QUALITY OF WATER BRANCH TECHNICAL MEMORANDUM NO. 90.04

 Analytical Methods--Turbidimetric Sulfate Method


                                           December 21, 1989



OFFICE OF WATER QUALITY TECHNICAL MEMORANDUM NO. 90.04

Subject:  Analytical Methods--Turbidimetric Sulfate Method

In early March 1989, the National Water-Quality Laboratory 
(NWQL) discovered a positive bias in the turbidimetric method 
for sulfate analysis.  The bias appears to result from not 
applying a blank correction to the measured values.  The 
magnitude of the bias is about 2 milligrams per liter (mg/L) 
median difference for concentrations less than 75 mg/L.  
The period of affected data is from October 1982 through 
July 1989.  Attachment 1 describes (1) the origin of the 
problem, (2) actions taken by the NWQL to address the 
problem, (3) implications for data stored in the water-
quality data base, and (4) the need for District-by-District 
evaluation of the bias.

Personnel who update the water-quality data base are 
instructed to make no adjustments to affected sulfate data 
until the Office of Water Quality (OWQ) and the Branch of 
Quality Assurance (BQA) have determined what adjustments can 
or should be made.  It is unlikely that a national adjustment 
to the affected sulfate values can be made because natural 
waters in the United States exhibit a complex range of 
dissolved ions (matrix) that probably will prevent 
identification of a nationally consistent correction for the 
bias.

To determine whether the magnitude of bias can be defined on 
a geographic basis, the NWQL will relate the values for 
affected samples to a wide array of chemical and water source 
factors that might cause the bias.  In addition, each 
District and project office is encouraged to make comparisons 
between blank corrected and uncorrected sulfate values.  The 
NWQL will assist this effort by providing paired analytical 
results until April 30, 1990.  Each office electing to make 
this comparison should follow the guidelines in attachment 2.  
Results of each comparison should be forwarded to Bill 
Shampine of the BQA by May 30, 1990.  Bill will coordinate 
review of the District comparisons with the NWQL, the BQA, 
and the Systems Analysis Branch.  Following review, 
individual offices and projects that have satisfactorily 
established a geographical pattern for the bias will be 
permitted to adjust values in subsequent investigative 
reports.

While the comparative evaluation of methods continues, a 
warning flag has been added to the WATSTORE data base for 
sulfate.  The flag reads:  "Sulfate values (00945) below 
75 mg/L have a median positive bias of 2 mg/L above the true 
value between 1982 and 1989.  See OWQ Memorandum 90.04."

If the comparative evaluations fail to enable a national 
correction, the WATSTORE data will remain uncorrected and the 
sulfate bias flag will become a permanent feature of WATSTORE 
tabling routines.




                              David A. Rickert
                              Chief, Office of Water Quality

Attachments

This memorandum does not supersede any previous memorandum.

Key words:  Laboratory analysis, sulfate

WRD Distribution:  A, B, S, FO, PO


QW90_04.txt

				ATTACHMENT 1



		United States Department of the Interior
			GEOLOGICAL SURVEY
			BOX 25046 M.S. _______
			DENVER FEDERAL CENTER
			DENVERt COLORADO 80225

IN REPLY REFER TO:

							December 18, 1989


Memorandum

To:		Chief, Office of Water Quality

From:		Chief, Branch of Analytical Services

Subject:	Positive Bias for Turbidimetric Sulfate Determinations


Problem.- In early March 1989, an apparent problem with the determination of
sulfate using the discrete-analyzer turbidimetric sulfate method was identified
by National Water-Quality Laboratory personnel. An evaluation of the problem
indicated that a significant positive bias exists in the 0-75 mg/L range. As far as
can be determined, the problem likely began in 1982 with the approval of the
discrete-analyzer turbidimetric sulfate method. At that time, it was determined
that data acquired by the turbidimetric method was comparable to ion
chromatography data (see Table 1). The comparison of data was acceptable
throughout the analytical range and therefore the method was approved for
routine use.

Over the past 18 months, however, a number of sulfate rerun requests on
samples with poor ion balances pointed to the possibility of an analytical
problem. When the analyses were rerun using blanking corrections, the sulfate
values were as much as 20 percent lower. Based on these preliminary results,
it was decided to examine the extent of the problem by analyzing samples for 1
week using the present turbidimetric method with a blank connection factor.
During that week, approximately 420 samples were analyzed for sulfate. The
concentrations of these samples ranged from 0.2 mg/L to 5,000 mg/L.

Plotting the blank-corrected values against the uncorrected values indicated a
positive bias for the uncorrected values below 75 mg/L (graph 1). Above 75
mg/L, the blanked and unblanked methods appear to give the same results.
From a more detailed look at data below 75 mg/L (graph 2), it can be seen that
there is a definite positive bias averaging approximately 2.0 mg/L over the
range of 0.2 to 75 mg/L, with a small number (1 S15 circled points) of the
samples having significantly greater positive bias. The sarnples with biases
significantly greater than 2 mg/L represent approximately 8.5 percent of the
samples analyzed during this 1-week time period. The 2.0 mg/L value is the
median difference between uncorrected and blank corrected data for all values
below 75 mg/L. Graph 3 shows a smooth fit line for the regular sulfate data
below 100 mg/L and gives an actual running bias concentration over the
complete range. When the median value of 2.0 mg/L is subtracted from the
regular sulfate values and the results are plotted against the blanked sulfate
value, a very good correlation is obtained (graph 4). The actual bias can be
seen (graph 5) to vary from 0-27 mg/L with the majority of the larger variations
falling between 5-15 mg/L.

Laboratory Action.- After comprehensive discussions with the Office of Water
Quality and the Branch of Quality Assurance, the NWQL decided to take the
following actions beginning July 1989:

	1. Samples containing less than 75 mg/L of sulfate have been
analyzed both by the blank-corrected procedure and by the uncorrected
method. The blank-corrected value is being reported using WATSTORE Code
00945, Method Code F. and is the accepted, approved value. For both 00945F
and 00945D (see Item 3), the Method Codes (F & D) can be found on the
WATLIST--in the last column under wMET," before wSIGw figures.

	2. The Uncorrected" value is being reported using WATSTORE Code
99890 (a special code assigned to the laboratory for such purposes), Method
Code A.

	3. Samples with sulfate values greater than 75 mg/L are not being
blank-corrected and are being analyzed according to the present turbidimetric
method. Data are being stored using WATSTORE Code 00945, Method Code
D.

	4. Analysis of samples with sulfate values below 76 mg/L by both the
present and the blank-corrected methods will continue until a completely new
method is implemented April 30, 1990.

Data Base Implications.- Parameter Code 99890 has been omitted from the
Amdahl Parameter Code Dictionary to prevent values stored under this code
from being uploaded to WATSTORE and STORET. These uncorrected data are
to be used for informational purposes only at the local level and are not to be
released. Any policy to the contrary will be made by the Office of Water Quality.
In order for the District users to access these data on the Prime, it is necessary
that they have the Parameter Code Dictionary (PCD) shipped with the 89.1
version of the NWIS software in place. Since this version of the PCD was not
available until August 1, 1989, users who try to access the data using older
versions will encounter messages indicating an error for Parameter Code
99890 in data retrievals. The error messages will continue when attempts are
made to upload 99890 data to the Amdahl.

District Evaluations.- Initially, it appeared that if 2.0 mg/L were subtracted from
sulfate values, the data base would produce a corrected national data set.
However, this is not applicable as a general rule because water matrices can
and do behave differently. Thus, the paired blank/nonblank analyses for sulfate
will allow comparisons of data. Some preliminary work has been completed by
John Izbicki, QW Specialist, Boston, Massachusetts. His data and analyses
indicate that by multiplying the nonblank data from his area by 0.88,
comparable results with the blank-corrected values obtained by the NWQL are
given. John had 76 samples rerun by the blank-corrected method. Of these
reruns, 67 had lower sulfate values, supporting the need for blank-correcting for
his particular water matrix.

Each office with affected sulfate data is encouraged to carry out its own
evaluation and determine whether an areal correction factor is viable. Policy
decisions involving the approach to be used in documenting and adjusting
erroneous historical data (not just sulfate) will be forthcoming from the Office of
Water Quality and the Branch of Quality Assurance. Until these policies are
formulated, no adjustments to the data other than those currently accepted
under existing quality assurance procedures are to be made.





							R.O. Hawkinson

Enclosures
===================================================================
A series of 5 graphs are presented in separate graphical documents.
===================================================================

 Table I. Comparison of Turbidimetric vs. Ion Chromatograph Methods
Done in 1982. (All results in mg/L S04)

	Sample			Turbidimetric		Ion Chromatrograph


	2055033				22			22
	2055049				42			40
	2057602				13			12
	2077090				16			15
	2083042				12			12
	2083043				13			12
	2083044				13			13

	2092017				22			20
	2092042				25			20
	2097025				24			25
	2109012				31			24
	2090036				10			10.1
	2090037				6.5			6.25

	2011012				35			36
	1341067				95			98
	2008001				38			39
	1309012				12			13
	1309019				11			12
	1309017				8.3			9.2

	1313012				73			73
	1341069				84			81
	1309011				63			64
	1309010				31			32
	1342022				16			12
	1309014				110			110



If you have problems or questions concerning this document
and would like clarification, please contact one of the
following individuals at the NWQL:

	Merle W. Shockey--FTS (776-5345)	 --(MWSHOCKEY)

	Harold Ardourel---FTS (776-5345)	 --(HDARDOUREL)

	Marvin Fishman----FTS (776-5345) 	 --(MJFISHMAN)





								 Attachment 2

       RECOMMENDED PROCEDURES FOR COMPARING METHOD DIFFERENCES


1. For the paired concentrations of a single sample, calculate the difference
(uncorrected S04 - corrected S04), and plot this difference versus -
corrected S04. Refer to this plot (Attachment 2 Figure 1) as "the scatterplot."
2. To emphasize the central pattern of this relationship, sort the data by
increasing corrected concentrations, and compute a lowess smooth of the
difference data. Plot this smooth as a line on the scatterplot. See figure 1
as an example. Software for computing a lowess smooth already resides on
most Primes. The compiled fortran code is also found on the RVARES
Prime, and can be ffle-transferred to your site using the following
command:

FTR COMMON>CLASS>smooth3.run FTS DEPOT>amooth 3.run -SS RVARES

This program prompts the user for the filename of (x,y) data pairs, and
outputs three files. The one named "datafile".mid, where "datafile" is the
file name of the input data set, is the middle or lowess smooth, consisting
of paired (x,y), a smoothed y for evry x value input.

3. Plot the differences between corrected and uncorrected concentrations
on a state map, looking for geographic consistency in biased values.
Determine if any patterns exist in the characteristics of the stations/wells
where bias exists or is large (see Figure 1 in a separate graphical 
document filed under this memorandum number).

4. Perform a multiple regression of differences versus several explanatory
variables such as type of sample (l=ground water, O=surface water, etc.),
color, other chemical constituents, etc., to determine whether any causative
factors for the bias can be isolated.

If you have questions concerning these procedures, please contact
Dennis Helsel at FTS 959-5713 (DHHELSEL)

============================================================