Search

Monthly streamflow forecasts have important practical applications in short-term water resources management, e.g., water allocation for different users, flooding prevention, and drought mitigation. This study focuses on developing categorical streamflow forecasts for flooding mitigation purpose, which can be used as a critical component in an early flood warning system. A Bayesian logistic regression approach is proposed to use antecedent streamflow and forecasted precipitation from General Circulation Models (GCMs) and derive the probability of streamflow greater than threshold streamflow. The logistic regression model is Hierarchical Bayesian Modeling that assumes Bernoulli distribution for monthly and Normal distribution for the parameters in the logistic function. To accommodate outliers in the modeling dataset, an additional parameter is added to the Bayesian modeling framework to make it a more robust approach. The Bayesian Logistic Regression is implemented in JAGS and posterior distributions of model parameters are estimated from Markov Chain Monte Carlo (MCM) chains. The proposed method is applied to a watershed in Hunan Province located in the middle south of China. Precipitation and streamflow data in the years 1960–2012 were used to estimate the model parameters' posterior distributions. The model's performance is tested for monthly streamflow data in the years 2013–2017, using one-month-ahead precipitation forecasts from GCMs. The model is superior to climatology, the reference model, in terms of the accuracy of hit rates. Potential improvement to the model is also discussed. Although the proposed method is demonstrated for the study area, it can be applied to other regions with similar applications. [ABSTRACT FROM AUTHOR]