Edited Version of December 6, 2000 Transcript
EIIP Virtual Classroom Presentation

"Chemical Accident Risks: A Preliminary Analysis of Risk Data."

Jim Belke
Environmental Engineer
US Environmental Protection Agency
Chemical Emergency Preparedness and Prevention Office (CEPPO)

Amy Sebring
EIIP Technical Projects Coordinator

The original unedited transcript of the December 6, 2000 online Virtual Classroom presentation is available in the EIIP Virtual Library Archives (http://www.emforum.org/vlibrary/livechat.htm). The following version of the transcript has been edited for easier reading and comprehension. Typos were corrected, date/time/names attributed by the software to each input were deleted but the content of questions and responses are as stated by each participant. Answers to participants’ questions are grouped beneath the appropriate question to facilitate meaning.


Amy Sebring: Welcome to the EIIP Virtual Classroom! Our topic today is "Chemical Accident Risks: A Preliminary Analysis of Risk Data."

You may or may not realize that access to the data that we are going to be hearing about today is severely restricted, due to limitations imposed by Congress. This is why we are especially pleased to present this unique opportunity and to welcome the author, Jim Belke, environmental engineer with EPA's Chemical Emergency Preparedness and Prevention Office (CEPPO). Jim's primary duties include development and implementation of federal regulations and technical guidance for EPA's Risk Management Program and working with industry and other private organizations to prevent hazardous chemical accidents. I would also like to mention that CEPPO is an EIIP Partner, and we are grateful to Armando Santiago for assisting with arranging this presentation.

Welcome Jim, and thank you for taking time to be with us today.

Jim Belke: Hello everyone. Thanks for attending my presentation today, I hope you find it interesting. This presentation is a synopsis of my paper, "Chemical accident risks in U.S. industry - A preliminary analysis of accident risk data from U.S. hazardous chemical facilities." The full paper is available at www.epa.gov/ceppo.

Before I discuss the paper, I'd like to very quickly review the Risk Management Program regulation. This EPA regulation requires that facilities holding more than specified threshold quantities of any of 140 listed toxic or flammable substances take a number of actions.

These include implementing an accident prevention program, conducting emergency planning, conducting an "offsite consequence analysis" (consisting of worst-case and alternative accident scenario modeling), reporting a five-year accident history, and sending a summary report, called the Risk Management Plan (RMP) to the EPA.

Approximately 15,000 RMPs have been submitted to EPA by hazardous chemical facilities in the United States. The database created from the information in these RMPs is called "RMP*Info". In the paper, I discuss the results of some preliminary analysis of the RMP*Info data.

The data include information about what types and quantities of hazardous chemicals are stored at facilities, along with information on each facility's accident prevention program, emergency response plan, accident history, and estimate of the consequences of hypothetical accident scenarios.

Much of this data is available to the public via EPA's internet Web site, but some (the worst-case and alternative scenario data) has been restricted from the public by Congress, due to concerns that it might enable terrorists to target facilities with the greatest potential to harm the public.

This presentation and the paper it is derived from do not reveal any information specifically restricted from the public by Congress.

Today, I'll provide an overview of the data with slides illustrating some basic descriptive statistics and some very preliminary analysis, and then I'll answer any questions. I should emphasize that my analysis is preliminary, and much more thorough analysis must be done before any firm conclusions about the data may be drawn.

I should also point out that some of the information I will present is derived from an earlier study conducted by the Wharton School of the University of Pennsylvania. Working under a cooperative agreement with my office, Wharton performed some basic analysis primarily of the accident history portion of the database, and I've extended their analysis somewhat.

The portions of this presentation not related to accident history is my own analysis, and any opinions I offer are my own, and do not necessarily reflect official EPA positions.

There are about 15,000 RMP facilities in the United States. My first slide will show their geographic distribution on a map of the US.

[Slide 1]

The large concentration of facilities in the Midwest is mainly due to the widespread use of ammonia for agriculture (both fertilizer production and direct land application) and refrigeration.

Another large concentration is present in central California, for similar reasons. Other concentrations are generally near heavily industrialized areas, such as the Houston Ship Channel, and represent a wide spectrum of toxic and flammable substances.

The next slide is a pie chart illustrating the percentages of RMP processes in the US that contain the various toxic and flammable chemicals. As you'll see on the slide, there are just over 20,000 covered chemical processes contained in the 15,000 regulated RMP facilities.

There are more processes than facilities, since some facilities have more than one regulated process. Likewise, a single process may contain more than one RMP chemical.

[Slide 2]

In calculating the percentages on this slide, a process was counted once for each chemical it contains. Therefore, while the total number of RMP processes in the US is just over 20,000, the total number of different "chemical streams" in those processes is actually over 25,000.

The top six chemicals, as measured by their frequency of occurrence in a process, are anhydrous ammonia, chlorine, flammable mixtures, propane, sulfur dioxide, and aqueous ammonia. They account for 75% of all RMP processes.

The next slide will show the percentages of RMP chemicals by quantity, instead of by number of processes. Comparing this slide to the previous one will show that while toxic chemical processes are more numerous than flammables, flammable RMP chemicals significantly outweigh toxic chemicals (literally speaking).

[Slide 3]

The next slide is derived from the Wharton accident history analysis, and shows the aggregate number of accidents reported for each North American Industrial Classification System (NAICS) code where more than 10 accidents were reported. You'll see that petroleum refineries rank first, followed by water supply, sewage treatment, and chemical manufacturing.

[Slide 4]

But this is a potentially misleading representation of accident data, because it isn't normalized to account for variations in the number of opportunities for accidents to occur across the various NAICS codes. In other words, a petroleum refinery is physically much larger and has more processes, employees and chemicals than facilities represented by just about every other NAICS code. So we need to try to account for this in order to compare the rates of accidents from different types of facilities on a level playing field.

The next slide does this by illustrating normalized accident data. The number of accidents in each sector is divided by the total number of chemical processes in that sector, and you get a completely different picture.

[Slide 5]

Note that refineries, water supply, and sewage treatment now rank closer to the bottom, and the top ranking sectors are animal slaughtering, pulp and paper, and alkalis and chlorine manufacturing.

The next two slides do the same sort of thing using chemical type, instead of NAICS code. The next slide shows the accident data broken out by chemical type, but not normalized. It shows that ammonia and chlorine are involved in more accidents by far than any other RMP chemical.

[Slide 6]

But again, this doesn't tell the whole story. There are lots more ammonia and chlorine processes out there than anything else. If you divide the number of accidents for each chemical by the number of RMP processes nationwide containing that chemical, you'll see that ammonia and chlorine processes are not the most likely to have an accident, on a per-process basis.

[Slide 7]

This slide shows that on a per-process basis, chlorine dioxide, H2S, HF, and HCl processes are more likely to suffer accidental releases.

Now I'm going to show a couple of graphs called frequency histograms. These are statistical graphs which model the shape of the distribution of a particular variable. The next slide is a histogram showing the number of facilities having a toxic endpoint distance in a particular interval. It shows that most toxic facilities have worst case endpoint distances of less than 3 miles, but that a significant number have longer - in some cases very long - endpoint distances.

[Slide 8]

Jim Belke: The numbers on the x-axis represent the top of a distance interval, and the y-axis is the number of facilities in that interval. In other words, the first bar indicates that there are about 3300 facilities with a toxic endpoint distance of between 0 and 1 mile. For this distribution, the median is 1.6 miles, the mean is 2.93 miles, and the mode is 1.3 miles. The median is the best measure of central tendency for this distribution, since it has high positive skewness.

The next histogram shows how many facilities have toxic worst case scenarios that could affect various population levels. The x-axis is a population interval, and the y-axis is the number of facilities with toxic WCS potentially affecting that population interval.

[Slide 9]

Unlike the last distribution, this one is plotted on a log scale, since the range of populations potentially affected by RMP toxic worst case scenarios is very large - over 7 orders of magnitude.

Note that there are a significant number of facilities with more than 1 million people in their worst case vulnerable zone. These are obviously located in densely populated urban areas. For this distribution, the median is 1500 people, the mean is 40247, and the mode is zero.

The last slide summarizes the basic descriptive statistics of all of the OCA information in the database. It's a little dense, but the one point I will make here is that all measures of central tendency decrease or remain the same as you move to the right across the chart.

In other words, toxic worst case scenarios are the most severe, followed by toxic alternative scenarios, followed by flammable worst case, followed by flammable alternative scenarios.

[Slide 10]

An interesting note about the last slide is that the median affected population for a flammable alternative scenario is zero. This means that more than half of the flammable alternative scenarios submitted to EPA do not project any potential public impact.

It's fair to ask what conclusions might be drawn from this analysis. As I indicated earlier, not much can be firmly concluded at this point, due to the preliminary nature of the analysis. Mostly, I think it raises more questions than answers and points out areas for further study.

EPA will continue to study the data, but my hope is that others will take on the task of performing more in-depth analyses and publishing their results. There is a provision in the law allowing "qualified researchers" to gain access to the full RMP*Info database, in order to perform exactly these sorts of in-depth analyses.

Such analyses could be useful in a number of ways. For example, they could be used by EPA to guide future regulatory decisions, such as chemical listing or de-listing decisions, threshold changes, and the like. They could also prompt industry to make safety improvements independent of regulatory action. And there are lots of other possible uses.

That concludes my prepared remarks. Thanks again for attending my presentation, and I'll now turn things back over to Amy Sebring for the Q&A session.

Amy Sebring: Thank you Jim. We now invite your questions/comments.

[Q&A with Audience]


John Pine: One intended purpose of the data was to make more information on risks available to communities. Do you believe that this has been accomplished?

Jim Belke: I think that this will take time and this sort of analysis is probably not what communities are interested in. But definitely the problem has been exacerbated by the restrictions placed on the data. Over time, I expect that communities will become more aware of the data.


J. P. DeMeritt: How much of this information might be available through LEPCs? If the concern is to keep the "bad guys" from accessing it, could they obtain raw data and construct OCAs themselves?

Jim Belke: The bad guys could definitely try to reconstruct the analysis and the LEPCs can give out any data for facilities in their locale, and neighboring ones. I recommend reading the EPA and DOJ assessments for a full discussion of this particular issue as well as the EPA regulation implementing Public Law 106-40.

Amy Sebring: Assessments referenced are linked from the session background page.


Dan Vaughn: Can we get around the restrictions for our communities?

Jim Belke: Yes, the LEPC or state officials can give you local data and, of course, the facilities themselves are permitted to give it out.


Amy Sebring: Jim, your report notes that it would be difficult, if not impossible, to draw any conclusions about the alternative release scenarios, yet they may represent a more "realistic" basis for a risk assessment in a community. Is any further analysis of the alternative scenario data planned?

Jim Belke: I would like to look at it further and I'm hopeful that Wharton and others will also do so. But as far as statistics go the rule left so much flexibility in defining an alternative scenario that it is really difficult to see how much can be done beyond what I have done.


Paul Bennett: Although OCA has been kept out of RMP Info, often the information is contained in the Executive Summary submitted by the company. Is there any thought to approaching these companies and asking them to reword their summaries to exclude this sensitive information?

Jim Belke: Hmm...some companies have already done so without our prompting. But I don't think we will formally ask all companies to do so because the executive summary information is not as searchable as the remainder of the database. So it doesn't present as much of a threat for someone desiring to target the worst facility.


Amy Sebring: Jim, the report notes that only rudimentary data validation has been performed during the RMP submittal process. Will there be a mechanism to validate and/or correct data through the audit process?

Jim Belke: Yes, we are working with Wharton and doing a great deal of in-house data validation to identify and eventually correct some of the worst problems.


Gerald Minor: I know that the EPA provides a Website that list which companies report under SARA Title III, however, is there a way of getting that information in one report for a county?

Jim Belke: There is no national database for SARA Title III reports that I am aware of other than TRI. But states and localities may have such data. I would recommend contacting state and local goverments.


J. P. DeMeritt: Two questions, if I may. First, what, if anything, do you think this data might say about someone's ability to cause a worst case release? Second, what do you think the future implications of this information might be?

Jim Belke: First, the OCA scenarios don't say anything about probability other than the general notion that they are all quite improbable. But the accident history information is real accident information so it may indeed be useful in assessing the likelihood of an accident from a particular sector.

On the second question, this type of information could have lots of uses. We might use it to guide new regulations. We might be able to draw correlations between certain characteristics of a facility and its likelihood of having an accident. And this type of analysis is currently ongoing at Wharton. It is called "epidemiological accident analysis."


David Graham: Assuming that greater access to OCA information is allowed in the future, do you think that the public is likely to be overly alarmed by press reporting on worst case scenarios as a result of not understanding that the probability of such events is low?

Jim Belke: I personally think that there is the potential for this simply because the media like to sell papers, etc. So it is in their interest to be somewhat inflammatory perhaps but we can also debate whether the net benefits of the data outweigh this concern. I think it probably does.


Avagene Moore: Jim, the restricted access to the data was based on concerns about the potential for terrorist acts. What guidance, if any, is EPA providing to industry regarding assessing vulnerability to terrorism?

Jim Belke: The Dept of Justice is required to do a site vulnerability study. This requirement was contained in Public Law 106-40 to see if we are doing sufficient things to ensure these sites are secure. But, to date, I believe that DOJ has not received funding for the study. They requested funding, but I believe they were denied. So nothing might happen. We'll see.


Paul Bennett: You say DOJ is going to study this, but to what extent? Many are private businesses. Will DOJ establish some base line level of security?

Jim Belke: Well, most are private businesses but that has never stopped DOJ before. I assume that they have ways of exercising their authority to do the study. We spoke with DOJ reps about this and they were intending to try to establish a baseline. But again, they didn't get requested funding yet, so had to put the effort on hold.


Amy Sebring: Jim, the report mentions that the RMP rule has had some impact already in that many facilities have either eliminated or replaced covered substances. Is there any way of estimating this indirect impact?

Jim Belke: Yes, we originally estimated about 66000 facilities would be covered. About 33000 or so were eliminated by Public Law 106-40 wich eliminated most propane facilities. So perhaps another 15,000 can be explained by either quantity reduction, chemical substitution or non-reporting.


David Graham: Weren't there hundreds of facilities that did not submit RMPs and will EPA take enforcement action against some of these?

Jim Belke: There were certainly some that are starting to be identified. I don't know about "hundreds" at this point but we will definitely take action against some of them. It depends on the individual circumstances. Some will get assistance in submitting; ohers will probably get enforcement actions taken against them.


Lloyd Bokman: Is there any thought being given to increasing the number of covered chemicals on the list to perhaps, including more EHS chemicals?

Jim Belke: We are thinking about listing/de-listing issues but no firm commitments have been made at this point. We are reviewing various petitions to list or de-list substances. Now, OSHA is doing something interesting; tey are considering changing the PSM list t match the RMP list better.


Amy Sebring: Jim, one of the things that disturbed me about the Worst Case Scenario approach is that it did not require facilities to assess their vulnerability to natural hazards, such as earthquakes, hurricanes, etc. Does OSHA or EPA have any guidance for facilities or LEPCs in this regard? Is there something already on loss of offsite power?

Jim Belke: It did not require any specific scenario to be considered. But, as a practical matter many facilities did consider such scenarios. And in our General Guidance we recommended, I believe, facilities subject to certain types of disasters because of their location, that they consider those events.


John Pine: Amy, good comment - exposure to risk from natural hazards - flooding - wind - freezing - is a real possibility. The facilities should anticipate these exposures.

Jim Belke: I agree.

Last Question:

Paul Bennett: Any plans to analyze RMP audits that are taking place?

Jim Belke: Umm... We are doing some interesting things with third party audits and of course each audit is looked at for potential enforcement action. I'm talking about government audits, not third party. But at this point we don't have a national audit database and have no plans for one. But I sure would like to have one! Thank you everyone.


Amy Sebring: Thank you for being with us today Jim. We very much appreciate your time and effort and we hope you enjoyed it. Please stand by a moment if you can while we take care of some announcements. First, I would like to announce that we have two Pledgers who have completed 12 months during November and are added to the Honor Roll at http://www.emforum.org/eiip/pledge.htm#HONOR . Claire Rubin! Rick Tobin! Thank you, Rick and Claire!

Avagene, can you tell us what's on for next week please?

Avagene Moore: Thank you, Amy. Jim, on behalf of the EIIP, we are grateful for your time and effort today. Fine session and of great interest!

Next Wednesday, December 13, 12:00 Noon EST, the EIIP Virtual Forum features "Effective Disaster Warnings: Report by the Working Group on Natural Disaster Information Systems." Our guest speaker will be Peter Ward, Working Group Chairman. Make your plans to be with us - next week will be our last session before the holidays. We will express best wishes to everyone for a happy holiday season while we are online. That's it for now, Amy.

Amy Sebring: Thank you, Ava. We will have a transcript of today's session posted later on this afternoon, which you can access via the Transcripts link on our home page, and the reformatted versions either Friday or Monday; also the slides so you can have a closer look.

Thanks to all our participants today. We will adjourn the session for now, and you no longer need to use question marks. Please join us in thanking Jim.