LAUSR

The curve number (CN) of a watershed varies spatially due to heterogeneity, and temporally due to changes in soil moisture, land cover, temperature, and other processes. The conventional event-scale lumped-parameter CN method lacks the capability to account for spatiotemporal variations, which diminishes the accuracy of its predictions. Heterogeneity causes several parameters of the CN method, including the initial abstraction (Ia), to vary with event rainfall (P), so one way to account for heterogeneity is to treat Ia as a function of P. This modification to the CN method gives rise to variable Ia models. Including antecedent moisture (M) is a common way to account for the temporal variation of CN. This paper presents an improved method of including M, which when used together with variable Ia can allow for accounting of both spatial and temporal variability. A suite of models that use M and/or variable Ia was evaluated using published event-scale data from several studies along with rainfall-runoff observations from two small watersheds in South Carolina, USA. Including M in the CN models significantly improved the accuracy of the runoff predictions, whereas including variable Ia alone resulted in modest improvements. The best performance, NSE > 0.8, was achieved when both variable Ia and M were included together. These modifications significantly improve runoff predictions while only modestly increasing the complexity of the CN method.