statistics and nutrient levels julie stahli metro wastewater reclamation district march 2010
TRANSCRIPT
Statistics and Nutrient Levels
Julie Stahli
Metro Wastewater Reclamation District
March 2010
MMI
Version 3 lacks a lot of the problems that previous versions contained.
Statistically, the model distinguishes reference sites from stressed sites.
The level chosen as the level of impaired is statistically generous.
MMI
MMI Issues (based on Metro Data) Comments should focus on implementation
Sites with low scores should be put on the M&E list and multiple samples should be taken.
Implementation should be based on median of all MMI scores over the course of five years.
We have seen substantial variation at a single site.
Big rivers need alternative indexes when sites are found in the “yellow zone”.
Sample DataType Year 31st 64th Up CC 88th 120th – H 124th 160th Ft Lupton Cty Rd 18
Kick 1993 25.98 27.59
Kick 2003 54.56 17.44
Kick 2004 16.82 21.61 26.33 31.36
Kick 2005 35.24 35.67
Kick 2006 17.76 44.56 18.96 19.52 17.52 28.61
Kick 2007 42.6 43.94 23.44
Kick 2008 32.1 19.84 41.65 35.54
Kick 2009 22.29 53.42 24.74 9.94 19.86 17.37 19.24 33.93 43.5
Med 22.3 44.3 19.8 18.2 30.8 19.6 26.3 33.9 39.6
Blue indicates attainment, red indicates impaired
Big Rivers
0
10
20
30
40
50
60
70
80
90
100
0.000 0.001 0.010 0.100 1.000 10.000
TP Median
MM
I
Large
Small
Unknown
Small Plains
Lets assume that the MMI is valid and can be related to nutrient concentrations to form TP and TN limits….
Regression analysis
Lots of types of regression analysis linear Multi-linear regression logistic regression etc
Regression analysis
Each type has restrictions on how it is used based on the assumptions we can make about the data.
Evaluate the assumptions to determine whether or not the model is valid or appropriate. Example assumptions
The relationship is linear in nature. The errors are independent and normally
distributed. The model is the best available for describing the
relationship.
Regression analysis
After the model is developed, we evaluate the model to see how useful it is
Tools for evaluation: R2 value Goodness of fit for data Confidence intervals Residual mean-squared error
Ideally, a model on which decisions are based are both valid and useful.
Quantile Regression
Breaks up the data into quantiles and forms an individual regression for each one.
Used with the wedge plot.
Quantile Regression
Typically, no assumptions are necessary. Lots of decisions are made during the modeling
process. Typically, goodness of fit is evaluated in two ways:
examining whether or not the slope is different from zero
The P value for the model (is it significant) Results are not predictive. Can’t really determine whether or not the model is
valid or useful.
Quantile Regression
Quantile Regression
Used to explore patterns in data that cannot be seen using more traditional methods.
Designed to be used in exactly the same way that CDPHE is using it – to illuminate an effect that may be muddied by lots of other variables.
There is no way to evaluate whether or not the model is right or wrong, good or bad and should be considered as part of a
weight of evidence approach.
Quantile Regression
What else do we use?
We don’t believe there is a better tool to solve the nutrient problem.
Especially for Phosphorus.
What??!?
The uncertainty in the model needs to be acknowledged. Using this method should not set
precedent for future usage where other options may be available.
In repeating their methods, we found that choices were made that are not scientifically valid.
CDPHE Review
Quantile Regression Slope
Problems
Using log scale. Using only the warm data to define the slope. Using the 90th percentile without biological
reason (85% is more commonly used). Nitrogen is difficult because of lack of data.
Unless otherwise noted, all limits would apply to Warm streams and rivers.
Log Scale Issues
Normal regression Using a log-scale
Quantile Slope
Does Not Cross Zero
Good P-value
Final Valuemg/L
0.05 -0.93683 Yes No N/A
0.25 -6.19878 Yes Yes 0.237
0.50 -10.9204 Yes Yes 0.143
0.75 -11.386 Yes Yes 0.139
0.85 -10.1523 Yes Yes 0.150
0.90 -10.4541 Yes Yes 0.147
0.95 -10.6691 Yes Yes 0.145
Quantile Slope
Does Not Cross Zero
Good P-value
Final Valuemg/L
0.05 -6.128 Yes No N/A
0.25 -16.018 Yes Yes 0.263
0.50 -25.208 Yes Yes 0.195
0.75 -26.127 Yes Yes 0.191
0.85 -24.666 Yes Yes 0.198
0.90 -25.764 Yes Yes 0.192
0.95 -26.07 No Yes N/A
Using only warm data
Using all data Using warm data
***Both using log scales to make comparable
Quantile Slope
Does Not Cross Zero
Good P-value
Final Valuemg/L
0.05 -0.93683 Yes No N/A
0.25 -6.19878 Yes Yes 0.237
0.50 -10.9204 Yes Yes 0.143
0.75 -11.386 Yes Yes 0.139
0.85 -10.1523 Yes Yes 0.150
0.90 -10.4541 Yes Yes 0.147
0.95 -10.6691 Yes Yes 0.145
Quantile Slope
Does Not Cross Zero
Good P-value
Final Valuemg/L
0.05 -2.461 No No N/A
0.25 -3.074 No No N/A
0.50 -5.205 Yes No N/A
0.75 -10.34 Yes Yes 0.148
0.85 ** did not run N/A
0.90 -11.96 Yes Yes 0.135
0.95 -9.722 Yes Yes 0.155
Cold data
Quantile SlopeDoes Not Cross Zero
Good P-value
Final Valuemg/L
0.05 1.7374 No No N/A
0.25 -3.6333 No No N/A
0.50 -1.9215 No No N/A
0.75 -3.9878 No No N/A
0.85 -5.4538 Yes Yes 0.202
0.90 -3.9919 No No N/A
0.95 -6.2549 No No N/A
Warm – 0.9 -11.96 Yes Yes 0.090
All – 0.9 -25.76 Yes Yes 0.1819
Recommendation
Quantile SlopeDoes Not Cross Zero
Good P-value
Final Valuemg/L
0.05 -6.128 Yes No N/A
0.25 -16.018 Yes Yes 0.263
0.50 -25.208 Yes Yes 0.195
0.75 -26.127 Yes Yes 0.191
0.85 -24.666 Yes Yes 0.198
0.90 -25.764 Yes Yes 0.192
0.95 -26.07 No Yes N/A
Nitrogen
Nitrogen
Valid models with the normal data if cold and warm are looked at together as recommended for Phosphorus.
Strongest scientific model (at 0.9 quantile) gives TN value of 2.06 mg/L.
Currently there is a paucity of data and I think we could argue for more information after the implementation of ammonia and nitrite criteria.
Caveats
I am not an expert in quantile regression and may have missed some qualifying assumptions.
All of the relationships we are discussing are tenuous at best, but may qualify as best current scientific opinion.
Thank you….
Questions???