Wednesday, February 6, 2019

Video Analysis of Vale Dam Collapse at Brumadinho

Below you will find annotated video analysis of the Vale Córrego do Feijão Dam I collapse at the iron ore mine near Brumadinho in Brazil. This is being done in an attempt to learn early lessons and identify takeaways. Close up in slow motion (0.50x and 0.25x speed) reveals initial slide movement, failure sequence and possible trigger.

Movement is seen initially at a seepage point in the toe of the dam near a "groin" area (where a starter dyke reinforcement extends outward at the center of the dam .... left side in the video). The movement is possibly a gush of water being ejected at the seepage point due to a change in pore pressure indicating that mobilized shear stress had overcome shear resisting force and initiated sliding along a critical slip surface. Based on this and other indications, the collapse of the dam is possibly yet another case of static liquefaction triggered by progressive deformation and increase in pore pressure, not (necessarily) seismic activity or other external disturbance. The dam's dimensions made a rotational slide involving most of the tailings possible and contributed to a large wall of mud and substantial volume of tailings rapidly descending downstream into the valley below the dam. See this explanation of static liquefaction in upstream embankment dams by Klohn Clippen Berger as well as here for potentially applicable principles of abrupt rotational failure.

In the video, the toe seepage area is being inspected by 2 personnel who notice the collapse first and begin to flee almost 2 seconds before animals and inspectors higher up the face of the dam notice anything is wrong. This is not surprising in a rotational slide as the back of the dam (not visible in the video) would begin to sink at the same time as the toe starts to deform and burst whereas the slope face and ridge line wouldn't undergo significant initial movement (the face and ridge are located closer to the rotational axis).

As a slow motion pixel analysis of the video reveals, however,  the central portion of the dam ridge starts to deform at the same moment that initial toe movement occurs. It takes another second before the entire crest line of the dam begins to slide as a single unit. Moreover, the area of initial movement on the ridge is right above a drill rig that is boring a monitoring well on the dam face. The monitoring well is likely investigating an area where piezometers have detected but not resolved a fluctuation or build up in pore pressure inside the dam. The location of the drill rig is along a vertical line with the areas of initial movement at the crest and toe (and also this line of initial movement likely continues to the back of the dam not visible in the video) strongly suggesting the involvement of these vertically-aligned areas in the triggering of the collapse.

Static liquefaction can possibly result in a rotational slide failure as observed at Brumadinho when a critical slip surface undergoes failure due to mobilized shear stress. The slip surface is the hypothetical (and real in case of actual collapse) curved plane where shear stress can overcome resisting forces and cause abrupt failure in a sliding motion. Shear stress along this sliding surface could accumulate (and fluctuate) over time due to a number of factors such as internal piping erosion, zones of frequent phreatic surface changes due to dry-wet cycles, the presence of sediment layers with potential to contract under shear, saturated sediments with high pore pressure especially if the phreatic surface extends into the dam's toe, poor drainage, etc. In addition, the sliding surface may become "lubricated" if the toe and upper portion of the dam have hydraulic conductivity enhancing and focusing the mobilized shear stress. The instantaneous and isolated movements at the toe and ridge of the dam prior to unitary sliding motion suggests that such hydraulic conductivity did in fact exist.

Some possible indicators for accumulation of shear strain could include:
  • deformation in the face or toe of the dam (as observed at the start of the video in the toe area before any sliding motion and also during the trigger moment at the central ridge area);
  • rise in pore pressure (which could potentially be detected ahead of time, except there are signs the piezometers at Brumadinho were not functioning correctly);
  • intermittent seepage especially sudden gushes of water from the dam's face (as possibly seen in the toe area during the trigger moment);
  • rise in water levels inside the dam with potential for increasing shear load and incursion of the phreatic surface into the dam's toe area.

The following animation is taken from two separate Google Earth satellite images from 2009 and 2018. The images selected are the best available match in terms of photographic angle and rotation to provide good overlap. As can be seen, there was significant depression/deformation at the toe of the starter dyke over the course of 9 years. This is likely due to subsidence in the dam fill materials used to construct the starter dyke. Natural drainage patterns resulted in piping erosion and possible cavitation from unmitigated incursions by the phreatic surface into the toe area.

A review of Google Earth images between 2009 and 2018 indicates most of the deformation occurred post-2015 when the dam was undergoing decommissioning. This procedure involved a system of surface pipes installed to help dewater the dam along its face to prevent surface erosion. No internal drainage (e.g. buried perforated pipes or pumps) was provided, however, and this could have contributed to the focusing of water flow within the dam into naturally-developed piping and channels that carried away some of the tailings. If this piping extended into the toe it could have possible formed hydraulic conductivity between the top/rear and bottom/front of the dam resulting in weakened layers of fill, saturation zones and even open cavities that could help mobilize shear stress along a critical slip surface involving substantial portions of the tailings in a rotational slide.

A further critical consideration in the potential risk of tailings dams is the geometry of the original land surface under the tailings and the consequent restrictions on the design of the dam in terms of its layout and geometry. As pointed out by Klohn (see link above), embankments constructed on top of tailings using the upstream method should not have a height exceeding 40 meters. The collapsed dam at Brumadinho had a height of 87 meters, or more than double this recommendation. Moreover, the width of the dam at the ridge was 694 meters and even included a protruding feature (the center sticks out from the natural topography along the sides) to maximize tailings capacity at 10+ million tonnes. Without this protrusion (which ostensibly explains the extension of the central toe area in at attempt to reinforce the structure), the dam might have held no more than a much more reasonable 3-5 million tonnes of tailings. These factors were strong, and in retrospect obvious, contributing factors in how this tragedy unfolded especially with respect to the enormous toll in human lives.

Indeed, instantaneous rotational slides may not create such extreme risk to life downstream in cases where the initial slump volume is not substantial. The reason why the slump volume was so large at Brumadinho is because the dam was built into a hillside not a topographical depression like most other large tailings dams in this part of Minas Gerais state. As a result, the dam was dimensionally compact meaning that its ratio of height to width to length was closer to one (1) in comparison to most other tailings dams. This doesn't mean that other dam geometries will not result in substantial spillage, risk to life and environmental damage, only that there might be more time to evacuate assuming an adequate monitoring, warning and alarm system are in place. Even if initially only the front portion of a dam is involved in a sudden rotational failure, the loss of confining force will probably cause liquefaction in rearward areas of tailings as well. Yet any delay or slowing of tailings movement downstream would extend the time window for evacuation. And this is precisely where the culpability of Vale management is perhaps the greatest and why in large part I have spent so much time analyzing this mining tragedy.

Simply put, the evacuation time window at Brumadinho was effectively non-existing due to the Dam I layout and geometry. This factor should have been taken into account when locating mine offices and buildings immediately downstream as well as residential and municipal buildings within the valley nearby. Even if monitoring and assessing the safety of the dam could be complex and difficult, moving facilities to higher ground was not (though there would be an obvious cost, explaining why it wasn't done). So we come back once more to a cost-saving decision that cost many people their lives. Deplorable.

Below is an image taken from a March 2016 presentation by Vale personnel in the aftermath of the Fundão tailings dam collapse. From this angle, the imposing nature of the Córrego do Feijão Dam I is readily apparent. It should be quite obvious even to the corporate apologist or intellectually disinclined that the very location of this dam protruding out of a hillside as well as its compact form were by themselves inherent risks of, and in, a potential catastrophic failure. The photo also places into proper context the reported statements of mine personnel and others who were fearful of an imminent collapse in the weeks and months before the tragedy.


  1. Lots of technical analysis I have no way of understanding. But conclusion is simple: if something is messed up enough any idiot can see it. In construction, we field people often see stuff with an engineers stamp that we know will fail. We often joke about the difference between theory and reality, but when there gets to be too great a difference between the two, reality can be no laughing matter.

    1. As I stated above, there is probably a critical aspect of a sudden failure risk assessment in dams of this type that has apparently been ignored by both engineers and operators: the "compactness" of dam geometry resulting in a wide dam face/crest, a large volume (millions of tonnes) of tailings being located in a potential slump zone, and a steep natural topography under the dykes and tailings. This might not be readily apparent and in fact based on the reporting on this dam failure so far it STILL isn't readily apparent to those reporting or commenting on the disaster.

  2. How was this damn built? I’ve heard it’s a natural valley that was walled off. Also, is it really a pond, or were the tailings dewatered before placement?

    1. There is a difference between building a dam on a hillside with a sloping topography vs. building a dam in a valley with a small (V-shaped or U-shaped) outlet that is "walled off". Many large tailings dams in Minas Gerais (and elsewhere) are built "walled off" and this might make them less likely to undergo a sudden rotational failure that involves large volumes if not most of the tailings behind the dam. In contrast, both the collapsed dams at Brumadinho and Mariana/Bento Rodrigues had a wide crest and were built into a hillside so that almost all the load of the tailings was acting on the dam itself.

      Tailings are typically not dewatered though might be thickened prior to being pumped to the storage facility. Usually tailings need to maintain a slurry consistency to be pumped (especially uphill) which means solid content rarely exceeds 50%.

  3. Great video Tom! The dam shell appears to kick out just like a river bank slump, exposing the contained saturated tailings. The question of what caused the movement will be extensively investigated, as it was with Mt Polley. Was it lacustrine material under the dam? Or was it something else? I'm hopeful the investigation report will be made public so that we all will know.

    1. At Mt. Polley the natural overconsolidated glacial soil under the foundation was a critical issue. It became normally consolidated under increased load and eventually failed due to contraction during undrained shearing. I don't think there is a similar problem with either natural or disturbed lateritic soils in this part of Brazil. See:

      I do think, however, that there was probably a mechanism at play here that is similar to Mt. Polley. Near the toe at base of the Brumadinho tailings there is a critical area (identified as "Layers" 6, 7, 8 and 9 here: ) that was found to be contractive. These layers correspond almost exactly to the elevation where you see the toe bursting in the video.

      Importantly, the study I referenced above may have made some critical mistakes beyond stating that safety factors under 1.5 are adequate. What it may have also missed is that the toe of this dam isn't one straight or parallel alignment but includes a downhill extension in the central area. This toe extension retreats to a steeper overall gradient along the sides with a resulting "groin". This groin area of the starter dyke is precisely where BOTH these contractive soils AND the phreatic surface could possibly extend critically close to the dam's toe. Therefore the generalized model of the dam used in the study is faulty. The only area where it might be correct is along the toe extension in the central part of the dam. In fact, this toe extension apparently did not undergo failure as you can see in the video where the upper benches of the dam essentially slide down "on top" of the central portion of the starter dyke.