Optimized techniques for landslide detection and characteristics using LiDAR data

Landslides are one of the major natural disasters that cause severe damage to lives and properties worldwide. Historically, landslide occurrences are usually mapped by taking inventory of location and magnitude of the landslide in a region. This information is used to examine and manage slope failures and distributions effectively. A good landslide inventory map is a prerequisite for analyzing landslide susceptibility, hazard, and risk. Field survey, optical remote sensing, and synthetic aperture radar techniques are traditional techniques use for landslide detection in tropical regions. However, such techniques are time consuming and costly. In addition, the dense vegetation in tropical forests affects the accuracy of the landslide inventory maps. Furthermore, it is difficult to distinguish different types of landslides due to geomorphological development along hillsides using the traditional approach. so, it necessary to develop more innovative approach that can resolve the aforementioned challenges. Therefore, in line with the objectives of this research, very- high-resolution LiDAR point cloud data and orthophotos image, have been utilized to map the landslide events in Cameron Highlands, Malaysia. The segmentation process was optimized using Fuzzy-based Segmentation Parameter. Also, six techniques: Ant Colony Optimization (ACO), Gain Ratio (GR), Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), Random forest (RF), and Correlation-based Feature Selection (CFS) were used for the feature selection. The locations of landslides were detected accurately by employing two Machine learning classifiers, namely, SVM and RF, decision rule and hierarchal rules sets were developed by applying decision tree (DT) algorithm to provide improved landslide inventory. In this task, two neural network algorithms, Recurrent Neural Networks (RNN) and Multi-Layer Perceptron Neural Networks (MLP-NN) were used and the hyper-parameters of the network architecture was optimized based on a systematic grid search. The performance of the outcome was validated based on the receiver operating characteristic (ROC) area under the curve (AUC) values, confusion matrix and Cross Validation method. Transferability of each of the models was verified by testing in another site for consistency. The overall accuracy of the Support Vector Machine SVM and Random Forest RF classifiers revealed that three of the six algorithms exhibited higher ranks in the landslide detection. The classification accuracy of the RF classifier is observed to be higher than that of SVM using either all features or only the optimal features. The proposed techniques performed well in detecting landslides in tropical area in Malaysia. Furthermore, the transferability indicates that the techniques can easily be extended to any region with similar characteristics. The result show that the accuracy of shallow and deep-seated landslides were 0.80 and 0.83, respectively. The intensity derived from the LiDAR data, geometric and texture features significantly affects the accuracy of differentiating shallow from deep-seated landslides. While, the results of shallow and deep using hierarchal rules set were observed to be 87.2%, and 90% respectively, for site A (Analysis area). More so, the hierarchal rules set were evaluated using another site named site B (Test area), and the accuracies of shallow and deep seated were found to be 86.4% and 80.8% respectively. This indicates that LiDAR data are highly efficient in detecting landslide characteristics in tropical forested areas. Furthermore, RNN and MLP-NN models in the test area showed 81.11%, and 74.56%, accuracy level, respectively. These results indicated that the proposed models with optimized hyper-parameters produced the accurate classification results. The LiDARderived data, orthophotos and textural features significantly affected the classification results. The results indicated that the proposed methods have the potential to produce accurate and appropriate landslide inventory in tropical regions such as Malaysia. Hopefully, this innovative method can be deployed in detecting landslide and distinguish between different types of landslides (shallow and deep-seated landslides) in the near future for landslide management due to its transferability capabilities to different environments