River Flow Forecasting

Overview

This project models and forecasts flow rates of the Bedon River using daily rainfall as a predictor, with data from hydrological stations in southern Colombia spanning January 2006 to April 2009. The work compares univariate and multivariate approaches across short (7-day), medium (30-day), and long (365-day) forecast horizons, motivated by water resource management decisions like reservoir releases, irrigation planning, and flood preparation. Different models perform best at different horizons: a multiple linear regression with AR(1) errors at the short horizon, an ensemble averaging ARMA, signal-plus-noise, VAR, MLR, and multivariate MLP forecasts at the medium horizon, and a signal-plus- noise model at the long horizon. In short, the best model depends on the forecasting horizon.

Skills

R · Time series analysis · Forecasting · ARIMA · VAR · Ensemble modeling