Identification of Debris Flow Initiation Zones Using Topographic Model and Airborne Laser Scanning Data

Abstract

Empirical multivariate predictive models represent an important tool to estimate debris flow initiation areas. Most of the approaches used in modelling debris flows propagation and deposit phases required identifying release (starting point) area or source area. Initiation areas offer a good overview to point out where field investigation should be conducted to establish a detailed hazard map. These zones, usually, are arbitrarily chosen which affect the model outputs; hence, there is a need to have accurate and automated means of identifying the release area. In addition to this, the resolution of the terrain dataset also affects the results of the detection of source areas. In this study, airborne laser scanning (ALS) data was used because of its robustness in providing detailed terrain attributes at high resolution. Primary and secondary conditioning parameters were derived from digital elevation model (DEM) as input into the modelling process. Three models were executed at different spatial resolution scales: 5, 10 and 15 m, respectively. MARSpline multivariate data mining predictive approach was implemented using morphometric indices and topographical derived parameter as independent variables. A statistics validation was calculated to estimate the optimal pixel size, 1200 randomly sample data were generated from existing inventory data. Debris flows and no-debris flows were categorized, and the transform to continuous integer (1 and 0), respectively. To achieve this, the data set was divided into two, 70% (840) for the training dataset and 30% (360) for validation. The best model was selected based on the model performance using the generalized cross validation (GCV) and the receiver operating characteristic (ROC) curve/area under curve (AUC) values. Conditioning parameters were numerically optimized to identify the arbitrarily maximum model basis function for eleven variables, using MARSplines analysis (algorithm). The three most influencing topographic parameters identified are topographic roughness index (TRI), slope angle, and specific catchment area (SCA) with the percentage values of participation in the model of 100, 93, and 86%, respectively. The chosen function appeared to describe the analysed correlation sufficiently well. Consequently, three stages of optimization were made to determine the optimized source areas is possible with 10 m pixel size, 200 maximum basis functions and 3 maximum interactions, resulting into 82% ROC train and 80% test, GCV 0.189 and 85% correlation coefficient. The result will be of great contribution to the advancement of a broad understanding of the dynamics of debris flows hazard and mitigations at regional level which; that is resourceful for comprehensive slope management for safe urban planning decisionmaking process and debris flow disaster management.

Data Mining and Statistical Approaches in Debris-Flow Susceptibility Modelling Using Airborne LiDAR Data

“…There are two raining seasons in the study area, which include; the September to December regime and the February to May regime but the peaks are recorded between March to May and November to December respectively. During the peak period, rainfall of between 88 mm to 100 mm in a single day usually triggers landslides [6,11,50,52,55]. When the local rainfall rate exceeds a distinctive intensity threshold, a runoff-generated debris flow may likely be initiated.…”

“…Tree canopies, dense forests and rugged terrains constitute a barrier to derive complex morphological and topographic features from microwave remote sensing data [40,44,46,69,70,71]. Unlike Synthetic Aperture Radar (SAR) data derived elevation model, LiDAR data is not affected by topographic shadowing because of the angle of inclination (<20° off-nadir) of LiDAR sensors [11,44,45,48,64,72,73]. In a recent study, a LiDAR-derived Digital Elevation Model (DEM) was considered the most favorab le terrain data used to derive topographic attributes integrated with appropriate modelling approach to detect and predict environmental disaster [67,74].…”

“…As an alternative, MARS structures a relation from established basis functions (BF) and coefficients, which are generally determined from the regression information ). Certainly, the areas in which this model have been applied include landslides [11,31,42,82,83,90,116,120,121,122,123,124,125,126]. The superiority of MARSpline over some MLAs have been demonstrated in other fields of research [81,125,127,128,129,130].…”

“…The event usually triggered by a number of factors such as heavy precipitation, lahar, earthquake, landslide and other anthropogenic activities [4,5,6,7,8,9,10]. Noteworthy, the prominent factors trigger debris-flow events in Malaysia is torrential rainfall that transforms landslides to a ribbon of surge liquid and solid sediments forming the debris flows [11]. Flows at high velocity along hill slopes and valleys exposing the regions to great risk of debris flow which results in a huge number of casualties and property damage along its path [8,12,13]; as the region suffers a frequent revisit period.…”

“…In Peninsular Malaysia, LiDAR technology has been used for landslide hazard mapping, modeling and other related phenomena e.g., [6,47,48,49,50,51,52]. However, only very few studies have been carried out on debris flow susceptibility mapping using LiDAR data in this region e.g., [11,47,53,54,55,56,57]. LiDAR DEM dataset allows retrieving primary and secondary landslides conditioning factors to predict susceptible and hazard zones.…”