Prospective scenarios of the saltwater intrusion in an estuary under climate change context using Bayesian neural networks.

In recent decades, saltwater intrusion over some low-lying coastal regions was deteriorated by rising sea-level and decreasing streamflow in the context of climate change. Though physically-based hydrodynamic models are the most detailed means to simulate salinity processes, they are commonly restricted by data insufficiency issues both in spatial resolution and temporal lasting. This motivates us to build a statistical model enable simulation and scenario analysis for coastal salinity change with limited observations. A Bayesian neural network (BNN) model is built hereby to simulate salinity. It offers more precise estimation compared with the conventional artificial neural network. Meanwhile, the model gives the uncertainty behaviors of the final salinity simulation which is not available for other methods. Future scenarios of salinity change are constructed and analyzed in different time periods on the basis of the validated BNN model. Results indicate that the water quality over lower Pearl River is degrading along with more significant uncertainties. Further analysis suggests that streamflow alteration has a more direct impact on salinity variations than the sea-level change does. The method allows a profound analysis of the potential influence on water quality degradation in coastal and low-lying regions in support of water management and adaptation toward global climate change. [ABSTRACT FROM AUTHOR]