Dedicated Lectures: Contemporary Landslide Risk Assessments: Applications for Machine Learning and Artificial Intelligence

Abstract by Kushanav Bhuyan
Multi-temporal landslide inventories are crucial for understanding the changing dynamics and states of activity of landslide masses. However, mapping landslides over space and time is challenging as it requires lots of time and resources to delineate landslide bodies for the affected areas. With the current advances in artificial intelligence models and acquisition of very high-resolution satellite images, the need to map landslides not just spatially, but also temporally, has become quite evident. Generating multi-spatiotemporal landslide inventories can allow us to improve our understanding of evolving landslides and landslide re-activations, addressing the changing susceptibilities, and the associated dynamic risks to elements-at-risk. Furthermore, as a result of having multi-temporal inventories, the temporal probability of occurrence of landslides can also be investigated with the help of envelope curves based on variables like rainfall duration, intensity, cumulative rainfall, antecedent rainfall. Therefore, in this endeavour, we have developed a modelling strategy that generates multi-temporal landslide inventories for some of the most affected landslide regions in Nepal (Gorkha earthquake of 2015), China (Wenchuan earthquake of 2008), Papua New Guinea (Porgera earthquake of 2018), and New Zealand (Kaikoura earthquake of 2016).

Abstract by Kamal Rana
Landslide hazard models aim at mitigating landslide impact by providing probabilistic forecasting, and the accuracy of these models hinges on landslide databases for model training and testing. Landslide databases at times lack information on the underlying triggering mechanism, making these inventories almost unusable in hazard models. We developed a Python-based library, landsifier, that contains three different Machine-Learning frameworks for assessing the likely triggering mechanisms of individual landslides or entire inventories based on landslide geometry. Two of these methods only use the 2D landslide planforms, and the third utilizes the 3D shape of landslides relying on an underlying Digital Elevation Model (DEM). The base method extracts geometric properties of landslide polygons as a feature space for the shallow learner—Random Forest (RF). An alternative method extracts topological properties of 3D landslides through Topological Data Analysis (TDA) and then feeds these properties as a feature space to the Random Forest classifier. The last framework relies on landslide-planform images as an input for the deep learning algorithm—Convolutional Neural Network (CNN). We tested all three interchangeable methods on several inventories with known triggers spread over the Japanese archipelago. Classification accuracies for different testing schemes vary between 70 % and 95 %. Finally, we implemented the three methods on an inventory without any triggering information to showcase a real-world application.