A principal component regression approach to simulate the bed-evolution of reservoirs

Long-term simulation of reservoir sedimentation suffers from process complexity, lack of data, as well as high computational cost. This work presents an efficient data-driven modeling approach to simulate the spatio-temporal dynamics of bed-evolution reservoirs using principal components regression. The daily bed-evolution of a validated numerical model was used as an input. The first four principal components contributed to some 90% of the total variance of bed-evolution. Multiple linear regression between the eigenvectors of the first four principal components with the inflow discharge, suspended sediment concentration, and differential discharge was able to reconstruct the spatio-temporal bed-evolution. Predictions with similar initial morphological condition performed reasonably. The work is a step forward to advance the assimilation of numerical and data-driven approaches in modeling long-term sedimentation of reservoirs.