150 likes | 237 Views
Shaken, not stirred – comparing lab and field filtering for TSS. Elgin Perry, Ph.D. Statistics consultant and Bill Romano MD Dept. of Natural Resources. Analytical Methods and Quality Assurance Workgroup. 3 October 2008. A little background.
E N D
Shaken, not stirred – comparing lab and field filtering for TSS Elgin Perry, Ph.D. Statistics consultant and Bill Romano MD Dept. of Natural Resources Analytical Methods and Quality Assurance Workgroup 3 October 2008
A little background • DHMH currently analyzes samples collected at 53 non-tidal stations that are part of Maryland’s CORE/Trend network (TSS is analyzed at the Cumberland lab for 17 of these stations) • In July 2005, the sampling protocol changed for nutrients from analyzing whole water samples to analyzing filter pads • This change did not include TSS, which is still analyzed as a whole water sample that is filtered in the lab at DHMH or Cumberland
A potential problem • DNR is planning to change the TSS sampling protocol to match what is now done for our tidal program, field filtering for nutrients and TSS • One would think that the analysis of TSS is so simple that even a statistician could do it • Experience with the change from DHMH to CBL with the Maryland tidal program and recent problems with the Virginia main Bay program have shown that analysis of TSS is not so straightforward • Subtle differences in how samples are handled and analyzed have resulted in step trends in the data
Yet another experiment(more work for the AFO and DHMH) • Compare field vs. lab filtered TSS • Field staff collected extra water at the 53 stations in March and July • DHMH supplied pre-weighed filter pads for all stations • All field-filtered pads were sent back to DHMH for analysis • 17 whole water samples were sent to Cumberland and 36 were sent to DHMH, which is the current protocol
Why this seems confusing (to us) • The March difference plot only shows large differences associated with high concentrations • This is supported by the March WMRL box plot, which had high concentrations of TSS because of recent storms • It may have been an “order effect” because the lab bottles were filled after the filter pad bottle
Confusion (continued) • TSS concentrations were much lower in July and we expected the bias to go away • The July difference plot shows a positive bias in the data (lab filtered exceeds field filtered) • The July box plot shows a reversal in the bias, with no bias at WMRL and a small positive bias at DHMH • The mean difference for March and July data is 3.9 mg/L, March only 6.9 mg/L and July only 1 mg/L
Comparing different data sets • CV split sample among sub-samples - 21% • CV split sample among labs - 37% • CV for DHMH lab replicates - 32% • CV for lab and field filtered - 44% • The CV for methods is a bit larger than the others and larger than we expected. Note, these CVs measure total variation = to bias + noise.
Questions to consider: • Should we continue the paired samples study? • Are there modifications of the experimental design that might yield better information?