# APPENDIX B. CASE STUDY II: SUPPORTING MATERIALS B.1. CASE ... Case Study II. The smooth lines are...

date post

24-Jul-2020Category

## Documents

view

7download

0

Embed Size (px)

### Transcript of APPENDIX B. CASE STUDY II: SUPPORTING MATERIALS B.1. CASE ... Case Study II. The smooth lines are...

B-1

APPENDIX B. CASE STUDY II: SUPPORTING MATERIALS

B.1. CASE STUDY II: MATRICES OF SCATTER PLOTS AND ABSOLUTE SPEARMAN CORRELATION COEFFICIENTS

Figure B-1. Anions. Matrix of scatter plots and absolute Spearman correlation coefficients between specific conductivity (μS/cm), alkalinity (mg/L), sulfate (mg/L), chloride (mg/L), and ion ([HCO3− + SO42−] mg/L) concentrations in Case Study II. All variables are logarithm transformed. The smooth lines are the locally weighted scatterplot smoothing (LOWESS) lines (span = 2/3).

B-2

Figure B-2. Cations. Matrix of scatter plots and absolute Spearman correlation coefficients between specific conductivity (μS/cm), ion ([HCO3− + SO42−] mg/L), hardness (mg/L), Mg (mg/L), and Ca (mg/L), in Case Study II. All variables are logarithm transformed. The smooth lines are the locally weighted scatter plot smoothing (LOWESS) lines (span = 2/3).

B-3

Figure B-3. Dissolved metals. Matrix of scatter plots and absolute Spearman correlation coefficients among specific conductivity (μS/cm), ion ([HCO3− + SO42−] mg/L), and dissolved metal concentrations (mg/L) in Case Study II. All variables are logarithm transformed. The smooth lines represent the locally weighted scatterplot smoothing (LOWESS) lines (span = 2/3).

B-4

Figure B-4. Other water-quality parameters. Matrix of scatter plots and absolute Spearman correlation coefficients between environmental variables in Case Study II. The smooth lines are locally weighted scatter plot smoothing (LOWESS) lines (span = 2/3). Specific conductivity is logarithm transformed specific conductivity (μS/cm); temp is water temperature (°C); Hab_Sc is habitat score from Rapid Bioassessment (Habitat) Protocol (Barbour et al., 1999) score (possible range from 0 to 200); fecal is logarithm transformed fecal coliform bacteria count (per 100 mL water); embeddedness is a parameter score from the Rapid Bioassessment Protocol (possible range from 0 to 20); DO is dissolved oxygen (mg/L); TP is logarithm transformed total phosphorus (mg/L); NO23 is logarithm-transformed nitrite [NO2−] plus nitrate [NO3−] in mg/L.

B-5

B.2. CASE STUDY II: ASSESSMENT OF POTENTIAL CONFOUNDERS Previous assessments of the factors potentially influencing the model of the causal

relationship between ionic concentration and extirpation of benthic invertebrates (Suter and

Cormier, 2013; U.S. EPA, 2011, Appendix B) indicated that the following factors did not

substantially confound the causal relationship between specific conductivity (SC) and benthic

macroinvertebrate assemblages: rapid bioassessment protocol (RBP) habitat scores (Barbour

et al., 1999), sampling date, organic enrichment, nutrients, deposited sediments, high pH,

selenium, heat (temperature), lack of headwaters, size of catchment area, settling ponds,

dissolved oxygen (DO), and metals. However, low pH could possibly affect the model (Suter

and Cormier, 2013; U.S. EPA, 2011, Appendix B) because its mode of action is associated with

increased solubility of metals which are toxic (e.g., Wren and Stephenson, 1991; Ormerod et al.,

1987). As a result, sampling sites with acidic waters (pH

B-6

Table B-1. An output table for two generalized linear models. The first is the simple model predicting the number of mayfly genera from specific conductivity. The second is a multivariate model with the additional covariates rapid bioassessment protocol (RBP) score, temperature, and fecal coliform count. These variables were chosen based on previous analyses as likely confounders that could co-occur and have combined effects. Fecal coliform count and specific conductivity were first log10 transformed to normalize the data, then all four variables were centered and scaled (subtracting the means and then dividing the centered values by their standard deviation) so that all four variables are at the same scales. The response variable is assumed to follow a Poisson distribution which appropriate for counts of occurrences.

Parameter Estimate Standard error Univariate model Intercept 0.848 0.017 Specific conductivity slope −0.852 0.018

Multivariate model Intercept 0.842 0.017 Specific conductivity slope −0.703 0.019 RBP slope 0.037 0.013 Temperature slope −0.202 0.013 Fecal coliform slope −0.077 0.013

B-7

Figure B-5. Scatter plot showing inter-relatedness of stream temperature and sampling date. The fitted line is a locally weighted scatterplot smoothing spline (LOWESS, quadratic polynomial, span = 0.75).

B.2.2. Influence of Poor Habitat and Organic Enrichment on the Hazardous Concentration (HC05) To assure that the genus extirpation concentration distribution (XCD) model was

detecting effects from SC and not a response to poor habitat, the HC05 was recalculated using the

example criterion-data set in which samples were removed with an RBP score

B-8

or HC05 (see Figure B-6). With this constrained data set (RBP score >130) the HC05 was

337 μS/cm (95% confidence interval [CI] 265−360 μS/cm). The confidence interval overlaps

with the HC05 for the example criterion continuous concentration (CCC; 338 μS/cm; 95% CI

272−365 μS/cm). Therefore, no correction was made for habitat quality or organic enrichment. Pr

op or

tio n

of e

xt irp

at ed

g en

er a

Figure B-6. Lower portion of genus extirpation concentration distribution with and without removal of sites with poor habitat. For both data sets the pH is >6, Rapid Bioassessment Protocol score is >130 and fecal coliform is ≤400 colonies/100 mL. The full (unconstrained, open circles) data set has 139 genera and the constrained data set has 88 genera. Habitat disturbance and organic enrichment have little influence; the hazardous concentration (HC05) for the constrained data set is 337 μS/cm (95% confidence interval [CI], 265−360 μS/cm), based on 88 genera (closed circles).

B-9

B.2.3. Potential Influence of Temperature on the Hazardous Concentration (HC05) To assure that the genus XCD model was detecting effects from SC and not a response to

warmer temperatures, the example criterion data set was constrained to samples with pH

B-10

Pr op

or tio

n of

e xt

irp at

ed g

en er

a

Figure B-7. Genus extirpation concentration (XC95) distributions for example criterion data set and temperature constrained data sets. Samples with pH ≤6 and (SO4 + HCO3)/Cl ≤1 were removed from all data sets. The example criterion (unconstrained, 0−32°C) data set has XC95 values for 139 genera (open black diamonds). The ≤17°C constrained data set has 95 genera (closed green diamonds [N = 658]). The ≥17°C constrained data set has 116 genera (open red triangles [n = 1,416]). For comparison the 5th centile is 338 μS/cm in the unconstrained data set, 425 μS/cm in the ≤17°C constrained data, and 315 μS/cm in the ≥17°C constrained data.

B.2.4. Potential Influence of Sampling Date on the Hazardous Concentration (HC05) To assess effects of date of sampling on the XCD model, three lines of evidence were

analyzed to address potential confounding by lack of seasonal observation of apparently

salt-intolerant genera. First, the HC05 using the spring (March to June) only data set was

compared with the full-year data set. (Seasons are defined by the phenology of the aquatic

insects and the changes in SC, not the conventional intervals.) The confidence bounds of the

B-11

spring HC05 overlap with the confidence bounds of the full-year data set (see Figure B-8). The

summer (July to October) only XCD lacks taxa known to be intolerant to ionic stress which can

be seen by the overall shift of the XCD to the right. The shift in the upper portion of the XCD in

spring is mostly due to the narrower SC sample range during the spring compared to the all year

data set.

Next, a scatter plot and regression model was developed for the relationship between

measurements of SC at the time of the biological sample and annual mean SC (see Figure B-9).

The annual geometric mean SC values were calculated from at least six water samples collected

before biological samples were taken. At least one spring and one summer sample were required

in order to be included in the data set. There were 325 sites with paired SC and biological data

(see Figure B-9) meeting these additional data requirements. On the x-axis is the SC value when

biological samples were collected and on the y-axis is the annual geometric mean value during

that rotating year for a site. A Model II Regression was fitted for this data set which takes into

account for error variance in both variables. The mean relationship between measurements of

SC at the time of the biological sample and annual mean SC is nearly 1:1. For example, when

SC is 340 μS/cm on the biological sampling date, the regression prediction for an annual mean

SC for the same site is 304 μS/cm.

B-12

Pr op

or tio

n of

e xt

irp at

ed g

en er

a

Figure B-8. Comparison of genus extirpation concentration distr