Growing population, hydroclimatic variability, and increasing frequency of droughts and floods are challenging the ability of water management strategies of matching competing water demands by effectively allocating water volumes in space and time. This motivates an increasing interest on improving streamflow predictions, which contribute in anticipating climate severe events and promptly implement adaptation measures. However, the complexity of natural systems, combined with all the social and physical interactions involved, may change or evolve over time and result in an extremely high number of drivers that might be potentially considered for streamflow predictions, especially for long lead times. Several researchers have been recently studying the influence of El Nino-Southern Oscillation (ENSO) events and ENSO teleconnection on the interannual variations of at-site hydrological processes. In fact, ENSO information can be used for extending the lead time of streamflow and weather forecast. Yet, despite in some locations (e.g., US, Australia) ENSO teleconnection is well studied, there is no consensus on how this teleconnection can be identified in other river basins. This thesis contributes a formalized procedure relying on Input Variable Selection techniques for detecting ENSO teleconnection and improving long-term streamflow predictions. In addition, we test and comparatively analyze four different nonlinear data-driven models (Artificial Neural Networks -ANN-, Radial Basis Funtion -RBF-,M5P and Extra-tress) to evaluate their prediction capabilities in the short and long-term. The procedure is applied in the Kemano hydropower system located in British Columbia, Canada. The hydrology of the system is mainly driven by snow-melt, which implies a high degree of interseasonal variability with maximum inflows during the snow-melting period and risk of flooding in the district of Vanderhoof. Results show that IVS outcomes are consistent with the system’s hydrology, with snow playing a relevant role in one and two months ahead streamflow prediction. The extension of the lead time shows that ENSO indexes become more and more important, ensuring an acceptable prediction accuracy also for lead times equal to four months. Among the different datadriven models, ANN, M5P and Extra-trees showed good performance for streamflow prediction in the short-term. However, for longer the lead-time, M5P model seems to be the most suitable for streamflow forecast.
ENSO detection for improving long term streamflow forecast in Kemano water system
TRIMIÑO BARBOSA, ANDREA CAROLINA;MESA MUNOZ, CAROLINA
2014/2015
Abstract
Growing population, hydroclimatic variability, and increasing frequency of droughts and floods are challenging the ability of water management strategies of matching competing water demands by effectively allocating water volumes in space and time. This motivates an increasing interest on improving streamflow predictions, which contribute in anticipating climate severe events and promptly implement adaptation measures. However, the complexity of natural systems, combined with all the social and physical interactions involved, may change or evolve over time and result in an extremely high number of drivers that might be potentially considered for streamflow predictions, especially for long lead times. Several researchers have been recently studying the influence of El Nino-Southern Oscillation (ENSO) events and ENSO teleconnection on the interannual variations of at-site hydrological processes. In fact, ENSO information can be used for extending the lead time of streamflow and weather forecast. Yet, despite in some locations (e.g., US, Australia) ENSO teleconnection is well studied, there is no consensus on how this teleconnection can be identified in other river basins. This thesis contributes a formalized procedure relying on Input Variable Selection techniques for detecting ENSO teleconnection and improving long-term streamflow predictions. In addition, we test and comparatively analyze four different nonlinear data-driven models (Artificial Neural Networks -ANN-, Radial Basis Funtion -RBF-,M5P and Extra-tress) to evaluate their prediction capabilities in the short and long-term. The procedure is applied in the Kemano hydropower system located in British Columbia, Canada. The hydrology of the system is mainly driven by snow-melt, which implies a high degree of interseasonal variability with maximum inflows during the snow-melting period and risk of flooding in the district of Vanderhoof. Results show that IVS outcomes are consistent with the system’s hydrology, with snow playing a relevant role in one and two months ahead streamflow prediction. The extension of the lead time shows that ENSO indexes become more and more important, ensuring an acceptable prediction accuracy also for lead times equal to four months. Among the different datadriven models, ANN, M5P and Extra-trees showed good performance for streamflow prediction in the short-term. However, for longer the lead-time, M5P model seems to be the most suitable for streamflow forecast.File | Dimensione | Formato | |
---|---|---|---|
thesis_Mesa_Trimino.pdf
accessibile in internet per tutti
Descrizione: Thesis ENSO detection
Dimensione
4.34 MB
Formato
Adobe PDF
|
4.34 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/116241