Development of a Machine Learning Interpretation Aid to Support Rockfall Hazard Forecasting

Geohazards present a substantial risk to Colorado’s transportation network. While overall geohazard risk can be assessed at the individual site/asset scale, corridor scale, or network scale, individual hazard event occurrences tend to be somewhat “random” in nature. What this means is that responses to individual hazard events occur on an emergency or urgent need basis, with no ability to proactively identify and prioritize mitigation measures based on real-time data. Several recent studies (e.g. Kromer et al., 2015; Kromer et al., 2018; Walton et al., 2023a) have demonstrated that many rockfall events exhibit precursors that are detectable by lidar or photogrammetric monitoring months or even years before the final point of failure. The advantage of such monitoring technologies relative to traditional in-ground monitoring devices is that they can cover large slope areas, meaning multiple source areas or an entire cut slope asset can feasibly be monitored. Accordingly, one can envision a future where large numbers of rock slopes are regularly monitored using remote sensing technologies, and potential emerging hazards are identified, forecast, and mitigated, thus minimizing the number of actual rockfall events that ultimately occur. Data collection and processing limitations that only a few years ago made it difficult to implement such monitoring at-scale have now been overcome. However, a new challenge now presents itself – with ever-increasing volumes of monitoring data now available, the process of manually evaluating spatially extensive change detection results to determine which subtle apparent ground movements over time represent actual failure precursors (as opposed to noise or data processing artifacts) has become a major bottleneck for potential failure forecasting efforts. As a result, at present, while some events that exhibit particularly large and long-lasting precursors can be detected and mitigated (e.g. Walton et al., 2023a), the vast majority are not identified prior to failure (e.g. Walton et al., 2023b). The focus of this research project is to develop a machine learning interpretation aid capable of performing a preliminary screening of a point-cloud-derived change detection result to identify potential areas of interest (i.e. where failure precursors may exist) that merit further detailed consideration by an expert. The development of such an interpretation aid would substantially reduce the effort required to effectively forecast rockfall occurrence, and would increase the reliability with which individual events could be successfully forecasted. The following research objectives shall be met: (1) develop a rockfall precursor database for Colorado slopes of interest using change detection results from historical monitoring; (2) explore basic statistical approaches to separate potential precursors from other changes that are not of interest; and (3) test machine learning approaches to identify areas of potential interest/concern meriting manual interpretation for forecasting purposes.

Language

• English

Project

• Status: Programmed
• Sponsor Organizations:

  Colorado Department of Transportation

  Applied Research and Innovation Branch
  Denver, CO  United States  80204
• Managing Organizations:

  Colorado Department of Transportation

  Applied Research and Innovation Branch
  Denver, CO  United States  80204
• Project Managers:

  Tran, Thien

• Performing Organizations:

  Colorado School of Mines, Golden

  1500 Illinois Street
  Golden, CO  United States  80401
• Principal Investigators:

  Walton, Gabriel

• Start Date: 20250601
• Expected Completion Date: 0
• Actual Completion Date: 0

Subject/Index Terms

• TRT Terms: Data analysis; Forecasting; Machine learning; Rockfalls; Statistical analysis
• Geographic Terms: Colorado
• Subject Areas: Data and Information Technology; Geotechnology; Highways; Planning and Forecasting;

Filing Info

• Accession Number: 01930632
• Record Type: Research project
• Source Agency: Colorado Department of Transportation
• Files: RIP, STATEDOT
• Created Date: Sep 16 2024 9:08AM