequency distributions in Irish rivers. A Generalised Extreme Value (GEV) type I distribution is recommended in Ireland for estimating flood quantiles in a single site flood frequency analysis. This paper presents the findings of an investigation that identified the GEV statistical distributions that best fit the annual maximum (AM) data series extracted from 172 gauging stations of 126 rivers in Ireland. Analysis of these data was undertaken to explore hydraulic and hydro-geological factors that influence flood frequency distributions. A hierarchical approach of increasing statistical power that used probability plots, moment and L-moment diagrams, the Hosking goodness of fit algorithm and a modified Anderson-Darling (A-D) statistical test was followed to determine whether a type I, type II or type III distribution was valid. Results of the Hosking et al. method indicated that of the 143 stations with flow records exceeding 25 yr, data for 95 (67%) was best represented by GEV type I distributions and a further 9 (6%) and 39 (27%) stations followed type II and type III distributions respectively. Type I, type II and type III distributions were determined for 83 (58%), 16 (11%) and 34 (24%) stations respectively using the modified A-D method (data from 10 stations was not represented by GEV family distributions). The influence of karst terrain on these flood frequency distributions was assessed by incorporating results on an Arc-GIS platform showing karst features and using Monte Carlo simulations to assess the significance of the number and clustering of the observed distributions. Floodplain effects were identified by using two-sample t-tests to identify statistical correlations between the distributions and catchment properties that are indicative of strong floodplain activity. The data reveals that type I distributions are spatially well represented throughout the country. While also well represented throughout the country, the majority of type III distributions appear in areas where attenuation influences from floodplains are likely. The majority of type II distributions appear in a single cluster in a region in the west of the country that is underlain by karst but importantly, is characterised by shallow of glacial drift with frequent exposures of rock outcrops. The presence of karst in river catchments would be expected to provide additional subsurface storage and in this regard, type III distributions might be expected. The prevalence of type II distributions in this area reflects the finite nature of this storage. For prolonged periods of rainfall, rising groundwater levels will fill karst voids, remove subsurface storage and contribute to recharge related sinkhole flooding. Situations where rainfall intensities exceed karst percolation rates also produce high levels of surface runoff (discharge related flooding) that can promote type II distributions in nearby river catchments. Results therefore indicate that in some instances, assuming type I distributions is incorrect and may result in erroneous estimates of flood quantiles at these locations. Where actual data follows a type II distribution, flood quantiles may be underestimated by in excess of 35% and for type III distributions, overestimates by over 25% can occur.