Identification of drivers of flood seasonality using Circular Generalized Linear Models

Abstract:

Seasonality of floods is important in assessment of flood risk. Identification of attributes that drive flood seasonality aids in understanding and quantifying contributions of flood generating mechanisms such as extreme precipitation, soil moisture excess and snowmelt. These attributes may be further used in formation of homogeneous regions to understand regional hydrological behaviours. In this study, we use a Circular Generalized Linear Model (GLM) framework to study changes in seasonality statistics of peak flow at 163 sites in the Northwest United States. Circular GLM is implemented in a Bayesian estimation approach. Flood seasonality defined in terms of timing of peak flow is characterised by the circular von Mises distribution that has a constant concentration parameter, while the mean direction is modeled to have a linear association with covariates such as seasonality of precipitation, soil moisture excess and snowmelt. Therefore, the Circular GLM-based framework considers non-stationarity in flood seasonality. The significance of the covariate is examined based on the 95% credible interval of the posterior distribution of the parameters. Soil moisture is identified to be the dominant factor for stations that show change in flood seasonality. Our analysis shows that antecedent conditions, and not extreme precipitation, drives changes in flood seasonality in Northwest United States.