Development of a regional-scale landslide early warning system applying statistical and physically based evaluation approaches in sequence

Abstract

In response to the sharp increase in gravity-induced geological disasters in Korea resulting from localized extreme rainfall events, this research proposes a landslide early warning methodology. The method is based on the sequential application of statistical and physically based hazard evaluation approaches and was devised by combining the strengths of the two mutually complementary approaches. Following a decision algorithm for the five phases of warning levels, a statistical evaluation stage in which two different rainfall thresholds and one fixed geo-property (landslide susceptibility) threshold are used determines a preliminary conservative warning level. In order to assess whether higher warning levels with higher certainties should be assigned, a physically based evaluation stage is applied on the precondition of the preliminary warning stage. This stage was accomplished by using the critical continuous rainfall that triggers slope instability derived from an advanced analysis for the physical modeling of landslide-triggering mechanisms. Having been developed with the conceptual motivation that the total rainfall amount that triggers failure upon being permeated and temporarily retained in a slope is constant regardless of rainfall intensity, critical continuous rainfall exhibited good performance as an index that forecasts shallow landslides on natural terrain. As a result of scripting the decision algorithm for warning levels in combination with the input parameter data, a landslide early warning simulation model based on the sequential evaluation approach was developed. With the establishment of a historical landslide database through an indirect field check method, the validity of the system was examined through discriminative and temporal performance analyses, which simulated the progress of warning levels over eight years (2009-2016). The performance analysis results provided quantitative information on various features of the warning capabilities of the proposed methodology. A citywide landslide early warning system using the proposed methodology in Busan, South Korea, exhibited several advantages in both spatial and temporal performance.