In the wake of the Mt. Polley tailings Impoundment failure on August 4, 2014, we have been working with the ICOLD/WISE data (see link at end for machine readable compilation) to assess the feasibility of risk pooling to fund the huge unfunded and presently unfundable public and environmental liability in every TSF failure. This requires reasonable predictions of the number and average cost of TSF failures for a given large group of TSF’s. Most within industry researchers working with this data and publishing technical reports with reference to it have looked at frequency and at causes of failure, i.e. descriptions of what has already happened. Few have worked specifically with the cost consequence of public and environmental liabilities or with predicting future failures.
Only two presentations we found in the literature spoke to prospective consequence and consequence trends. Rico/Benito/Diego (2008) used actual ICOLD/WISE data supplemented with additional research to create a complete uniform data base. They developed regressions to predict volume of spill for a TSF of a given height and capacity and for calculation of run out in distance (size of the area potentially impacted)
“simple estimations can be performed based on generic empirical relationships. In these equations, key hydrological parameters associated with dam failures (e.g. outflow volume, peak discharge, mine waste run-out distance) can be estimated from pre-failure physical characteristics of the dam (dam height, reservoir volume, etc.), based on reported historic dam failures. This approach has been successfully applied to estimate peak discharge and flood volume resulting from water-dam failures [8,9,10].” .
V f=0.354*Vt1.008 r2=0.86 where Vf= spilled volume vt=Total volume
Dmax=1.612*(H*Vf).0655 R2=.057 where h=height;Dmax-= run out flow in Km”
The above equation shows, that in average, one third of the tailings and water at the decant pond is released during dam failures” http://www.academia.edu/6662478/Floods_from_tailings_dam_failures
Using an estimate of total volume at time of failure based on verified dimensions and stated freeboard( (59.6 million cubic meters) we did the failure volume calculation for Mt. Polley) It came out very close to actual: predicted failure volume 21.9 v 24.5 actual)
Our search for what is needed to evaluate the feasibility of risk pools for public and environmental liabilities, which are now largely both unfunded and unfundable, lead us to some insights that also have a bearing on descriptive interpretations of the failure data. Our findings mainly affect the interpretation of the failures per decade and the popular claim that the data show improved performance in the last two decades attributable to better knowledge and better practice. These claims have been based on frequency data alone which is not a complete or even best indicator of potential liability Further both the numerator and denominator of those numbers are fuzzy and frail. The denominator “3500” is really just an informed best guess that has been used by virtually all other users of the ICOLD/WISE data as “gospel” The numerator, # of failures per decade is not a complete inventory even post 60’s but again its the best number available so it too is treated as “gospel”. In the world of the statistics that we need to rely on to give some realistic shape to public liability exposure these are very “soft and fuzzy” numbers. Coupled with the irresponsibility of citing only frequency it’s not much to “stake a claim” of improved industry performance on.
Taking Dr. Robertsons lead and going beyond the limitations of the ICOLD/ WISE data to the actual numbers the industry and its investors and analysts rely on it is very clear that public liability loss exposure is shaped by two numbers: ore grade and production volume which have been spreading from each other in opposite directions since about 1950. These two numbers in turn are what is driving the characteristics of risk in the TSF’s them selves: Height and more importantly TSF total capacity. As dropping ore grades force more and more volume of ore production for essentially a flat line production of refined product , more waste is generated and larger TSF’s are created mostly we assume, as at Mt. Polley through expansion of existing TSF’s.. The correlations between Total number of failures (tfail) and ore grade (oregrd) is -0.654). The correlation between magnitude of failure and ore production is 0.835.(4)
Correlation Matrix of Key TSF Failure DescrIptors/indicators
Bowker Associates Science & Research In The Public Interest October 2014
The two variables MXPRIC and PRICYC were an exploration of price as a predictor and our data element PRICYC does have a strong correlation, -0.496, with the number of large failures. We coded PRICYC to try and capture the character of the price climate over the decade L 0=no change, 1= 1 decade of upward price, 2 =2 decades f upward price,3 = 3 decades of upward price and -1, -2,-3 for downward trends. MAXPRIC was the highest price per ton for ore attained in the decade. MAXPRIC had very low correlations with the non price variables explained by a line almost parallel with downward grade of downward cost per ton for production(3) essentially nullifying price as a factor in predicting decade trends for all TSFS’s. ( Although of course on an individual mine basis price often makes a particular mine infeasible and results in either going on standby as at Mt Polley or not going forward with a mine application.)
We were interested to see how this “maps” for the whole of failures as the price of copper ( in constant$2010) maps exactly to the failure histogram. We believe on further analysis it will prove a useful indicator of operating stage of TSF life. Specifically, we hypothesize that the combination of price/price trend and production will define periods of intense active use of TSF’s where a higher failure rate with higher consequence are most likely to occur. and for identifying “at risk ” TSF’s. The 0.496 correlation of PRICYC with consequence ( LRGFAIL=# of failures gt 1 million cubic meters)suggests that as well.
The bottom line is that when both frequency & consequence are taken into account the modern era performance is lower than at any other time in history and continuing in that direction.
We explored the discriminating/predictive power of expressing TSF failure rate in terms of global mine production ( TSF Failures per million tons of ore produced) but found it had lower correlations and performed less well in trial baloon regressions than failings/tsf. We will look at this again when we have more data. Data on mine production is more solid and established over the entire 100 year period whereas exact data on inventory of TSF’s is not and probably never will be historically. More importantly there is no comparison between a typically small pre 1950 TSF and a modern TSF like Mt. Polley. Using mine production as a base for failure rates includes that difference to some extent. On a per million ton basis( scaled by 100 ) the rate of failure is actually the same pre-1940 as compared with post 1990 (.0020 ). On a per TSF basis adjusted for the smaller number of TSF’s pre 1950 the frequency is actually greater in the two most recent decades.005 as compared with .001 pre-1940. Consequence has been constantly escalating since 1960 therefore overall performance either way is significantly lower.
By definition, by the way, failure rates over this entire history as at least one other researcher has recognized, are well above what is considered reasonably attributable to chance. Rates this high over the entire history of mining by definition indicate human error at work ( ie an established pattern of failure to take reasonable precautions to control off site damages with increasingly grave consequences). Even the overall failure rate for water dams, .0001 ,is just barely in the range of what could be considered attributable to chance.
Azam/Li (2010) arrayed the ICOLD/WISE data into this histogram of incidents per decade from 1900 to 2000 for a total of . 218 failure incidents . http://www.infomine.com/library/publications/docs/Azam2010.pdf
The stat most frequently cited is that the overall failure rate for the century is 1.2% expressed with reference to the total number of mine sites, 18401.
By custom the pre 1960 decades are not usually included in analysis. rates. According to TailSafe 1136 TSF’s were built before 1950 and had failure rate of .0220. That number is probably every bit as “solid” as the “3500 and any comparisons on a per TSF basis should use 1136 as the basis. We think it is important to understand TSF failures in the context of the entire century. 1960 in fact is key.It marks the beginning of “post peak” metal extractions where demand for metals is met with ever increasing production volumes relative to final output of refined metals.It also marks the beginning of the ever increasing spread between ore grade and total production. and a dramatic change in the profile of mines generating production from a very few large mines pre 1960 to many more mine sites of widely varying size. We are looking to use as much reliable data over the longest term possible.
The rates (2) cited as evidence of improved performace by the industry(failures/3500) are .01467/decade for the 60’s 70’s and 80’s and .0052/decade for the two most recent decades shown on the chart This interpretation seemed to warrant a closer examination First, because frequency alone does not measure outcome. Frequency and severity together measure outcome. ICOLD/WISE have no data on cost of failures and only spotty data on the size of the failure which most agree is a good surrogate for severity of TSF failures. There is general consensus within the industry that pursuit of profit at ever decreasing average ore grades results in larger and larger volumes of waste and therefore larger and larger TSF’s with a greater consequence in the event of failure. It is universally recognized that impoundment size is a principal driver/ indicator of potential consequence..
In a keynote address at a 2011 Tailings & Waste conference, A.MacG. Robertson ( 2011 )http://www.infomine.com/library/publications/docs/Robertson2011c.pdf looked at the entire century beyond the failure incident data just in terms of changes in “potential risk” over time as indicated in the upper limits of height and volume achieved per decade He estimated that the volume of “potential waste” per 1/3 century had increased 10 fold accomodated by a 2 fold increase in achieved maximum height and a 5 fold increase in achieved maximum volume of TSFs per 1/3 century. While pointing in the right direction and framing an excellent theoretical model for looking at “consequence of failure”over the entire century, it is not useable as an actual measure of consequence over the century nor did he intend that it be used that way.
This is the formula for “expected loss” (size of group,* frequency of loss* average loss for group). So with more data on both the standing inventory of TSF’s and on failed dams this is exactly the formula for estimating total liability in any group of dams. We hope to be able to explore development of a Risk score for TSF’s based on that and other data not yet compiled by any known source. We are hoping to “round out” information on at least 28 more of already failed dams and develop a data base of at least 50 dams representative of the standing inventory.
Reading the dam committee reports at WISE / ICOLDand surveying all the literature on TSF failures available online it was apparent that the greatest magnitude of loss for any given TSF failure was in periods of active production of ore as at Mt. Polley. Errors in the deposition of tailings and in the rate and size of raises were of concern throughout the literature indentifying active operations as the most critical period in a TSF where “best practices/best knowledge is most important in preventing TSF failures. So we set about looking for some published and reliable data that might provide a more complete framing of these 218 TSF failure incidents against the periods in which TSF potential consequence is highest, in periods of actual production and in periods of price upswings.
The first issue we considered is what basis to use for frequency. The table below compares three ways of looking at frequency. Per TSF is customary in all literature we have located via on line search and of course the most “normal” if we actually had “census” on TSF’s which we don’t.. So we wanted to explore other approaches. The table below, compares frequencees per TSF ( with a correction for the smaller number of mines pre 60’s) with frequency per mine site and frequencies based on production volume. Per mine site is often used to cite century performance, 1.2% is the most frequently cited number. However this is not as sensitive to the likely inventory of TSF’s over time so shouldn’t be used as a basis for stating overall performance either. Overall performance should be stated on the same basis as per decade analysis.With the pre 1960 adjustment the failure rate per TSF is .015 pre-1960 v..055 60-s through 2010. Much higher post 60. Most researchers though only cite as “modern performance as the 90’s and 2000’s and the failure rate 0.0057.
On an ore production basis the failure rate over the century is an atsronomical .0449 mainly determined by the extremely high rates in the 60’s 70’s and 80’s.
|Comparison Of TSF Failure Frequencies Per TSF V.Per Mine siteV.Per Unit Production|
|Decade||TSFincidents||Per/TSF*N1=1136 N2=3500||Per/SiteN=18401||Per/10 6 tons/10a CU mine production a to create closer nominal scale|
|Bowker Associates Science & Research In The Public Interest October 2014|
Knowing that the production at mines in operation is frequently interrupted by falling prices which can affect most of a decade we first looked to data on copper prices over the entire century 1900 to 2000 and found that it mapped exactly into the shape of the hisorgram suggesting that peaks in failure incidents in the histogram in the 60s 70s 80s were in a period of price increase and that the two decades of the 90’s and 2000’s were in a period of general price decline where it would be expected there would be a higher proportion of mines and TSFs in standby mode, ie not actively extracting. This was the case at Mt. Polley which was reopened in 2005 after a 4 year period of no extraction due to falling market prices.
These periods of long upswings and long downswings in copper prices also presumably affect inventory of standing mines and inventory of standing TSF’s with the possibility of additions to both on long periods of upswing and the possibility of permanent shifts from “active” life phase to “closed” during long downward trends in price. Here in Maine a long expensive history of exploring Bald Mountain, a small, low grade, high risk VMS deposit had reached a point in 1990 where Boliden was looking at the possibility of active extraction. As the possible operation was too small and too uncertain they passed it off to a Denison subsidiary whose application was withdrawn in 1997 again citing falling metal prices. If that deposit had been a higher grade( and not had such extreme risk characteristics) it might have shifted into the “active life” phase with a small new “TSF” adding one new mine and one new TSF to inventory. On a down swing if a deposit is close to mined out it might just go into earlier than planned permanent closure of the mine site and of the TSF.
We are now in a period of continued sustained upswing in copper prices and as a result of that we would expect both a higher frequency of failure, exceeding those of the middle three decades (60’s,70s, 80s) and significantly greater magnitude as grade has continued to decline over the past 1/3 century and the size of the standing inventory of TSF’s has pushed to greater heights and greater volume as compared to the size of these same facilities in the middle decade.(3)
This is very much the case at Mt. Polley and in the exit letter the designer of the original TSF who had continuously served as consulting engineer to the mine owner expressed concern about the size of the facility. Further specific details about their concerns were found in the recently released 2009 annual inspection report submitted by the consulting engineer to the mine owner/operator.
What this all means is that degree of risk in any given standing TSF has a tendency to increase over time if it remains in active use because as production continues, the TSF grows in both height and total volume.
Looking at TSF failures against global copper production 1900-2010 yields perhaps better data and more insight on actual trends in TSF failures and consequence implications especially when taken in conjunction with the world bank graph of copper production, ore grade and ore production over this same period. As a pedictive variable we found very high correlations with magnitude of failure but we found the best correlation and the lowest redundancy and highest independence between potential Y,X variables using failure/per tsf and using volume of production as a surrogate for “average impoundment size”
Looking at the world bank graph below we see that the spread between the blue line ( ore extracted (=waste volume)) compared with the flat refined copper line (red line) continually increases after 1960. The blue line ( total production of ore to produce the same level of refined copper), in effect is the same as the magnitude of consequence and more or less graphs what Dr. Robertson was conveying in his 2011 key note tailings conference address.
Mapping the histogram of 218 TSF failure events (Azam/Li (2010) onto this graph, expressing it on an incident per million tonnes of production basis, shows that the two decades pre 1960 had a failure rate of .002 per million tonnes produced ( adjusted for China which is not reflected in the TSF failure incident data) . and almost the same for the two most recent decades.( see table above)
Given the difference in magnitude of consequence over the post ’60 period ( larger and higher impoundments to handle more and more waste per unit of final metal) this is obviously a considerably worse risk management performance by the industry as a whole as compared with the pre- 1960 era ( looking only at TSF’s).
The middle period (60s, 70s, 80s) , of maximum TSF failure incidents was at a time of frenzy chasing an almost continuous upward trend in copper prices ( in constant $2010) and as expected during such a big push on production the pressure on safety of TSF’s shows in much higher failures rate of .00878 per million tonnes(scaled by 100).
The big price push over this period was driven mainly by demand from China who were both stockpiling and using at very high levels and by electrical infrastructure demands in developing/modernizing nations.. By 2010 China was a significant producer of refined metal in its own right with 25% of the global refining production and 10% of global production through mining.(after a very bad record of TSF failures not reflected in the ICOLD/WISE data)
This preliminary analysis does not factor in the growing trade in copper concentrate solutions(secondary refining) at about 16% of total mine production in 2009 or the increasing role .and the increasing role of SW/Ex which was almost 20% of all production in 2009. Both of these changes in the profile of “mine production” have implications for TSF utilization and expansion
End Notes & Links
(1)Here is a link to Tailing.Info’s excellent codification of the ICOLD/WSE data. It is not current with that data. ICOLD/WISE is immediately updated and revised as each new failure occurs so does not include Mt. Polley.Its consistent codification though does significantly improve the possibilities of using the ICOLD/WISE for systematic analysis.
Thank you Eric A. Tuttle for volunteering to transfer the Tailings.Info codification into useable spread sheet form.
(2) Number of Standing TSF’s .The industry seems to have adopted “3500” as the “semi official” inventory of TSF’s and it seems, as far as we can see to originate form this paper by Davies et. al. citing a 1998 paper by Morgenstern.
We have not yet checked the Morgenstern to see what he intended/believed the number represents. It is not clear whether they intend their created number of “3500 TSF’s globally” to mean all TSF’s between “Put in service” and ” closure” or also closed TSF’s. . Everyone cites Azam/LI (2010) as the source who in turn cite this paper.
(3)Schodde, Richard “100 Years of Resource Growth For Copper Impacts Of Cost. Grade and Technology” ( http://www.minexconsulting.com/publications/Growth%20Factors%20for%20Copper%20SME-MEMS%20March%202010.pdf) has some fascinating insights on HOW miners continued production as grades fell and prices varied. His main point is that costs of production went steadily down making it possible to profitably mine lower and lower grades.and that a 6 fold lowering ore grades allowed a 3 fold output in refined copper. He says “price is input” . We don’t believe this is a relevant predictor of TSF failures but his analysis is compelling. His slide presentation also has fascinating bar graph showing that pre 1960 almost all copper production was out of a few very large mines where as post1960 there is a greater number of mines and more diversity in the size of the mines.