*3.2. Model Validation*

To validate the HEC-RAS model with the SRC technique, it was run for another four years up to 2017. The output was compared with observed sediment deposits of 2017 and is presented in Figure 9.

**Figure 9.** Comparison of observed and SRC simulated bed levels during validation for 2017: (**a**) along the Tarbela Reservoir; (**b**) R/Line 65; (**c**) R/Line 41; (**d**) R/Line 20; (**e**) R/Line 11.

The R<sup>2</sup> and NSE in the validation process were 0.950 and 0.893, respectively. The observed standard deviation was at 0.041. In a recent study [46], the HEC-RAS model was validated for the Tarbela Reservoir by simulating it only for one year, and an approximately 20-m difference between the observed and simulated river beds for the sediment delta in the year 2000 was found. However, in the present study, the difference of four years of simulation was only 4–5 m in the whole longitudinal profile (Figure 9). A better modeling performance might be due to more accurate sediment load boundary conditions generated using a long-term data series, i.e., 1969–2014, whereas [46] used only a 28-year data series, i.e., 1979–2006.

To validate the HEC-RAS model with the above calibrated WA-ANN sediment series, it was run for another four years up to 2017, similar to the SRC model. The output was compared with observed sediment deposits of 2017 and presented in Figure 10. The R<sup>2</sup> and NSE in the validation process were 0.968 and 0.959, respectively. The observed standard deviation was at 0.025.

**Figure 10.** Comparison of observed and WA-ANN-simulated bed levels during validation for 2017: (**a**) along Tarbela Reservoir; (**b**) R/Line 65; (**c**) R/Line 41; (**d**) R/Line 20; (**e**) R/Line 11.
