Improving Streamflow Prediction Using Uncertainty Analysis and Bayesian Model Averaging

Abstract

Hydrological modeling has been used worldwide as an important tool to evaluate the consequences of land cover and land use change on hydrological processes. However, the lack of spatial-temporal rainfall and runoff data have compromised the reliability of the results in several regions of Brazil. In this study, the authors investigated the use of uncertainty analysis and Bayesian model averaging (BMA) as a tool for improving streamflow estimates in the Ribeirão da Onça Basin (ROB), located in southeastern Brazil. They used a set of two precipitation data sources (ground and remote sensing data) and different spatial interpolation schemes as input data for the Soil and Water Assessment Tool (SWAT) model, resulting in five model configurations. These models were submitted to automatic calibration and uncertainty analysis through the sequential uncertainty fitting ver-2 (SUFI-2) method. Then, the BMA method was used to merge those different model configuration results into a single probabilistic prediction, thereafter compared among themselves. An analysis of the accuracy and precision of all simulations produced by the precipitation ensemble members against the BMA simulation supports the use of the latter as a suitable framework for streamflow simulations at the ROB. Furthermore, the approaches evaluated in this study may be used to improve streamflow predictions in ungauged or data-scarce basins.