Development of Discharge Ratings for Low-Slope Streams Under Tidal Effects Using Non-Parametric and Data-Driven Models

Estimation of flow discharge in natural streams has been traditionally obtained using simple stage-discharge relations known as rating curves. However, in situations where streams are characterized by low gradients and are subject to tidal flow reversal effects, no simple stage-discharge relation can be developed. Instead, a more complex relation is required in which discharge is related to the stage and other relevant hydraulic variables. The present study investigates this issue using advanced data-driven computational techniques such as neural networks and non-parametric regression analysis. The proposed techniques will be applied to low-gradient streams in the Vermillion river basin in southwestern Louisiana. The results of the predicted discharge estimates are validated using actual discharge measurements in order to assess the performance of the proposed models and their prediction accuracy. The study also addresses issues such as selection of number of input variables, sample size requirements, computational efficiency, and uncertainty bounds of the developed models.