You also need to consider the biological component of arsenopyrite oxidation by iron oxidizing bacteria. For example, see the following which returned from a Google search on the key words "biotic arsenopyrite oxidation":

XPS study of Bioleached Arsenopyrite by Acidithiobacillusferrooxidans

This mechanism will likely overpower any a biotic reactions that the simulations consider. I'm no geochemist, so these models might allow a biological kinetic component. Good luck.

Have you tried to take a look to the Phreeqc USGS web page? There are some interesting links to FAQ, to the mail archive or to Appelo’s home page where you can find ideas to solve your case.

Sounds like a fun project. In the user manual of PHREEQC there is a kinetic expression to model pyrite oxidation using RATES. As a first cut you could start with that or simplify that expression by retaining the rate (-10.19) and A/V terms. Run the model and calibrate the kinetics by modifying the two unknowns.

If you want to do a forward model, with the calibration step needed to match the history; checking the archives is good advice, too. But you have a physical model of the oxidation, so you can determine the empirical oxidation rate, which will incorporate both the biotic and any biotic components of the reactions. Since I presume you have more than one, you could adopt the rate expression that works best for, say, your median case, and use that to explore the behaviors in the other columns, this producing a better understanding of the range of behaviors that could be expected for your system, which almost surely will not be homogeneous and isotropic.

Also, be sure to check your mineralogy to be sure that the only As-bearing phase is Apy. In materials derived from a partially-weathered sulfide ore body, you may very well have both residual Apy and a distinct ferric oxide with sorbed/co-precipitated As (or even, depending on your actual systematic, scorodite). If this were so, then you will have to partial out the two classes of reactions. You may want to look at PHREEQC's capacity to do quite rigorous inverse mass-balance modeling as part of evaluating the overall system if there are both oxides and sulfides present.

Thank you all for the suggestions. You guys gave me a lot of fun ideas for passing the weekend :). I will evaluate what can I do and in case I have any further question/consideration I get back to you.

I am curious, are there any case-studies where the PHREEQC's has successfully projected the empirical outcome of a constructed waste pile/tailings?

This is a critical question, of course. I believe the answer is Yes, provided one has a fair view of what counts as successfully projecting outcomes. If the issue is whether modeling can properly determine the likelihood that a pile will (or will not) become net acid-generating, then the answer is yes. If the question is whether the modeling can successfully anticipate whether water treatment will be needed and what sorts of water treatment may be effective, again the answer is Yes. If the question, however, is whether PHREEQC (or nay other mathematical model) can predict the specific concentrations of every component of potential concern to say, +/- 5% or less of future conditions, specifically across the time series that may be anticipated over the life cycle of a waste pile/impoundment, then I would say that the answer is No - *in my experience at least*.

The Yes answers are further qualified by noting that to achieve even these levels of modeling; we are critically dependent on at least three matters:

•The sampling that went into the test work that is being used must be representative of the full-scale facility that is to be modeled.

•The test procedures and their analyses must meet the appropriate conceptual model of the pile, must be scaled in some reasonable manner, and must be conducted sufficiently long that the range of behaviors that are central to qualitative behavior can express themselves.

•The modeler is sufficiently experienced in mineralogy and geochemistry and is sufficiently familiar with the project at hand that good judgments are made with respect to many special modeling issues.

For example, how will the forward model address pH if the system being modeled includes sources that include both alkalinity and acidity? This may seem simple, but my experience is that it is not, specifically for open and irreversible systems.

The kind of modeling we do for mine waste geochemistry ought not - in my opinion - to be thought of as "strong" modeling in the sense of predicting the return of Halley's Comet. We do not have such clear and well documented constraints on the system. Understanding the limits of what it is credible to do and say and expressing outcomes with respect to the full range of uncertainties (including conceptual uncertainties) is essential if we are to produce defensible geochemical models.

Sorry the answer is not clear-cut, but there you are, in our current state.

Rather than try a forward predictive model first, I would suggest you look at the inverse modeling options in PHREEQC. But be careful as these "models" are not controlled by thermodynamics, they are just simple mass balances you could.

That is very helpful. If I am understanding the boiled-down point that both of you are making is that while effort and experience are required for developing a good model, which is possible, the real work is in proper sampling and sample testing, which would be fairly substantial if done properly. This leads us back to the old adage of modeling: "garbage-in, garbage-out." Estimating geochemical conditions based on a few random samples is not likely to lead to realistic model results. Then begging the question, how substantial will the sampling, analysis, and empirical lab testing have to be?

There is no way to give a firm rule on sampling. If the median being sampled is close to homogeneous and isotropic than only a few samples are needed. But of course, this is rarely true for ore deposits, even for tailing (because of ranging characteristics of the ore feed). For waste rock it is a very great difficulty, especially if there is a need for highly precise estimates of water quality over time.

But we often get too focused on detail and too consumed by the fun we are having with modeling and the bells and whistles capabilities of the newest version. . For example, if I knew that the effluent would have a pH in the range of say 6.5 to 8.5, a TDS < 500 mg/L, and trace metals in the low ppb range; or pH 5000 mg/L, and trace metals in the ppm range, would I not have sufficient information to understand the nature of my problem and the nature of the challenges for water treatment I must face? Would it matter whether the pH were 6.9 of 7.4 for the forts case and 3.2 or 2.8 for the second? [I use pH here as a short-hand for how I draw the classes - that of course needs thought and could well differ from mine to mine and regulatory environment to regulatory environment.] We need to strive for *accuracy* between classes of effluents, but ordinarily we do not need great precision within those classes. If our estimate is close to a class boundary, we can either accept the uncertainty and plan on "over"-engineering, or we can justify to the Project the need to improve the precision to be clear where we sit.

Look, I use geochemical models a lot, and have for a LONG time (I'm a really old guy). They have many great uses, but they are not, and must never be used as, Black Boxes that produce The Answer in some independent manner. There is a lot that needs to be done before one starts modeling, and a lot that needs to be done after the first round of calculations have appeared. Computer models are part of our tool-kit, but only a part. The hardest bit is always defining the actual problem and then determining what would count as an answer to that problem - where the answer has to count not only for the modeler, but also for his Client, and equally for the other stakeholders who have standing.

My background offers me zero experience with geochemical modeling beyond mass-balances, so everything that has been said is learning for me.

I too in general agree that a range is probably adequate and certainly provides for a higher confidence level than trying to predict specific water chemistry within a few mg/l or less. It adds practicality all-around and I am more a practical thinking engineer than a precision-thinking scientist. That said, both have their place!

Generally you want to use a model to get insights into a specific process. Models are great teaching tools to understand overall processes, I continue to learn every time I get into an interesting modeling project; and I do a lot of models with PHREEQC, GWB and several other programs that are out there.

In some cases, models can be used to compare alternative processes for treatment, and a conclusion such as, "This option for pit lake management or some other treatment is better is better than the alternative because (insert reasons here)".

In the original case that started this thread one can use the inverse model to better estimate different oxygen consumption rates as the column ages. Perhaps different treatments options set up in different columns can be better quantified and compared. Or maybe you will see armoring of arsenopyrite with ferric hydroxide is occurring because of the high neutralizing capacity. This would be indicated by the increase in sulfate indicate pyrite or arsenopyrite oxidation and the low amounts of iron that actually are coming out in the leachate.

Any really quantitative evaluation requires lots of supporting data, and in that case the models can help point out what to sample and maybe how to sample. But those predictive models that say arsenic will be exactly 60 ppb in some plume at some location, or in a pit lake in fifty years are more likely to be incorrect. Rather an insight drawn from a model that indicates that arsenic will be a problem is usually the best result. This is the role of using geochemical models as a management tool; such an approach can be valuable.

As mentioned above a conclusion such as "This treatment option is the better choice based upon model comparisons" can be helpful in planning and implementation. Models can also be used to help conceptualize remedies. If you add X amount of such and such reagent to this pit lake some benefit will occur. Then usually one goes into a lab and back to the field to confirm.

Also a model supported with monitoring results can be valuable to help identify potential issues such as faster migration of a target chemical.

Not that we don't all try the more numeric predictive models for solute transport applications, but in situations where those models are really required a combined effort of modeling and laboratory testing is generally performed.

Finally you have a whole range of database issues. One can run the exact same model with different thermodynamic databases and get vastly different conclusions. That is another long discussion though.

You really have to have a good and thorough understanding of the system you are modeling before you can develop a good model. You still have to be in charge. That usually entails a lot of what if scenarios, many of which never make it to the final report, but you (as the modeler) are still learning about the system under study, and the final conclusions will be better for that effort.

Basically, any forward type model that relies upon thermodynamics is just calculating the activities of a basis set of components. In the end, you are selecting or for components that are not deliberately fixed at least accepting that set of activities. In PHREEQC this is often done through the equilibrium phases keyword. One on can set the pH, or a redox condition such as oxygen gas partial pressure, or setting the CO2 gas pressure as an open or letting the program do it as closed system, or though the presence of certain phases such as calcite or quartz. You are defining or accepting those activities and then the rest is summed up by the program to get the final concentrations, but the model really comes down to how the modeler defined those activities.

I did not expect that many answers to my questions. Thanks again. The objective of my question related to PHREEQC was for curiosity purposes more than anything else. I’m just wondering that if we could simulate somehow the kinetics and the long term behave from a leachate of a column experiment based on its mineralogy composition, we might decrease the requirement of doing lots of columns or kinetic tests.

The costs of doing a lot of leach columns for different projects are getting expensive in the long term and several specialists around the world has their own preferences regarding the best methodology (leach columns, humidity cells, “barrel” tests). If we could simulate these columns on PHREEQC, which are much simpler systems than a tailings dam environment for example, we could eventually add this as an operational tool instead of constructing new columns for every single different situation.

Of course that each site has its own particularities but “calibrating” a particular model for a site in a way that an engineer could change simple conditions according to his needs, could be an operational tool rather than a “black box”. When I talk “change conditions” I’m referring to sulphur and carbonates levels at the very maximum and not any geochemical or thermodynamic variable that, of course, could only be done by an expertise geochemistry.

But then again, if modeling simple kinetic tests provides more questions and uncertainties than answers, I would guess that modeling tailings dam environments are much more critical in terms of trustworthy.

A well constructed modeling program such as you suggest is possible as a matter of geochemical modeling. The essential, steps would be to establish what tolerances you will accept for "calibration" and then to evaluate the range of variations in terms of composition and such other matters (i.e. degree of saturation, particle-size distribution, and temperature) you may need to consider. You pretty surely will want to use much or all of the currently available data from your testing program, sampling within your available test data to include one set of tests that will be your basis for establishing your initial model and working out the initial calibration, and then another see of tests against which you will test (and probably refine) your model. You may need a third set, also, that can continue to run some period into the future so that you can extend your calibration tests and demonstrate the long-term robustness of the model

But there are some other complications that you may need to consider. The most important is what the relationship(s) may be between the behaviour of column tests and the behaviour of the mine-waste units you actually care about. After all, in terms of project performance, it is the field behaviour, not the bench-scale test, that matter. This is where some of those other modeling factors such as water-rock ratio, surface-area effects, and your knowledge of life-of-mine mineralogy and chemistry start to become really important.

There are some modellers here, from whom you probably will hear, who that “probabilistic modeling” are of the view (e.g. Monte-Carlo-based approaches) are the way to go. I have great respect for some of the efforts that have been made in this direction.