Along with global warming, marine plastic pollution has been described as one of the most pressing matters that our oceans will face in the coming decades. In this context, as part of the Decade for Ocean Science for sustainable Development (2021-2030), the Sustainable Development Goal 14.1 targets the prevention and significant reduction of all kinds of marine pollution by 2025, particularly from land based activities, including marine litter. However, accurate observations of the sources, composition and densities of floating marine litter in the world’s ocean are sparse and lacking. Remote sensing can play a significant role in the detection and monitoring of marine litter, and to this extent adequate in situ observations of calibration and validation data are essential.

Towards this goal, we have since 2018 launched a series of experimental field campaigns, the Plastic Litter Projects, that aim to enrich the scientific community’s understanding of floating marine litter’s (FML) spectral properties and behavior. By developing, constructing and deploying artificial floating targets containing various types of FML, we aim to produce a comprehensive remote sensing image database which can be used for the development, calibration and validation of FML detection algorithms. Throughout the years, we have used various types of marine litter items such as PET bottles and HDPE bags, as well as natural floating materials such as reeds, in the construction of artificial floating targets. Our efforts have been focusing mostly on the ESA-deployed, multispectral satellite Sentinel-2, mainly due to its frequent revisit intervals, relatively high spatial resolution and very importantly, the open access data provision.
In the past two years, we have shifted our focus away from the small-scale re-deployable artificial floating targets, and started moving towards semi-permanent target infrastructure. We have also focused on answering the scientific community’s call for replicability in terms of the experimental field campaigns, both in regards to the materials being used but also the experimental set-up. To that extent, in 2020 under the scope of the ESA funded OSIP project “Plastic Litter Project: Detection and monitoring of artificial plastic targets with satellite imagery and UAV”, we investigated the use of a single reference material that can act as a representative proxy to replace the use of different FML items in artificial floating targets. In addition, we developed and deployed two different types of prototype targets, in order to assess the possibility of long-term deployment.

Moving on to 2021 we constructed two large long-term deployment artificial floating targets that were deployed in the Gulf of Gera, in the Island of Lesvos in Greece. The first target consisted of a circular 28 m diameter HDPE pipe frame, on which we fastened a series of white HDPE mesh sheets, creating an effective target area of about 600 m2. The second target was made out of more than 350 3 m wooden planks, in order to represent natural floating marine debris, approximating the same effective target area. Deployment of such large surface area targets guaranties the acquisition of at least one Sentinel-2 10x10 m pixel that contains solely the target material. However, both the HDPE mesh and the wooden planks targets do not completely cover the water surface, hence part of the signal response is attributed to water leaving reflectance, resulting in a more realistic set-up.
The targets were deployed in the Gulf of Gera for a period of 4 months, from June to September 2021, resulting in a total of 22 cloud-free Sentinel-2 acquisitions. In addition to the Sentinel-2 data, we acquired very-high resolution UAV images for the production of orthophoto maps, hyperspectral UAV measurements in the range of 400 to 900 nm, in situ spectrometer measurements in the same spectral range, as well as ancillary metadata including wind speed, water turbidity using a Secchi disk and incident light intensity measurements. The experimental set-up and the large number of acquisitions allowed for the acquisition of data under different conditions and target configurations, providing the opportunity to examine the effect that a number of parameters, such as biofouling and degree of submersion have on the spectral response of the target materials. Combining the two targets for a number of Sentinel-2 overpasses allowed for a number of acquisitions with mixed FML and natural debris scenarios.

Biofouling accumulations can potentially affect the spectral signature of FML, both in the visible and infrared range. Deployment of the floating targets in productive waters such as those of the Gulf of Gera translates into the fact that the target materials are susceptible to bio-fouling. The long-term deployment of the large surface area HDPE target during the PLP2021 acquisition campaign, has allowed us to produce images of the target with and without biofouling accumulations, thus presenting the possibility to assess the effect of biofouling on FML reflectance. Preliminary analysis results show that biofouling mostly affects the intensity of the FML signal and specifically on the near-infrared range, without major effects to the signal shape.

Together with the above mentioned parameters, turbidity and wind velocity are also factors that can influence the water leaving reflectance and FML signal, affecting spectral classification methodologies. Turbidity is especially important for semi-enclosed basins with very productive and turbid waters, making it especially challenging in terms of spectral analysis. However, the somewhat varying turbidity of the Gulf of Gera waters presents the possibility to correlate detection accuracy with degree of turbidity. This can be especially useful in operational detection applications in river outflow situations in estuaries and river mouths, from where a high percentage of FML originally reaches the ocean.

Initial results suggest that spectral classification methodologies such as partial unmixing algorithms, can successfully be used for FML detection using the mean spectral signature acquired during the PLP2021 deployment period. Further analysis of the correlation between the above mentioned parameters and the resultant signal of FML can prove especially useful in operational scenarios of FML detection and monitoring.

Oceans receive solid waste from anthropogenic activities. A significant amount of the produced solid waste is made of plastics. The amount of plastic debris in the ocean and coastal areas is steadily increasing and is now a global major environmental problem. Accumulation of marine debris poses considerable threats to aquatic species, ecosystems and human beings too, as microplastics are eaten by fish and shellfish and consequently enter our food chain. At the global scale, the 2030 Agenda for Sustainable Development, adopted by the United Nations in 2015, calls for action to conserve and sustainably use the oceans, seas and marine resources with the Sustainable Development Goal (SDG) No. 14. Among the SDG 14 targets, the 14.1 calls to prevent and significantly reduce marine pollution of all kinds, in particular from land-based activities, including marine debris and nutrient pollution. From the European perspective, the Marine Strategy Framework Directive (MSFD) requires the EU Member States to ensure that "properties and quantities of marine litter do not cause harm to the coastal and marine environment".
For monitoring marine plastic litter, ground-based monitoring systems and/or field campaigns provide precise information on the quantity and quality of the marine litter. Nevertheless, ground-based monitoring campaigns for collecting data on marine litter are limited, time-consuming, expensive, require great organisational efforts, and provide very limited information in terms of spatial and temporal dynamics of marine debris.
Earth Observation by satellite has the potential to contribute significantly to marine plastic litter monitoring thanks to its global synoptic point of view. However, remote sensing of marine plastic litter is in its infancy, and it is a significant scientific and technological challenge.
In 2020, a consortium led by Planetek Italia (Italy), including the National Technical University of Athens (Greece) and the Environmental Prevention and Protection Agency of Puglia Region (Italy), participated in an ESA call for ideas, the Discovery Campaign on Remote Sensing of Plastic Marine Litter, with a novel idea for assessing the feasibility of marine plastic litter detection from space, which was selected.
The project was titled "Crowdsourcing, Copernicus and Hyperspectral Satellite Data for Marine Plastic Litter Detection, Quantification and Tracking" or REACT.
The REACT project focused on providing a proof-of-concept on remote sensing of marine plastic litter by developing the following methodology to detect plastic litter offshore and onshore. The methodology exploited data fusion of multispectral (MS) satellite data from Sentinel-2 and WorldView and hyperspectral (HS) satellite data from PRISMA, together with in situ data collection, and took advantage of two different approaches. The first was based on Spectral Signature Unmixing (SSU), and the second was based on Machine Learning (ML) methodologies.
The main objectives of the project were to:
• Assess how plastic litter can be detected and possibly quantified with current and future remote sensing technology;
• Develop adaptive indices insensible to biases induced by sunglint on satellite radiometric products and indicate the constraints of current satellite missions under various atmospheric and illumination conditions;
• Exploit data fusion methods between remote sensing data with high spectral (PRISMA hyperspectral, Sentinel-2) and high spatial resolution (PRISMA panchromatic, WorldView) to increase the sensor detectability of marine plastic litter;
• Explore SSU methodology for sub-pixel detection of floating marine plastic debris;
• Explore ML techniques for detecting plastic litter;
• Conduct controlled experiments under real conditions to better understand the effect of the atmosphere and the illumination conditions on the spectral properties of marine plastics in visible and infrared wavelengths.
The key target user of the REACT project was the Environmental Prevention and Protection Agency of Puglia Region, Italy (ARPA Puglia), which is in charge of detecting and monitoring marine plastic litter in the framework of the European legislation (i.e., MSFD).
The critical target users' needs can be summarised in reaching the capacity to:
• support field campaigns by environmental agencies to implement data collection plans by field operators;
• perform regular monitoring of marine litter over broad areas;
• provide spatial and temporal distribution of marine litter;
• forecast paths of floating litter;
• identify potential sources of plastic litter into the marine environment and forecasting possibly places of beached litter;
• achieve a cost-efficient, repeatable, and flexible methodology, from local to the national level.
During the project, some controlled experiments were performed to understand better the effect of the atmosphere and the illumination conditions on the spectral properties of marine plastics in visible and infrared wavelengths. Twelve floating plastic targets were constructed for the experiment. Their sizes were selected according to the spatial resolution of data expected to be achieved by image fusion. Targets were realised with four types/compositions of plastic materials with various colours: 1) low-density polyethylene (tarps in white, yellow and green colour), 2) polyethylene terephthalate (transparent water bottles, green oil bottles), 3) polystyrene (sheets for building insulation in cyan colour), and 4) all the above materials in equal surface extent. The 12 targets were placed offshore and onshore during satellite passages. In-situ measurements using a spectro-radiometer were also carried out during the controlled experiments. The controlled experiments were realised in Mytilini and Koplos Geras, on the Greek island of Lesvos, with the support of the remote sensing group of the University of Aegean.
The MS and HS satellite imagery collected during the controlled experiments were processed with image fusion techniques to obtain merged images with higher spatial resolution than the initial high spectral resolution images. On the fused images, SSU techniques and ML algorithms were used.
SSU contributes to the extraction of information at the sub-pixel level. Its main scope is to detect the distinct spectra in the fused MS and HS scenes, which can represent different materials, and to estimate their apparent quantification in a pixel in terms of a fraction. Endmembers correspond to different signals. In contrast, abundances refer to the fractions of these endmembers within a mixed pixel. The project achieved marine plastic litter detection by separating endmember spectra that best characterise plastic materials and waters. The abundance maps of these plastic material spectra led to plastic targets detection.
Plastic targets were also detected by supervised and unsupervised ML algorithms. The output was a probability map representing the probability that a pixel contains or not plastic.
The main findings of the project are summarised below:
• Plastic and water spectral signatures are significantly correlated in the original images and the fused HS data. The Principle Component Analysis (PCA) and the Gram-Schmidt Adaptive (GSA) algorithm gave better results in terms of spectral discrimination between water and plastics;
• Sunglint correction provided no beneficial in spectral discrimination between water and plastics;
• Image fusion based on matrix factorisation performed better than the deep learning methods. The Coupled Nonnegative Matrix Factorisation (CNMF) method produced better results when plastic targets were placed offshore. On the other hand, the Hyperspectral Superresolution (HySure) method presented slightly better results with targets placed onshore;
• SSU was able to detect plastic targets, although some manual tuning is necessary. Similar conclusions in the case of plastic spectral indexes were used;
• PS and LDPE were detectable, while PET was not;
• SSU on fused PRISMA data efficiently detected floating plastic accumulations up to 2.40x2.40m (about 1/2 of the panchromatic PRISMA band resolution);
• SSU on fused Sentinel-2 and WorldView data efficiently detected floating plastic accumulations of 0.60x0.60m (about 1/4 of WorldView resolution);
• ML algorithms provided promising results in plastic detection, although with a small dataset;
• Both SSU and ML require land and shallow waters masking;
• No significant results were highlighted with the targets onshore with both SSU and ML.
This work was supported in part by the Discovery Element of the European Space Agency's Basic Activities under ESA Contract 4000131235/20/NL/GLC (REACT Project: Crowdsourcing, Copernicus and Hyperspectral Satellite Data for Marine Plastic Litter Detection, Quantification and Tracking).

Over the past few years, spectral signature of marine litter has been studied both for virgin and weathered plastic samples and several spectral techniques based on indices have been developed to detect them. In this research, we present artificial intelligence (AI) results based on RGB, multispectral and short wave InfraRed (SWIR) for detection of floating marine litter using different sensors ranging from fixed camera, drone, airborne and high resolution satellite based multispectral data. All the data from the different cameras have been preprocessed considering the type of camera and the installation set-up. To reach this aim, several modifications have been made to the VITO’s cloud processing workflow for UAV data named MAPEO.
Part of the study has been done through making synthetic plastic accumulation zone in Belgium, while the other part was done in one of the hotspot source of marine litter in Hanoi, Vietnam which is part of an ESA project called “Artificial Intelligence and drones supporting the detection and mapping of floating aquatic plastic litter (AIDMAP)”. For fixed cameras, Detectron2 from Facebook AI research which provides state of the art detection and segmentation has been used for detecting individual floating litters. Faster Region based Convolutional Neural Network (Faster-RCNN) has been applied and 88.74 percent accuracy obtained at intersection of unit 50.
For detecting litter accumulation zones from multispectral UAV images, Decision Tree, Random Forest and Support Vector Machines (SVM) applied for classification of litter versus no litter using several widely used spectral indices, band ratios and normalized band ratios. Random forest classification showed better performance compared to the other algorithms.
Finally, first tests in an outdoor condition have been performed with a multispectral SWIR camera with six spectral bands that have been tuned to be sensitive at specific wavelengths. These wavelengths correspond to spectral features that are of particular interest to marine plastic based on the literature. The SWIR camera was installed as a fixed camera over a bridge in Belgium and its performance in detecting plastic objects was tested through several campaigns, here we present the detection results for the first time to the community.

Pristine and wilderness landscape of Patagonia is continoulsy threated by antropoghenic marine debris. Here, we presented a new approach to detect marine litter over beaches using UAV multiespectral imagery. The eBee UAV using the Parrot SQ ® were used to detect marine litter in two different beaches over the Patagonian Fjords located in the North part (Playa Muerta) and southern part close to Sn Rafael glaciers (Playa Leopardo). The Multispectral imagery were processed and calibrated in order to estimate the Surface reflectance. Then, support vector machine, Random forest and automatic digital classification were used to tratin and detect several litter such as styrofoam and buoys. Results shown 3 tons of litter per 16 km2 with an accuray about 6%. Color buoys are better classified than styrofoam which might be correlated with brightning pixel such as branches, shells and rocs. The combination of machine learning and very high spatial resolution imagery provided by UAV is validated to provide better information about marine litter in Patagonia mainly produced by the acquaculture systems.

In the past decade, plastic marine litter (PML) research has grown into a serious endeavour, exposing various aspects of plastic marine litter. For example, several studies have investigated and described the diversity and complexity of PML as well as the need to address not only floating or semi-floating plastic marine litter, but also plastic debris distributed in the first meters of the water column. Although remote sensing techniques could offer unique opportunities for addressing PML from local to global scale, it is agreed by the scientific community that a single sensor cannot provide all the required information since PML can vary considerably as for type, size, shape, chemical composition, buoyancy, and the way it manifests in the environment. While several optical passive remote sensing techniques (e.g. panchromatic cameras, multi-/hyperspectral sensors, thermal infrared cameras) have already been applied to the remote sensing of plastics either in controlled experiments or in real marine scenarios, the potential of LIDAR techniques to address marine litter issue is still almost unexplored.
The fluorescence LIDAR is an active remote sensing technique that can be exploited to acquire information on the chemical-physical characteristics of a volume target by using a pulsed laser. Fluorescence, actually, is an inelastic process due to the spontaneous emission of photons after the absorption of the incident radiation by the material system, given that such radiation belongs to an absorption line or band of the target components. The excited level decays after a characteristic time , called the fluorescence lifetime, in lower energy levels and in the fundamental level. The signal acquired by the LIDAR system can be analysed either in the time domain, e.g. by using a streak camera, in order to perform lifetime spectroscopy, or in the frequency domain, e.g. by using a spectrometer coupled to a linear array detector, in order to perform fluorescence spectroscopy. In both cases, the fluorescence LIDAR system can provide information on the chemical-physical properties of the target.
In this paper, we present a set of experimental tests, carried out in the laboratory, by using an in-house developed hyperspectral fluorescence LIDAR system. Different types of raw plastics and ocean-harvested objects have been investigated as for their fluorescence properties. The LIDAR sensor was operated from a distance of about 10 m while the samples were measured both in dry conditions and while immersed in a simulated water column, down to a depth of about 1 m. Besides the results of the experiments for the detection and characterisation of plastic marine litter, several factors that can affect the detection and characterisation of plastic marine litter are discussed: these includes photobleaching effects on plastics, fluorescence of dissolved organic matter and Raman scattering due to the water molecules.
This study was funded by the Discovery Element of the ESA's Basic Activities, contract no. 4000132184/20/NL/GLC.

Since the 1950’s, positively buoyant plastic objects have been accumulating at the surface of the oceans, transported by currents, wind and waves. Small millimeter-sized pieces (less than 4.75 mm), known as microplastics, count in trillions at global scale and pose an increasing risk to marine biota. Floating microplastics concentrate along large-scale convergence zones associated with Ekman dynamics in the five major ocean basins, but a comprehensive analysis of the spatial and temporal distributions is lacking and the monitoring tools are not well developed to assess global distributions. Through our recently funded NASA project Spaceborne Quantification of Ocean MicrO-Plastics (SQOOP), we are conducting a feasibility study of remote detection of surface microplastics, in the context of different surface particle properties and uncertainties in atmospheric correction with multiple, advanced detection techniques including hyperspectral and polarimetric approaches.

Here we will present preliminary results evaluating: 1) Geospatial and temporal trends in existing ocean color products across hot spots that may be related to enhanced reflectance from plastics; 2) Simulations of ocean surface hyperspectral reflectance using simple mixed pixel models to the Top of the Atmosphere (TOA) under different microplastic concentrations and atmospheric conditions; and 3) Simulations of microplastics using robust vector radiative transfer models for coupled ocean-atmosphere systems that include polarization. Detectability of floating marine microplastics will be conducted using statistical information content assessment in terms of current and future instrument characteristics, microplastic quantity and nature, and external conditions, such as observation geometry and atmospheric state. The sensitivity analysis will compare simulated and ocean color data to historic data of microplastic concentrations measured in surface net tows across the world ocean. The influence of floating microplastics on atmospheric correction and on standard ocean color retrievals of aerosols and of other ocean properties will also be explored.

In collaboration with visual artist Oskar Landi, we are producing an unconventional and engaging art exhibit to engage the public further in this pressing environmental problem. This unique art and science collaboration also provides new insights and informs our scientific models of sea surface dynamics including sun glint, sea foam, and floating plastics.