16.05.2015 Views

1.1 MB pdf - Bolsa Chica Lowlands Restoration Project

1.1 MB pdf - Bolsa Chica Lowlands Restoration Project

1.1 MB pdf - Bolsa Chica Lowlands Restoration Project

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

SECTION 3: ANALYSIS<br />

Amphipod (Eohaustorius estuarius) Regression Analysis<br />

Results of the amphipod toxicity bioassay consisted of counts of individuals surviving and<br />

counts of individuals exhibiting reburial behavior. Regression analyses were not performed<br />

for reburial behavior because few significant effects were observed; in more than 50 samples<br />

for which bioassays were performed, significantly decreased reburial was observed in only<br />

3 (see Appendix F).<br />

Sample sizes within each replicate were constant among replicates (i.e., N=20). Therefore,<br />

regression analyses were performed with the count data as the dependent variable. The<br />

independent variables consisted of concentrations of chemicals in sediment (as mg/Kg or<br />

µg/Kg). (Note: TOC-normalized concentrations of organics were evaluated in initial data<br />

screens. As these data did not improve the predictive quality of the regression analyses,<br />

they were not considered further). Because environmental chemistry data are frequently<br />

log-normally distributed (Burmaster and Hull, 1997), analyses were also performed with<br />

natural-log-transformed concentration data as the independent variable. Simple linear<br />

regression analyses were performed using SAS (1994) PROC REG. All models were<br />

considered significant if p#0.05.<br />

Of 75 compounds detected in sediment samples used for amphipod toxicity bioassays based<br />

on all test media adjustment groups combined, significant linear relationships between<br />

untransformed concentrations and survival were observed for 39 (Table 3-15). This<br />

relationship (e.g., slope) was negative for 38 of the 39 compounds. The amount of variation<br />

(r 2 ) explained by these models ranged from 3.7 percent (e.g., r 2 =0.037; lead) to 99 percent<br />

(e.g., r 2 =0.99; total phenol and PCB-028). Natural-log-transformation of the concentrations<br />

resulted in significant fits for 43 of 75 compounds; 39 of them were negative (Table 3-15).<br />

Among these models, r 2 also ranged from 0.033 (e.g., 4,4’-DDD) to 0.99 (e.g., total phenol<br />

and PCB-028). Comparison of the results obtained based on transformed and untransformed<br />

data indicates that, in general, better model fits (e.g., higher r 2 values) were obtained from<br />

the transformed concentration data; untransformed data produced the best fit for<br />

18 analytes whereas 24 analytes were fit best by transformed data (Table 3-15). If total<br />

sample size among all four test media adjustment groups exceeded 100, further regression<br />

analyses were performed for each group for each analyte. A total of 24 analytes met this<br />

criterion (Table 3-19). Although quality of model fits differed by test media adjustment<br />

group for each analyte , a significant fit was obtained for at least one group within each<br />

analyte. In addition, although a significant model fit had been obtained for all analytes with<br />

n>100 when test media adjustment groups where pooled (Table 3-15), better model fits<br />

(e.g., higher r 2 values) were obtained for at least one test media adjustment group within<br />

each analyte, except for chromium (Table 3-16).<br />

F-tests (Draper and Smith 1981) were performed to compare regression results among the<br />

four test media adjustment groups and among wet/dry sediment or the presence/absence<br />

of salinity adjustment. Differences were considered significant if p#0.05. A summary of the<br />

results is presented in Table 3-17. Significant differences between regression models among<br />

all four test media adjustment groups were observed for 47 of the 75 analytes detected in<br />

sediments used for amphipod bioassays (Table 3-17). In addition, regression models for wet<br />

vs. dry sediment differed significantly for 29 analytes; and 35 analytes differed significantly<br />

by presence/absence of salinity adjustment.<br />

SAC/143368(003.DOC) 3-33 ERA REPORT<br />

7/31/02

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!