- LATEST BLOG POST
- echo $post_date ?>
- Cadia tailings facility failure expert review and risk considerations allow us to discuss InSAR application and low seismicity considerations for…
- Read More
100 years 289 tailings dams failures history is the result brought in by a very detailed research (www.csp2.org) including a freely downloadable database and several explanatory and interpretative papers by Bowker and Chambers.
The database contains (on October 7th 2016) 289 failures occurred in 2016-1915= 101 years. Additionally the database contains all reported, failures in the period, from irrelevant/small released volumes, short runout distances, no victims, to the recent large failures (Mount Polley and Samarco). Finally, the database also records some very large cases most people have forgotten (32Mm3 in 1992 and 28Mm3 in 1982 and various releases in the 10Mm3 each). And, of course, some highly mortiferous events (despite very small volumes) like Stava, in Italy.
Unfortunately there is fragmentary or nonexistent (to date) information related to:
The failed dam were present in mines extracting a wide array of materials. From Copper to Coal, all base and precious metals, Uranium. But also sand, limestone and sodium, vermiculite, gypsum, oil sands, etc.
The actual number of tailings dams in the world is also still the object of research. In prior papers we used a highly disputed “average value” of active dams estimated at 3,500. When the census of dams will be completed it will be interesting to know where we stand.
One needs to exert major caution when “claiming time correlations” or stating that:
The proof comes from occurrences of large failures in the past. For example the two appx 30Mm3 failures we cited above. Thos were comparable to Mount Polley and Samarco in terms of released volume.
Riskope has stated this in several occasions in the past.
Most of the time these statements are just the result of human perception on shorter observation spans. Furthermore misinformation is lurking everywhere and many people make hyperbolic statements just for the thrill of it or to please the media.
In 100 years 289 tailings dams failures history there is little doubt that the root cause of failures is generally human error. That’s a point we used as a basis for our most recent paper (Oboni, F., Oboni, C., A systemic look at tailings dams failure process, Tailings and Mine Waste 2016, Keystone, Colorado, USA, October 2-5, 2016). Bowker and Chambers advocate the root cause is also lying at a higher law and policy level and use a sophisticated set of statistical algorithms to prove a correlation with copper ore production. However, as engineers, we will stay at our “lower level” of analysis.
As we all know by now, and the last failures inquiries proved it, no dam failure, no accident, generally occurs for one single cause. Key human causes of dam incidents include design errors, faulty construction, operational errors, maintenance-caused incidents, and system failures that encompass several causes. It is the “dice draw of the day” leading a sum of conscious (or unconscious) deviations to generate a failure.
Though most dam incidents are human-caused and therefore largely preventable, they are also, like human behaviour, largely unpredictable. Due diligence, peer review, monitoring, and planning can reduce this unpredictability but not eliminate it (de Rubertis, K., van Donkelaar, C., LOOK BOTH WAYS, Changing Times — The Challenges and Risks of Managing Aging Infrastructure Under a New Financial Reality, 33rd Annual USSD Conference Phoenix, Arizona, February 11-15, 2013).
As engineers we are interested in and have to deliver replies to the questions detailed in the sections below, even if, to do so, we have to “bend some rules” of rigorous mathematics.
If we take all the failures in the 100 years 289 tailings dams failures history we can evaluate an average frequency of 289/101= 2.86 failures per year over one century. This long term average on all failures, ignoring the type of mined material, etc. is obviously fraught by significant uncertainties.
In the future we will attempt to “clean and filter” up the data, add details, but at this time this is all we can do. If we were to exclude smaller failures (thus limit, for example to significant and very significant events, the frequency would of course reduce slightly). When we will analyze further the data we will work using decades, as we did in our prior papers, as the averaging over a century masks “decennial spikes”.
If we use the average frequency and apply the Poisson model to derive the probability of occurrence of failures in a future time t (one year, n years, etc.) we get the two graphs below, respectively for “next year” and for “the next decade”.
On the horizontal axis of both figures there is the number of failures -n- and on the vertical axis the probability -p- associated to it. With a frequency of 2.86 there is p=0.055 chances that there will be no dam failures next year, respectively p=0.17 that there will be exactly one, etc. Note that the p value for exactly 4 is almost identical to the p value of exactly one. Two to three failures next year are the most likely numbers.
Below you have the Poisson distribution for the next ten years. The reading logic is the same as above.
Using the database it could also be possible to derive a “rate of failure”. That of course provided we know the number of (active) dams present in the world-wide portfolio. Of course we also have to ensure some kind of compatibility (avoid comparing “apples and bananas”). If the service life of each dam (from start to failure) was known, the numbers would be more correct. However we do not have that duration for many dams.
Again, by working “by decade” the error decreases, but certainly does not disappear. In 2013 we performed a first attempt of evaluation (Oboni, C., Oboni F., Factual and Foreseeable Reliability of Tailings Dams and Nuclear Reactors -a Societal Acceptability Perspective,Tailings and Mine Waste 2013, Banff, AB, November 6 to 9, 2013) using a frequently referenced number of 3,500 active dams in the world which is “approximate at best”, as discussed above.
We also know about the tendency to under-report failures. However, larger releases receive more attention than smaller failures. With respect to the first half of last century, due to increased public awareness, news and social channels we believe underreporting is diminishing.
Keeping in mind these “self-balancing” omissions and uncertainties, we doubt the results would significantly differ from those in our 2013 paper. Those were 10-3 to 10-4 failures per dam per year (1/1,000 to 1/10,000). That, incidentally, is in the same order of magnitude of Bowker and Chambers results.
This is the simplest way to bring the numbers under an umbrella everyone should understand. Furthermore it also bring the number under risk tolerance thresholds like Morgan and Whitman or ANCOLD. These use annual rates/probabilities of events.
We perfectly understand that Bowker and Chamber actuarial approach has shown a correlation of failures with mining activities (and price of copper). We also salute their valuable contribution to actuarial science. However, our purpose is to be able to talk to the public and miners making ourselves understood. Thus we need to derive more obvious metrics (see also note on risk tolerance earlier).
We are firm believers in rates of failures (failures per dam per year), annual probabilities etc. Those are way simpler, familiar indicators, although often difficult enough for the public to properly understand. That’s why, while asserting we need to look at consequences of failures holistically, we concentrated our effort in showing how hazardous these structures are to human life.
That’s a language people understand. Additionally,, there is a solid body of knowledge linked to loss of life tolerance in hazardous industries. To this day we have clearly shown that tailings dams are way more hazardous to human life than, for example, processes in the hazardous chemical industry.
At the end of the day what we all want is:
In the future we will delve in the database and try to extract/complete as much information as possible. Of course, if we knew the number of the world’s active dams, we could push the analyses way further.