## Changes to the assessment methodolgy

Changes made since the 2013 Assessment are described below:

### 2017 Assessment

#### Assessment Concentrations for organo-bromines

Organo-bromine concentrations were assessed for status for the first time. Canadian Federal Environmental Quality Guidelines (FEQGs) were used as EAC equivalents for biota and sediment. Background Assessment Concentrations (BACs) were trialled for BDE47 in biota and sediment.

### 2016 Assessment

#### New biological effects

Stress on stress, comet assay, micronucleus assay, neutral red retention time and lysosomal labilisation period were assessed for the first time. However, as there was no time series with more than two years of data (for any of these effects), no models were fitted and only an ad-hoc assessment of status was possible.

#### Regional assessments

The individual time series assessments were synthesised in a meta analysis to make regional assessments of metals, PAHs, chlorobiphenyls, organo-bromines, organo-metals (sediment only) and imposex. The regional assessments can be found in the links under ‘More information’ to the right of the maps.

### 2015 Assessment

#### Modelling of contaminants and biological effects

There were major changes in the way contaminant and biological effects time series were assessed. These included

• modelling the original data, rather than annual indices derived from the data
• using a linear mixed model that estimated the variance components in the data, rather than a loess smoother applied to the annual indices
• correctly incorporating the analytical variation in the data (supplied as uncertainties), rather than using an ad-hoc ‘scaled weight’ to measure analytical quality
• adapting the likelihood so that less-than measurements are treated as left-censored observations

The changes were so wide-ranging that, to understand them properly, it is probably best to compare the current help files with the 2014 help files:

#### Modelling of imposex (VDS)

There were also major changes in the assessment of imposex time series when submitted as individual VDS measurements. These included

• modelling the individual measurements, rather than annual indices, using a proportional odds model
• considering smooth changes in VDS levels over time
• considering change-point models in which VDS levels suddenly begin to change; the change-point is constrained to a year in the period 2004-2008, when the ban on the use of TBT was being implemented

Again, to undestand the changes properly, it is best to compare the current help files with the 2014 help file: assessment of imposex.

### 2014 Assessment

#### Type and width of smoothing neighbourhood

Loess smoothers are used to model smooth changes in contaminant concentrations (for both biota and sediment) and biological effects measurements (apart from imposex) when there are 7+ years of data. The amount of smoothing is determined by the type and width of the neighbourhood of contaminant indices that is used to estimate each $$f(t)$$ as $$t$$ runs from 1 to $$T$$. Previously, a fixed-width neighbourhood (Fryer & Nicholson, 1999) was used with, for example, a width of 9 meaning that only the indices in the 9 years closest to $$t$$ were used to estimate $$f(t)$$. This worked well if there was annual monitoring, but was less effective when monitoring was less frequent since some parts of the fit were sometimes based on only a few indices. This has been replaced by a neighbourhood in which a fixed number of indices are used to estimate each $$f(t)$$. For example, a neighbourhood of 9 now uses the 9 indices that are closest to $$t$$ to estimate $$f(t)$$. The fit in year $$t$$ can now be influenced by indices from years relatively distant to $$t$$, but the fit is always based on the same number of indices. This type of neighbourhood was used in the original development of loess smoothers (Cleveland, 1979).

A greater range of neighbourhood widths are also now considered. Previously, widths of 7, 9, and 11 years were considered, with the final choice being the width giving the smallest Akaike’s Information Criterion corrected for small sample size (AICc). Now, widths of 7, 9, 11 up to $$T$$ (if $$T$$ is odd) or $$T+1$$ (if $$T$$ is even) are considered, with the final choice again based on AICc. However, if there is no evidence of nonlinearity in the data (i.e. if the AICc of the linear model is lower than that of the best smoother) then the linear model
$$f(t) = \mu + \beta t$$ is used instead.

Cleveland WS, 1979. Robust locally-weighted regression and smoothing scatterplots. Journal of the American Statistical Association 74: 829-836.

Fryer RJ & Nicholson MD, 1999. Using smoothers for comprehensive assessments of contaminant time series in marine biota. ICES Journal of Marine Science 56: 779-790.

#### Determinands

Persistent organic pollutants were introduced as a group of contaminants for biota.

Scope for growth and glutathionine transferase were introduced as biological effects for biota. For both, high values indicate healthy organisms. Glutathionine transferase is assessed in exactly the same way as chemical contaminants in biota, except that the lower confidence limit on the fitted value in the last monitoring year is used to assess status. For scope for growth, the annual indices are the median values of the scope for growth measurements in each year (there is no log transformation). This is because scope for growth can be negative (with negative values indicating bad status). The annual indices are then modelled in the same way as chemical contaminant indices, except that the lower confidence limit is used to assess status.

#### Assessment criteria

The ERLs for C1-naphthalene, C2-naphthalene, C1-phenanthrene, C2-phenanthrene and C1-dibenzothiophene were not used as it was not possible to find sufficient justification for them in the literature.

#### Assessing status of imposex

Previously, environmental status of imposex levels was assessed using the model fitted to the annual indices. The upper one-sided 95% confidence limit on the fitted value in the most recent monitoring year was compared to the available assessment criteria. However, in many time series, imposex levels have declined so rapidly that the linear models used to assess trends cannot track the change completely. The linear models correctly show evidence of a decline, but over-estimate imposex levels in the final monitoring year suggesting that environmental status is worse than it actually is. To overcome this, an alternative test of status is now used when there are individual measurements in the final monitoring year. A proportional odds model is fitted to the individual measurements and used to place an upper one-sided 95% confidence limit on the annual index in the final monitoring year. This confidence limit is then compared to the available assessment criteria.