Uncertainty in Flow Time-Series Predictions in a Tropical Monsoon-Dominated Catchment in Northern Thailand

Abstract

Flow time-series data are crucial for water resource and flood management. In catchments where flow observations are not available or data are of poor quality, regionalization methods are needed to generate flow time series. Explicit attention to data quality and uncertainty can give insight into the value of information, model predictive capability, and priorities for new and better data. While such analyses are becoming more common in the literature, there is a particular lack of knowledge about suitable modeling approaches and data priorities in tropical, monsoon-dominated regions. To address this, a case study of 44 subcatchments of the upper Ping catchment in Thailand is reported. Three regionalized rainfall-runoff indices— rainfall-runoff coefficient, base flow index, and rainfall-runoff elasticity—are used to condition the rainfall-runoff model using a Bayesian approach. The model performance is tested at daily, monthly, and seasonal timescales in terms of accuracy, reliability, and sharpness in predicting observed flows. The regionalized model was considered to be imprecise at the daily timescale and thus unsuitable for flood studies. The model is more suitable for supporting water resources planning at the monthly and seasonal timescales, with recognition of its tendency to underestimate flows at the start and end of the monsoon. The model showed poor reliability in a few catchments, implying that assumptions made in the Bayes approach were not useful in these cases and a different approach to uncertainty analysis may be justified along with further efforts to improve the rainfall-runoff model and data sets used.