To measure the performance and calibrate the model, we computed the model bias, i.e. the relative difference between predictions and observations, for the eight subcatchments. kb was the only calibration parameter, following the work of Vigiak et al. (2012): the authors suggest that CD 1530 IC0 is landscape independent so that calibration should be based on kb only. We selected the value that minimized the average bias for all catchments, and compared with the calibrated value for individual subcatchments. We then compared absolute predictions to observed data for both uncalibrated and calibrated runs of the model performance, to assess the degree to which calibration improved the average performance.
3.2.2. Accounting for additional sediment sources and sinks for absolute predictions
Reviewing the local literature confirmed that large uncertainties remain around the relative effect of stream bank erosion and instream deposition. Because these processes are a function of stream length, we assessed whether there was any correlation between model performance and this catchment characteristic; this analysis aims to test the hypothesis that the simple representation of stream bank erosion and instream retention, assumed to be negligible or to compensate each other, results in a model bias. We discuss the effect of these uncertainties in 3.3 and 5.2, in the context of other parameter uncertainties.