303(d) Listing Methodology September 25, 2012 Application of the - - PowerPoint PPT Presentation

Application of the Integrated Impact Analysis Tool

303(d) Listing Methodology September 25, 2012

• MMI provides a score to characterize

macroinvertebrate health.

• What does it mean if my segment has a low score?
• What do I do if I get on the 303(d) list because of my

MMI score?

• What is really causing the change in biological health

that I am seeing in my segment?

• Is the nutrient level in my segment impacting the

biology or is it the fact that my stream is wide, straight, and nearly dry half of the year?

Setting the Stage

• WERF project (partly funded by EPA) to distinguish the

relative impact of chemistry and habitat on aquatic life (Project 98-WSM-1).

• Answered the question: How do I determine the

relative impact of chemical and habitat stressors on the biology in my stream?

• Teamed with GEI (Chadwick) and Risk Sciences (Tim

Moore).

• Has been used in Santa Ana UAA, Arid West work,

Southeastern Wisconsin Watershed Trust, and by GEI

• ften in their biological work.

Integrated Impact Analysis

• How would this model be used to detect impairments?
• Would it be more useful for stressor identification after

an impairment has been identified?

• Would the model be the basis for stand-alone listing

decisions or would it be used in conjunction with other lines of evidence?

Topics to Cover

Brown and Caldwell 4

Integrated Impact Analysis Chemical Physical Biological

+ =

• Uses existing statistical methods
• Principle Components Analysis
• All Possible Regressions
• Chi-Square Automatic Interaction Detection (CHAID)
• Results identify key stressors and their relative impact.
• Don’t try this at home without a keen interest in

statistics . . .

IIA – The Gist

Basic Steps of IIA

MODEL

Apply Basic Statistics Identify Key Stressor and Response Variables Rank Variables According to Relative Impact Repeat Cycle if More Variables Are Needed Fit Equation to Describe Interactions Between Stressors and Response

• Perform basic descriptive statistics and

develop graphics

• Normalize data as needed – develop new

descriptive statistics and new graphics

• Compile a correlation matrix

Step 1 Apply Basic Statistics

Water Chemistry Basic Statistics

Habitat Basic Statistics

• Looking for correlations (≥0.6).
• Correlated variables can act as surrogates for each
• ther.

• Identify key independent stressor variables

and relationships between variables using:

• Principle Components Analysis (PCA)
• All Possible Regressions
• Chi-square Automatic Interaction Detection (CHAID)
• Iterative process that systematically removes

variables from the larger pool of variables.

Step 2 Identify Key Stressor and Response Variables

Principal Components Analysis

• Useful in determining how variables relate to
• ne another – how they move in space.
• If variables generally move together, one

variable can act as a surrogate for the other(s).

Principal Components Analysis

• Look at values with an absolute value ≥0.6.
• A component in PCA is a group of variables that move in the same

direction.

• Generally, the variable with the highest score is identified as the

surrogate.

• Rerun to limit number of stressors to 6 and responses to 2 or 3.

strongest

Principal Components Analysis

All Possible Regressions

• Combines one response with many stressor

variables into models using all combinations of the stressor variables.

• Look at all combinations to see what

combinations explain the greatest amount of variance with the lowest error.

• Look for lowest variable count that explains the

most variance.

All Possible Regressions

All Possible Regressions

Chi-Squared Automatic Interaction Detection (CHAID)

• Identifies both linear and non-linear

relationships between variables.

• Non-parametric technique, so data should not

be transformed.

• Robust to missing data points.

NH3 range Macroinvertebrates P/Channel Range Macroinvertebrates

• Develop matrix of key independent stressor

variables and relationships found in Step 2

• Repeat Steps 2 and 3 until the two most

influential independent stressor variables are identified for each dependent response variable

Step 3 Rank Variables According to Relative Impact

Rank Variables According to Relative Impact

• Develop a matrix of response variables and

their corresponding “important” stressor variables for each of the 3 analyses.

• Look for stressors identified by multiple

analyses and sort by number of analyses in common.

• To help with sorting, refer back to the analyses

and how strong the relationships are between the stressor and response variables.

Matrix of Ranked Stressors for Each Response

• Use three-dimensional modeling program to

identify specific non-linear relationship transformations

Step 4 Fit Equation to Describe How Stressors Interact

Fit Equation

• Enter the residuals calculated for the response

variable into a new column in dataset.

• The residuals are the remainder of the of the

response variable after the variability caused by the 2 stressor variables is removed.

• Repeat steps 2-4 to identify the next 2

important stressor variables.

Step 5 Repeat Cycle if More Variables Are Needed

Residuals From Equation Fitting

IIA Gives You . . .

• An ordered list of stressors that are causing an

impact on the response variables.

• A model to help predict how the response

variables will change based on a change in the stressor.

• An understanding of whether habitat is playing

a role in limiting the response variable.

• A means to make sense of what the MMI

metrics are showing.

• How would this model be used to detect impairments?
• Would it be more useful for stressor identification after

an impairment has been identified?

• Would the model be the basis for stand-alone listing

decisions or would it be used in conjunction with other lines of evidence?

Back to the Topics . . .

Brown and Caldwell 33

Questions