April 15, 1991 MEMORANDUM To: District Water-Quality Specialists, WRD Regional Water-Quality Specialists, NR, CR, WR, SR From: National Networks Coordinator, Reston, VA Subject: WATER QUALITY--Results of analysis of fecal streptococcal bacteria data collected using 0.45 um and 0.7 um pore-size filters During the 1990 water year, all Districts were requested to do concurrent determinations of fecal streptococcal bacteria using 0.45 um and 0.7 um pore-size filters at National Network stations. These data would be used to determine if there is a significant difference between recoveries using the 0.45 um filter that is recommended for use in the 1987 revision of TWRI Book 5, Chapter A4 and the 0.7 um filter that had been used at many stations in the past. Twenty-five Districts participated in this effort and concurrent determinations were made at 162 NASQAN and 22 Benchmark stations. The data were analyzed by Dave Peterson of the Wyoming District. The results of the data analysis showed that the "adoption of the 0.45 micron pore- size for fecal streptococcal determinations will not adversely affect the utility of the data set for long-term analysis." Therefore, the 0.45 um filter is to be used for all fecal streptococcal bacteria determinations and results can be stored under parameter code 31673, as noted in the 1987 TWRI. Dave's report on the results of the data analysis is attached to this memorandum. This data analysis also showed that about 10 percent of the NASQAN and 20 percent of the Benchmark counts had remark codes, primarily 'K' for non-ideal counts. Every effort should be made to make sure that at least one of the plates has a count in the ideal range of 20 to 100 colonies per 100 mL. If counts are consistently low, and sediment concentrations are very low, a larger volume of sample should be filtered. It is not necessary to limit the sample volume to 100 mL if there is no interference from suspended sediment. If counts are consistently higher than 100 colonies, dilutions must be made. Dilutions are described in the TWRI. All individuals processing bacteria samples should have access to historical records of volumes filtered and counts for the station. The fecal strep data currently stored under parameter code 99908 should be moved to the District quality-assurance data base (QADATA) and stored under parameter code 31673. The samples should be designated with medium code R (quality- assurance sample-surface water) and sample type 7 (replicate). Include parameter code 81352 (FILTER PORE SIZE (micro-meters)) with the appropriate pore size used for each sample. This parameter code should be entered for both sets of samples. Once the 99908 values are moved to the QADATA file, they can be deleted from the QWFILE. This data set will also be stored in the Branch of Quality Assurance's National Quality Assurance Data Base. Thanks to the Districts that participated in this effort. Please call or send edoc if you have any questions or comments on this memo. Tim Miller INFLUENCE OF FILTER PORE SIZE ON BACTERIA COUNTS FROM THE NASQAN AND BENCHMARK NETWORKS, WATER YEAR 1990 by David A. Peterson INTRODUCTION The membrane filter technique has been used for determination of fecal streptococcal bacteria for many years. The 1977 TWRI (book 5, chapter A4, p. 59) allowed use of either 0.45 or 0.7 micron filters for fecal streptococcal determinations. The 1987 version of the TWRI recommended use of the 0.45 micron pore size, to be consistent with USEPA recommendations. An experiment using both pore sizes was conducted during water year 1990, because of the need for consistent data collection with time and across the United States within the NASQAN and Benchmark networks. PURPOSE The purpose of this analysis is to determine whether the change in pore size has the potential to affect the utility of the fecal streptococcal data from the network samples. APPROACH The 1990 NASQAN/Benchmark implementation memo asked Districts to use parameter code 31673 for 0.7 micron filter results and District user code 99908 for 0.45 micron results. The data were pulled from the District PRlMES by Kathy Fitzgerald (Wisconsin District). The data were analyzed with PSTAT by Dave Peterson in the Wyoming District. A nonparametric paired Eta test of the ranks was used to compare the data from the two filter sizes, because of the paired nature of the data and the skewed distribution typically associated with bacterial data. The data were analyzed as whole data sets for each network and as subsets divided by several variables which had the potential to affect the results. Some of the original data were discarded from this analysis because of remark codes, primarily "K" for non-ideal count. About 10 percent of the NASQAN observations and 20 percent of the Benchmark observations were rejected for remark codes. RESULTS A total of 599 pairs of determinations from the NASQAN network and 63 pairs from the Benchmark network were used (table 1). The mean concentrations were larger in the NASQAN network than the Benchmark network, as might be expected considering the different objectives of the two networks. In both networks, the standard deviation was considerably larger than the respective mean. Table 1. Summary of raw data from the NASQAN and Benchmark networks. [Pore size in microns; mean, standard deviation (S.D.), Min. (minimum), and Max. (maximum), in colonies per 100 milliliters] Code Pore N Mean S.D. Min. Max. NASQAN P31673 0.7 599 658 2,490 0. 44,000 P99908 0.45 599 673 2,785 0. 51,000 Benchmark P31673 0.7 63 124 338 1.0 2,300 P99908 0.45 63 119 304 1.0 2,000 NASQAN Network The paired t-test indicated there is not a significant difference between the results of the two pore sizes, when considering all 599 data pairs. Results of the test are summarized in table 2. The mean and standard deviation of the ranks were identical. The t value was zero, indicating the mean was at the center of the distribution, with a probability of 1.0, indicating a very high probability (100 percent) of selecting that t again. The results indicate the data sets, based on ranks, are virtually identical. Table 2. Summary of paired t-test data from NASQAN network. [Pore size in microns; mean, standard deviation (S.D.),-by rank] Code Pore N Mean S.D. t Prob. P31673 0.7 599 300 173 0. 1.0 P99908 0.45 599 300 173 0. 1.0 Subsets The NASQAN data were divided into 4 subsets by range of concentration. The quartile ranges were as follows: 0-20 cols/lOOml, 21-81 cols/100 ml, 82- 305 cols/lOOml, and 306 to 51,000 colstlOOml. After the data were divided into subsets, the ranks were assigned, and the paired t-test computed. Results of the test indicated no significant difference between the two pore sizes in all 4 subsets. The mean and standard deviation were virtually identical, the t was zero and the probability was 1.0 for all 4 subsets. This is similar to the results presented previously in table 2 for the entire data set; the same pattern is repeated in all of the subsets described below. The quartile ranges of discharge were used to divide the data into 4 subsets. The discharge ranges were: 0.01-103 ft3/s, 104-639 ft3/s, 640-5,240 ft3/s, and 5,241-484,000 ft3/s. Paired t-tests of the ranks of the 4 subsets indicated no significant difference between the pore sizes. No significant difference was found between pore sizes, in subsets divided by specific conductance. The ranges used were: 19-157 us/cm, 158-350 us/cm, 351-782 us/cm, and 783- 73,400 us/cm. Season did not have a significant effect on results. The data were divided into 4 subsets: Oct. 1 to Dec. 31, Jan. 1 to March 31, April 1 to June 30, and July 1 to Sept. 30. Paired t-tests were performed on the ranks. The data were also divided into 8 subsets based on geographic area. The subareas were: North and South Atlantic slopes (N=32, stations 01540500 to 02075500), Ohio River basin (N=64, stations 03112510 to 03593005), St. Lawrence, Hudson Bay, and Upper Mississippi River basins (N=84, stations 04014500 to 05514500), Missouri River basin (N=84, stations 06228000 to 06934500), lower Mississippi and western Gulf of Mexico (N=138, stations 07019000 to 08447410), Colorado River and Great Basin (N=46, stations 09180000 to 10356500), Pacific slope (N=128, stations 11042000 to 13353200), and Pacific islands (N=23, stations 16213000 to 16713000). Paired t-tests of the ranks of each subset showed no significant difference between pore sizes. Benchmark Network Paired t-tests of the 63 points in the Benchmark data set indicated no significant difference between the pore sizes. As with the NASQAN data, the mean and standard deviation of the ranks were virtually identical, t was zero, and the probability was 1.0. Tests of selected subsets of the Benchmark data and the results of the NASQAN testing indicate little likelihood that the pore size results would be different if the data set were further subdivided. CONCLUSIONS Adoption of the 0.45 micron pore size for fecal streptococcal determinations will not adversely affect the utility of the data set for long term analysis. Paired t- tests of experimental data for 0.45 and 0.7 micron pore sizes collected in water year 1990 indicated no significant difference between the two pore sizes. This relation held for the complete data set, as well as for subsets divided by concentration, discharge, specific conductance, season, and geographic area. ================================================================== ================================================================== A Re-analysis of the 1990 Fecal Streptococci Paired Data Set Richard B. Alexander, U.S. Geological Survey, 1996 Because of concerns that the earlier paired t statistic of exactly zero (Table 2 above) may have resulted from coding errors in the data used for statistical analysis, a re-analysis of the original data was conducted. The summary statistics show some differences, but the conclusions are still identical--that there is no statistically significant difference in analytical results of the two filter types. Much of the variability in the differences in counts of the two methods appears to reflect random error intrinsic to the fecal streptococci method. A median difference of zero percent is obtained (mean difference of -5 %), and the distribution of differences is highly symmetric (25th and 75th percentiles of -16 and 15, respectively), suggesting that no upward or downward biases exist between the two methods (see Table 1 below). Table 1. Statistical summary of 1990 paired fecal streptococci data. Data associated with nonideal remark codes ('K') are deleted (42% of the observations) and observations with remark codes of '<' are assigned one half of the reporting limit. --------------------------------------------------------------------------- Code Pore Size N Mean S.D. Min. Max. Median 25th 75th ---------------------------------------------------------------------------- NASQAN ------ P31673 0.7 388 938 3,032 0 44,000 170 61 600 P99908 0.45 388 954 3,397 0 51,000 173 59 535 Dif* -- 387 -5.2 45 -400 100 0 -16 15 Benchmark --------- P31673 0.7 27 191 260 1 1,100 99 43 190 P99908 0.45 27 189 253 1 970 80 32 300 Dif* -- 27 -10.6 63 -283 100 -7 -17 7 NASQAN and Benchmark -------------------- P31673 0.7 415 888 2,938 0 44,000 160 60 560 P99908 0.45 415 904 3,291 0 51,000 164 58 480 Dif* -- 414 -5. 46 -400 100 0 -16 15 -------------------------------------------------------------- * Percent difference between fecal bacteria counts, (P31673-P99908)/P31673. ================================================================== Results of Statistical Tests: ----------------------------- Results of the application of the Wilcoxon Signed-Ranks Test for differences in all analytical results (NASQAN and Benchmark): Number of observations: 415 Two-sided p-value = 0.9369 Conclusion: The distributions of differences are not significantly different from one another. No significant differences exist in the analytical results of the two filter types. ==================================================================