| Rainfall-induced landslides cause a large number of casualties as well as severe economic losses worldwide every year.Particularly in recent years,global climate change and urbanization have led to a prolonged high incidence and widespread occurrence of rainfall-induced landslides,poseing a significant threat to human society.Such a diffuse risk cannot be prevented and mitigated only by means of structural measures.Thus,implementing early warning systems for landslides over a large area through the monitoring of meteorological parameters is of significant value in addressing the complex risks posed by rainfall-induced landslides.Currently,most of works on early warning of landslides at regional scale are carried out based on defining rainfall thresholds for rainfall-induced landslides.Despite abundant results indicating that incorporating hydrological processes on slopes before landslide initiation aids in predicting landslides,research on integrating hydro-meteorological parameters into warning model still lacks in-depth exploration.This thesis aims at proposing an effective methodology to enhance the performance of warning models for rainfall-induced landslides at regional scale.To this aim,Lueyang county,Shaanxi Province and the central portion of Campania region,Italy were taken as the study area.The works on how to define the effective rainfall thresholds,hydro-meteorological thresholds,as well as warning model for rainfallinduced landslides were carried out for the two distinct landslide cataolgues.The main contents and results are highlighted as follows:(1)The potential of defining rainfall thresholds by means of adpoting rainfall events not responsible for landslides was explored.Given that more than 90% of the recorded landslides in Lueyang County concentrate on two dates,an innovative approach was proposed to define the rainfall thresholds by seeking the upper limit of non-triggering rainfall events using the negative Frequentist approach.The results highlight that the negative thresholds are more effective than the positive thresholds defined by the traditional positive Frequentis approach in the case of Lueyang county.The proposed approach is valuable to apply in areas with few landslide records and strongly imbalanced recorded landslide occurrences vs.non-occurrences.(2)A procedure designed to define probabilistic rainfall thresholds adopting innovative rainfall variables was proposed.Rainfall variables that are more significant than the conventional cumulated rainfall,rainfall intensity,and duration were derived through the automatic extraction of numerous rainfall variables from reconstructed rainfall events and the importance analysis of extracted variables.Then,adopting the Bayesian approach,the most significant rainfall variables are used to define a set of probabilistic thresholds in one,two,and three dimensions.The results highlight,in both cases,the probabilistic thresholds defined by the novel variables perform better than the ones adopting the conventional rainfall variables.The proposed new procedure provides a powerful option to define efficient rainfall thresholds.(3)The potential of mining soil moisture-related variables to characterize soil wetness condition was explored,and a new methodology to define probabilistic hydrometeorological thresholds adpoting multiple rainfall and hydrological variables was proposed.Using hourly and multi-layer soil moisture information from ERA5-Land Reanalysis data,derived variables with trends in soil moisture were deeply mined and analyzed for their significance in characterizing soil wetness status.Then,the hydrometeorological thresholds were defined based on the derivation of indicators representing rainfall severity and soil wetness.The results highlight the derived variables of soil water content variation are more significant in representing soil wetness than transient soil moisture,and the performance of hydro-meteorological thresholds defined based on the derived variables is better.In addition,the proposed methodology ensures statistical significance and the simplicity while integrating multivariate information.(4)A demonstrative reserarch of landslide early warning based on hydrometeorological parameters was carried out.The hydro-meteorological warning models were defined for Camp-3 SE zone and Lueyang county within the framework of the opeartional early warning systems in Campania region and Shaanxi province,taking into account geo-environmental factors in the implicit and explicit ways,respectively.In Camp-3 SE area,a warning model was defined with precipitation and soil moisture at depths of 0–7 cm as inputs and four warning levels at the municipal level as outputs,by implicitly considering geo-environmental factors.In Lueyang county,a warning model was defined with soil moisture at depths of 28–100 cm and 100–289 cm as well as precipitation as monitoring inputs and four warning levels at the township level as outputs,by explicitly considering geo-twoenvironmental factors.Subsequently,the practical performance of the defined warning models was evaluated for two severe areal landslide events in each study area.The results highlight that the warning model for Camp-3 SE achieved 83% and 69% of correct alert rates under two validation cases in2019,respectively.The warning model for Lueyang County achieved correct alert rates of 82% and 76% in two validation cases in July 2018,respectively.In summary,within the current framework of landslide early warning,the thesis preliminarily developed a methodology to improve the performance of warning models for rainfall-induced landslides,which provides the foundation to improve the effectiveness of territorial landslide early warning systems.The innovative approaches on defining rainfall and hydro-meteorological thresholds provide references for analysts to define effective thresholds for rainfall-ineduced landslides.The practice of landslide early warning by means of monitoring hydro-meteorological parameters provides an evidence for the introduction of hydrological information into territorial landslide early warning systems,which is of significant importance for prevention and mitigation of rainfall-induced landslide risks. |
