This newsletter covers the analytical limits for Methods 25 and 25-C, both real and theoretical. Depending on how each type of limit is defined, the options are numerous and the implications more so.
Everyone is probably familiar with the old Method Detection Limit (MDL) for Method 25, which set the use guideline at any concentration over 50 ppmC. That is where a lot of the discussions begin, but there are some issues with such a limit since the two fractions being sampled have other limits which come into play. One such recent discussion involves audits, blanks, and their impacts on one another in a proposed alternate method. Working from a blank as a basis is one way to determine some form of limit, but that only works well if the blank is constant and consistent in the effect on samples. The old data from the U.S. EPA audit program showed an average blank subtracted from the audits in the <150 ppm range slightly decreased the pass rate. Newer data from a series of private audits in a round-robin-ish comparison showed an increase in pass rate for field audits and no real change for laboratory audits. These data on audits and blanks seemed to indicate a possible lower limit for quantification from the previous MDL of 50 ppm.
The question then came up as to what the statistical detection limit would be based on a 100 ppmC audit over a period of years. The thought was to essentially do the same type of statistical analysis on these audits as is performed on an analyzer to determine an MDL. As it turned out, we had some 159 audits at ~100 ppmC performed on a regular basis for over a decade and an additional 61 audits at ~150 ppmC performed in the early to mid 1990s. There were more of the audits taken but some of the recoveries were performed to check a system to confirm or deny contamination or in the attempt to remove contamination. Those were excluded because they would clearly bias the results, but it was understood that some of the audits used might also have had some small amount of contamination that was not noted on the copy I was using. There is also the possibility of a typo or transposition since the information was entered into a spreadsheet by hand and I was known to have made an error in that regard once. I believe it was a Tuesday, but I digress.
Since I was unsure whether to add the 150 ppmC audits or not, I did not perform the statistical analysis. If I included the 150 ppmC audits and the results were skewed, I would have a problem with justification for removal and if I did not include them I would have a similar concern. So, I allowed someone else to make that determination on the addition or not.
The data set was pretty tight since most of the audits had passed the 20% requirement of our SOP, but the results of the statistical were unbelievably low.
The standard deviation of the differences of this data indicate that you should be able to tell the difference between zero and a sample as low as about 0.4 ppm and have some confidence in the quantitative results above 1 ppm.
This is actually below the combined analytical MDL for the method. The analytical MDL for just the condensible fraction, which is measured as CO2, of 0.11 ppmC and the condensible fraction, which is measured as propane, of 0.75 ppmC. These analytical MDLs are determined by the analysis of seven replicates of 3.270 ppm CO2 and 2.040 ppm propane rather than the combined analysis of a couple of hundred 100/150 ppmC audits. The results of these data are different from the initial attempts when a higher concentration standard was used. It could be that if the concentration in the audits had been lower, the standard deviation would have also expanded to a more expected range. Thus, the two statistical methods do not seem to agree. The base 0.86 ppm from the combined analytical MDL for both fractions does not include any dilution, which would normally triple or quadruple the MDL when compared to the report limit (RL) after all of the dilution factors were applied. This would give a rough MDL of ~3.5 ppm for the combined analytical RL for both fractions. Even increasing the amount to ~9 ppm to have some confidence in the ability to determine what is sample from noise still leaves a significant difference between the current use limit and potential use limit. If a much lower use level or combined MDL were to be used, thus compressing that differential, we might need to consider another approach with the RL, such as reporting flagged concentrations between the analytical MDL and first calibration point, but that does not seem to be very likely at this point.
There were also concerns that although the audits were collected as they would be in the field and using the lowest possible sample volume to allow for the maximum potential impact from equipment, etc, the benefits of collecting them in the laboratory may have limited the deviation between audits. If this is the case, the problems with the audits being collected in the field may be more numerous and/or severe than initially expected. Methanol is reactive and the standards used were only certified for six months, however the cylinders used in these audits gave consistent concentrations as carbon for years even after the certification had expired. The thought was that if there started to be a pattern of failures <80% of the expected concentration, that would be an indication of a problem associated with the short certification period. This too was surprising in that no impact was ever documented concerning any issue with the expired standard .The potential loss of sample due to the reactivity also should have made the data more scattered rather than less scattered.
As it is, we will continue to report a RL for each compound in each sample based upon the lowest calibration point and the amount of dilution applied to the sample during recovery or in order to get any analyte within our calibration range. Given the use limit had been set at 50 ppmC by the U.S. EPA, the use of the lowest standard normally indicates a RL which is well below that 50 ppmC level. The only exclusion is the Method 25-C results with very high concentrations of both CO2 and CH4, which can result in an NMOC RL of well over 100 ppmC due to the dilution factors applied. As always the RL is listed on the data page for each and every compound for each and every sample.
Triangle Environmental Services, Inc.
There were many questions generated by the last newsletter in such a short amount of time, so I felt a supplement might be beneficial to everyone. The two major questions were related. The first was the impact of blanks on the detection limits and the second was how could a reasonable detection limit be developed for the method if there were so many questions and contradictions.
The latter is the easier of the two questions, since the procedure has been used for some time. The use of the standard deviation for a large enough series of samples should give a reasonable method detection limit (MDL). I believe the sample concentration may have to be closer to the detection limit to get sufficient deviation, since the internal audit samples we have give an MDL which is below that of our combined analytical MDL. The composition of the sample may be significant too. The breakdown of the two halves of the sample could be developed separately and the total combined, much as I referenced in our Report Limit (RL) discussion. Another aspect which should be considered is the collection of the samples to be used. The collection in the lab would give the best theoretical limit, while having actual field collection with various crews would give a good practical limit for general field use. The better experienced crews should do better than non-experienced crews, but which would be more reasonable for the regulatory community? That is a question which requires some thought and probably funding.
The other question concerned blanks and their use in the development of the MDL . One thought on an MDL was a multiple of the blank, but there is a concern in that regard. The blank may not be cumulative in these samples. For example, the blank levels for the blanks collected during the same time frame as the internal audits indicated that if the blank had a cumulative effect, the 100 ppmC audits should fail more often than they did given the significant percentage the blank should have been added. Since we collected a zero air blank using the same procedures and sample volumes as the internal audits, the concentration comparisons should have been very close. The problem with the internal audits is the lack of significant failures attributed to blank addition. In fact, the failure rate would probably go up if there had been any type of blank subtraction. This phenomenon was also seen in the U.S. EPA audits administered by ERG in the 50-150 ppmC range. If an average blank were subtracted from each of the results, there was a higher failure rate than with no subtraction. A later set of mostly laboratory audits showed a better pass rate with blank subtraction than without. Thus, the determination of the cumulative aspect of a blank is still in question, especially for the older data sets. The newer data set may be the outlier as well, but only time and more data will tell.
There are also the anecdotal cases of a few field samples which were reported at significantly lower concentrations than the “normal” blanks we had collected. When there are three samples from different crews at different times which indicate a consistent concentration over three runs which is significantly lower than the average blanks being collected, there is the possibility there is some matrix effect. I do not know what it might be about having a sample, which removes some of the response seen with a sample gas rather than zero air with no components other than oxygen, nitrogen, and a few ppmC carbon dioxide. There is insufficient data to draw a firm conclusion, but there is enough anecdotal information that causes me to wonder how these factors interact.
I realize these answers do not cover the question of just what the MDL for Method 25 actually may be. That lack of an answer is because there is really too many conflicting sets of data, which may or may not be correctly applied to this method. If the EPA audit data is correct, there may be no significant impact from the blank. If the anecdotal information from the field samples is correct, there may be no significant impact from the blank. If our internal audit data on the ~100 ppmC range samples is correct, there may be no significant impact from the blanks. If the new round robin audit study data is correct, there may be some impact from the blanks. If the common understanding of what a blank is and how it works is correct, there may be an impact from the Method 25 blanks, but what is truly known is there is a lot of room for discussion and further research on the subject.
Triangle Environmental Services, Inc.