Bayesian Framework for Uncertainty Quantification and Bias Correction of Projected Streamflow in Climate Change Impact Assessment

The study focuses on the uncertainty quantification and bias correction of hydrological projections using Bayesian applications. The climate change impact assessment on streamflow has been done using Soil and Water Assessment Tool (SWAT) model in Bharathapuzha river basin, India. The uncertainty quantification has been done by using Generalised Likelihood Uncertainty Estimation (GLUE) algorithm and the ensemble spread in the streamflow projections is quantified as the total uncertainty. A Hierarchical Bayesian Algorithm is adopted in the current study to remove the systematic bias in the projections of extreme streamflow. The approach established a probabilistic correction to the projected streamflow based on the biases in daily scale hindcast streamflow simulations with the corresponding observed historical streamflow data. The procedure is applied to the ensemble streamflow predictions for the Bharathapuzha catchment and over 10 times reduction in RMSE is observed in the bias corrected streamflow. The skill of the procedure in correcting the streamflow across different terciles is studied using the concept of reliability and significant improvement is observed in the reliability of high and medium flow ranges. The average width of the ensemble streamflow simulation band for the period 2021–2030 is seen to reduce from 5560 cumec to 2188 cumec after the correction procedure is applied. Highlight The parametric uncertainty in the streamflow projection is accounted by calibrating the hydrological model using the GLUE procedure. A Hierarchical Bayesian approach has been adopted for bias correction of the streamflow projections. A higher reliability is observed in the bias-corrected streamflow, especially in high and medium flow ranges. The average width of the ensemble streamflow simulation band has reduced from 5560 cumec to 2188 cumec after the bias correction.