Long-short term memory (LSTM) neural network integrated with satellite datasets to simulate streamflow in transboundary river basins

Abstract:

One of the key challenges in simulating streamflow in transboundary river is the lack of spatial-temporal understanding of hydrological variables. Satellite observation provides an excellent way to obtain various hydrological elements for such transboundary basins. Therefore, this study aims to examine capacity of remotely sensed datasets in several transboundary rivers in Indochina. These datasets include GPM IMERG precipitation, MODIS evapotranspiration, MODIS NDVI, and SRTM DEM. Two studies were carried out: (1) Understanding spatio-temporal variability in the remotely-sensed datasets in different parts of the transboundary river basins; (2) Streamflow simulation using long-short term memory (LSTM) neural network with input data as remotely-sensed observations. The advantage of LSTM is its ability to learn long-term dependences between the inputs and outputs of the network, which are very useful for prediction of streamflow time series. This study, therefore, can provide benefit for water resources management in cross-multiple countries rivers.