The MArine Debris hyperspectral reference Library collection (MADLib)
Abstract. Marine debris is a ubiquitous and growing threat to environmental and human health. Efforts to monitor and mitigate marine debris pollution face many challenges. A primary limitation is the absence of standardized methodologies for monitoring capabilities due to the complex and diverse physical and chemical properties of marine debris. Variabilities include object size, apparent color, polymer type, weathering, and aqueous state. Despite the challenges in object characteristics, advances in remote sensing methods are showing promise for detecting marine debris across local to global scales. Algorithms are needed to link remotely sensed observations with relevant characteristics of marine debris to fully realize this potential. Although more optical measurements of marine debris reflectance are becoming available for algorithm development, inconsistencies in data curation remains an obstacle. Variations in data processing and inconsistent metadata hinder efforts to develop robust, generalizable algorithms for marine debris detection. To address this, we present the well-curated MArine Debris hyperspectral reference Library collection (MADLib) containing 24889 spectra from 3032 samples. All optical measurements are available in open access via https://doi.org/10.4121/059551d3-2383-4e20-af2d-011c9a59d3ac (Ohall et al., 2025). MADLib demonstrates the importance of open-science and open-access datasets, as it compiles and harmonizes spectral data collected from publicly accessible datasets and individual research projects. Consistent methods were applied for data standardization, quality assurance, and integration. We also propose a robust protocol for generating metadata tailored to marine debris and ocean color remote sensing applications. MADLib possesses spectra of a wide range of marine debris materials including different polymer types, color, size, weathering, and aqueous states. Here, we analyze the metadata associated with the spectra to identify sampling gaps and propose considerations for future work. By providing open-access and standardized data, MADLib is expected to support the development of robust marine debris detection algorithms.
General comments:
The manuscript addresses an important topic, compiling a comprehensive dataset of mostly plastic litter reflectance spectra into a spectral library for use in plastic litter detection algorithm development and assessment. The structure is generally good, the writing is clear, and the figures are helpful in presenting results. However, some aspects require clarification and minor additions to improve transparency and interpretation of results, specifically with regard to spectral analysis and interpretation.
- A more thorough comparison between datasets for the same samples would be useful. When significant variation is present, this should be examined and reported. See comment on figure 5 below.
- Comparison of findings from this study to other studies assessing the same effects of weathering and submersion on reflectance can be improved.
- Most current sensors lack the capacity of distinguishing narrow SWIR features. The authors state that SWIR should be used for algorithm development. How would that be put in practice? Are the authors referring specifically to VHR hyperspectral platforms?
- Additionally, most polymers exhibit similar absorption features, which are characteristic of plastics in general, which we can also see from figure 5 in this study. How useful and/or practical is it to be looking at/for specific polymer types? Would a mean spectrum for dry, wet, submerged, biofouled, weathered etc. spectrum for plastics in general be useful for algorithm development?
Specific comments:
- Line 142: Shouldn't atmospheric absorption only matter when dealing with dense layers of the atmosphere? Since measurements are taken with instrument at less than a meter from sample, instrument noise can be present but atmospheric effects should be negligible. As a matter of fact, this is the reason we usually don't perform any atmospheric correction when dealing with UAV data. And this is where these spectra areas would come in handy.
- 2.4.4 weathering: Besides the type of weathering, the degree of weathering can also affect reflectance properties significantly. Perhaps authors should consider adding a degree of weathering metadata field, or state that all samples were similarly weathered if that is the case.
- 2.4.5 aqueous state: I can see from the metadata file that the depth of submersion is reported. This should be reflected in the text.
- Lines 216-218: Does pixel coverage affect reflectance spectra magnitude? Have you taken this into account when presenting spectra?
- Lines 218-219: Same with the glass, have you examined how this could affect measurements and when presenting spectra?
- Figure 5: mean reflectance spectra of HDPE (5c) and PP (5d) samples present some pronounced variation between datasets, specifically in the visible range. Is this due to different colours? If so this should be reflected in the text. I can see that the authors touch on this in the discussion section. However as the differences are quite pronounced, I believe this warrants further examination or explanation and should be given more weight in the results section as well. If this is not due to colour, can authors hypothesize as to what it might be? Could it mean there are measurement validity issues?
- Figure 5: Normalization methodology could be better defined.
- Figure 10: There is a significant variation in reflectance magnitude of floating PP samples between dataset 1 (figure 10b) and dataset 5 (figure 10d) in the visible part of the spectrum. Have authors examined why this is the case?
- Lines 312-313: Aqueous state affects reflectance magnitude across the full range of spectrum, both in the NIR/SWIR and visible (although less pronounced). If authors here mean the shape of the spectrum this should be clearly stated.
Technical corrections:
- Data availability is missing section numbering
- Figure 1 resolution should be improved
- Table 2 formatting for line stroke
- Conclusions is missing section numbering