The reservoir landslides are characterized by repeated phases of acceleration and isokinetic de-formation under long-term periodic external forces. The state-of-the-art research lacks reliable prediction methods and judgment of their evolution stages. This work promotes the application of the deep learning algorithm and landslide evolution model in long-term warning systems. The test site is the Sifangbei landslide in the Three Gorges reservoir area of China. The main innova-tive features are: (i) the displacement of the landslide is considered as the prediction target, and the optimal model (i.e., conditioning factors and hyper-parameters combination) driven by the deep learning framework is used for spatial prediction; (ii) different warning methods (from both literature and current practice) are compared to single out the one that can best describe the evo-lution stage of the reservoir landslide; and (iii) deep learning model and adaptive evolution model are combined to analyze the temporal-spatial kinematic characteristics and evolution trend of the landslide under extreme scenarios related to rainfall and reservoir water levels. The results show that the predicted displacements of the lower and central part of the landslide are re-spectively controlled by reservoir water level and rainfall; the five-stage evolution model can cap-ture the long-term evolution trend of the Sifangbei landslide; under extreme scenarios, landslide deformation exhibits step-like characteristics and is more likely to start from the middle of the lower portion of the unstable area. These models represent the up-to-date steps of a long-term re-search plan. The gathered knowledge can be used to analyze the spatial evolution characteristics of landslides and promote the setup of long-term warning systems. Furthermore, the results show that combining the proposed deep learning and evolution methods provides forecast information that could help adjusting short-term warning strategies in such a complex risk area as the Three Gorges Reservoir.

Deep learning powered long-term warning systems for reservoir landslides

Peduto, D
2023-01-01

Abstract

The reservoir landslides are characterized by repeated phases of acceleration and isokinetic de-formation under long-term periodic external forces. The state-of-the-art research lacks reliable prediction methods and judgment of their evolution stages. This work promotes the application of the deep learning algorithm and landslide evolution model in long-term warning systems. The test site is the Sifangbei landslide in the Three Gorges reservoir area of China. The main innova-tive features are: (i) the displacement of the landslide is considered as the prediction target, and the optimal model (i.e., conditioning factors and hyper-parameters combination) driven by the deep learning framework is used for spatial prediction; (ii) different warning methods (from both literature and current practice) are compared to single out the one that can best describe the evo-lution stage of the reservoir landslide; and (iii) deep learning model and adaptive evolution model are combined to analyze the temporal-spatial kinematic characteristics and evolution trend of the landslide under extreme scenarios related to rainfall and reservoir water levels. The results show that the predicted displacements of the lower and central part of the landslide are re-spectively controlled by reservoir water level and rainfall; the five-stage evolution model can cap-ture the long-term evolution trend of the Sifangbei landslide; under extreme scenarios, landslide deformation exhibits step-like characteristics and is more likely to start from the middle of the lower portion of the unstable area. These models represent the up-to-date steps of a long-term re-search plan. The gathered knowledge can be used to analyze the spatial evolution characteristics of landslides and promote the setup of long-term warning systems. Furthermore, the results show that combining the proposed deep learning and evolution methods provides forecast information that could help adjusting short-term warning strategies in such a complex risk area as the Three Gorges Reservoir.
2023
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11386/4843431
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 13
  • ???jsp.display-item.citation.isi??? 11
social impact