Rapidly Evolving Controls of Landslides After a Strong Earthquake and Implications for Hazard Assessments. Issue 1 (13th January 2021)

• Record Type: Journal Article
• Authors: Fan, Xuanmei; Yunus, Ali P.; Scaringi, Gianvito; Catani, Filippo; Siva Subramanian, Srikrishnan; Xu, Qiang; Huang, Runqui
• Is Part Of: Geophysical research letters. Volume 48:Issue 1(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-01-13
• Subjects: controlling factor -- earthquake‐induced landslide -- hazard assessment -- machine learning -- susceptibility assessment -- Wenchuan Earthquake; Geophysics -- Periodicals; Planets -- Periodicals; Lunar geology -- Periodicals; 550
• DOI: 10.1029/2020GL090509
• Languages: English
• ISSNs: 0094-8276

Abstract

Abstract: Strong earthquakes, especially on mountain slopes, can generate large amounts of unconsolidated deposits, prone to remobilization by aftershocks and rainstorms. Assessing the hazard they pose and what drives their movement in the years following the mainshock has not yet been attempted, primarily because multitemporal landslide inventories are lacking. By exploiting a multitemporal inventory (2005–2018) covering the epicentral region of the 2008 Wenchuan Earthquake and a set of conditioning factors (seismic, topographic, and hydrological), we perform statistical tests to understand the temporal evolution of these factors affecting debris remobilizations. Our analyses, supported by a random‐forest susceptibility assessment model, reveal a prediction capability of seismic‐related variables declining with time, as opposed to hydro‐topographic parameters gaining importance and becoming predominant within a decade. These results may have important implications on the way conventional susceptibility/hazard assessment models should be employed in areas where coseismic landslides are the main sediment production mechanism on slopes. Plain Language Summary: Strong earthquakes in mountain regions can trigger thousands of landslides, forming deposits of rock and soil debris along steep slopes. Months to years later, rainstorms may generate debris flows—destructive water‐debris mixtures that rush downslope and flood valleys. Scientists use models to estimate the hazard of landslides and debris flows, which are based on accurate maps of the slopes, the type of rock or soil, inventories of known landslides, rainfall trends, and more. Susceptibility and hazard maps are the main product of these models. They are used to predict the probability of a hazardous event occurring at a given location in a given time span. These maps are usually "static, " in the sense that they are thought to remain valid for a long time because the data they are based upon (such as the shape of slopes) do not vary much. However, postearthquake landscapes are very dynamic: debris moves downslope, carried by rain or shaken by aftershocks; meanwhile, new landslides occur on some slopes, while others revegetate and stabilize. The overall picture is complex as many variables are involved. We use machine learning to demonstrate that static hazard maps become unable to predict landslides after just a few years, and advocate for the use of frequently updated maps linked to fresh inputs, tracking the location and activity of debris deposits, and old and new landslides. Key Points: We explored the rapidly evolving controls of postseismic debris remobilization The increasing importance of hydrological factors and a large amount of channel deposits cause persisting postseismic debris flow occurrences Hazard models need to be frequently updated by timely inventories in order to assess the hazard evolution of postseismic landslides