Background Check: The Why and How of Background Threshold Values

Melanie Veltman: Hello and welcome. While we wait for more participates to flow on the platform, on screen is the ERIS content  disclaimer. This webinar is being recorded and  will be posted on the ERIS website in the next few days at erisinfo.com. Today’s presentation will be about 1 hour in length. There will be time dedicated to Q&A after the formal presentation. So please enter any questions you have in the Q&A section of the platform. That’s a separate section of the platform than the chat section. So the Q&A will be monitored and the chat section will not so if you have a question you want answered put it in the Q&A, please. Alright, it’s time to get started. I am Melanie Veltman. I’m the director of research and data here at ERIS. ERIS, as you know, is an environmental risk data and information services company providing high quality reports for environmental site assessors conducting due diligence and site assessments of real property. ERIS uses the most advanced platforms and tools to help environmental professionals review and analyze critical environmental data records and historical products. Our latest platform, Scriva, brings ERIS Project information about your site. Seamlessly into the report writing phase. And this can all start during your onsite visit with our ERIS mobile app. But we’re here today for the webinar. So today’s webinar, focuses on background threshold values. That represent the concentrations of how hazardous substances used as defensible reference points to determine whether a release has occurred and why knowing these background threshold values, BTVs, is important to the risk assessment process. We’re very fortunate to have with us today Dr. Kenneth Tramm, the principal of Modern Geosciences. Doctor, Dr. Tramm has more than 25 years of experience working on environmental assessment and remediation projects across the US. So now’s the time without further ado. I’ll turn it over to you, Kenneth. To tell us, the why and how of background threshold values.

Dr. Kenneth Tramm:  Okay, thank you. Can you hear me all right?

Melanie Veltman: Yes

Dr. Kenneth Tramm: Just to double check before we go and how’s my pointer on the screen.

Melanie Veltman: The laser pointer…

Dr. Kenneth Tramm:Is it my laser point? Is it pointing?

Melanie Veltman: Yes, it is. It’s glowing and pointing and you sound great.

Dr. Kenneth Tramm: Well. Alright, then we’re go for launch. Yeah. Alright, let me advance it down here. All right. Well, thank you very much, everybody, for coming and thank you for that introduction, Melanie. A few things I just want to mention that I would like to get to. There’s a lot packed into this presentation and so that I can hopefully cover all the topics well enough. We can dive deeper perhaps in the Q&A if that’s necessary, but I’ve said a little bit of refresher on this term that is risk-based corrective action. It obviously means different things in different states perhaps, but usually it’s the cost-effective way to go forward to fix your problem. And then I want to actually put a scenario together so you guys can walk with me through what is why is the BTV. And then we’ll talk about some tools to get it and we’ll use it will choose a state in this case I’ve just chosen Oklahoma to walk through that. There are 2 journals papers that came about from this research. One is kind of focused on the why and that one’s out and the DOI is down below and there will be a QR code at the end if you are interested in that and then the how is one that just hit the streets recently and that’s some of what we’re really going to talk about today because I want to pass on the how. Alright, so if we’re going to talk about risk at all, we have to give it a definition. Right the most common accepted way to kind of describe risk as the one who’s taught in this area for a long time is simply the toxicity times the exposure. Or speaking now as I bring it into a regulatory con. Contacts, I suppose, like cancer risk. Right, so one person in 10,000, another additional incidence of cancer, one in 10,000 being a very high level of risk. And perhaps one in a million. One in 1 million. Being a lesser risk. So some regulatory standard typically will drive how we back calculate these values we’d leave in place. And they’ll be driven by the toxicity of the compounds of chemical specific and then the exposure which has different elements to it, right? The actual concentration you anticipate that would be at the point of exposure and then all the assumptions that go into the modeling of what that exposure means. Within there, you have different routes, ingestion, dermal contact, groundwater ingestion, things like that. If I move that into risk based corrective action. Essentially, they’re driven by the idea that you have a source. Right? A release of some hazardous material. I’ve chosen that word specifically to be inclusive of more things and just hazardous substances perhaps a pathway Right, so we’re gonna hail it ingest it, perhaps abdormal contact or another exposure pathway. And then we have some receptor assumed to be exposed. When these 3 are together, that’s truly when you’ll have your risk. Risk based corrective action then recognizes these 3 principal players and looks at ways to mitigate or manage that risk. We can manage the pathway. We can manage the receptor, right, putting limitations on what can be used at a piece of property. Or we can actually address the source itself and decontaminate a property to reach some established goal. So if I take Right, so we’ve gone through risk, risk base, corrective action. If we move that into Kind of a more conceptual way to look at this. The EPA embraced holistically in the mid ninetys this idea of risk-based corrective action making some decisions about we want to see a decrease in some concentration of chemicals over time. So if we put effort in so for example on the x-axis I’ve got some remedial process is happening. Right, so we’d see a lowering of some concentration. And perhaps we would have a risk goal that’s then calculated by those things we just talked about. We have to be mindful of what a detection limit might be. Right, so sometimes the detection limit may be higher than the goal. That’s when we need better methods, right? So we were living through some of that in the P fast generation now, right? That’s all tempered against. Okay, what’s the cost of the different remedial options that we might choose? And I’m obviously an optimum one would meet just this right here. The point where we would meet the risk goal. And a decrease of the chemical concentrations maximizing every dollar we spent to hit that goal. So what I’m going to talk about today is an added element to this, which is the fact that some of the compounds that we get involved in remediating exist in nature. So there will be a background concentration. So sometimes those existing background concentrations here before and I don’t mean anthropogenic in this case but they could be you could be in an environment where there’s been impact by mankind that’s affected everything. So urban areas often have pHs, some elevated or inorganics as well. There you go to the northwest. We may have orchard areas where we’ve got impacts from arsenic and other pesticides that have been applied. But I’m really speaking truly to an existing Non-anthropogenic background from my discussion here. You can obviously move into anthropogenic if you want to. But that said, there is times when our risk goal is lower. Than that background number. It’s also possible that we can never see down to what the existing concentration is for certain, let’s say metals. I’m just gonna pull it into metals, but it could be anything that exists. Naturally in nature and so that’s where we spend A lot of time screening and kind of get snagged into screening things in a little bit early. So that’s a lot of what I’m going to talk about today. Now, for risk based corrective action, which came about because we had a lot of things to fix and only so much money to do it, we want every dollar to count. We have 2 frameworks in the United States. That we use for that. One. Is the EPA sole screening guidance came out in 1996. It is actually informed by things that came before it, such as the rags document, the risk assessment guidance for Superfund, of which there are many. Starting from 1989 on they resulted in a generic risk-based soul screening levels and the formulas and the assumptions that went into them actually inform most programs in the United States, even if they are state-specific, they’re still informed by these. Today they’ve been expanded into what is accepted as the regional screening levels. And we’ll talk about that in a second. But this is the original document from 96 that all my students have had to learn and memorize. The second part to the nuanced understanding of risk-based corrective action comes from ASTM. Right, so we had an emergency standard which came out in 94 in response to really petroleum releases and the funds that fed into getting them addressed. That led into a petroleum centric risk-based corrective action one being finalized that lived from 95 to 2,024 and it has passed now. I’m exclusively have the just chemical specific one that’s lived on since 2,000. That’s it over there. But this gives you the terms such as tier one, tier 2, tier 3, mindset. And that approach is incorporated into many state programs today. So these are your 2 frameworks. In the universe of risk-based corrective action to be aware of both of them. Mention background. Now, it’s hard when you’re making a program that can be used anywhere. Like with ASTM anywhere means anywhere with EPA means United States, but to give you hard background numbers. So they both mentioned this importance, but leave it to other things. To specify how to get there and to perform them. So it’s good just to note that there’s a placeholder. In this world of risk based corrective action for background. So, knowing that. How we use background is important. So if I looked at a scale and I want to say which one’s going to drive my risk program going forward or maybe a screening level, let’s say. Right on one side. I’ve got risk criteria, which represents usually the lowest. Of all calculated values. So in this case, it could be something by a county, a state, or the EPA itself, in those regional screening levels we talked about. And that’ll be based on the risk criteria that’s acceptable to them. Hence that’s the guy with the question mark next to him and a few of the things we’ve talked about before. And so that would be your driving. Risk criteria unless There is a background value. A defensible background value, right? So then this value would come in and become usually something that trumps. That risk criteria. That’s how most risk-based programs are established and set up to function. So I’ve now shrunk the whole universe of possible. Compounds we’re going to talk about under the scenario of having a background. To those that exist in nature, right? So I’m not talking about perk or TCE or many of the PFAS compounds, although we could talk about anthropogenic background with some of those too. But the lowest of all your grist criteria or background. So with all that said, I want to you to walk with me through a scenario, right? So you get to play consultant for a minute. Many of those on the call are. And I have done this exact scenario. Probably many times in their career. So let’s say we were tasked. Whatever led up to this, maybe it was a phase one. Maybe it was just somebody wanting to know this answer, but we were asked to actually go evaluate a piece of property and we determined, hey, we need 5 samples to represent this small area we’re looking at. And lead was our concern. That’s right, so I’m gonna take some samples and if I was just interested in knowing what the lead concentration is, I’ll obviously get my data back from the lab. In this case, my scenario here, I have a range from 22. The 27 milligrams per kilogram. So the real question is knowing that do I have a release or not? Right, so, all right, let’s look to what we would look at to know that. Now, you know some of that because we just talked about it, right? So I need to have some risk criteria. I have none for my state, I can go pull the regional screening levels from the EPA. My state has criteria I could pull it from them and the lowest of those risk criteria drives my risk unless There’s a background value, so again I’m talking lead so we know that exists naturally. So having this will be very important, especially when coupled with this. And then with those 2 together, I can actually get my screening value to then screen and say do I have a release or not, right? So relatively simple. All right, so one thing to know about. If we looked at LED, this is lead specific, right? So I mentioned. That 1996 sole screening guidance document from. EPA. If we went back in time a little bit, we can actually pull up. And see the values that go with it. So. If we just highlighted lead here as I have, I essentially have an ingestion based number. And inhalation a particulate based number and then I have a soil to groundwater leaching number right so it will migrate from soil to groundwater and then the number of groundwater would exceed something we would drink. Now, it’s designed the EPA framework to have a dilution attenuation factor. Don’t go to sleep on me yet. Right of 20. If it is a half acre or less. Or one if it is more than a half acre. So in this scenario though I’m sure you’ve by now looked down here and said well I see a lot of dashes. Right. So what feeds into that is the fact that when EPA was putting together for the original compounds in the nineties, so 96 specifically. They noted, hey, we actually have guidance on this and let’s not highly mobile and under typical soul condition. So we’re actually going to not put a number there. We want people to focus on this. Immediate risk number, ingestion based. Right, direct contact of 400. Now earlier this year, hopefully some people have already said this to me or at least they’re talking to their screen like I do right that hey that was updated to 200 100 if you have 2 sources or more than one source. That is true. So today, if we looked at this, that number would be 200. Right? So, okay, that’s kind of the EPA federal level. Good to know that. If I actually went and googled regional screening levels and I pull up the The default table that Fred and his team at Oakridge National Laboratories has prepared for us. Shout out to Fred. Right. We can look there and we’ll find that 200 that is current today. And we’d find that there is a groundwater number. Now there’s not, if you looked here, right, there’s not an RSL, a risk. Screening, regional screening level for risk-based numbers. But there is one. That is based on the MCL. And it’s only given for that assumption that’s great than a half acre. But nonetheless, that’s where that number is. Now I mentioned that if you’re less than a half acre, right, that would actually be a different number. So if we wanted to know that the table doesn’t give us that so we’ve got to look under the hood. I’ve got to introduce. The a new idea to you which is okay if I’m half acre or less this partitioning coefficient. Plays a big role in understanding what that leaching is going to look like. So I’ve got to introduce that to you. But this is it right here. I won’t force you to do the math, right? If we take our MCL and we plug it in. Here’s our KD. I’ll keep it red just because it’s important in its role. Alright, there’s some properties about. Water content in the soil, air content in the soil, bulk density of the soil. We run those through. It’s 13.5. That’s where this 14. Comes from. And if we knew that we were half acre or less, we could do the next piece here, which is multiply it out to get our 2 70. Right, so that 2 70 would be the number that would live still here. If we had the less than a half acres. We’ve we’ve now entered a A fork in the road. Do I know that it’s more than a half acre? Or is it less than a half acre? Well, I’ve got to carry both those numbers with me for the moment. That source size becomes real important in kind of making a release determination. It’s easy. If it’s a chemical that we don’t think is naturally occurring in nature or we have a good history of a site and we can identify where the releases, that’s easy. It’s really these other scenarios where we’re screening, perhaps the history has changed and we’re trying to make a decision over here. Maybe there was a historic. Point of a release. That’s affected here. This the reason that we have to have a realistic and inclusive background number to make that decision. All right, let’s get back into our scenario here. Right? So I’ve just pulled in. The EPA regional screening levels here. Right, we talked about that 14. That 2 70. And I’m going to refer to the other pathways together as human health. They’re all human health. But I’m going to isolate groundwater protection from it for our scenario, but we’ve got our 200. And we’ve got our groundwater protection of 14 or 2 70 if I knew my source size. There’s not a background number for LED from the EPA, a default one for you to use anywhere. So we’re stuck with those to start from. So I’ve either got 14 or I’ve got 200 because you remember my scale, right? So 270 being higher than the 200 200 wins. Alright, so I’ve carried that one so I can’t totally answer that question yet. It’s always good to look at other states and just different approaches. So if I went to Pennsylvania, the groundwater protection base number for LED, different approaches. So if I went to Pennsylvania, the groundwater protection base number for LED is 450, the human health again being the other pathways. The lowest is 500. So from that the number would be 450. If we were in Pennsylvania and we had a 22 to 27, we’d feel pretty good. I’ll bring up Texas just because that’s where I am right now. Our groundwater protection number in Texas is 3. And all these are in milligrams per kilogram by the way. Human health is 500. We do get a background number in Texas which is helpful. But it is a published median value, which means it’s the 50th percentile of the entire state of data. So that could be problematic and we’ll talk about that in a second. But this would lead me with either the 3. Or on the other side of the scale. The 15. So 15 would be my driver here. To look at another state that approaches these things differently just to give us context again if we were in Hawaii. Where they like to make fun of us mainlanders in our risk approach. They’ll do batch testing to answer this question. They already use the human health value of 200. They had that in place before EPA changed their numbers, but they actually use what’s called a background threshold value. Right? They did an independent study. Looked at the data sets that came back and said, all right, let’s calculate the high end of the curve of what is anticipated to be naturally occurring. And it’s 73. So in that scenario, right, my lowest would be. 73. So we’d screen out. So I’ve given you 2 ways to screen out. One is purely on the risk side. And one is driven by the background threshold value side or a background number. And then we have Texas with these low numbers. So just to emphasize where we are, right, so general EPA I could apply anywhere. Pennsylvania, Texas, and Hawaii. So one thing to note is how these numbers stack. And there’s a decent range between them. So it’s good to know the background to these pieces of the puzzle so it can be helpful to you. So as we look at Texas, what I want to do is say, all right, well, what was the EPA default? Right, so this is that sole screening level. Calculated again, same numbers. If I plug that in and we just moved here to Texas, it’s our soul to groundwater PCL as we call it here, but the groundwater protection based number. We use a 10 where that 900 is the recommended KD value. Right, for, estimating the partitioning coefficient. If we play our delusion attenuation factors, if we used a 1, it would be point 1 5. But we get to use the 20 in Texas. We call it a lateral dilution factor here. That would get us the 3 that we use. As you look at the KD, value for lead. Sometimes it’s helpful to. Give you things that are in context. Right, so you’re not very mobile once you get to a higher number of 50. Or more of a centimeters cubed per gram. But you’ll be familiar with things. Here’s 1 of my P fast friends that we chase often. But Be Tax, TCE things that commonly in the soil, we would anticipate. Would be highly mobile. And we should check groundwater. As we move up to less mobile species, you see we get different classifications. And so this is where the Texas KD value that we assume for our model would live. And this is where that 900 sits. Pennsylvania just being different, likes to use 8 90. Almost everybody else likes to use that 900. There are some states that have done their own. Testing to demonstrate where that threshold is and you’ll find those are usually 10,000. Plus. So, LED is an interesting one because it occurs commonly. It’s, it’s naturally occurring. Some places higher than than others. It also has risk criteria to it that we are concerned about because it’s just been part of our urban environment for a long time and it has direct health implications. So it’s just an interesting area we found ourselves in that we just need to know more about it and make some good decisions. Anyway, the The assumptions that went into that 10 if you’re interested. Is the assumption that we have a 4.9 pH which turns out to be very little of Texas. That is less than the value the EPA points to and I see less than most the surface area. The second part is an assumption on the sole type. Being SAN versus what EPA recommends versus Probably about 84 shows up in these 5 classes. Here. So the second part, so that was the risk part, right? Because that was my driver at 3 is that background part. And now this one’s important because what we learn here can apply anywhere. So this is the original, 81 data set and these are the points from a USGS study and that informed that median value. There is something very impressive data sets related to a 2,013. Study that a lot of states have started to use as their main reference point. It’s up here in the right, if you’re curious, just Google US GS data series 8 0 1. With a look, all the geospatial data is available to you. But we move from having 119 samples in Texas to Oh, I think roughly 1,200 or so, so 1237. And the QAQC component to this, of course, is very impressive as well because they specifically stay away from major highways, rural roads, buildings. Industrial activities. So an intent. To be non-anthropogenic but to be as natural as we can get because we keep building out. We keep affecting our environment. If you wanted to see these 2 data sets, Collectively, the original 81 data set unfortunately had very limited detection. So you were either 30, 2015, 10. And this is half of the 10. I’ve just added it here to finish out my histogram. And this is where your 3 would sit there. This is the 2013 USGS data shown in a histogram, right? This is frequency of occurrence, number of samples that meet this criteria, and then these are the actual concentrations. And I have actually the 5 cm horizon. Or interval and then the A horizon and the C horizon. With good equal distribution amongst those. But you’ll quickly look if you said we chose the median, which by its name you knew we were probably going to penalize half the state. Right, so there’s this tail that we would expect to exist that we’re going to miss in that. So that can be problematic. One thing when we were doing this research was to reach out to Municipalities and the reason for that is those municipalities always are where they are, right? So if you happen to be in that upper 50%, you’re gonna feel it more than maybe others would. And so some people don’t pay attention to things until they see it cost them money. So we sent some surveys out to a few of the cities that are here in Texas and got their annual cost of demonstrating background again and again or unfortunately sometimes making remedial decisions based on that median value because we can’t use the 3 so we go to the 15. And and then that policy has been in place for 20 years. And so you get to millions of dollars over just these 6 cities. Right? It’s not to emphasize this, but others just to say it’s impactful. Right, so seeing this, let us do, okay, what are some options? Right? Is there a better way? To represent background. So the answer is yes. Let’s go back to. The original intent and the calling and the directions of like, what are we supposed to do with the way that we sample and our protocols? So if we go back to 92 and this really pulls, of course I’m an Aggie from A and M. So I just buy some hissing out there at that moment, but. One of the A and M professors is actually quoted in an EPA document and it’s pretty rare for me to see them to call out somebody’s quote for direction. But in this case, emphasizing when this began of, then this, I mean, how do we assess properties and how do we assess the in this case, soil specifically. So spatial variability. Is not an academic question, right? It exists whether we want to argue about it or not. What we need to understand is the magnitude of that and understand that we can’t put a single value to it unless that single value is or more specifically. We accommodate the existence of that variability in whatever models we use. So as a driving force or a directive. That informed the way EPA put together a lot of its background thinking. So ITRC did some fantastic work. In 2,021, 2,022. Just on the development of background threshold values, if you’re not familiar with it, I highly recommend you pull some of those up. A lot of work and thoughtful process from many, many states went into that. Really, how do we arrive at this number for screening purposes that allows us to do exactly what we try to do in my scenario here. Meanwhile, EPA improved its tools. Right, so anyone not familiar with pro UCL, that’s your next thing to write down. And Google, it’s freeware paid by your tax dollars and it’s a fantastic tool to like you to touch and evaluate and review your data sets. So. From the EPA though, we definitely know. The sources of data that we can get includes the USGS data or more localized information when available. But we want to represent the upper. Threshold of an expected population. So that means I’m gonna pull you into statistics for a minute. So staying hang with me. We need some tools. Right, so just about everybody has worked very hard to purge their statistical learning that they did in school and I don’t blame you. But it’s coming back and it turns out these are very helpful just to understand as we live in the age of big data. Right, so the mean itself, obviously self-explanatory, arithmetic average most often. The median, we’ve already spoken about, the 50th percentile, the modes a harder one when we look at data sets with limited resolution. As you get more resolution, it’s hard to get to your mode, but limited resolution. We may find this would be a poor statistical term. But the standard deviation you should be familiar with, in the coefficient of variance, which is simply your standard deviation divided by your mean. Very helpful. Till anyone familiar with statistics will know this is what most. programs. Point 2 1st to say, hey, is your data normal enough to make some decisions? I’m going to add to that the relative percent difference between the mean and the median only because I want to see how those 2 come together. All right, so those are tools or maybe more factors, true tools. Would be these. Hopefully you, the crowd listening to this is familiar with the 95% UCL and I mean the 95% when I talk about any of these but the UCL the UPL the UTL. The USL. Right, so these are in order of conservative most most conservative at the top, least conservative at the bottom. Right, so I’ll always have to have a very impressive data set to even propose a USL. The most common background threshold value tool is probably the 95% UTL. And a conservative one less realistic. But used, is the UPL. But those are the 3 that I want to talk about. But first, st I can’t start statistics without showing you a curve. Right, so hopefully you remember from school. That within a Gaussian distribution and there’s Carl, you can look at him if you’re tired of looking at me, right within one standard deviation each way from the mean we should have 68% of our population in a theoretical perfect world of a Gaussian distribution. Within 2 standard deviations we’ll have 95%. And then within 3, which not drawn on here, would be 99.7. So now what I have to do when we’re doing background concentration, so something that has a non-negative number is do a 1-tailed evaluation. So if I was going to take that away. And give you a 1 tail. Right, this is actually where about your 95th percentile sits. If it was perfect. I’d say it’s never perfect with the mean, median in the mode are all the same number. But we try to get as close as we can so we can make determinations from that. Or we use more elaborate statistical approaches. Right, so I’m just gonna draw these out. This where the mean is. One standard deviation 2 and then 3 would sit somewhere about here. And now that we’re doing a 1 tail, you see that my inclusiveness in the population has changed a little bit. But that’s how it looks for the one tail test. Alright, so let’s see where those tools that we just spoke about. Would sit, right? So the 95% UCL sits about here. UPL, UTL. And the USL, right? So the USL. Probably sits close to and just under that 3.rd Standard deviation being added to the mean. And I say that having gone through enough data to see performatively. How do we see where these things sit as we play with normal distribution and normal data sets? So most conservative probably least likely to represent any background. Then we head to these 3 tools right here and those are the ones that we looked at in our research. All right, the only other thing I want to do is I want to leave the Gaussian. Because Nothing’s ever perfectly a Gaussian and distribution rather. The real world has things with tails. Right, and your most common one for naturally occurring. Concentrations of, let’s say, compounds or metals is going to be a right tailed set. There’s very few that are high, but they are high. And then there’s gonna be more mass or common occurrence within a central tendency will be within these 3 right here. One thing to note So if I anticipate right tail data to be the most common that I won’t counter in the natural environment, The median is always less. Than the mean. Right, so shown over here. Right, so this and just looking at quartiles. I’m gonna actually not represent the 50.th Or 50% of the likely concentrations that I’m actually penalizing. The most common occurrence of data sets by choosing the median. Now the median is a fine tool if I want to know what is least disturbed by tailed or skewness in data sets, but it is probably not the right choice because it’s just overly conservative versus the mean in a naturally occurring data set. Now left tailed, you’ll see it does the opposite. Left, less common that you would have this in natural data sets. Fact I typically find this when I have over censored. Data right when the my other peers here like you’ve censored your data too much. So anyway I just want to mention those specifics and that’s how the performance of the mean and the median are in normal right tail data sets. And so, right, you may remember before we got into the UPL UTL USL. I mentioned the relative percent between the median and the mean as those become closer, closer together, I know that the curve is starting to get more where it has equal sides. And it’s more Gaussian in nature. So that becomes a tool that we look to in a second. Alright, what data? Obviously local data is always preferred in a background study. So if I have a site where I can get undisturbed samples and actually derive my own background values, that’s preferred. In a scenario where I’ve just taken my 5 samples and I just want to say is it here or not, that’s probably not, it would cost more to do the second part than even the 1st part. But if I’m in a regulatory program, that may be something that’s useful, unless there is defensible data to work with. Moving from local data, we have state and then maybe some other agency has signed off on or performed in a way that would be acceptable to a regulatory entity. And the 3rd and probably the most common used and referenced by EPA and others is to use USGS data. Or something again that’s collected maybe academically or in our CS or something similar to that. So I have up here this Kind of, 1,975 vintage is what that map is from. It’s from the shaklet study that’s probably referenced more often than not in your state guidance. At least if your state guidance is pre-twenty 15 or so. And so that’s limited. In data set quality, you’ll see if you stare at it long enough, you’ll start to see highways. Because it really informed where some of those points were collected. Nonetheless, decent data. Covered good. Compounds and spatially at least hit all the contaminants United States. There’s more focused area data as you head to the nearer study, but that is not as rich in soil as it is in sediment. There’s some NRS NC and RCS studies that are helpful but limited in value only because look at the density. There’s a second one done. It’s the part of the National Geochemical Survey. Again, this is limited in soil content to specific areas. But it was meant to kind of patch. You see some patchiness where we’re filling in data sets there. And then there’s another NRCS. One that was done more recently. But that 2013 one that I referenced. This is what that looked like. And this is why people have turned to it because they improved both their methodology so they have representation on a good spatial coverage, but their quality control was about as good as I’ve ever seen in any of these USGS studies and many EPA studies. So give us something to start with. So if I was gonna start with something, I would begin there. Alright, so we’ve mentioned the UPL. We’ve mentioned the UTL, we’ve mentioned the USL. So now why would I choose one over the other? Right? So this is the effort that was done in our second of those 2 paper, right? So I had the why and this is the how. So what we did is we calibrated some decision points based on its performance of normality, defensiveability when we did visual review of data sets to determine when is it more likely than not that we have no outliers a hard ask or perform so normally in our distribution. At the UTL makes sense or it’s acceptable. But it’s not great. So in that case, if I just put these together, right, the most conservative being the UPL, the most inclusive. Being USL. We have a minimum sample set size. And then we have some requirements for our correlation coefficient and then our CV. Coefficient of coefficient of variance. Right? So this right here, this criteria, only the CV is typically the threshold required by most states and by the EPA. So really everybody’s in. This number. We’ve just added more conservative criteria to it. As we’ve made some decisions about censoring and so they’re based on the CV, the R, and that relative percent difference. Between the mean and the median. So it’ll make more sense if we actually like put it to use. Right? So for example, I told you we’d look at Oklahoma. So welcome to Oklahoma. A friendly state. If we looked at lead concentrations in that 2,013 USGS data, right? So just look, go with me on this column right here. Right. If I did no censoring whatsoever. I have a sample set of 333 samples. The minimum is 5.3 parts per 1 million or milligrams per kilogram. The maximums 122. Here’s your mean, here’s your median, right? So from this, you already see that our mean is to like you already can feel the right tail, right? You can feel it already. The relative percent difference is a little bit high because we’re shooting for 10% before we’d ever think about a USL being on the table. Here’s our standard deviation. Our CV and our R. So in this case, my CV actually is well below the one. Right, so typical fashion if I just use EPA guidance on making a decision of putting a UTL together, that would satisfy and we could grab it, which would be down here at 35. Now we’ve added the R, which we’ll talk about in a second. I really haven’t given you much data on that. And then each of the other columns is where I’m censoring. Only the right tail. Right, so I’m chopping off at a hundred. Nobody in here is above a hundred ppm, 50, 45 30 to see how the performance changed. Right, so over here on the right is kind of my steps to that. Actually find your data. Then process it through pro UCL. So show you what that looks like in a second and then start doing your censoring to see if this data set presents as having an outlier. So looking at your data, evaluating it under different scenarios, there’s nothing better than that. So if I put this data together, now I didn’t give you each of the concentrations because that would be too long for this page. But if we ran it through, you can see that’s where I got my 3 33 from my maximum my mean my median My coefficient of variance and then I’ve got my 3 possible estimators. Right, so we ran it through here and we’re going to make the assumption that everything’s behaving with a normal distribution. Now, It certainly is worthwhile looking at any data set to say, hey, this is log normal. This is non-parametric and their pro UCL can do those things too. What I’m doing in this process is making a conservative approach. That would have increased likelihood of regulator acceptance. Right? Let’s stay conservative and stay more conservative than the EPA has put forward for our purposes. So. But that said, I’ll remove that. All right, so let’s talk about how I got that R, right? So again, in pro UCL, but for those not familiar with QQ plots, the these are QQ plots right here next to them, right? So they’re on a point to those are these frequency histograms, right? So part of my reason to show you the lead ones for Texas was I want you to see what they look like. So this is about as Gaussian, you know, it’s a little bit blocky here. We get more heavy tailed here. We get right tailed here and we a little bit left tail there. So the correlating QQ plots, right? So I’m looking for the rise over run essentially. Of each of these to have an understanding of our correlation. So this is the one that’s probably the most common one you’ll see. The right skewed data set. And if you’ll notice it makes a little bit of a smiley face here. So from that, we can get the R value. And this will be calculated for you. So you can make your own, they’re usually done in 6 quantiles, right? So I’ll have concentrations over here and I’ll have quantiles of occurrence. On the x-axis. So this is what they look like. Straight out of pro UCL. So for example, I’m going to use our Texas one where we used lead. And I used the 1237 or so that are in the state. Put them all together. And this is what it came out as. So from this, you can make the decision probably visually. It looks like we’ve got some decent gaps here. Let’s go ahead and call those outliers. You know, on my team as well as when we deal with even with some regulators, they’re like, well, it’s probably not outliers is part of a heavy tail. It’s totally true, but. For our purposes of just censoring so we can get the most inclusive estimator. We’ll want to remove those. And this is what that looks like when you do it. And you can see my smile kind of forming there. So in this case, this original data set. I had a CV of less than one, so if I just was using the blanket requirements from EPA, it would meet numerically, but the R that we proposed, our correlation coefficient here was point 6 5 7. So it didn’t meet my 7. Requirement. As soon as we censor it, you can see I’ve dropped my CB down even further to point 4, which is very good. And then my R has improved just not quite one, but for a data set like this is pretty impressive. So this then would allow the use of the USL from this data set. And if you’re curious about kind of EPA’s opinion on outliers, if you googled. I love googling, a request from me today. LED, background, EPA, you’ll find in their website lists for different states. If you click on Texas, you’ll see the full data set. And they’ve actually chosen to censor it at 41 as well. So this is actually there. So this is their QQ plot that they present. Anyway. That’s about censoring so I could get my highest R value. To work with. All right, back to Oklahoma. Everybody still with me? Good. Right, so this right here I’ve taken in pro UCL and I’ve just taken the data sets that we all went through, right? So that 333, no censoring one, until I censored it, 100. I censored at 50. I censored at 45 and I censored at 30 and it’s just giving me the data output for each of those but they’re all on top. I’m gonna put in the corner over here each one by itself. But essentially, right, the CV was good enough, the R was good enough. I could choose a UPL of 34.0 5 from that original data set. It met criteria and it looked like this. By itself. Not horrible, but a decent gap. So if we actually censored it at 100, you can see I now get improved performance. And I become decent enough. Right to meet my threshold criteria to allow a U. Right, that would be a 29.1 7. If we sensor again. At 50 I get a minimal improvement if we sensor again. At 45, still minimal improvement. And if we sensor again At 30, we get a very tight data set that now would allow the use of a USL, right? It now meets our CV, our our, and our relative percent difference between. The mean and the media. Right, so what we’re doing is trying to make some choices. Normality performance goals, right, that’s our term. Not catchy, I know, but, we’re not creative in that way. This is the most inclusive. This is the most conservative. Choice of an estimator. Now, I’ll pull you back into Texas just for a second. There’s another way to use these 2. Once you have different variables to play off at each other. You’re probably familiar with a quadrant graph. This is a normality quadrant graph. Where we simply took data from USGS and we ran it through different ecoregions of Texas, right? I didn’t define these regions. These come from the TCQ themselves. So we just ran it through GIS popped out the data sets and said how did LED perform? Right, so within the range. For the use of a UPL, 2 of them fell without any censoring. We just ran it through. I think one more, the central planes, which is number 3, so over here, and then several of them. Fit in the USL just by themselves. So having data that performs so normally is pretty impressive. And this also is a helpful tool if you need to censor your data and see it visually. So this was done just led in different regions. So if I take this Go back to Oklahoma and I’m doing the exact same data set where I took the full data set. This is where that sat. This is where I censored it at 100 at 50. At 45 and at 30 right and this is just comparing the CV to the R from the QQ plot. Alright, that’s 2 ways to look at your data, my favorite way to look at your data and understand this is actually a histogram of already shown you those but since we’re talking in Oklahoma this is Oklahoma right off the bat and I’ve got up in the right kind of our statistics that go with this. And so this is a pretty crude looking heavy right tail because I’ve got 2 points that are so small. You don’t seem over here, but it would allow the UPL in here, but it wouldn’t allow a USL and it wouldn’t allow a UTL. If we censor it one time Again at 100 you can see the immediate improvement. Right, we have to our histogram. If we censor it a second time. We see improvement, but not enough. A 3rd time at 45. Hey, it takes again to that 30 before we now see a decent Gaussian-like distribution. And under that circumstance we meet the USL goals to choose because our median and mean are now close enough. To choose a UPL. The UTL or in this case. Now going back to the full data set, which is a full data set presented here, essentially this is our population and obviously would be inclusive of sub populations as you saw in that Texas example with different regions. This is where the UPL would have put it. So you can see it sits on the end of the right tail and likely is a reasonable estimator. That actually is more inclusive, a higher value. Then what we got by censoring further. But you never know that until you go through the process. So if we took that same process. We applied it because we did as part of our calibration of these choices for the United States. That brings us back to our scenario. Right? So I’ve got my 22 to 27 lead. I’m still trying to answer that question. Does this look like I have a release? So, right, if you remember where we left our story, Pennsylvania, I knew I didn’t have a problem just because I was below the risk criteria and it wouldn’t drive anything. I actually now know that there’s higher background values to be had there. Although if I was in. Florida less so. And maybe if I had to North Dakota less so. So I wouldn’t expect to see these concentrations there. A second compound that we did it for, metal, metalloid, really. As sole scientists, I’m forced to say that. Right, this is arsenic. So just to give you an idea of using that same process and you let the estimator that is the most inclusive but represents the individual data set. This is what you would come out with for those. Points, doing some, you know, anecdotal review of different states and things they’ve got to. Most of this is not Inconsistent, most of it is consistent. With kind of the background values, most the states use. Not Texas, different, no longer story, working on that. So the only other thing I’d want to point out with the time I’ve got for few seconds because I want to leave some QA time is doing this for regional work. Things to point out is that the USGS data sets and the analysis methods are slightly different. And then the EPA methods. So if you’re doing it for risk purposes, you’ll want to make sure that the end output is consistent with if I did it by an EPA method, right? So if I was going to run by 60 1060, 2071 41 for mercury. Things we know from prior studies is essentially a aluminum extremely high so I wouldn’t choose anything higher than UCL if I was going to use that data set. Barium chromium and vanadium. Little high. There definitely seems to be a bias. You see they’re based on the grain size choices or the acids for digestion. And likewise, mercury will be biased low. Obviously putting it in an oven for a while is gonna cook off some mercury. This is your data set. That’s USGS put together. Thank you to Daniel Smith. And his team and all his work. This is the 5 cm where he actually has a kind of color coordinated so you can see where the arsenic has just some commonality in certain places and it’s lower in certain places, right? This matches up with some of what I just showed you before. Each of those can be pulled in. This is actually in Google Earth. I just pulled it in so I could play with it. And you can see the individual concentrations by clicking through and you can develop your own data set to work backwards from. I’ll give you an example for Texas because that’s where I’m in sitting in right now. So I had done one for the Dallas Fort Worth area, which really is the surrounding counties as well. Something to note is, you know, there was in this large area that you see here, there was only 4 samples in that 1981 data set that at 119 across Texas but there are 48 individual samples. So up to 3 per. From across the area. So we have a higher resolution today but if I were just to run through that same process I just did for you. And we look at the performance of our CV. By now, because you are all now statisticians, I feel like you’ve survived to me throwing a lot of acronyms that you should send you all certificates. And the our values are pretty darn impressive too. I have 2 of those QQ plots that we ended up with here. The relative percent difference between Right, the median and the mean, right, if you look at these. Really good performance, meaning they’re coming together close enough that we actually will take on a normal distribution even if we’re not. That would lead you to a 19.5 and a 36 essentially and I’ll wave at you from where I am. So with that, I know I’ve thrown a whole bunch out there and there’s a lot more to talk about. But I hope you enjoyed the journey. And I’ll say thank you and ask if there’s any questions. Hopefully I’ve left some time, Melanie.

Melanie Veltman: Wow. Yeah, we do have a few moments. I’ll just start by saying thank you, Kenneth. Yeah. That felt like a semester’s worth of information, and very valuable, very thorough. So. Yeah, I’ll get started with a few of the questions that came in.

Kenneth Tramm: Okay.

Melanie Veltman: Certainly appreciate those questions coming from the audience. Okay. Hopefully one of them is quick. Kenneth, what software were those QQ plots and histograms generated with?

Kenneth Tramm: Sure, so, pro UCL. And right so, Use your search engine of choice. I’ll say Google more as a verb, but right? Pro UCL, I think they’re on version 5.2 right now and you’ll find the EPA website. It’s downloadable. Get the 2 documents that go with it on use and limitations. But very useful tool and it’s very regulator accepted, right? It has, I’m a fan of our and SAS is what I learned in school, but pro UCL does the things we need for this specific purpose and has regulatory acceptance. So it’s as I tell my students in school when we use it, it’s more like a car wash, right? So you’re gonna roll up your windows, you’re gonna have everything prepared and you roll on through and it will do the rest of it for you.

Melanie Veltman: Awesome. Thank you. So Kenneth, you had mentioned regular regulator acceptance and we had a question that came in through the registrations that seem similar to that, I guess.

Kenneth Tramm: Okay.

Melanie Veltman: How would the validity of the background threshold values be test or measured.

Kenneth Tramm: Well, and that’s part of it is obviously the data sets come any data set you need to define its entire provenance right so why were these sample did they avoid sources So all those things I emphasize on that one. So whether you’re collecting your own. Or you are using data sets collected by others. I’ll say you USGS because that’s available to all of us. And when I went through that data set, Melanie, that same data set extends into Mexico and extends into Canada. Right, so cooperation from all those governments exist and so there’s a larger data set. So The provenance of how it was collected is vital. The methodology in which it was analyzed to get your data in the quality control vital. So I can speak to those 2 from USGS. Obviously if you’re developing your own program, match those. The last part is once you know the data represents what it’s supposed to represent was collected from where it was supposed to be collected, so spatially relevant. Then is the analysis portion of it. And so what I’ve done in this second paper that we wrote is specifically focus on the analysis portion. Right, so taking that data and looking at it defensively and conservatively so that a regulator should have acceptance. And I say that meaning I’ve… We are book ended by being more conservative than the EPA allows us to use. And I’ve used in many other states in developing background work. I I always aired towards the more conservative. So when you show up at the end of the day you’re not dealing with numbers that are drastically different. So that’s a hard question to answer other than you have to go through that entire story and make sure you can defend your data.

Melanie Veltman: Got it. That makes sense. Thank you, Kenneth. There’s a an attendee who has asked if you can just very briefly re-explain the concept of censoring where you take some of those data points out because you might have to defend those choices as well, right?

Kenneth Tramm: Correct. And so in traditional censoring, so There’s a lot of different treatments of data. So on the low end will be at what is your decision-making on non-detects. Right, so I’m incorporating more into that question than they did. Ask but in our sense, if it’s less than 15%, I believe is our threshold, we’ll just do half because that’s been accepted by more regulators than not. If you’re in a regulatory program, all these conversations happen with them on, do they want us to have a kind of a ramp approach to that data set? Do we want to assume all of it is the maximum, which some have, which I think is not very conservative. But we leave that part alone besides choosing half the detection limit. We only sensor on the high end, right, as outliers, which is actually again the more conservative. I would truly want to sensor from both sides in if I’m in a regulatory program and making these decisions. But that said, being conservative, we treat them as outliers only for the iterative effort of trying to see what happens when it is censored and does the data suddenly perform differently. And you’ll be amazed at the performance elements. And so if you’re willing to censor enough, like we’ll have to arm wrestle in the office often on calling that an outlier or not because like I live in a state where there are formations that have arsenic upwards of 200 milligrams per kilogram plus. Right, so it exists here. But for my data set for Texas, I’m chopping it. I’m sorry, you’re out. I’ll deal with those… those areas uniquely when I’m dealing with trying to develop a number that’s not so for a large state, right? It’s hard to develop for a state because I have sub and have to include them in some way. Did I answer a question in there at all?

Melanie Veltman: Got it. Got it. I think so. It did for me. Kenneth, I have approval to go until like 2:05 if you’re okay because we do have some more questions.

Kenneth Tramm: Okay. Sure.

Melanie Veltman: Great awesome. There was a question about interactions of multiple talksins. With the idea that there could be synergistic effects. Compounding the toxicity. So our interactions with multiple toxins are those included anywhere in those risk calculations or in the assessment.

Kenneth Tramm: Sure, so on the if you’ll remember I had the scale with the like risk criteria and then background. Now the side with the risk criteria absolutely multiple contaminants needs to be weighed and that’s usually within different programs. So some programs consider additivity. Or Whether there’s 1, 2 or 3, some programs begin additivity adjustments and waiting when there’s 10 or more. So it just depends programmatically your regulatory body will make some decisions of that so the regional screening levels offer you some options in the modeling. And you can, you know, play with those by getting not the default tables, but actually calculating your own specifically and putting your COCs in. But I am the talk today is exclusively. On the background side, right? Just saying, I wonder. If this is does this exist in a normal distribution or not, if so what what decisions can I make because I want the other side of that question so there’s no true additivity other than if we include arsenic in because it is a release well then it should be judged on now I pull it over on the wrist side right because it’s a contaminant and it has a number because we have risk criteria for that. So that all happens there, but we first need to answer the question, was there release? Does it exceed a background value? And so to that I’d say we regulatory specific on the answer because you need to look like your program that you’re going to close something in or address release but I need to answer the 1st question which is was there a release?

Melanie Veltman: Got it. Great. And then going back, I guess, to some of those earlier slides, there is a question asking. If you happen to know or if you have more context for the EPA’s decision to distinguish at less than half an acre or more than half an acre because that changes your calculations right?

Kenneth Tramm: Sure, so well you’re pulling me over to my my other side I spend way more time on the risk calculation side so that’s fine with me. So the decision to use a dilution attenuation factor in that model has to do with the anticipated. Truly the A in that that word or the acronym right the attenuation so knowing that something as in soil is going to leach And so not a 100% mass can can make it out of that model. We know it will be less than you know one and so the assumption of a acre size gives you an indication of source strength. And so in all modeling, risk calculation, source strength plays a role in understanding. And so obviously the I can’t get back into the minds of what they were thinking in the early 90s when they calculated it for a bunch of different sizes, but most states look at that half acre. Because it came to them from EPA. And so all I can say is small source. We assume it’s half acre or less, large source. We assume that it has no attenuation. Let’s be conservative, even more important for you to actually get your background values then, right? Because in this case, the regional screening level had 13.5 rounded up to 14 for the table. vs 270. So So I sat in between with these in the twentys. That’s why I needed the background in my cartoon scenario that I gave you there.

Melanie Veltman: Got it. Another attendee asks if you were updating a site specific background, how would you go about that?

Kenneth Tramm: That gets very site specific, right? So do I have a lot of different lithologies? Do I have a lot of different horizons that are unique or am I gonna group them together? Those decisions probably stem towards the type of release I’m investigating, but I’ll say generally true general truisms to mention probably is be in an unaffected area collect enough samples so to statistically significant so lowest numbers 8 as you saw from our estimators. I probably would get up to the 20 if you can because you get a lot of value being able to play with your potential outliers in there and understand it better. And then do exactly what we just did in the presentation.

Melanie Veltman: Okay. Okay. Kind of this has been great. I’m gonna ask one final question because there are actually a few that we didn’t get to but this one I think is a nice wrap up. Or there are actually a few that we didn’t get to, but this one I think is a nice wrap up. Are there other resources?

Kenneth Tramm: Sure. I’d say If you want to know really good background guidance documents, start with the Pro UCL website. 1st because the guidance document that there’s a technical guidance document go to that. The second is go to ITRC I’ll give them a plug Great people at ITRC and they have a background threshold value document. So go there first. Because when you’re starting with current mindset and then you’ll work backwards Or you could look at both of our papers and there’s tons of references in that too.

Melanie Veltman: Excellent. Wow. That’s where we’re gonna wrap up for today. I can see in the chat that this has been well appreciated. Thank you so much for sharing your expertise, Kenneth, and thank you to our attendees for those interesting questions. So yeah, that’s all the time we’ve got for today and anyone who’s interested in reading more about environmental due diligence is encouraged to visit ERIS’ infoHub at erisinfo.com where we have curated articles podcasts and past webinars. Again, I’ll remind everyone that this webinar has been recorded and it will be posted on the ERIS website in the next few days. So on behalf of ERIS, my colleagues who’ve participated and our audience, I would like to give a big thank you to Kenneth for sharing your wealth of knowledge and experience with us today and again to the audience. Thank you for your attention. And your questions. Please everyone enjoy the rest of your day.