Downscaling Water Storage: Training Long Short-Term Memory (LSTM) model to Downscale Nonlinear Storage-Discharge Relation in Rum River Watershed, MN
Downscaling Water Storage: Training Long Short-Term Memory (LSTM) model to Downscale Nonlinear Storage-Discharge Relation in Rum River Watershed, MN
Session ID#: 121619
Session Description:
Flooding is one of the most financially devastating natural hazards in the world. When current hydrological models emphasize rainfall-runoff models, studying storage-discharge relation can potentially improve existing flood forecast. This presentation will assess the relation between daily water storage (S) and discharge (Q) by utilizing physical-based hydrological modeling and long short-term memory (LSTM) network; and evaluate the improvement of predicting discharge by downscaling total water storage. Machine learning (ML) algorithms have been used for predicting the outputs that represent arbitrary non-linear functions between predictors and predictands. LSTM, the time-series deep learning neural network that has been used for improving the accuracy of non-linear rainfall-runoff prediction, will be applied to improve non-linear predictions between S and Q. This experiment will evaluate the improvement between downscaled and total water storage by training two sets of LSTM networks storage-discharge relationships with unbiased hydrological variables simulated by the semi-distributed Hydrological Simulated Program-Fortran (HSPF).
Index Terms:
1805 Computational hydrology [HYDROLOGY]
1846 Model calibration [HYDROLOGY]
1847 Modeling [HYDROLOGY]
1942 Machine learning [INFORMATICS]
Primary Convener: Pai-Feng Victor Teng, University of Minnesota, Twin Cities, St. Paul, MN, United States
Convener: John Little Nieber, Univ of Minnesota, Twin Cities, St. Paul, MN, United States
Primary Liaison: Pai-Feng Victor Teng, University of Minnesota, Twin Cities, St. Paul, MN, United States
See more of: Hydrology