Finding a parsimonious suspended sediment transport model structure

Prediction of sediment load in rivers is crucial for addressing several theoretical and practical problems. While purely mechanistic models of sediment load transport suffer from the limitation of being impractical for application due to constraints in finding adequate field data, as well as computational resources, purely data-based models are not fully reliable due to fitting and interpretability issues. Conceptual approaches provide a middle ground to develop models that capture the best of the two above approaches - incorporating key physical processes while exploiting empirical observations. This study adopts the unit sediment graph method, which decouples sediment generation and sediment routing, akin to the unit hydrograph method which decouples runoff generation and runoff routing. Banasik and Walling (1996) expressed effective sediment (ES), the amount of sediment that ultimately exits in the catchment, as a general function of precipitation (R) as well as effective precipitation (ER): ES = X center dot R(t)m center dot ER(t)n. For the sake of simplification, either m or n can be assumed to be zero, giving us two derivative models. In this study, we assume that both m and n are equal to 1, leaving only X free for calibration. Sediment routing is performed using the sediment rating curve as well as by directly routing effective sediment with two parallel linear reservoirs. Effective rainfall and discharge are estimated with the HBV hydrological model. We tested the eight generation-routing model combinations considering daily sediment load data from USGS and precipitation data from NOAA for 18 catchments. The proposed sediment generation model in combination with the direct routing type appears to be the most reliable, with a median NSE and R2 equal to 0.47 and 0.63 respectively. The results here support the idea that basin-scale sediment load prediction can be performed with a simple, parsimonious model applicable to a wide variety of geographical regions.