Evaluating the potential of underwater television to contribute to marine litter assessments alongside bottom trawling

Abstract

Marine litter presents a global threat to marine ecosystems, human health, and safety. Therefore, it is important to increase our knowledge about spatiotemporal trends of litter in the environment. Bottom trawl surveys provide a practical method for monitoring seafloor litter on the continental shelf, but can have severe negative impacts on the environment. Here we evaluate the potential of an ongoing underwater television survey (UWTV) to also collect litter density data, and develop model-based indices of litter densities integrating coastal and offshore trawl survey data using geostatistical models. Based on our case study along the Swedish west coast, we find that UWTV in its current format may be limited as an alternative to trawling in areas with relatively low densities. There are also clear spatial trends in litter, with the highest densities in near-shores areas currently only included in the national monitoring program. Our results illustrate the potential of combining data, but also the importance of careful sampling designing for marine litter monitoring.

Fig 1. Sampling locations over time.

The Coastal Trawl Survey (CTS) is depicted in green, the International Bottom Trawl Survey (IBTS) in orange, and the Underwater TV survey (UWTV) in purple. The IBTS is conducted in Kattegat, Skagerrak and parts of the North Sea but in this study only stations within the area covered by the UWTV survey in 2024 are included (S1 Fig). The dotted line in the topleft panel depicts the Skagerrak/Kattegat border.

Fig 2. Image from filmed transect.

Image of the seafloor with a litter object taken from a transect filmed with an UWTV in 2023 in ICES subarea 4 (Kattegat). The distance between red laser dots is approximately 80 cm. Due to turbidity, it is difficult to say if the object is A2 = plastic sheet or A3 = plastic bag according to the ICES manual [24]. Foto SLU-Aqua, P. Jonsson.

Fig 3. Example of a single replicate of a randomly filled spatial grid.

Litter densities are 100, 500, 1000, 5000 items per ${\text{km}}^2$. Black grid cells indicate presence of litter. The pink line corresponds to a randomly placed straight UWTV transect. For visualization purposes, we have used relatively high litter densities, zoomed in on a 50$\times$50 m portion of the full grid, and divided the transect by 4 (hence, in the simulation experiment, the UWTV transect would be 4 times as long).

Fig 4. Results from the simulation experiment.

Each panel (A–F) corresponds to a litter density scenario, and each blue point represents the estimated mean density for that sample size (number of hauls) (x-axis) and iteration. To avoid overplotting, we randomly sampled 30 of the 1000 blue points and added a small horizontal and vertical jitter. The pink circles correspond to the mean litter density across all 1000 replicates. The horizontal pink line depicts the true litter density in the simulation (also indicated in the panel title). The number on the top corresponds to the proportion of the 1000 simulations that did not catch a single litter item in that sample size scenario.

Fig 5. Spatiotemporal random effects for the binomial model (top row) and the Gamma model (bottom row) for selected years (2015–2018).

Fig 6. Predicted litter densities from the spatiotemporal model for the years 2012–2024.

To better visualize the spatial patterns, values greater than the 99% quantile (479 items per ${\text{km}}_2$) are set to the highest color.

Fig 7. Conditional predictions of litter density for the IBTS level (points) and 95% confidence interval (vertical lines) without random effects.

The purple line illustrates trends in annual estimates of mean litter densities and is the prediction from a GAM year modelled as a penalized spline and the inverse of the CV for annual predictions as weights.

Fig 8. Effect of sample size on upper confidence intervals of probability of detection.

Illustration of how the upper confidence interval (95% in green and 99% in orange) for the probability of litter being present in a given haul (A) and the estimated density that corresponds to (B) change as a function of sample size if no hauls record any litter, using the “rule of three” and a sampled area of $148{\text{m}}^2$. The lower confidence interval is always zero.