Hydrographical Flow Modelling of the River Severn Using Particle Swarm Optimization

Abstract

Abstract A model is presented to model hydrographical flow, which we apply to flood forecasting in the River Severn catchment area. The approach uses Particle Swarm Optimization (PSO), a swarm computation heuristic, to produce a predictive model of hydrographical flow. Hydrological flow data from 1980 to 1990 are considered, comprising the daily average flow through the River Severn and its tributaries. PSO models are developed from each year of data and are applied to predict flow in the other 10 years; model performance is shown to be largely independent of the training year, suggesting the catchment system is stable and the approach is robust. Importantly, and in contrast to most of the existing alternatives, flow is derived from data measurements taken 2 days previously, as demanded for early-warning flood prediction. The cross-validated model for prediction of extreme (Q95) events R2 = 0.96, significantly improving upon multiple linear regression R2 = 0.93, the best performing of current existing methods.