Over 1,500 dams have been removed in the United States within the last 30 years, predominantly due to the economic constraints and as part of riverine restoration efforts. At the same time, the number of dam removals is expected to further increase in coming years owing to aging infrastructure and high rehabilitation costs. Coupled with the limited number of removals that has been closely monitored and documented, there is a need to develop innovative approaches that enhance our predictive and quantitative understanding of river response to dam removal. In this study, we developed a modeling framework to estimate reflectance-based turbidity in rivers, which can be used as a proxy for understanding sediment transport processes. The accuracy and applicability of the framework were assessed when reconstructing the fluvial sediment pulses generated after the well-documented removal of two dams on the Elwha River in Washington, USA. We found that inclusion of physically based predictors helped in increasing both the framework's accuracy and transferability. Three data splitting approaches (referred to as M1, M2, and M3) were considered based on model application and tested to reconstruct the sediment pulse following the dam removals. M1 was the most accurate during the testing phase (NSE: 0.67 RSR: 0.60), followed by M2 (NSE: 0.3 RSR: 0.8), and finally M3 (NSE: 0.1 RSR: 0.9). For the Elwha River dam removals, the three approaches were successful in simulating mean turbidity values, but under and over predicted peak and low values, respectively. Future work will incorporate a physics-based numerical model to complement the data-driven approach, all while accounting for the different sources of uncertainty during the intermediary steps of the integrated approach.
