The suitability of the physically based model SHETRAN for simulating sediment generation and delivery with a high degree of spatial (20 m) and temporal (sub-hourly) resolution was assessed through application of the model to a 167-km2 catchment leading to an estuary in New Zealand. By subdividing the catchment and conducting calculations on a computer cluster for a 6-month hydrology initialisation period, it was possible to simulate a large rainfall event and its antecedent conditions in 24 h of computation time. The model was calibrated satisfactorily to catchment outlet flow and sediment flux for a large rainfall event in two subcatchments (~2 km2). Validation for a separate subcatchment was successful for flow (Nash–Sutcliff efficiency of 0.84) with a factor 2.1 over-prediction for sediment load. Validation for sediment at full catchment scale using parameters from the subcatchment scale was good for flow but poor for sediment, with gross under-estimation of the dominant stream sources of sediment. After recalibration at catchment scale, validation for a separate event gave good results for flow (Nash–Sutcliff efficiency of 0.93) and sediment load within a factor of two of measurements. An exploratory spatially explicit landslide model was added to SHETRAN, but it was not possible to test this fully because no landslides were observed in the study period. Application to climate change highlighted the non-linear response to extreme rainfall. However, full exploration of land use and climate change and the evaluation of uncertainty were severely constrained by computational limitations. Subdivision of the catchment with separate stream routing is suggested as a way forward to overcome these limitations. Copyright © 2011 John Wiley & Sons, Ltd.