Data-based mechanistic modelling and forecasting of hydrological systems

The paper presents a data-driven approach to the modelling and forecasting of hydrological systems based on nonlinear time-series analysis. Time varying parameters are estimated using a combined Kalman filter and fixed-interval-smoother, and state-dependent parameter relations are identified leading to nonlinear extensions to common time-series models such as the autoregressive exogenous (ARX) and general transfer function (TF). This nonlinear time-series technique is used as part of a data-based mechanistic modelling methodology where models are objectively identified from the data, but are only accepted as a reasonable representation of the system if they have a valid mechanistic interpretation. To this end it is shown that the TF model can represent a general linear storage model that subsumes many common hydrological flow forecasting models, and that the rainfall-runoff process can be represented using a nonlinear input transformation in combination with a TF model. One advantage of the forecasting models produced is that the Kalman filter can be used for real-time state updating leading to improved forecasts and an estimate of associated forecast uncertainty. Rainfall-runoff and flood routing case studies are included to demonstrate the power of the modelling and forecasting methods. One important conclusion is that optimal system identification techniques are required to objectively identify parallel flow pathways.