the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Detecting most effective cleanup locations using network theory to reduce marine plastic debris: A case study in the Galapagos Marine Reserve
- 1Institute for Marine and Atmospheric Research Utrecht, Department of Physics, Utrecht University, Utrecht 3584 CS, Netherlands
- 2Ocean and Earth Science, University of Southampton, National Oceanography Centre, Southampton SO14 3ZH, UK
- 3Galapagos Conservation Trust, 7-14 Great Dover Street, London, SE1 4YR, UK
- 1Institute for Marine and Atmospheric Research Utrecht, Department of Physics, Utrecht University, Utrecht 3584 CS, Netherlands
- 2Ocean and Earth Science, University of Southampton, National Oceanography Centre, Southampton SO14 3ZH, UK
- 3Galapagos Conservation Trust, 7-14 Great Dover Street, London, SE1 4YR, UK
Abstract. The Galapagos Marine Reserve was established in 1986 to ensure protection of the islands' unique biodiversity. Unfortunately, the islands are polluted by marine plastic debris and the island authorities face the challenge to effectively remove plastic from its shorelines due to limited resources. To optimise efforts, we have identified the most effective cleanup locations on the Galapagos Islands using network theory. A network is constructed from a Lagrangian simulation describing the flow of macroplastic between the various islands within the Galapagos Marine Reserve, where the nodes represent locations along the coastline and the edges the likelihood for plastic to travel from one location and beach at another. We have found four network centralities that provide the best coastline ranking to optimise the cleanup effort based on various impact metrics. In particular locations with a high retention rate are favourable for cleanup. The results indicate that using the most effective centrality for finding cleanup locations is a good strategy for heavily polluted regions if the distribution of marine plastic debris on the coastlines is unknown and limited cleanup resources are available.
Stefanie Leonore Ypma et al.
Status: open (until 11 Aug 2022)
-
CC1: 'Comment on egusphere-2022-426', Noam Vogt-Vincent, 16 Jun 2022
reply
Firstly, I think this is a great manuscript and I enjoyed reading it – a novel idea with potentially very useful results for those involved in marine debris management efforts on Galapagos (and other remote islands if this methodology were implemented elsewhere). The effectiveness of the clean-up strategies proposed in this manuscript will depend on the veracity of the assumptions a significant proportion of debris undergoes resuspension, but this is clearly stated in the conclusion. So overall, I think this manuscript will be valuable for island managers, and researchers working in this field. I do have some technical questions/requests for clarification though:
- I wonder whether you carried out a sensitivity analysis to test whether you released sufficient particles? With such high resolution hydrodynamical data forcing your particle-tracking, significant dispersion can occur over a 60-day integration time with an original separation of <4km (e.g. the off-diagonal cells in the transition matrix are ‘noisy’, and this is probably why). This in itself is not an issue since practitioners will probably not be using your transition matrix, but it would be nice to know how robust, say, Figure 7 is to particle number. I’m guessing that you were limited to 700k particles due to storage (since you were saving particle positions with a very high output frequency) but, if it is tractable, a quick sensitivity test might provide some assurance.
- I’m struggling to completely follow section 2.5. If I understand correctly, the equation on line 162 models the distribution of debris on coastlines in the limit of 100% resuspension. You simulate clean-up as removing the outgoing nodes of a clean-up target cell. You then say that “particles will accumulate at the target cleanup node”, but how can particles accumulate if you’re constantly removing them?
- It’s also not clear to me why you based your definition of steady state on the number of particles in the ocean – does the vector vtnot reach steady state?
- I’m also finding the “benefit” metric quite difficult to follow. “Zero connectivity between different nodes”, to me, implies that 100% of resuspended debris enters the ocean, but I don’t think this is what you meant?
- I’m not 100% convinced by your sensitivity tests to the initial macroplastic distribution (3.3). You’ve tested how robust your method is to uncertainty in the initial distribution by using a completely random initial distribution, i.e. assuming that the mass of plastic on beaches is completely decorrelated across length scales L > 4km. Is this realistic? Your result that the efficacy of node rankings remains the same with a random initial distribution is not surprising to me, since mesoscale ocean structures are much larger than 4km so will on average still see a ‘uniform’ distribution of resuspended debris. But given that van Sebille et al. (2019) showed that most debris incident to Galapagos arrives from the East, and that there could be large-scale effects from wind shadows, wake eddies, etc., I’d have expected that there would be some large-scale structure in the distribution of debris. I wonder if a more realistic way to model an uncertain initial macroplastic distribution might be by generating perlin noise with a wavelength larger than 4km (e.g. maybe the length scale of an island).
- This is not a criticism, but I wonder to what extent using a time-mean transition matrix affects your results. For instance, if we have sites A, B and C each releasing 1 unit of debris per time-step with probability P(A→B)=1/6 and P(B→C)=1/6, the probability of transition P(A→C) is 1/36 (over two timesteps). After 12 time-steps, 1/3 units would have been transported from A→C. But if these transition probabilities were time-varying and turned out to be P(A→B)= P(B→C)=1 during time-steps 1-2 and 0 otherwise, 1 unit would have been transported A→C – 3x more than the time-mean case, even though P(A→B)= P(B→C)=1/6 in both cases when averaged over 12 time-steps. This is obviously an artificially bad case and I completely accept that this goes beyond the scope of your study, but I didn’t see a mention of this in your discussion of limitations so I was wondering whether you thought this was a limitation (or if there is evidence that this time variability in the transition matrix is not important).
-
RC1: 'Comment on egusphere-2022-426', Anonymous Referee #1, 01 Jul 2022
reply
Review for : Detecting most effective cleanup locations using network theory to reduce marine plastic debris: A case study in the Galapagos Marine Reserve
The present work proposes a new methodology for improving beach clean up, to reduce marine plastic debris. The case study is dedicated to the Galapagos islands and the methodology is using a network to select the optimum criteria for the beach clean up purposes.
The manuscript is of very good quality, well organized with no specific structure problem or methodological problem. Although quite dense for the part explaining the criteria, it is possible for non specialist to understand the goal of the selected criteria.
In summary, The overall quality of this work is up to the expected standards and to my point of view can be published as it is ( after some typo and minor possible mistakes). I emphasize the fact that such decision (to my experience of reviewer) is very rare. I therefore thank the authors for having so well prepared their work before submission.
However, in order to improve the manuscript, or give ideas for new works, I wish to share some questions or comments :
- a very basic but tricky one to start with : could the authors imagine what the results could be with an irregular grid with a strong refinement close to the coast, or with a high resolution regular grid of , let say, 300 m in resolution on the horizontal ? I reckon that this point could be discussed using some literature. My main concern is that , the present study does not prove that the results are not strongly dependent on the OGCM resolution used. If it were, the overall methodology might provide different scores in terms of criteria ?
- the beaching probability: despite the fact that this has been already published , I am still very skeptical about the tuning of this probability. I am pretty sure that replacing this rather “adamant” formula by a probability calculated on the average wind direction versus the coastline layout in the area would be more realistic...Of course, I understand that there has to be a beaching time-scale of some sort to compensate the lack of grid resolution, but just days, regardless of general weather conditions in the area is strange to me. This is of course just my point of view, but this is what we can see when deploying drifters in general. Of course, this is complicated as the results can be different when looking at different scales, from large to very local scale (at the scale of a beach for example), but for a statistical approach, using regional winds might be ok.
- A remark concerning the overall network method: according to some literature, connectivity and oceanic distance can be calculated out of the Lagrangian model “quite” simply, I mean without using a complex machinery. Therefore, it could have been interesting to show some comparisons with some basic diagnostics (overall stranding rate / source identification / oceanographic distance) , to better prove the real added value of this complex work presented here. In other words, one might think that rather simple diagnostics could deliver nearly the same message or present similar general results , with a computational cost or calculation cost quite reduced. I understand that the authors might reply that their diagnostics are way refined and accurate, which is probably true, but the fact remains that, in the end, when it comes to mobilizing a cleaning team to go on site, maybe simple diagnostics can already deliver the necessary information for the management? (ok I am teasing a bit here, but never underestimate the robustness of simplicity ..!)
- Concerning the seeding of the particles, the random seeding is one interesting case , but when targeting real cases, I am quite surprised that the authors have not tried to perform test cases for which the seeding was increased after heavy rains. If so, they would have put themselves in the position of delivering a real connectivity between islands with identified real sources, i.e. the beaches receiving inland waste through rivers or waste management pipes just after the rain events ?
- In addition, I think that, as the final goal is to deliver advice for where and when to go for beach cleaning, one missing part can be the part looking at the correlation between the weather conditions and the stranding or accumulation close to the shore. This is a temporal advice that can be more efficient than the cleaning frequency advice that can be deduced from the present work. Indeed, as mentioned in their introduction, waste distribution is highly heterogeneous, and one cause is the high variability of regional to local weather conditions. Therefore, one could think that the general cleaning plan could be quite different if the weather conditions variations were selected as one of the key parameter.
Stefanie Leonore Ypma et al.
Data sets
Data and code for analysis and figures Stefanie L. Ypma, Q. Bohte, E. van Sebille https://github.com/OceanParcels/Galapagos_connectivity
Stefanie Leonore Ypma et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 146 | 38 | 4 | 188 | 1 | 2 |
- HTML: 146
- PDF: 38
- XML: 4
- Total: 188
- BibTeX: 1
- EndNote: 2
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1