2.3.2. Model Performance Evaluation Using SWAT-CUP

SWAT-CUP (SWAT Calibration and Uncertainty Procedure) is a SWAT-compatible model. When compared to the old approach of manual correction by trial and error, the SWAT model's sensitive variable analysis, calibration, and validation procedures have more flexibility and take less time. The outcome of altering the sensitivity variable will serve as a guide for the best calibration and adjustment of the solution(s) between the SWAT generated results and the station data. The following are five approaches for determining the proper values: (1) Generalized Likelihood Uncertainty Estimation (GLUE), (2) Particle Swarm Optimization (PSO), (3) Parameter Solution (Parasol), (4) Mark Chain Monte Carlo (MCMC), and (5) Sequential Uncertainty Fitting (SUFI-2) [38]. For this study, the use of the SUFI-2 technique was selected to apply in the operation. The SUFI-2 technique is

uncertainty analysis consisting of predictive P-factors representing the actual measured values that appear in the simulation results for 95% of the uncertainty of the simulation. The prediction (95% prediction uncertainty; 95PPU) and R-factor are calculated as the ratio of the mean amplitude range of the 95PPU to the standard variance of the actual data. The calculated 95PPU values were positioned at 2.5% and 97.5% of the cumulative probability distribution of the variables considered. Using Latin hypercube sampling [38] as this technique requires the least number of sensitivity variables but can produce the best results compared to other methods [39]. Eight parameters from the most vulnerable model types were chosen for examination in this study. Eight parameters from the most vulnerable model types were chosen for examination in this study. The results of the modification of the parameters that calculated streamflow from the model closest to the data from the measurement station are shown in Table 2.


**Table 2.** Adjusted Model Sensitivity Parameters.

Then, the results were compared with the data from the measurement station, and the efficiency was assessed using two statistical indices to check the accuracy of the results [40], which showed the level of accuracy of the monthly streamflow comparison results. It is divided into four levels as shown in Table 3 [41].

**Table 3.** Typical performance level for accepted statistics in monthly time step.


1. The Coefficient of Determination (R2), as shown in Equation (6), is between 0–1, with values greater than 0.6 indicating that the two data are correlated at a level of reliability.

2. The Nash Sutcliffe efficiency (NSE) coefficient, as shown in Equation (7), is between −∞ and 1, with values greater than 0.5 indicating that the two data are correlated at a level of reliability.

$$\mathbf{R}^2 = \left[ \left( \frac{\sum\_{i=1}^n (\mathbf{Q}\_{\rm{oi}} - \mathbf{Q}\_{\rm{oa}})(\mathbf{Q}\_{\rm{si}} - \mathbf{Q}\_{\rm{sa}})}{\sqrt{\sum\_{i=1}^n \left(\mathbf{Q}\_{\rm{oi}} - \mathbf{Q}\_{\rm{oa}}\right)^2} \sqrt{\sum\_{i=1}^n \left(\mathbf{Q}\_{\rm{si}} - \mathbf{Q}\_{\rm{sa}}\right)^2}} \right) \right]^2 \tag{6}$$

$$\mathbf{E\_{ns}} = 1 - \left(\frac{\sum\_{i=1}^{n} \left(\mathbf{Q\_o} - \mathbf{Q\_s}\right)^2}{\sum\_{i=1}^{n} \left(\mathbf{Q\_o} - \mathbf{Q\_{sa}}\right)^2}\right) \tag{7}$$

where n is the total number of data. Qoi is the i-order value, Qoa is the mean from all measurements, Qsi is the i-order model, Qsa is the i-order value from all models, Qs is the calculated value from the model, and Qo is the measurement value.
