410 likes | 522 Views
Method Adjustment Analyses for VA DEQ Nutrient Determinations. Introduction. 1984 until 1994 TN calculated as TKNW + NO 23 F(NO 2 F + NO 3 F) TP measured directly (analyzed by DCLS). January 1994 TN calculated as TDN + PN (analyzed by VIMS).
E N D
Method Adjustment Analyses forVA DEQ Nutrient Determinations
Introduction • 1984 until 1994 TN calculated as TKNW + NO23F(NO2F + NO3F) TP measured directly (analyzed by DCLS). • January 1994 TN calculated as TDN + PN (analyzed by VIMS). TP calculated as TDP + PP (analyzed by VIMS). • February 1995 to the present DCLS adopts VIMS methods for PC, PN, PP, and TDN DCLS uses EPA method 365.2 for TDP. • Changes resulted in step trends in both parameters. • Adversely affects statistical analyses.
Pre-Method Change Post-Method Change Example – TN at Station LE5.4
Methods • 33 Tidal Monitoring Stations (1998-2003). • 46 Pfiesteria Monitoring stations (1998-2002). • Focus on Tidal Monitoring data only but Pfiesteria Monitoring data also analyzed. • For CBP TP, there was an average of 33 with a min. of 11 and max. of 42 samples per station. • For CBP TN, there was an average of 28 samples per station with a min. of 10 and max. of 75 samples per station. • Sample collection and processing reflect historical methods used except: change in instrumentation.
Definition of Bias • New Method – Old Method = Difference Between Methods. • A negative value indicates Old Method is biased high relative to the New Method i.e. consistent with historical bias. • A positive value indicates Old Method is biased low relative to the New Method in contrast with the historical bias.
TN - Paired Comparisons • Difference Between Methods significantly different from zero: Student’s t= -26.66; Prob. >| t | <0.01, Wilcoxon Signed Rank S=-500812; Pr >|S| <0.01.
TN - Screening Analyses + 2 Std Mean Difference • Mean difference between methods: -0.320.52. • 75% of all observations at or below 0.0 mg/L. • Old Method biased high. • There were two distinct groups of values for the differences. - 2 Std
TN – Screening Analyses • First group of differences Mean: 0.00 mg/L Range: –0.50 to 0.80 mg/L Range constant regardless of mean conc. • Second group of differences Mean: -1.00 mg/L Range: -0.50 to –2.00 Difference decreases with increasing mean conc.
TN - Screening Analyses • Frequency histogram confirms two groups in bias.
TN – Source of Groups? • Procedural Effects: • Two Monitoring Programs • Three DEQ Offices (TRO, PRO, NRO) • Spatial Effects • Temporal Effects • Environmental Effects • Combinations
Mean diff. = -0.22 Positive Bias CBP Pfiesteria TN – Monitoring Program Effect Mean diff. = -0.41 Positive Bias
Mean diff. =-0.480.45 Old Method Biased High Mean diff. =0.080.18 Old Method Biased Low TN – Collection Office Effect (CBP) Mean diff. =0.030.33 Old Method Biased Low • Was collection agency (PRO, TRO, NRO) responsible? • Grouping persists in TRO data with Old Method biased high. • Grouping disappears in PRO data but now Old Method biased low. • Grouping disappears in NRO data but now Old Method biased low.
TN – Spatial Effects on Bias (CBP) • Negative bias at most Tidal Fresh and Oligohaline stations. • PRO and NRO responsible for collection at most of these stations. • Positive bias at higher salinity (mostly TRO) stations.
TN – Spatial and Temporal Effects (CBP) • Spatial pattern persists between years. • Difference between methods higher in 2002 ( 0.00 mg/l).
TN – Temporal Effects (CBP - TRO) Mean diff. =-0.130.30 Old Method Biased High • Two groups observed appear to be two different time periods. • Old Method is biased high for both time periods. Mean diff. =-0.650.43 Old Method Biased High
TN – Environmental Effects (CBP - TRO) Shown are Pearson’s |R| and associated p values. All correlations based on > 400 observations except CHL a (221). • Several significant correlations but none explain patterns observed. • Other potential predictors?
TN – Environmental Effects (CBP - TRO) • Prior to 2003, difference was much higher even in high salinities. • For 2003, values of difference at 0 salinities were higher. • Significant but slight correlation with salinity (|R|=-0.12;p<0.01).
TN – Component Variables (CBP - TRO) • Plots of component variables indicate TKNW as problem. • Plot of bias confirms this observation. • Correlation analysis revealed no apparent environmental causes.
TN – Component Variables (CBP - TRO) • When TKNW < 0.8 then Old Method bias is negative. • When TKNW >=0.8 then Old Method bias is positive.
|R| = NS |R| = 0.25 |R| = 0.29 |R| = 0.08 TKNW – Environmental Effects (All Data)
TN - Conclusions • Overall significant difference between methods with Old Method biased high relative to the New Method. • Two groups of differences in the data set. • Two groups were not due to differences between monitoring programs or DEQ Offices. • Old Method was biased low at most Tidal Fresh and Oligohaline stations and at low concentrations of TKNW. • Old Method was biased high when concentrations of TKNW were high. • No clear effect of environmental factors on results.
TN - Conclusions • Method correction development is not recommended using these data • Bias in Old Method low at Tidal Fresh/Oligohaline areas • Bias is inconsistent in all other areas • Bias changes in relation to TKNW • No apparent environmental predictor of this change • Changes in instrumentation are still at issue.
TN - Recommendations • Use Blocked Seasonal Kendall until questions are answered. • No other recommendations… • Are there additional analytical approaches that might be useful for exploring these data? • Are there any explanations for the patterns observed?
TP - Paired Comparisons (CBP Only) • Mean difference between methods significantly different from zero: Student’s t=8.00; Prob. >| t | <0.01 Wilcoxon Signed Rank S value =219432; Prob. >|S| <0.01 • Mean difference between methods = 0.020.035 mg/L. • 75% of all differences at or above 0.00.
+ 2 Std Mean Difference - 2 Std TP - Screening Analyses (CBP Only) • Mean difference between methods: 0.020.035 mg/L. • Old TP Method biased low relative to New TP Method up to mg/L. • This conflicts with historical pattern in the data.
TP - Conclusions • Significant difference between methods but Old Method biased low relative to the New Method. • Opposite of pattern in historical data. • Spatial, temporal or environmental effects do not explain the difference between current and historical bias. • Difference may be due to a change in instrumentation and/or other procedures. • Current bias may be due to difference in accuracy between methods at high levels of TSS.
TP – Recommendations • Data are not conducive to method adjustment. • Blocked Seasonal Kendall for TP until additional studies are available. • Another paired study? • Use old instrumentation or more trouble than it is worth? • Do we control for season or not? • Do we control locations, TSS, salinity or not? • Are there other effects?
CSSP Data • Collected at Three Stations • Two Tidal Fresh stations PMS10 and Wilcox Wharf • One Mesohaline station CB4.4. • May ’95 through June ’04. • 328 total observations. • Surface measurements only.
CSSP TP Data – Screening • Mean difference between methods • significantly different from zero: • Student’s t: • t value =3.79; Prob. >| t |=0.0002 • Wilcoxon Signed Rank: • S value =-4076; Prob. >|S|=0.0002 • Mean difference between methods = • - 0.010.036 mg/L. • 70% of all differences at or below 0.00 with 25% = 0.00. • No significant effect on results by removing outliers. • Data appear to be comparable with slight positive bias for Old Method. + 2 STD Mean - 2 STD
CSSP TP Data – Station PMS10 Only • Mean difference between methods • NOT significantly different from zero: • Student’s t: • t value =-0.968; Prob. >| t |=0.3367 • Wilcoxon Signed Rank: • S value =119; Prob. >|S|=0.1741 • Mean difference between methods = • - 0.010.064 mg/L. • Bias of Old Method positive. • 70% of all differences at or above 0.00 with 29% = 0.00. • Three probable outliers. + 2 STD Mean - 2 STD
CSSP TP Data – Station PMS10 Only • Remove outliers and there is a significant difference: • Student’s t: • t value =2.562; Prob. >| t |=0.013 • Wilcoxon Signed Rank: • S value =264; Prob. >|S|=0.0180 • Mean Difference between methods= 0.010.015 mg/L. • With outliers removed, bias for Old Method is negative. + 2 STD Mean - 2 STD
CSSP TP Data – Station CB4.4 Only • Mean difference between methods significantly different from zero: • Student’s t: • t value =2.562; Prob. >| t |=0.013 • Wilcoxon Signed Rank: • S value =264; Prob. >|S|=0.0180 • Mean Difference between methods= • -0.010.025 mg/L. • No significant effect on results by removing outliers. + 2 STD Mean - 2 STD
CSSP TP Data – Wilcox Wharf • Mean difference was significantly difference from zero: • Student’s t: • t value =-3.44; Prob. >| t |=0.0013 • Wilcoxon Signed Rank: • S value =261; Prob. >|S|=0.0015 • Mean Difference between methods= • -0.010.021 mg/L.
CSSP - Conclusions • Data indicate a slight positive bias in Old Method. • Bias was consistently positive at all stations except at PMS10 when outliers were removed. • Bias was very small (–0.01 mg/L) and does not change in relation to concentration. • Data are not compatible with a method correction.
DCLS Instrument Comparison Studies • Data were part of a larger study comparing 8 variables between the Technicon AA and Skalar AA: • 94 total phosphorus samples • 176 PO4F samples. • Documentation indicates: • Study Conducted in the Summer of 1997 • No information on collection locations.
DCLS Instrument Comparison for TP • Mean difference was NOT significantly different from zero: • Student’s t: t value =-1.58; Prob. >| t |=0.1173 • Wilcoxon Signed Rank:S value =196.5; Prob. >|S|=0.0155. • Mean difference =0.000.01 with most values around 0.00 mg/L. • No bias for the Old Method.
DCLS Instrument Comparison for PO4F • Mean difference between methods • significantly different from zero: • Student’s t: • t value =-11.58; Prob. >| t | <0.0001 • Wilcoxon Signed Rank: • S value =-6341; Prob. >|S| <0.0001. • Mean difference was -0.010.01 mg/L • 83% percent of all observations were at or below zero.