Application of generalized additive models in regional flood frequency analysis : a case study for Victoria, Australia

![CDATA[Design flow estimation for ungauged catchments is considered to be a challenging task in hydrology. Regional Flood Frequency Analysis (RFFA) can be used to estimate design flow for ungauged catchments. Commonly adopted RFFA methods include the index flood method, the rational method and the quantile regression technique. This paper examines the applicability of the generalized additive model (GAM) in RFFA. GAM establishes a well parameterized modelling framework which describes the multivariate and nonlinear characteristics of a complex dataset. This approach allows flexible specification of reg ession splines to represent the functional relationships between a response variable (i.e. flood quantile) and a suite of temporal and spatial covariates that can be continuous or discrete. This is done using a link function and smooth functions of the covariates such as catchment characteristics. In this study, both the GAM and log-linear model (LLM) are applied in RFFA to a data set of 114 catchments from Victoria State in Australia. Based on independent testing, it was found that GAM generally provides more accurate flood quantile estimates than the LLM Further study is needed to confirm this finding.]]