Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Using attributes of ungauged basins to improve regional regression equations for flood estimation: a deep learning approach
Regional quantile regression is used for estimating flood quantiles at ungauged sites. This study presents an application of stacked autoencoder (SAE), a deep learning algorithm, for developing regional quantile regression equation. Data from the conterminous United States are used for developing SAE-based regression equations and for comparing them against standard feed-forward neural networks. The deep learning algorithm improves the estimation of flood quantiles at 100 test sites (median mean square error (MSE) dropped by 10% and median coefficient of determination (R2) improved by 6%). The most impressive improvement is obtained when basin attributes of ungauged basins are used in SAE-based regression equations (median MSE dropped by 42% and median R2 improved by 19%). The results from the study show that information at ungauged basins, when employed in the deep learning framework, can significantly improve hydrological predictions.
Using attributes of ungauged basins to improve regional regression equations for flood estimation: a deep learning approach
Regional quantile regression is used for estimating flood quantiles at ungauged sites. This study presents an application of stacked autoencoder (SAE), a deep learning algorithm, for developing regional quantile regression equation. Data from the conterminous United States are used for developing SAE-based regression equations and for comparing them against standard feed-forward neural networks. The deep learning algorithm improves the estimation of flood quantiles at 100 test sites (median mean square error (MSE) dropped by 10% and median coefficient of determination (R2) improved by 6%). The most impressive improvement is obtained when basin attributes of ungauged basins are used in SAE-based regression equations (median MSE dropped by 42% and median R2 improved by 19%). The results from the study show that information at ungauged basins, when employed in the deep learning framework, can significantly improve hydrological predictions.
Using attributes of ungauged basins to improve regional regression equations for flood estimation: a deep learning approach
Ojha, Richa (Autor:in) / Tripathi, Shivam (Autor:in)
ISH Journal of Hydraulic Engineering ; 24 ; 239-248
04.05.2018
10 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
| Elsevier | 2017
Probable Flood Predictions in Ungauged Coastal Basins of El Salvador
| Online Contents | 2008
Probable Flood Predictions in Ungauged Coastal Basins of El Salvador
| British Library Online Contents | 2008
Probable Flood Predictions in Ungauged Coastal Basins of El Salvador
| British Library Conference Proceedings | 2008