Landslide Dams

Image Credit: Martin Mergili

A landslide that blocks a river creates a lake with no spillway, and when the dam gives way the flood wave can reach towns far downstream. Detection today depends on someone noticing. We will compare monthly satellite water maps against the long-term baseline to flag new lakes in steep valleys, and train a model to separate landslide dams from new reservoirs and mining pits using the shape of the valley and the lake. We will test it on the landslide dams already on record, and the output is a detector that could run every month.

Research questions

1. Can new lakes in steep valleys be flagged automatically from monthly satellite water maps, and how soon after they form?

2. Can a model tell a landslide-dammed lake from a new reservoir or mining pit using the shape of the valley and the lake, and how many false alarms does it raise?

Datasets

• Sentinel-2 water index composites, monthly, for one mountain region

• JRC Global Surface Water, as the baseline of where water normally is

• The Fan et al. global inventory of landslide dams, for the examples to learn from and test on

• Copernicus DEM, for valley width, slope and whether the lake sits behind a constriction

• Global Reservoir and Dam database and mining-site maps, for the look-alikes

• OpenStreetMap roads and WorldPop, for who is downstream

Machine learning

• A gradient boosting classifier that labels each newly flagged water body as landslide dam, reservoir or pit from valley width, slope above the lake, lake shape and whether a river enters without leaving, tested on dams the model never saw

GIS

• Compare each month's water mask with the long-term baseline to flag new water bodies, limited to steep valleys from the elevation model

• Measure valley width, slope above the lake and lake shape, and check whether a river enters the lake without leaving it

• Trace the river downstream of each flagged lake and count the roads and people on the valley

References

• Wu, Trigg, Murphy et al. (2024). A global-scale applicable framework of landslide dam formation susceptibility. Landslides, 21.

• Fan et al. (2020). The formation and impact of landslide dams: state of the art. Earth-Science Reviews, 203; includes a global database of 410 landslide dams larger than one million cubic meters formed since 1900.

• Pekel, Cottam, Gorelick and Belward (2016). High-resolution mapping of global surface water and its long-term changes. Nature.