Landslides in Kentucky

Machine Learning and Remote Sensing Show Where and When Landslides are Most Likely

More Rain, More Landslides

Landslides are primarily triggered by extreme or prolonged rainfall.

With climate change, some regions will likely be at more risk for landslides due to increasing extremes and human activities such as road construction, mining, and expanding development into vulnerable locations.

Landslide Impacts

Landslide hazard modeling requires information on a region’s landscape, the subsurface, and hydrologic conditions.

Landslides cause numerous fatalities and billions of dollars in damage each year. With advanced modeling, we can characterize where and when landslides may impact infrastructure and populations.

<< Landslide Fatalities from the Global Landslide Catalog (GLC)

NASA and Kentucky Studying Landslides

In Kentucky, steep slopes, weak rocks and development cause landslides, costing the state over $10M each year in landslide mitigation and repairs.

NASA is working with the Kentucky Geological Survey (KGS) to identify locations susceptible to landslides using satellite information, model data and local observations.

>> Landslide locations identified by KGS

Large, rotational slump along Old Ky. 10 Credit: Kentucky Transportation Cabinet, Matt Crawford

ML for Landslide Hazard Assessment

Using Machine Learning for Landslide Hazard Assessment

With landslide data provided by the KGS, we have created a state-wide machine learning (ML) model based on the  Landslide Hazard Assessment for Situational Awareness (LHASA)  system.

This tool incorporates GPM IMERG precipitation, soil moisture estimates from LIS, local geology maps, and landslide locations to estimate the relative probability of landslide hazard at a 1 km spatial resolution. 

>> Kentucky-wide probabilities of landslides modeled using machine learning. Left - Landslide on April 24, 2011, Right - Pre-landslide on April 23, 2011. Black dots show landslides, red indicates high hazard areas and blue indicates low hazard areas.

Seasonality of Landslides in Kentucky

When aggregated over time and space, the model outputs can reveal long-term trends and patterns in landslide hazard.

Landslide activity follows a seasonal cycle over 2015-2020. Most landslides occur in the winter and spring due to changes in subsurface water storage driven by precipitation, solar radiation, and water transport.

>> Comparison of statewide monthly mean of model outputs (blue) against total number of landslides recorded in Kentucky (red) for the period 2015-2020

Conclusions

  • Landslides are common and cause major damage, both in Kentucky and throughout Appalachia.
  • This case study shows how regional landslide hazard can be derived from remote sensing, hydrologic modeling, and geologic field research.
  • Machine learning can tell us not only about landslide hazard at a point in time but also about long-term patterns.

Large, rotational slump along Old Ky. 10 Credit: Kentucky Transportation Cabinet, Matt Crawford