Comparison of physically-based hydrological model (SWAT) with artificial neural network (ANN) model in different climatic regions | |
| Author | Pradhan, Pragya |
| Call Number | AIT THesis no.WM-19-15 |
| Subject(s) | Hydrological--Climatic factors Neural networks (Computer science) Watershed hydrology |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management |
| Publisher | Asian Institute of Technology |
| Abstract | Artificial Neural Network (ANNs) is a science used by hydrologists and engineers to evaluate streamflow of river basins. In this study, a physically-based hydrological model, soil and water assessment tool (SWAT) and three types of artificial neural network model i.e. feed forward backward propagation (FFBP), general regression neural network (GRNN), multilayer perceptron (MLP) were used to simulate the daily streamflow and their results were compared in order to analyze their competence. The study was carried out in three different river basins with three different climatic characteristics namely; West-Seti river basin of sub-tropical (partially wet) climatic region, Srepok river basin of tropical (wet) climatic region and Hari rod river basin of semi-arid (dry) climatic region. The performance of SWAT and ANN models were evaluated using statistical performance indicators such as correlation coefficient (R2), Nash and Sutcliffe efficiency (NSE), Percentage Bias coefficient (PBIAS). It was found that the performance of ANN model was very good with both R² and NSE value greater than 0.95 for both training and validation period in West-seti river basin and Srepok river basin. Whereas, in the Hari rod river basin, the performance of SWAT model was good with both R² and NSE value greater than 0.60 for both calibration and validation period. Similarly, the performance of SWAT and ANN model were evaluated based on hydrological indicators (i.e. annual discharge, base flow, Q5, Q95, Qdry, Qwet). For complete evaluation and accuracy of these model, their performances were evaluated during different periods of flows (very high flow to very low flow) using flow duration curves (FDCs). The performance of different phases of flow was also evaluated based on R². This segmentation of flow analysis is very important to analyze the model performance based on the simulation of streamflow. SWAT was found to be better at low flow simulation and ANNs model performance was better at high flow simulation in all the three-river basin. The result indicates that SWAT and ANN were good models for the streamflow modelling. However, ANNs model performance is better for tropical and sub-tropical climatic regions while SWAT model performance is better for semi-arid climatic region. Being the data driven model, ANN does not need any physical characteristics of the river basin which makes ANN easier to use compared to SWAT model. However, the ANN model cannot yield proper results where there is change in the physical characteristics of the river basin. In this case the SWAT model yields better results. |
| Year | 2019 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Tawatchai Tingsanchali;Shrestha, Sangam |
| Examination Committee(s) | Ekasit Kositsakulchai;Datta, Avishek ; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2019 |