Sea-floor litter

Description of data

Available sea-floor litter data were downloaded from the International Council for the Exploration of the Sea (ICES) Database of Trawl Surveys (DATRAS) as ‘Litter Exchange Data’ products on 15 April 2020. The downloaded data comprised the North Sea International Bottom Trawl Survey (code ‘NS-IBTS’), the Scottish West Coast Groundfish Survey, (‘SCOWCGFS’) and the Scottish Rockall Survey (‘SCOROC’).

The data comprise counts of observed litter items recovered from the trawling gear of ICES coordinated international scientific survey trawls assessing commercial demersal fish stocks. Potentially relevant data from the Irish Groundfish Survey (‘IE-IGFS’) and beam trawl survey (‘BTS’) were not downloaded, the former because they do not include counts of litter items, and the latter because they are obtained using a different trawling gear. Litter items are additionally categorised into types and linked to the date of trawl, trawl shoot and haul coordinates, distance of trawl, wing-spread of gear, and sampling nation.

Trawls east of -4° longitude, hereafter referred to as ‘east-coast trawls’ and available using survey code ‘NS-IBTS’, are from spatially stratified grids of 1° longitude and 0.5° latitude. Several nations, including Scotland, contribute to this survey and participants annually sample one of several possible localities within pre-allocated grids.

Trawls west of -4° longitude, hereafter referred to as ‘west-coast trawls’ and available using survey codes ‘SCOWCGFS’ and ‘SCOROC’, are from semi-randomised locations within a series of a priori strata. These surveys are undertaken by Scotland. The spatial coverage of the west-coast trawls is not as complete as the east-coast trawls. Further details of the survey designs are available (ICES, 2012, 2017).

Trawls undertaken by MRV Scotia, which nominally occur at depths of between 20 and 500 m, are 30 minutes long and utilise Chalut de Fond Grande Overture Verticale (GOV) 36/47m gear, including an eight metre long cod end lined with 20 mm mesh. Different ground-gear are used for the west-coast (ground-gear ‘D’) and east-coast trawls (ground-gears ‘A’ and ‘B’). Further details of the gear are available (ICES, 2012, 2017). The area swept by each trawl, calculated from the distance of a trawl and wingspread, averaged 64 m² and included a minimum of 22 m² and a maximum of 85 m².

The litter from each trawl was processed in accordance with the ICES recommendations current at the time. Currently, this involves visually checking the fish catch and gear for items of litter which are then placed into baskets. The litter items are then categorised and counted into litter categories as a subtype of one of ‘plastic’, ‘metal’, ‘rubber’, ‘glass/ceramics’, ‘natural products/clothes’, or ‘miscellaneous’ (ICES, 2018).

The plastic category includes the subtypes of ‘bottle’, ‘sheet’, ‘bag’, ‘caps/lids’, ‘fishing line (monofilament)’, ‘fishing line (entangled)’, ‘synthetic rope’, ‘fishing net’, ‘cable ties’, ‘strapping bands’, ‘crates and containers’, ‘plastic diapers’, ‘sanitary towels/ tampon’, and ‘other’ (ICES, 2018). While several pieces of litter are counted as a single item if they are ‘known’ to have originated from a single object (e.g. a broken glass bottle for which the parts fit together), they are counted as separate items even if they are ‘suspected’ of having derived from a single item. An exception to this are entangled litter which are counted as a single item and assigned to the dominant litter subtype (Figure b). Counts include individual pieces of ‘twine’ used to mend the gear and displaced during the trawl.

The absence of litter in a trawl is explicitly noted and, therefore, all trawls regarded as ‘valid’ provide information on the presence, counts and categories of litter. An effort is made to ensure that procedures remain consistent over time and between personnel. On return to base (Marine Scotland Marine Laboratory, Aberdeen) the litter data are quality checked and submitted to ICES for inclusion into DATRAS at the individual litter item level.

The trawl and counting procedures are specified by ICES Marine Litter Working Group guidelines (ICES, 2018). Differences in procedure between nations arising from different implementations of the ICES guidelines occur.

Figure b: Entangled litter (excluding the underlying hose) counted as a single item of ‘synthetic rope’.

 

Litter counts are obtained from scientific surveys with the primary purpose of fish stock assessment, rather than being specifically designed to estimate sea-floor litter densities. Trawls are only undertaken at localities where demersal fishing is practical, omitting deep seas, coastlines and rocky areas. The gear is designed to sample demersal fish and will underestimate, at the very least, litter items less than the gear mesh size as these will likely pass through the trawl net. Additionally, the different ground-gears will have different litter capturing capabilities.

Data preparation

The downloaded data were assembled and prepared for analysis within the R statistical environment version 3.5.2 (R Core Team, 2018), accessing geographic information system capabilities with the supplementary R package rgdal version 1.3-9.

Data for 2012 to 2018 inclusive within a modelling area comprising a polygon of minima and maxima longitudes -16° to 7.4° and latitudes 54° and 64° were retained for analysis (Figure c). Data for 2019 and 2020 were excluded because they were not yet sufficiently complete. The modelling area incorporates all MRV Scotia east-coast and west-coast trawl localities, and includes the Scottish Zone which constitutes the primary focus of the analysis (Figure c). The Scottish Zone is defined as the sea adjacent to Scotland (Great Britain, 1998) within an area defined by The Fishery Limits Act 1976 (Great Britain, 1976), as amended by relevant Fishery Limits Orders (Great Britain, 1997, 1999).  Data from the east-coast trawls of other nations within the modelling area were also retained.

Further data edits prior to analysis comprised:

• deleting trawls flagged within ICES DATRAS as ‘invalid’;
• deleting scientific surveys, comprising nation and year quarter combinations, failing to record trawls containing no litter;
• the trawls of some non-UK nations for some years utilise a previous litter categorisation scheme. An algorithm identifying these and converting them to the current scheme (ICES, 2018) was used to improve the consistency of litter categories;
• litter counts from different litter categories were aggregated to the relevant trawl;
• the area swept by each trawl was calculated as the product of the distance of each trawl and the wingspread of the GOV gear;
• the midpoint of each trawl was estimated from shoot and haul coordinates and used to characterise the location of the trawl;
• trawl locations were converted from coordinates to isotropic grid references.

The edited data comprises 2,484 trawls within the modelling area of which 1,635 are within the Scottish Zone (Figure c). Within the modelling area, 79% of trawls were undertaken by MRV Scotia. Additional trawls, also using GOV gear, were undertaken by England (13%, 2012 to 2014 and 2016 to 2018), Denmark (5%, 2014 to 2018), France (1%, 2015 and 2017 to 2018), The Netherlands (1%, 2015 and 2016) and Sweden (<1%, 2018).

Figure c: Isotropic statistical model set-up.

The modelling area is denoted by a thick continuous boundary line, the Scottish Zone by a thin continuous boundary line, the midpoint of trawls from 2012 to 2018 inclusive by small blue points, and ‘knots’ around which the smooths are based by larger red points.

Descriptive statistics

A table comprising a total of counts and percent relative frequency for each litter category within the Scottish Zone across the years 2012 to 2018 inclusive was prepared.

Inferential statistics

Statistical modelling was used because it generates estimates of sea-floor litter density throughout the Scottish Zone.

The data for nine valid trawls were deleted prior to statistical modelling. These trawls were for nation and year combinations contributing fewer than 10 trawls within the modelling area. The rationale for these deletions is that statistical estimates of nation effects, used to standardise for potential methodological differences in the propensity of nations to detect litter, would have been poor.

The spatial variation in litter densities was modelled using a generalized additive mixed-effect model utilising soap-film smoothers (Wood, Bravington & Hedley, 2008). This approach avoids smoothing litter densities across the land-mass of the north of Great Britain. The modelling area and ‘knots’ are presented (Figure c).

Statistical modelling involved fitting the dependent variables of litter item count for each trawl and the area swept by each trawl, generating a density.  Counts were modelled because they represent a richer source of information than presence/absence. Mass was not used because occasional large items composed of dense material dominate, generating less informative and less tractable analyses. Separate models were fitted for each year.

The full statistical model comprised:

where y_ij = count of items at location i trawled by nation j assuming error distribution Pois(y_ij); a_ij = area trawled at location i by nation j; β₀ = estimated fixed effect intercept; f = estimated spatial smoother functions f across locations x_i; β₁ = estimated fixed effect of nation x_j relative to MRV Scotia; and u_ij = random effect of unexplained variation at location i by nation j. Maximum likelihood estimates of β_0, β₁, f, integrated over random effect u_i using the Laplace approximation, were calculated within the R supplementary package gamm4 0.2-5 (Wood, 2017). Model outputs were used to generate annual predicted sea-floor litter densities (km^-2) within the Scottish Zone. Modelling precisions were estimated as the inverse of predicted density coefficients of variation, and are relative to the median precision across all years. Additionally, a policy relevant spatial summary plot, averaged over 2016 to 2018 inclusive to reduce the impact of annual variations, was generated. Averages are weighted mean densities, with weights comprising the inverse of the variance of annual densities. The summary plot is for 2016 to 2018 inclusive only because this is the period most relevant to present policy.

Annual mean litter densities and percentile 95% confidence intervals were estimated for the Scottish Zone and those parts of five biogeographic regions (United Kingdom Marine Monitoring and Assessment Strategy, 2010) within the Scottish Zone. These comprise a part of the Northern North Sea, the Atlantic North-West Approaches, the Scottish Continental Shelf, the Minches and Western Scotland, and a part of the Irish Sea. The 95% confidence intervals were estimated using a parametric bootstrap (Efron, 1979) utilising the lpmatrix option of the gamm4 predict function. Biogeographic regions were used in preference to Scottish Marine Regions (Scotland, 2015) because sampling locations are not optimal for coastal areas and it is uncertain to what extent the statistical smoothers are able to adjust to local differences.

Temporal trends from 2012 to 2018 inclusive for the Scottish Zone and each constituent biogeographic region were estimated using a weighted linear mixed models. The model comprised:

where y_i = the annual mean predicted density for the relevant region in year i assuming error distribution ~N(0,σ_i²). The value of σ_i is the standard deviation of the annual mean predicted density estimated from the previous analysis and is assumed to be known; β₀ = estimated fixed effect intercept; β₁ = estimated coefficient of linear trend over years x_i; u_i = estimated random effect of year; ε_i = unexplained variation of year i estimated from the previous bootstrap analysis. The model was fitted by residual maximum likelihood (Patterson & Thompson, 1971). Plots for the Scottish Zone and each biogeographic region were generated.

Values of p are interpreted as a level of support for ‘no-effect’ and are regarded as ‘statistically significant’, indicating a lack of support for the null hypothesis, if no greater than 0.050.