Abstract:

4D Landslide Hazard Sensing

​

Landslides continue to pose substantial threats to lives, infrastructure, and the environment worldwide. For decades, engineering geologists have relied on proven traditional methods, such as field mapping, geotechnical surveys, borehole sampling, and manual monitoring of slope stability, to assess and mitigate landslide hazards. While these techniques remain essential for ground truthing and understanding site-specific processes, they often struggle to capture the spatial and temporal complexity of rapidly evolving slopes. This presentation explores the synergy between traditional ground-based approaches and advanced 4D remote sensing technologies, which collect three-dimensional data over time to map and monitor landslides with unprecedented detail. Modern tools, including satellite imagery, UAS-based photogrammetry, and terrestrial LiDAR, when combined with sensor networks, provide continuous, high-resolution information on slope movement, deformation, and triggering events. By integrating these datasets with object-based frameworks, we enhance our ability to characterize landslide mechanisms and support comprehensive hazard assessment. The true potential is unlocked through artificial intelligence: machine learning algorithms ingest and analyze these multi-modal, multi-temporal data to automate hazard classification, identify complex precursors, and predict future landslide activity.

Abstract:

Digging Deeper with Data: Artificial Intelligence as a Force Multiplier in Environmental and Engineering Geology

​

Artificial intelligence is reshaping environmental and engineering geology by enabling more accurate, scalable, and timely analysis of subsurface conditions, geological hazards, and earth-surface processes critical to infrastructure planning and risk management. Machine learning and deep learning approaches now support landslide susceptibility mapping, ground deformation monitoring, and soil characterization at spatial scales and resolutions previously unattainable through conventional field-based or numerical methods. In environmental geology, AI-driven remote sensing platforms such as Google Earth Engine facilitate continuous monitoring of surface water quality, soil moisture dynamics, contamination indicators, and land-cover change, providing actionable intelligence for groundwater protection, remediation planning, and regulatory compliance. Physics-informed neural networks and hybrid models that integrate geomechanical and hydrogeological domain knowledge into data-driven frameworks offer physically consistent predictions suited to the interpretability demands of engineering practice. However, translating AI research into operational geotechnical and environmental workflows presents persistent challenges, including limited labeled training data in geologically complex settings, model transferability across site conditions, and the need for uncertainty quantification in safety-critical applications. Advancing practical AI for environmental and engineering geology requires standardized benchmark datasets, reproducible end-to-end pipelines, and governance frameworks ensuring that model outputs meet the accountability and reliability standards expected in professional geoscience practice.​

​

Abstract:

Innovative Hazard Assessment in Engineering Geology

Utilizing EdgeAI and Vision Language Models

 

Even without the use of AI, geosensing and geospatial systems have been experiencing a rapid evolution in the past few years with smallsats, drone lidar, digital twins, gaussian splatting, and many other innovations. This talk discusses two AI-based advances that can add additional usefulness to these systems: 1) EdgeAI (also referred to as TinyML) and 2) Vision Language Models (VLMs).  EdgeAI is the process of putting trained AI models (including Generative AI and Foundation Models) onto small edge devices such as ground sensors, drones, mobile devices and wearables. Advantages of putting the AI onto the edge device are many and include privacy, energy consumption, latency, and cost. VLMs are Large Language Models (LLMs) with the addition of a vision encoder that allows images (and sounds) to be analyzed and interpreted. In terms of hazard monitoring and decision-making, VLMs can perform much like humans. For example, given a picture of a large block partially blocking a highway, the vision encoder detects the rock on the highway and the LLM part of the VLM understands that this is a hazard and so communicates the severity of the hazard to appropriate channels. If the VLM detects a newly dislodged rock block in a ditch, on the other hand, the LLM part of the VLM understands and communicates it as a lower priority event. Based on the author’s own studies and experience with EdgeAI and VLMs, this talk will cover examples that include rockfall, landslides, rock mass characterization and flooding. In addition, the author discusses the advantages of combining these two technologies to produce cost-effective, far-reaching hazard-assessment systems for the future.