Modelling of optimized hybrid debris flow using airborne laser scanning data in Malaysia

Despite the well-reported havoc caused by debris flows in Malaysia especially mountain and foothill communities, it received little attention from researchers. It has therefore, become imperative to explore the nature of the disaster in the tropical Malaysia. The general objective of the study was the development of optimized hybrid debris flow models using airborne laser scanning data and Machine learning algorithms in Malaysia. The specific objectives are to identify the optimized geomorphological, topographic parameters derived from LiDAR data source for the tropical area; map the debris flow susceptible areas using the LiDAR data; and develop a hybrid RAMMS (Rapids Mass Movements) debris flow model for tropical countries. The quality of spatial data required and approaches adopted in acquiring the data is directly related to the level of analyses accuracy involve and pixel size. A high-resolution vertical accuracy (15 cm) airborne laser scanning data (LiDAR) discrete-return, echoes, and intensity was used to generate DEM; invariably used to derive the debris flow conditioning factors for the spatial prediction and modelling of debris flow. The topographic and geomorphological conditioning factors includes slope angle, slope aspect, total curvature, plane curvature, profile curvature, relative stream power index, topographic wetness index, stream catchment area, topographic roughness index, and topographic position index). Other determinants were velocity and rheological parameters data that is influencing debris flows run-out. In this study, an existing inventory data that depicts a number of debris flow locations was utilize for binary features selection with high-resolution airborne laser scanning data. The features were categorized into two “debris flows present” (1) and “debris flow absent” (0). Six hundred randomly selected sample points for each category was generated gives 640 sample points. The sample data of the area was randomly divided into a training dataset: 70 % (448) for training the models and 30% (192) for validation. Spearman Correlation was used to checked multi-collinearity effect on debris flow conditioning factors; evaluations factors of Information Value (IV), Crammer V were assessed.Wrapper feature subset selection technique was used, different metaheuristic search algorithms (e.g. Cuckoo search), and evaluator or model inducing algorithms (e.g SVM) were utilized for feature subset selection, which further compared to select the optimal conditioning factors subset. At the initial stage, heuristic optimisation techniques were employed in identifying the global best latent SVM and MARS hyperparameter values selection used for debris flow prediction modelling. A susceptibility debris map is the combination of debris flow source area and run out model, this is achieved by emergent of revolutionary advancement in MLA, two optimized-data mining techniques (BFO-SVM and PSO- MARS) were amalgamated. The resultant susceptibility mapping and models strength were subjected to statistical accuracy evaluation metrics using Receiver Operating Characteristic (ROC) curve and Area Under Curve (AUC), Mean Asolute Error (MAE), Root Mean Square Error (RMSE), coefficient of determination (R²) and Generalized Cross Validation (GCV) methods. To simulate debris flow run-out pattern, a friction resistance model (Voellmy model) RAMMS-dbf was modified by fusing erosion model; this improve the model results in reality. The model is capable of ameliorating decision-making process in planning and environmental risk- hazard mitigation and management. Results have shown that integrated Cuckoo search and induced SVM learning algorithm produced the best-selected feature subset with 99% coefficient of determination, lowest RMSE and MAE of 0.081 and 0.0132 respectively.

Text FK 2019 129 - ir.pdf Download (1MB)

Actions (login required)

View Item