Author Topic: Inverse Modelling_Uncertainty issues and Mineral dissolution (Read 7300 times)

rfembilejr · « **on:** 14/03/17 05:00 »

Hi David,

Apologies for bugging you again.
So I have been working on inverse modelling using tailings from a magmatic Ni sulfide deposit. I have pure water as input solution and one solution per week that represents the leachate collected. This is data from kinetic testing. I have attached the code for 70 weeks. I thought I have my models working but realized that having 0.8 uncertainty for some elements is totally not good.
This and other issues have been bugging me and would like to ask if you have some suggestions/remarks.
The main isssues are the following.

1. In the uncertainty, some balances had to be changed to 0.8 (ie. Inverse Models 50 and above) in order for the model to work. This is really bad. I am not quite sure how to reduce this value while still producing model(s). Does this have something to do with the phases or minerals I am reacting? Or am I missing some elements/minerals that should precipitate/dissolve?
2. As you can see, my main dissolving minerals are mainly forsterite (55%) and pyrrhotite (2%). The other minerals present in the tailings are pyroxenes (enstatite), amphiboles (tremolite) and some plagioclase feldspars (ie. anorthite). I have excluded these three silicates since their dissolution rate is 3x less than that of forsterite. Basically, we assume that most Mg is coming from forsterite dissolution. The forsterite also contains Fe and Ni based on microprobe data, so we also think that these metals are coming from forsterite. We also assume that most Fe is coming from pyrrhotite dissolution. Chalcopyrite is also present but in a very small amount ( ~<1%). Should I be actually be adding these 3 silicates in the phases? I am just not into this idea because in the TRANSPORT and KINETICS code, I was able to match actual kinetic testing data with the forward modelling results with just forsterite and pyrrhotite in the reaction.
3. In the solution spread, we do not have actual Si data so we just put 10 for Si. Is this fine? The rest of the data are actual leachate data from our analysis.

Thank you in advance for any advice.

Rodrigo

dlparkhurst · « **Reply #1 on:** 14/03/17 15:45 »

The first thing to consider is why you have a bias toward negative charge balance (in general, -10 to -60 percent), with larger negative values from solution 55-70. Is this an error in one of the analyses (SO4 for example) or is it because you do not have analyses for Na? If it is a known error in an analysis, then you can increase the uncertainty on that analysis. If it is a missing cation (Na), then perhaps your assumption ignoring feldspar is incorrect.

I only looked at solution 70 in detail. It requires an uncertainty of about 0.4 applied to all elements to be able to meet the charge balance constraint used in inverse modeling.

rfembilejr · « **Reply #2 on:** 14/03/17 22:13 »

When you say negative charge balance, you mean the percent error? Yes, that is why I am mostly increasing the uncertainty of SO4 even if I don't think there is an error with our analyses. The solution does indeed look more negative because of the very high SO4 concentration as compared to the sum of the cations. We do have Na analysis but did not use it because we think it does not have a significant effect on our mineral phases ie. since we don't have anorthite. What is the allowed percent error value in this case?

As suggested, I tried adding Na in the solution spread but it did not really help in lowering the negative charge balance. I also added Na in the balances but that did not help either. I also played with adding anorthite in the phases but with no Na in the solution spread (that's how I understood from your previous suggestion) but did not help either. By doing all these, there are no models produced. In the attached input file, Na is added in the solution spread in case you may want to check how it affects the models. Anorthite and Na are also added as part of the phases and balances respectively. I am disabling gypsum, epsomite and tenorite from week6 onwards since we think these will least likely form in the column after the initial 5 weeks of rinsing. At this point is it reasonable to ignore enstatite and (magnesio)hornblende?

I tried using an uncertainty 0.4 for all elements for solution 70 but I did not get any model. I am using wateq4f database.

rfembilejr · « **Reply #3 on:** 14/03/17 23:12 »

I just realized that adding Na would suggest having a Na-source which could be a Na-rich feldspar ie. albite. But based on the mineralogy of the deposit, anorthite is more likely to occur. There must be other sources of Na?

dlparkhurst · « **Reply #4 on:** 14/03/17 23:33 »

You will have to decide on plausible mineralogy.

The first problem is the charge balance. As part of inverse modeling, PHREEQC creates charge balance by adjusting the cations and (or) anions, so if you have a large charge imbalance, it will have a big effect on your inverse model. In the attached file, I charge balanced on Na in the first inverse modeling and SO4 in the second inverse modeling. I have added albite and anorthite plus gibbsite in the first, and anorthite plus gibbsite in the second.

The second problem is that Forsterite is the only source of both Mg and Ni and no minerals are sinks for Mg or Ni. Unless the final solution has exactly the same ratio as your Forsterite composition, then something has to be adjusted. I think there is much less Ni in solution relative to Mg compared to your forsterite. I have added a forsterite with no Ni, to allow a variation in the ratio.

Models are produced in both cases, so you can consider how to deal with charge balance and start to winnow the models based on your understanding of the mineralogy.

I have only run solution 70. You can determine how to deal with the others.

rfembilejr · « **Reply #5 on:** 15/03/17 01:43 »

Thank you David. I highly appreciate your help.
Looking at the code, I have some questions:

1. I noticed that you added Mackinawite in the phases block but exclude it in the inverse modelling blocks. May I know why.

2. In the "-uncertainty" line, there seems to be no value for solution 2, only 0.1 was entered. I assume 0.1 also applies to solution 2 as per the usgs example (https://wwwbrr.cr.usgs.gov/projects/GWC_coupled/phreeqc/phreeqc3-html/phreeqc3-20.htm).

This uncertainty also applies to the elements under "-balances" since no uncertainty limit was specified?

3. The code works but it made me wonder because I thought by adding gibssite, the program will look for Al in the input solution, but it did not. That is interesting. Where would Al come from?

I like the idea of adding a forsterite without Ni because there are actually some samples with very small Ni content.
I also played with Epsomite as possible sink for Mg but this should only be for the first 5 weeks so I ended up not using them in the inverse models.

I guess I might end up putting "charge" to some elements/ions ie. Na or SO4 just to keep the charge balance as low as possible. But is it ok to do this for different elements in the weekly solutions ie. Na for week 50 and then SO4 for week 51 as long as I can get the charge balance lower? I actually did this before but it is quite daunting since I have 9 columns for seventy weeks (630) and sometimes charge balancing is different every week, I noticed.

rfembilejr · « **Reply #6 on:** 15/03/17 02:02 »

I forgot to ask, you used llnl.dat for the code to work, right? I tried with other databases but I am getting the following error:

ERROR: Elements in species have not been tabulated, SiO2.
ERROR: Reaction for species has not been defined, SiO2.
ERROR: Calculations terminating due to input errors.

So llnl.dat is the best to use in this case?

dlparkhurst · « **Reply #7 on:** 15/03/17 02:46 »

1. I added mackinawite to investigate whether the composition of the iron sulfide was important. It is an iterative process to find a set of reactions that you like. The first step is to get something that produces some models that you can then revise.

2. The single 0.1 will be applied to all elemental uncertainties. The last value on the -uncertainty line will apply to any additional solutions for which values were not given. The 0.1 will also apply to the elements listed in -balances if no values are given there. Any values listed in -balance for an element will supersede the value for the solution listed in -uncertainty. If one value is given for an element in -balances, then it applies to all solutions.

3. The addition of gibbsite introduces a mole-balance equation for Al. If Al is not defined in the solutions, then the concentration is assumed to be zero. In this case, all Al concentrations are zero, which implies that Al is conserved in the minerals.

4. You could increase the uncertainties on Na and SO4, which would force these species to account for the charge balance. I would not want to go through week by week and pick which element to charge balance with; it would be tedious and arbitrary. The charge balances tend to be negative, so I think there is a systematic error. I would prefer to be consistent throughout. You'll have to consider whether what you know about the source of the charge imbalance.

5. llnl.dat has SiO2 as the silica master species. All other databases have H4SiO4 as the master species. The only problem is in the minerals you define in PHASES. If you use llnl.dat you must use SiO2. If you use other databases, you must use H4SiO4 in the chemical equations (and balance with H2O).

rfembilejr · « **Reply #8 on:** 15/03/17 04:00 »

Thanks again! At least it is a relief from not having to check every week, also now it is more clear.
This should keep me busy and hopefully get everything working.

rfembilejr · « **Reply #9 on:** 15/03/17 05:23 »

Quick question, since we have two forsterites (one with Ni and the other without), is it safe to assume, that if there are no other Mg sources, I can just add up the moles from these two forsterites to come up with the total moles of forsterite dissolved?

dlparkhurst · « **Reply #10 on:** 15/03/17 14:58 »

Yes, you can add the two forterites. All it is doing is allowing some flexibility in the ratio of Mg to Ni.

I still wonder about the charge balance of your solutions. If you used IC to measure sulfate, it should be a pretty good number. Are you missing some other major cation? Maybe Zn or Mn?

rfembilejr · « **Reply #11 on:** 15/03/17 16:18 »

yeah, but we cannot indicate how much of each forsterite is present in the tailings, right? So basically they are like 50/50? It would be nice if I can vary that ratio.

Yes, we used IC for major ions. We do have Mn (using AAS) but only for a few columns mainly due to time constraints of analysis and also detection limits. And this will have to be accounted for in the modelling because others do not have Mn on their solution spread.

What I did so far is to deal with the charge balance by doing the charge balance on SO4 for all of the columns. I think it worked fine. I decided to ignore Na since we do not have a lot of it and that albite dissolution is also much much slower than forsterite.

rfembilejr · « **Reply #12 on:** 21/03/17 14:36 »

Hi David,

I was able to generate models using the phases below. However, epsomite should be undersaturated (based on the solution speciation) and should not precipitate. The models produced, which is only one model per week (except for Week1-5), are suggesting that it should precipitate. We don't think that is the case. So we thought of adding the epsomite value (negative moles) to the total forsterite to account all the Mg that is released in solution. Is this assumption/step correct? Or is there a better way solve this issue?

If I remove epsomite in the phases, then it would be the case of trying to play with the uncertainty for each week which oftentimes can reach 50-80% for most species. And sometimes, no models are produced at all.

Attached is the sample code.

Forsterite     dis
Forsterite1 dis
Pyrrhotite dis
Chalcopyrite dis
Anorthite dis
Goethite
Epsomite
     Gibbsite
Quartz
     O2(g)
CO2(g)

dlparkhurst · « **Reply #13 on:** 21/03/17 15:34 »

I doubt epsomite is present to dissolve, and it is very soluble, so unlikely to precipitate. I would remove it from your list of minerals.

If you do that, the problem is still Mg and Ni are both coming from the two fosterites. I think some of the waters have Ni/Mg ratios greater than the fosterite stoichiometry. With no sink for Mg and no other source for Ni, a model cannot be found. If the Ni content of the end-member fosterite were higher, then I think models would be found for most weeks.

rfembilejr · « **Reply #14 on:** 21/03/17 15:44 »

Totally agree about epsomite. For now I removed epsomite from the phases and it is almost 90-100% uncertainty before getting a model, if I get a model. Just for the sake of getting models, keeping epsomite in the phases does work with 10% uncertainty, but as mentioned, it is precipitating. So would adding the precipitated epsomite to forsterite sounds alright?

I guess I will also have to rethink about Ni and Mg sinks/sources. I may have missed some mineral phases.

dlparkhurst · « **Reply #15 on:** 21/03/17 16:32 »

I added .06 Ni to the Forsterite formula at the expense of Mg. I think almost all weeks have models.

rfembilejr · « **Reply #16 on:** 21/03/17 16:49 »

Awesome! Did not think of doing that. Although, I was also thinking of adding Ni into Pyrrhotite but not sure how to go about that yet. But this one works perfect for now.
Thanks a lot! I will see if this same approach works for the other 8 columns.

rfembilejr · « **Reply #17 on:** 28/04/17 05:37 »

Hi David,

So there were some changes in the forsterite formula that I was using and I was trying to incorporate that in the script but it turns out I don't get a model unless I increase the uncertainty values to about 0.5 or higher.
How did you arrive into the 0.06 Ni that was added to the forsterite? Was this just an arbitrary value?
Thanks.

Rodrigo

dlparkhurst · « **Reply #18 on:** 30/04/17 19:09 »

You should have asked sooner. I don't remember, but I suspect the amount of Ni was arbitrary.

rfembilejr · « **Reply #19 on:** 30/04/17 23:39 »

I got fixed now. I just used an arbitrary value.
Thanks.