Estimating flood statistics from basin characteristics in Scotland
Acreman, Michael Charles
PublisherUniversity of St Andrews
MetadataShow full item record
Estimation of the probability of occurrence of future flood events at a site is frequently required for the design of bridges, culverts, dams and other river engineering works. This study considers a method for estimating the flood frequency distribution from the physical characteristics of the drainage basin for use in Scotland when adequate records of river discharge are not available. The data base collated includes 3071 station years of annual maximum flood peaks for 168 high quality gauging stations and 12 physical characteristics for each drainage basin. A linear regression model is derived which explains 91% of the variation in the average magnitude of floods using five physical characteristics indexing drainage area, rainfall, stream density, soil type and lake storage. This model appears robust over the range of basin types and shows no improvement when shrinkage or ridge regression is employed. Five physically homogeneous subsets of basins are derived using a clustering algorithm (NORMIX) and the same five characteristics, with the addition of an index of channel slope. For each of subsets 1, 3, 4 and 5, the individual dimensionless flood frequency distributions for each station are not significantly different from a single GEV distribution derived for that subset. Consequently these subsets are considered to be hydrologically homogeneous in addition to their physical homogeneity. Dimensionless regional flood frequency distributions are produced for each subset which allow the estimated average flood magnitude to be scaled to estimate floods of less frequent occurrence. These regional 'growth curves' imply a larger return period for a given magnitude flood than those from the Natural Environment Research Council Flood Studies Report (NERC, 1975). When the floods are described by a lognormal model which allows for cross-correlation between stations the respective return periods are smaller.