Abstract

Modeling of sediment transport in relation to changing land-surface conditions against a background of considerable natural variability is a challenging area in hydrology. Bayesian dynamic linear models (DLMs) however, offer opportunities to account for non-stationarity in relationships between hydrologic input and basin response variables. Hydrologic data are from a 40 years long record (1951–1990) from the 5905 km² Yadkin River basin in North Carolina, USA. DLM regressions were estimated between log-transformed volume-weighted sediment concentration as a response and log-transformed rainfall erosivity and river flow, respectively, as input variables. A similar regression between log-transformed river flow and log-transformed basin averaged rainfall was also analyzed. The dynamic regression coefficient which reflects the erodibility of the basin decreased significantly between 1951 and 1970, followed by a slowly rising trend. These trends are consistent with observed land-use shifts in the basin. Bayesian DLMs represent a substantial improvement over traditional monotonic trend analysis. Extensions to incorporate multiple regression and seasonality are recommended for future applications in hydrology.