Skip to Main Content

The simulated discharges obtained using hyperparameter tuning of the three ML models are compared with the observed discharge. It is further evaluated using R2, RMSE, NSE, and PBIAS metrics, and the related discussion follows (refer to Table 4). Relevant observations (in addition to remarks) are presented in Table 4. WNN-Gaussian wavelet has displayed similar model performance in the training period compared with the WNN-Mexican hat. The model has fallen short in the validation period based on R2 obtained. WNN-Mexican hat displayed better performance than the WNN-Gaussian wavelet in the training and validation periods with relatively lower RMSE values. Despite a higher learning rate, WNN-Morlet is found to dominate all other WNN models in the training period. However, it failed to show consistency in the validation period compared with other WNN models.

Table 4

Information on metrics of Bi-LSTM, WNN, and XGBoost in the training and validation periods

S.NoModelTraining
Validation
Remarks
R2RMSENSEPBIAS (%)R2RMSENSEPBIAS (%)
Bi-LSTM 0.60 1.73 0.59 12.4 0.58 1.96 0.57 13.0 This model has shown a satisfactory performance with an R2 of 0.6 and 0.58 in the training and validation periods. This model structure falls short compared with XGBoost, although it proved better than WNN. 
WNN (Gaussian) 0.23 2.12 0.23 0.3 0.22 2.20 0.22 1.0 The model has shown unsatisfactory performance both in the training and validation periods. 
WNN (Mexican hat) 0.22 2.06 0.23 0.3 0.22 2.13 0.22 1.0 Relatively low metrics were displayed by Mexican hat in the present study compared with other WNN models. Overall, the model has been unsatisfactory during training and validation periods. 
 WNN (Shannon) 0.26 2.30 0.26 1.14 0.25 2.31 0.25 1.46 This wavelet has shown a better consistency in the validation period than the other employed mother wavelets, although the overall model performance is unsatisfactory in both the training and validation periods. 
 WNN (Morlet) 0.29 2.06 0.29 0.9 0.22 2.39 0.21 1.8 Morlet wavelet is slightly better than other mother wavelets in training but lacks consistency in validation. Overall model performance can be rated as unsatisfactory for the present study. 
XGBoost 0.88 1.49 0.86 29.3 0.86 1.63 0.85 28.5 The model is exceptional, with an R2 of 0.875 in training. It has also displayed a great consistency in validation with an R2 of 0.863. Thus, this model helps derive reliable streamflow forecasts compared with other models employed in this study. 
S.NoModelTraining
Validation
Remarks
R2RMSENSEPBIAS (%)R2RMSENSEPBIAS (%)
Bi-LSTM 0.60 1.73 0.59 12.4 0.58 1.96 0.57 13.0 This model has shown a satisfactory performance with an R2 of 0.6 and 0.58 in the training and validation periods. This model structure falls short compared with XGBoost, although it proved better than WNN. 
WNN (Gaussian) 0.23 2.12 0.23 0.3 0.22 2.20 0.22 1.0 The model has shown unsatisfactory performance both in the training and validation periods. 
WNN (Mexican hat) 0.22 2.06 0.23 0.3 0.22 2.13 0.22 1.0 Relatively low metrics were displayed by Mexican hat in the present study compared with other WNN models. Overall, the model has been unsatisfactory during training and validation periods. 
 WNN (Shannon) 0.26 2.30 0.26 1.14 0.25 2.31 0.25 1.46 This wavelet has shown a better consistency in the validation period than the other employed mother wavelets, although the overall model performance is unsatisfactory in both the training and validation periods. 
 WNN (Morlet) 0.29 2.06 0.29 0.9 0.22 2.39 0.21 1.8 Morlet wavelet is slightly better than other mother wavelets in training but lacks consistency in validation. Overall model performance can be rated as unsatisfactory for the present study. 
XGBoost 0.88 1.49 0.86 29.3 0.86 1.63 0.85 28.5 The model is exceptional, with an R2 of 0.875 in training. It has also displayed a great consistency in validation with an R2 of 0.863. Thus, this model helps derive reliable streamflow forecasts compared with other models employed in this study. 

Close Modal

or Create an Account

Close Modal
Close Modal