Statistical Analysis

A statistical analysis of soil vapor methane and hydrogen sulfide data collected over two periods: (1) after July 1, 1998 and (2) before July 1, 1998 has been performed taking into account instrument variability/ accuracy/temperature.  See “Appendix A: Data Statistics” for an explanation of statistical methods employed and charts generated.  In the statistical trends investigated, a " 95% significance level" indicates that the probability of the observed data being due to pure chance is less than 5%; a "99% significance level" means that the probability of the observed data being due to pure chance is less than 1%. There are several kinds of “T” tests.  The one used here is a "T” test for the equality of the means of two populations with unknown variances.   In this case the two populations are the soil vapor levels before and after the cutoff date.  Samples from each population (in this case, the readings before and after the cutoff date) are taken and the sample mean and standard deviation computed.  These are entered into an algorithm to produce a significance level.  The test takes into account both the difference in means and the amount of internal variation in each sample.  Other things being equal, the greater the difference between the means of the samples, the higher the significance level.  However, other things being equal, the greater the standard deviation of one or both samples, the lower the significance level of the test.

 

Of 184 soil vapor tube locations within the exterior boundaries of the Southern Ute Indian Reservation, sufficient data to permit statistical analysis had been collected at 133 sites. Significant increases in methane concentration (99% confidence level) were reflected at 54 sites; 7 additional sites showed less significant methane increases (95% confidence level).  Fourteen sites indicated a substantial decrease in methane concentration (99% confidence level) and 17 additional sites showed a less significant decrease (95% confidence level).

 

T-tests using log (LEL) showed very similar results with several more sites showing upward trends and several fewer showing trends of decline.  Sen and Mann-Kendall statistical indicator tests were not dependent upon an arbitrary cut-off date, but captured increases and decreases over the entire period of measurement.  These tests are also less sensitive to details such as changes in magnitude introduced by the use of different measuring instruments.  The Sen and Mann-Kendall analyses depicted almost identical results when compared to the “T” tests.

 

There were 97 reservation sites with sufficient measurements of hydrogen sulfide to compute T-tests.  The hydrogen sulfide component of the soil vapor increased at 7 sites with a 99% confidence level and 3 more increased when the confidence level was reduced to 95%.   There were no hydrogen sulfide decreases at the 99% level and only 2 at the 95% confidence rating (Chart 14d below).  Soil vapor emission rate comparisons at 133 sites over the same time span showed that flow increased dramatically at 6 sites (99% confidence level), with 4 additional sites exhibiting a less impressive increase in flow (95% confidence level).  Substantial flow decreases occurred at 5 sites (99% confidence level), while 16 more sites exhibited a less significant decrease in flow volume (95% confidence level).  The remainder of the sites either presented insufficient data to draw a statistical conclusion or the data did not depict a significant trend.


Chart 14d

 


It is important to note that a decrease in methane concentration does not necessarily suggest a mitigation effect.  A decrease in methane concentration would be expected to accompany greatly reduced reservoir pressure.  Methane is the first gas to be released from the host coal following reduced reservoir pressure.  Carbon dioxide has a greater affinity for the coal and is only released after greater pressure reduction.  As reservoir pressure within the coalbed(s) is decreased, more carbon dioxide and proportionately less methane are released according to the coalgas isotherm.  The baseline for carbon dioxide monitoring was established by testing within the exterior boundaries of the Southern Ute Indian Reservation during October 1999.  The establishment of a baseline on private, State of Colorado, and lands north of the Southern Ute Indian Reservation is scheduled for November 1999.

 

Of the sites located on public and private lands north of the Southern Ute Indian Reservation exterior boundary, sufficient data to permit statistical analysis had been collected at 147 of 162 total locations.  No sites showed a statistical “T-test” increase in soil gas concentrations of methane.  At the 99% confidence level 2 sites showed a definite methane component decrease, while no additional decreases were noted when the 95% confidence level criteria was applied.  Sen estimators and Mann-Kendall statistics show methane concentration increases at 10-12 sites (at 95% confidence) and decreases at 1-5 sites.  A single site showed the hydrogen sulfide concentration to be on the increase at the 95% confidence level, but none under the 99% criteria using the “T test”.  No decreases in hydrogen sulfide gas components were observed at either 95% or 99% confidence levels.  Gas flow at one site statistically decreased, given the 95% boundary, but no other statistically viable flow changes were apparent.

 

In understanding the foregoing statistical analysis, several cautions must be considered.  Due to the shorter duration (one-year) of monitoring at the 50 northern-most stations, statistical tests would be inconclusive; increases/decreases may be seasonal.  Therefore statistical tests on these sites have been delayed until additional monitoring is complete.  Other interpretative discretion must also be applied.  Statistically significant changes may not always reflect actual trends.  For instance, a change from 0 ppm to 1 ppm or even 100 ppm could be reported as a significant statistical change, while in reality such data variation might only reflect instrument inaccuracy at or near the detection limits for a particular gas.  This is especially true when measuring methane concentration with the Industrial Scientific “ATX-620”, which has a lowest detection limit of 50 ppm.   Sites that rapidly increased from low values to a crescendo in the first year and maintained at similar concentration might escape detection as statistical increases. Decreases in methane content could be attributable to early mitigation attempts, but more likely are the result of methane replacement by carbon dioxide as predicted by the coalgas isotherm at reduced reservoir pressure. Further monitoring of carbon dioxide concentrations will serve to clarify the issue (Appendix C: Chart 12).  Marked increases generally correlate to basal and intermediate coalbeds located up-dip of CBM wells producing considerable quantities of water.

 

Water well methane concentration variations over the past decade were also reviewed for statistical comparisons. Due to the infrequent testing of water wells, meaningful statistical analyses could not be attained.  Nevertheless, preliminary indications from 1998 testing of 117 water wells located in proximity to remediated gas wells show an order-of-magnitude decrease in methane content from prior tests at 32 wells.  Order-of-magnitude increases were documented at 11 wells.  Less variation was noted at 67 wells.  Of these latter wells, the latest test values were lower at 44 sites, higher at 14 wells and essentially identical to baseline values at 15 locations.  Six wells had no pre-existing baseline data for comparison.  (See Appendix C: Chart 11.