Streamflow forecasting using functional regression

TL;DR

This paper introduces functional linear models for streamflow forecasting, enabling the prediction of entire flow curves using meteorological data, and compares their performance with neural networks.

Contribution

It adapts functional linear models to hydrological forecasting, allowing analysis of entire streamflow curves instead of discrete points, and discusses their advantages over neural networks.

Findings

Functional models effectively analyze whole streamflow curves.

They reveal features hard to detect with traditional methods.

Comparison shows strengths and limitations relative to neural networks.

Abstract

Streamflow, as a natural phenomenon, is continuous in time and so are the meteorological variables which influence its variability. In practice, it can be of interest to forecast the whole flow curve instead of points (daily or hourly). To this end, this paper introduces the functional linear models and adapts it to hydrological forecasting. More precisely, functional linear models are regression models based on curves instead of single values. They allow to consider the whole process instead of a limited number of time points or features. We apply these models to analyse the flow volume and the whole streamflow curve during a given period by using precipitations curves. The functional model is shown to lead to encouraging results. The potential of functional linear models to detect special features that would have been hard to see otherwise is pointed out. The functional model is also compared to the artificial neural network approach and the advantages and disadvantages of both models are discussed. Finally, future research directions involving the functional model in hydrology are presented.