Modelling the Brumadinho failure with publicly available data

In version v2019.3.1 of Muk3D some simple tools for dealing with dambreaks were added. These tools are suited to a particular type of beach where there isn’t a large volume of water mobilising tailings and instead the tailings mass flows by gravity alone. These may be referred to as ‘sunny day’ failures.

In January 2019 a tailings dam near the town of Brumadinho failed and killed hundreds of people. Without a large pond in the TSF, the liquefied tailings flowed downstream and formed essentially a deposition surface downstream of the failed dam.

In this video, we try and recreate the footprint of the tailings runout surface using publicly available data (SRTM survey for the surface retrieved from USGS Earth Explorer), an estimate of the maximum runout volume, and aerial imagery from Google Earth.

Note that the approach for modelling tailings runout in Muk3D is a simplified geometric approach. The inputs are simply the at-rest slope of the tailings mass and the volume of tailings released. There are many factors that might need to be considered when using more advanced modelling (e.g. Flow2D) but this approach is a quick and easy way to get some first pass estimates almost immediately, and for applying a reality check to the results of more detailed modelling.

We also note that this approach is not applicable for all dam break scenarios. If you have a large volume of water in the pond at failure and this mobilises lots of the tailings solids then this is likely not a valid way to estimate the runout behavior.

There are some things we could have done to improve on the modelling in this video. The first thing would have been to use the Watershed tools to remove some small depressions in the topography caused by some noise in the SRTM dataset. This would improve the speed and stability of the deposition models.

Also, given the scale and level of detail we’re working at, the dataset could have also been thinned – larger grid spacing (less points) will yield faster results without necessarily impacting accuracy.