*3.3. Model Validation*

Using calibrated parameters, we validated the model performance on the rainfall event of 28 August 1982. The return period of the event is one year. It started at 23:30 and lasted 4.5 h with an average rainfall intensity of 10.4 mm/h. The simulation was initialized on 1 January 1981 to assure the model reached a relatively stable condition before the rainfall event. Figure 6 shows the simulated and measured hydrographs at the outlet. The model was able to reproduce the measured hydrograph. The statistical results shown in Table 6 indicate satisfactory simulation skill of the model with a high NSE (0.86). We notice that the simulated rising limb starts earlier than the measurement and the falling limb drops slowly and lasts longer. The simulated peak is within 20 min of the measured one and the simulated peak value (27 m3/s) is 26 percent lower than the measurement (36.4 m3/s). Simulated water discharge at the outlet is within 10 percent of the measurement.

**Figure 6.** Measured (dash line) and simulated (solid line) streamflow at MSGC01 for validation event.

The comparison between simulated and measured sediment concentration and sediment flux is shown in Figure 7. The model can catch the shape of the observed sediment graphs but with an overall overestimation for both sediment concentration and flux. Sediment flux is better simulated than sediment concentration with NSE and KGE 0.09 (vs. −2.26 for concentration) and 0.34 (vs. −0.15 for concentration), respectively. A comparable model performance in sediment concentration simulation is also reported by Elliot [35], which was partially attributed to the fact that simulation of sediment concentration involves an error in both sediment simulation and runoff simulation. Due to the absence of observation data before 2:00, 28 August 1982, it is impossible to calculate the sediment yield during the entire event. Yet it is reasonable to assume that the real sediment yield is larger than the estimation based on the available data (9100 t). Thus, we consider that our model-simulated sediment yield (14,950 t) is acceptable.

**Table 6.** Performance of the simulation of streamflow, sediment concentration, and sediment flux for validation event.


In spite of the acceptable performance of the model in simulating the sediment yield for both events, the model tends to overestimate the sediment yield for validation events while underestimating it for calibration events. This difference in model behavior should be partly due to the difference in initial soil conditions of the two events. According to Rojas [23], the initial soil condition could be much wetter during the validation event, since a series of preceding rainfall events brought >110 mm rain in the previous month. Yet there was only one preceding rainfall event bringing ~15 mm rainfall before the calibration event. In this case, the soil erodibility factor calibrated for the calibration event might be too large for the validation event, as soil erodibility tends to be smaller in wet conditions than in dry conditions [36]. Thus, the sediment yield during the validation event is overestimated by the model.

**Figure 7.** Measured (dash line) and simulated (solid line) sediment concentration (**a**), and sediment flux (**b**) at MSGC01 for validation event.
