A novel approach for river discharge prediction in Lancang Mekong River Basin : incorporation of multisource remote sensing and LSTM model | |
| Author | Aryal, Tilasmi |
| Call Number | AIT Thesis no.WM-23-08 |
| Subject(s) | Hydrology--Remote sensing--Lancang Mekong River Basin Streamflow--Lancang Mekong River Basin--Measurement |
| 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 | The spatiotemporal variation of river discharge plays a crucial role in effective water resources management, as it provides crucial information for decision making in various domains such as irrigation, hydropower, flood forecasting, and environmental conservation. Traditional ground-based gauging stations, although providing accurate and reliable measurements, are typically limited in their spatial coverage and prone to data gaps. Satellite based remote sensing can provide a feasible alternative method for river monitoring that can be utilized to infer river water level and discharge. In this context, this study aimed to develop a framework using a Long Short-Term Memory (LSTM) model that incorporates multiple satellite missions to predict daily river discharge and evaluate its effectiveness for ungauged locations. First, water level at eight reach of Lancang Mekong River was obtained utilizing altimetric water level measurements from Jason-1/2, ENVISAT, and Saral altimetry missions. Optical images from MODIS Aqua and Terra were used to obtain the time series surface reflectance ratio of dry to wet pixels as a proxy for water discharge for each location. However, careful selection of the location of the dry and wet pixels near the gauging station is essential for accurate prediction, and the study determined the best location by computing the correlation between surface reflectance ratio and discharge for each pixel. LSTM models were developed for each station and trained using single and combined input features; surface reflectance ratio from MODIS Aqua and Terra and altimetry derived water level to assess their influence on the accuracy of daily discharge prediction. The inclusion of features from multi mission satellite showed improved and better performance in discharge prediction, with predictive accuracy ranging from 0.54 to 0.92 in terms of NSE. The spatial transferability of the LSTM model to predict daily river discharge at ungauged locations was further assessed. The result showed model performs better for certain reaches of the river, but performance decreases while moving farther from the trained reach indicating the model can be developed for certain reaches of the river that can predict discharge at ungauged locations within that reach with good and reliable accuracy. |
| Year | 2023 |
| 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) | Shrestha, Sangam |
| Examination Committee(s) | Shanmugam, Mohana Sundaram;Ho Huu Loc |
| Scholarship Donor(s) | Asian Development Bank-Japan Scholarship Program (ADB-JSP) |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2023 |