A review of remote sensing technology for plastic waste monitoring

Abstract

Plastic waste pollution has become a critical environmental challenge that requires innovative monitoring approaches to support effective environmental management. This systematic review synthesizes recent advancements in remote sensing (RS) technologies for plastic waste detection, analyzing 84 studies published between 2018 and 2024 following PRISMA guidelines. The review evaluates RS platforms, sensor types, spectral ranges, classification methods, and polymer identification across diverse environmental settings. Satellite platforms dominate large-scale marine monitoring (45% of studies), while unmanned aerial vehicles (UAVs) excelled in high-resolution coastal applications (23%). Correspondence analysis identified four distinct research clusters optimized for specific platform-environment combinations. Supervised learning was most prevalent (50%), though deep learning approaches and hybrid models show emerging promise. Polyethylene was most frequently detected across platforms. Limitation of the research field includes geographic bias towards European sites (> 50%), focus on controlled conditions rather than operational deployment, inability to detect microplastics, and lack of standardized protocols. The review also highlights emerging developments in RS technologies, including spectral mechanisms that support polymer discrimination and ongoing gaps in plastic monitoring. An integrated framework is proposed that combines multi-platform Earth Observation (EO), machine learning, and citizen science to enable scalable plastic waste monitoring. The findings provide theoretical and practical insights to guide future sensor design, algorithm development, and global monitoring strategies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11356-025-37347-7.

Introduction

Mismanaged and littered plastic waste continues to leak into the environment, particularly aquatic ecosystems, with projections indicating an increase from 6.1 million tons per year in 2019 to 11.6 million tons per year by 2060 (OECD 2022). Despite concerted global efforts, an estimated 710 million metric tons of plastic waste has already accumulated in both aquatic and terrestrial ecosystems (Lau et al. 2020). Land-based and sea-based activities (Hwang 2020; Chitaka et al. 2022), as well as runoff and riverine outflows (Hadiuzzaman et al. 2022; Wang et al. 2022), constitute major sources and transport pathways of plastic waste into natural environments. It is estimated that up to 80% of plastic found in the ocean originates from land-based sources and is transported via rivers (Ritchie and Roser 2023). The impacts of plastic waste are profound and multifaceted, affecting climate, ecosystems, socioeconomic systems, and human health (Tian et al. 2022). Throughout its life cycle, plastic significantly contributes to global greenhouse gas emissions, thereby intensifying the carbon footprint and accelerating climate change (Shen et al. 2020). Discarded plastic also poses severe threats to wildlife and ecosystems through habitat disruption, entanglement, and ingestion (Casabianca et al. 2019). Addressing plastic leakage is therefore critical to mitigating these broad and interconnected environmental challenges.

In response, the United Nations Environment Assembly (UNEA) has adopted several resolutions aimed at reducing and preventing plastic pollution (Stöfen-O’brien 2022). One such initiative involves the development of monitoring and assessment frameworks by the Joint Group of Experts on the Scientific Aspects of Marine Environmental Protection (GESAMP), which advocates for repeated litter monitoring to assess item quantities, types, and temporal trends (GESAMP 2019). Furthermore, the fourth meeting of the Ad Hoc Open-Ended Expert Group on marine litter and microplastics (AHEG-4) underscored the significance of leveraging innovative technologies—including Earth Observation (EO)—for monitoring plastic waste from source to sea, managing legacy waste, and preventing future pollution (UNEP 2020). Plastic waste monitoring methods are diverse and can be broadly classified into five categories: plastic tracking, active sampling, passive sampling, visual observations, and citizen science (van Emmerik and Schwarz 2020). Among these approaches, emerging technologies play a vital role in both preventing plastic leakage into aquatic systems and identifying and removing existing plastic pollution (Schmaltz et al. 2020).

Recent advancements in environmental monitoring technologies have significantly enhanced the detection, identification, and quantification of plastic waste across diverse ecosystems. Remote sensing (RS), both satellite- and UAV-based, has emerged as a powerful tool for observing floating marine litter, plasticulture farmlands, and coastal debris at various spatial and spectral scales (Gnann et al. 2022; Veettil et al. 2023). Techniques such as multispectral and hyperspectral imaging, coupled with AI-based classification algorithms—including convolutional neural networks (CNNs) and object segmentation models—offer promising capabilities for automating plastic detection while reducing labor-intensive manual surveys (Gnann et al. 2022; Yuan et al. 2023). Simultaneously, integrated frameworks that combine RS with in situ observations and citizen science have demonstrated value in addressing spatial–temporal gaps and improving validation processes for plastic monitoring efforts (Maximenko et al. 2019; Topouzelis et al. 2021). For instance, Sentinel-2 (S-2) imagery has been widely applied for detecting macroplastic patches in open waters, while unmanned aerial vehicles (UAVs) and close-range sensors have enabled high-resolution identification of plastic materials in riverine and agricultural settings. Additionally, unmanned vehicle platforms such as unmanned surface vehicles, autonomous underwater vehicles, and gliders are increasingly deployed to complement satellite data, especially in inaccessible or hazardous regions (Yuan et al. 2023). Despite promising developments, challenges persist regarding spectral ambiguity, classification accuracy, and standardization across platforms. This review synthesizes the methodological parameters, sensing systems, and computational models used in recent studies to inform the design of more robust, integrated, and scalable frameworks for plastic waste monitoring in both natural and human-modified environments.

Despite growing interest in applying RS to monitor plastic pollution, existing studies often remain fragmented by platform capabilities, sensor limitations, and regional focus. Building upon the increasing body of research—such as the spectral and platform-based classifications highlighted by Salgado-Hernanz et al. (2021)—this review aims to provide a comprehensive assessment of plastic waste detection through EO. Specifically, we classify and evaluate RS platforms, sensor types, spectral ranges, and classification methods in relation to environmental settings, land cover, and polymer characteristics. Furthermore, we identify persistent gaps in geographic representation and technological limitations, such as the challenges of detecting microplastics or integrating multi-source data, which restrict the development of scalable monitoring solutions. Finally, this review proposes a methodological framework that leverages emerging EO technologies, machine learning algorithms, and participatory citizen science practices to enhance global plastic waste surveillance. These efforts aim to inform future research directions and guide the development of more inclusive, scalable, and data-integrated plastic monitoring systems.

Methodology

This systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, which are applicable to both original and updated systematic reviews (Page et al. 2021). This review focuses on the application of RS technology for plastic waste monitoring. The methodological process is illustrated in Fig. 1 and described in detail below.

Fig. 1.

Systematic review process for remote sensing (RS) technologies in plastic waste monitoring

Literature search

The literature search was conducted using two major English-language databases, Scopus and Web of Science, which are widely recognized for their comprehensive coverage of scientific publications in society, environment, and technology. The search aimed to identify relevant studies on plastic waste and RS by applying keyword queries across all searchable fields. Two keyword combinations, “plastic litter”/“plastic waste” and “remote sensing,” were used in the advanced search.

In Scopus, the search strings were:

TITLE-ABS-KEY(plastic* AND litter*) AND TITLE-ABS-KEY(“remote* sensing”) and TITLE-ABS-KEY(plastic* AND waste*) AND TITLE-ABS-KEY(“remote* sensing”).

In the Web of Science database, the corresponding queries were:

ALL = (plastic* AND litter* AND “remote* sensing”) and ALL = (plastic* AND waste* AND “remote* sensing”).

These searches yielded 370 documents from Scopus and 331 documents from Web of Science, resulting in a total of 701 records. All Web of Science records were exported in full, whereas for Scopus, only citation information, bibliographic details, abstracts, and keywords were extracted. The literature search was completed in November 2025.

Selection criteria

Subsequently, the initial sample of 701 articles was filtered using three specific criteria. First, unique identifiers—Web of Science Unique Identifier (UI) and Scopus Electronic Identifier (EID)—were used to identify and eliminate duplicate records across both databases. A PivotTable function in Microsoft Excel was employed to cross-check and remove 113 duplicate entries. Following this, a title screening process was conducted, resulting in the exclusion of 180 additional records with duplicate or non-applicable titles. Lastly, the document type was filtered to include only journal articles, resulting in 253 articles for further analysis. Second, a publication year filter was applied based on the earliest appearance of relevant keywords in 1978, defining the inclusion range from 1978 to 2025. The number of publications has grown notably since 2018, reflecting the growing attention from global bodies such as GESAMP and UNEP, which have highlighted EO tools as innovative approaches for plastic-pollution monitoring. Since 2018, 228 relevant studies have been published. Third, studies were assessed using categorical variables related to RS characteristics and plastic detection features during both initial screening and full-text assessment. These criteria were established to reflect the regional scope of this review, focusing on land and water body plastic litter detection via RS technologies. For example, Maximenko et al. (2019) proposed the Integrated Marine Debris Observing System (IMDOS), encompassing multiple observation platforms and sensors with appropriate spatial and temporal resolution. Similarly, Topouzelis et al. (2021) reviewed the detection of floating marine litter using satellite, aircraft, and drone imagery, classified by RS data types, classifiers, and processing techniques. Farré (2020) examined the use of UAVs and related sensors, while Gnann et al. (2022) focused on the classification of plastic types and polymers using diverse algorithms and datasets related to terrestrial and aquatic environments. Based on these comprehensive parameters, a final set of 82 articles was selected for inclusion in this review.

Remote sensing component assessment

A total of 82 studies were deemed eligible for qualitative and statistical analysis in this stage of the systematic review. The process followed key principles of systematic review methodology, including the formulation of clear research questions, identification of relevant literature, quality assessment, and evidence synthesis (Khan et al. 2003). To facilitate qualitative comparison, componential analysis was employed—a technique that uses matrices and tables to compare and contrast subcomponents across various domains from the reviewed literature (Onwuegbuzie et al. 2012). A comprehensive category database was developed based on the identified RS characteristics and plastic waste detection domains. The frequency of each variable was organized into a contingency table, where RS platforms were listed as rows and detection-related parameters as columns. To further examine associations among these variables, a correspondence analysis (CA) was conducted. CA is a multivariate statistical technique that evaluates relationships among categorical variables and provides graphical representations of their distribution patterns (Gilula et al. 2001). The analysis was performed using R, with RS platforms treated as the dependent variable and the remaining detection parameters as independent variables.

Limitations

This review focuses exclusively on academic publications related to plastic-waste monitoring using RS data, published between 2018 and November 2025. While this timeframe captures recent technological advancements and aligns with global initiatives on marine-litter monitoring (GESAMP 2019; UNEP 2020), it may exclude earlier foundational studies that contributed to the development of RS-based detection methods. The search strategy also relied on English-language keyword queries in Scopus and Web of Science, which may have introduced selection bias by omitting relevant studies published in other languages or indexed in additional databases. Moreover, this review excluded review articles, conference papers, and non-peer-reviewed documents, ensuring a focus on original research but potentially overlooking insights synthesized in this body of literature. Another limitation is the exclusion of studies that relied solely on laboratory measurements without spatial remote-sensing data. While necessary to maintain the focus on RS-based monitoring, this criterion may have omitted controlled experiments that could provide important spectral information for polymer identification. The review prioritizes research targeting macroplastics (25–1000 mm) and megaplastics (> 1 m), given the current limitations of RS technologies in detecting smaller plastic debris. Consequently, studies focusing on microplastics (< 5 mm) or mesoplastics (5–25 mm) were not included, which may limit the scope of findings across the full plastic size spectrum. Although care was taken to ensure accurate screening and categorization, the interpretation of RS characteristics and plastic-detection domains relied partly on metadata, abstracts, and reported methodological details. These sources may not fully capture the complexity of the original studies, introducing the potential for misinterpretation.

Results

Parametric components for plastic waste monitoring

The reviewed studies reveal that RS technologies used to detect and monitor plastic waste in the environment involve a wide range of components. RS observation technologies are applied to track sources of plastic pollution and the leakage of litter across diverse environments (Garello et al. 2019). This review classifies plastics based on their occurrence in land, water, or mixed environments, considering both the characteristics of RS technologies and plastic sample types. The location of plastic waste is a key spatial parameter, with each setting exhibiting different land cover types that influence the effectiveness of RS-based classification. On land, primary land cover types include beaches, floodplains, agricultural land, and urban areas (e.g., grasslands and dumpsites), while aquatic environments comprise oceans, rivers, and lakes. When study areas span both land and water, they are grouped under mixed land cover. Depending on the monitoring context, different RS platforms are used to detect plastic waste (El Mahrad et al. 2020; Veettil et al. 2022), which are categorized into five types: satellite remote sensing (SRS), aerial remote sensing (ARS), unmanned aerial vehicles (UAVs), static sensors (SS), and combination platforms (CP).

Each platform is equipped with sensors that collect electromagnetic spectrum signatures from surface objects. These sensors are classified as passive or active, with some studies employing both. The electromagnetic spectrum includes a range of wavelengths divided into spectral bands such as visible (VIS), infrared (IR), microwave (MW), and others (Le Moigne et al. 2011). The spectral data captured by different sensors are analyzed using various classification methods, which vary according to the RS platforms, spectral range, and number of images used. Plastic waste detection techniques identified in the reviewed literature range from visual interpretation to advanced machine learning approaches, including deep learning (Salgado-Hernanz et al. 2021; Mehmood et al. 2022; Veettil et al. 2022). These classification methods can be grouped into visual sighting, photointerpretation (PI), unsupervised learning (UL), supervised learning (SL), deep learning (DL), and hybrid techniques. Another parameter considered is polymer type, derived from classification results. This review highlights commonly studied polymers such as polyethylene (PE), polyethylene terephthalate (PET), polystyrene (PS), polypropylene (PP), and other polymers (OTH), which represent the dominant plastic types leaking into the environment (Lusher et al. 2017). Based on these considerations, the final dataset was categorized according to key parameters including location, land cover, platform, sensor type, spectral range, classification method, and polymer type, as summarized in Tables 1 and 2.

Table 1.

Platform–environment parameters and categories used to identify and verify remote-sensing technologies for plastic waste monitoring. (Values in parentheses indicate the number of research studies in each category)

Parameter	Platform–environment focus (no. of studies)
Location mapping	Waterbody (43), land (31), waterbody and land (8)
Land cover	Ocean (33), river (7), lake (2), beach (16), agriculture (8), urban (6), floodplain (1), mixed (9)
Platform	SRS (37), UAVs (19), ARS (5), SS (8), CP (13)

Table 2.

Analytical parameters and target polymer categories used to identify and verify remote-sensing technologies for plastic waste monitoring. (Values in parentheses indicate the number of research studies in each category)

Parameter	Algorithm–polymer focus (no. of studies)
Sensor type	Passive (74), active (4), passive and active (4)
Spectral range	VIS (23), VIR (49), IR (2), MW (3), VIR and MW (4), other (1)
Classification method	SL (41), DL (11), PI (8), UL (1), CM (21)
Polymer type	PE (35), PET (19), PP (12), PS (11), OTH (9)

VIR visible to infrared, CM combination methods

This review focused on monitoring plastic waste across diverse environmental settings, drawing on 82 studies that examined the complete pathway of plastic leakage—from sources to final accumulation—using RS technologies for detection and classification. Accordingly, the location or extent of plastic waste accumulation on Earth’s surface, whether in water bodies, on land, or across both, was considered as a key parameter in the analysis.

Location mapping

All 82 studies included in this review were categorized into three sampling location types: water bodies, land, and a combination of both. The majority (43 studies, 52%) focused on aquatic environments, such as oceans, rivers, and lakes. For example, Sannigrahi et al. (2022) investigated marine plastic pollution in Greece, Cyprus, Italy, and Lebanon; Simpson et al. (2022) monitored plastic waste in rivers across Serbia and Bosnia and Herzegovina; and Baburaj et al. (2023) conducted a study on plastic pollution in Ukkadam Lake, India. The second largest group (31 studies, 38%) concentrated on terrestrial environments, including beaches, agricultural land, floodplains, and urban areas. Escobar-Sánchez et al. (2021), for instance, assessed beach plastic pollution along the southern Baltic Sea (Germany and Lithuania), while Acharki and Kozhikkodan Veettil (2023) compared plastic debris on agricultural land in Morocco and Vietnam. Sakti et al. (2023) focused on riverbank litter in Indonesia, and Liu et al. (2024) examined plastic pollution in urban areas of China. Only eight studies (10%) assessed both land and water environments within the same research scope. For example, Zhou et al. (2021) conducted a multi-country study covering Spain, Germany, Ghana, and Egypt, investigating plastic waste in diverse settings such as lakes, floodplains, agricultural fields, and urban areas.

Land cover types

The corpus of studies was categorized into eight distinct land cover groups: ocean, river, lake, beach, agriculture, urban, floodplain, and mixed. Ocean environments were the most frequently studied, with 33 studies (40% of the total) primarily aiming to identify various types of plastic waste, particularly in the Mediterranean Sea (19 studies). For example, Topouzelis et al. (2019) and Kremezi et al. (2021) focused on floating plastic debris off the coast of Lesvos Island, Greece, while Themistocleous (2021) assessed plastic waste originating from fishing boats and fish farms in Cyprus. Other studies addressed marine plastic pollution in the North Pacific (4 studies), North Atlantic (4 studies), and other oceans (6 studies), including disaster debris in the Gulf of Mexico (Hu et al. 2023), marine litter in Porto Pim Bay, Portugal (Freitas et al. 2021), and floating debris across various oceans (Duarte and Azevedo 2023).

Rivers were investigated in seven studies across Asia (4 studies), Europe (2 studies), and North America (1 study), focusing on plastic quantities and movement. For instance, Luo et al. (2022) identified plastic items in China’s Longhe River, Flores et al. (2022) tested novel sensors to detect floating plastic in the Waal River (Netherlands), while Vadivel et al. (2025) monitored plastic debris accumulated in Mexico’s Coatzacoalcos River. Only two studies focused on lakes: Fronkova et al. (2024) analyzed plastic waste in Whitlingham Greater Broad (United Kingdom), and Baburaj et al. (2023) examined sewage-related plastic waste in Ukkadam Lake, India.

Sixteen studies were dedicated to beaches along the world’s coastlines, with targets along the Pacific (4 studies), Indian (3 studies), Atlantic (3 studies), Mediterranean (2 studies), Baltic Sea (2 studies), and other oceans (2 studies). Notable examples include plastic accumulation on beaches in Hawaii (Winans et al. 2023), the Red Sea coast of Saudi Arabia (Martin et al. 2018), Leirosa Beach, Portugal (Pinto et al. 2021), San Rossore, Italy (Merlino et al. 2021), the Southern Baltic Sea (Escobar-Sánchez et al. 2021), and the Caribbean Sea (Manzolli and Portz 2024a). Eight studies explored agricultural land as a source of plastic leakage in Asia (3 studies), Europe (2 studies), South America (2 studies), and across Africa and Asia (1 study). For instance, Song et al. (2025) assessed greenhouse plastic waste management in China. Six studies addressed plastic pollution in urban areas, mostly in Asian and European cities, with Mo et al. (2021) reporting microplastic contamination linked to construction in China and Rettig et al. (2025) focusing on litter in Germany. Floodplains were studied in only one case, involving plastic litter along the Citarum River in Indonesia (Sakti et al. 2023). Finally, nine studies focused on mixed land cover types, examining plastic pollution across multiple landscapes and countries. For example, Zhou et al. (2021) analyzed plastic clusters across floodplains, agricultural land, urban areas, and lakes in Spain, Ghana, West Africa, Egypt, and Germany; Balsi et al. (2021) compared marine debris on various beaches in Italy; and Estrela et al. (2025) analyzed the spatial extent and distribution of plastic-covered surfaces in urban, agricultural, and marine settings across several countries.

Remote sensing platforms

The corpus of studies was classified into five remote sensing platform categories: SRS, ARS, UAVs, SS, and CP. Each platform presents distinct technical, financial, and temporal advantages and limitations. The majority of studies (37, 45%) employed satellite remote sensing for data collection. Of these, 23 studies used single satellite datasets to monitor plastic waste, leveraging key features of satellite platforms such as data availability, spatial resolution, revisit frequency, and consistent area coverage. In terms of data availability, satellite remote sensing provides reliable, open-access information for plastic waste detection at low cost, making it a widely used approach in the literature (Basu et al. 2021; Trinh et al. 2022). Due to their consistent spatial resolution, high-resolution satellite images are commonly employed in plastic monitoring studies. For instance, Sannigrahi et al. (2022) and Mikeli et al. (2022) utilized the Sentinel-2 satellite archive, offering 10-m resolution, to detect floating marine plastics. Similarly, Kremezi et al. (2021) highlighted the potential of PRecursore IperSpettrale della Missione Applicativa (PRISMA) imagery, with 5-m resolution, to identify plastic waste across various target sizes. Moreover, the high revisit frequency of SRS enables monitoring of plastic waste over multiple time periods, depending on weather conditions. For example, Papageorgiou et al. (2022) and Valente et al. (2023) accessed the S-2 data series—with a 5-day revisit cycle—throughout all seasons of a year to estimate the accumulation of floating plastics. Likewise, David et al. (2023) employed monthly S-2 imagery to identify areas covered by plastic silage bags. Each satellite’s observing area is defined by a specific swath width, which allows for combination of multiple image tiles to achieve wider spatial coverage. The S-2 constellation, for instance, has been used to detect floating plastic debris in coastal waters extending up to 20 km from the shoreline of Greece (Duarte and Azevedo 2023). Based on S-2 data, Berto et al. (2025) identified seasonal and temporal patterns of floating litter in the South Atlantic Ocean. Similarly, NASA’s Earth Surface Mineral Dust Source Investigation (EMIT) detected spectral signatures of these two plastic types across multiple continents, highlighting their association with land-based sources of plastic pollution (Estrela et al. 2025).

Fourteen studies employed multiple satellite datasets to support long-term plastic waste monitoring and improve classification efficiency. For example, Kikaki et al. (2020) examined the sources and transport of marine plastic debris over a five-year period (2014–2019), using high-resolution imagery from Planet, S-2, and Landsat-8 (LS-8). Similarly, Hu et al. (2023) utilized medium-resolution, multi-band imagery from the MODIS and Medium Resolution Imaging Spectrometer (MERIS) satellites to quantify the size, distribution, and temporal variation of plastic debris patches.

The second group of studies (19, 23%) focused on the use of UAVs, which provide very high-resolution data with spatial coverage over a few square kilometers at low-altitude ranges. These studies generally fell into two subcategories: optimization of flight altitude for plastic detection and comparison of UAV efficiency against visual census (VC) methods. In the first subgroup, the optimal flight height of UAVs was evaluated to improve plastic identification accuracy, taking into account spatial resolution, weather conditions, and local regulatory constraints. For example, Geraeds et al. (2019) tested three different altitudes under stable weather conditions to serve different monitoring purposes, such as visual counting, identification of river transport hotspots, and detection of stranded plastics along riverbanks. Similarly, Escobar-Sánchez et al. (2021) assessed optimal spatial resolutions at three altitudes, under European Union regulatory frameworks, for detecting plastic beach litter at meso- and macro-scales. The second subgroup focused on millimeter-level data resolution and presented comparisons between UAV surveys and citizen scientist operators (CSOs) in terms of classification efficiency. For instance, Andriolo et al. (2021) and Manzolli and Portz (2024b) evaluated plastic waste characteristics, such as type, material, color, and size, by comparing UAV image screening with VC observations. Martin et al. (2018), Escobar-Sánchez et al. (2021), and Almeida et al. (2023) compared the cost and time efficiency of UAV-based processing against manual annotation methods for plastic classification.

SS platforms, installed at ground-level sites, were featured in the third group of studies (8, 10%). Five of these studies employed cameras as primary devices, offering panoramic views for plastic detection. These included both surveillance and specialized cameras. For example, Armitage et al. (2022) used footage from low-cost security cameras to quantify plastic waste in a port area, while De Giglio et al. (2021) analyzed multispectral images captured with a hand-held camera, positioned on a footbridge, to detect plastic litter in a river. Other SS platform devices used include an X-band radar installed on a rooftop (Serafino and Bianco 2021) and ground-based sonar (Flores et al. 2022), both of which were used to detect plastic debris in marine and riverine environments, respectively.

Despite offering excellent detail and wide-area coverage, ARS platforms were utilized in only five studies (6%), forming the fourth group. ARS provides ultra-high-resolution imagery covering hundreds of square kilometers from medium-altitude platforms. For instance, Moy et al. (2018), Garcia-Garin et al. (2020a, b), and Winans et al. (2023) generated natural-color mosaics with centimeter-level resolution to investigate plastic objects across diverse coastal areas under varying local legislative and meteorological conditions. Additionally, ARS platforms incorporated hyperspectral imaging system to distinguish different types of plastic litter in marine environments, such as open-ocean areas in the North Pacific, USA (Garaba et al. 2018), and coastal bays around Faial Island in the North Atlantic Ocean, Portugal (Freitas et al. 2022).

The remaining group consisted of 13 studies (16%) that employed a combination of platforms for plastic classification. These studies focused on multi-platform imaging approaches to evaluate the proportion of plastic pixels detected and to optimize spectral characteristics within the detection process. For instance, SRS and UAV data sources were integrated to estimate the fraction of plastic abundance at the pixel level in marine environments (Topouzelis et al. 2019, 2020a) or urban settings (Sakti et al. 2023). Additionally, spectral characteristics were compared across platforms to enhance plastic detection accuracy. These comparisons involved UAV and other platforms, such as ARS (Freitas et al. 2021; Rettig et al. 2025), SS (Iordache et al. 2022), and SRS–ARS combinations (Zhou et al. 2022), to improve the identification of plastic items.

Sensor type and spectral range

RS sensors, which detect and measure electromagnetic radiation, can be categorized into two main types: passive and active, each operating across different spectral ranges. In the reviewed literature, studies were grouped into three categories based on sensor type: passive (74 studies, 90%), active (4 studies, 5%), and a combination of both (4 studies, 5%). Notably, 90% of the studies employed passive sensors, which capture energy reflected or emitted by objects, primarily within three spectral ranges: VIR (49 studies), VIS only (23 studies), and IR only (2 studies).

Among the passive sensor studies using VIR data, three subgroups were identified based on spectral resolution: multispectral imagery (38 studies), hyperspectral imagery (8 studies), and a combination of both (4 studies). Multispectral sensors capture images in multiple bands across the visible and invisible light spectrum. Most studies using multispectral satellite data relied on the analysis of spectral signatures in different reflectance bands to distinguish plastic waste from other surface materials. For instance, G.N. and Muthu (2024) and Pathira Arachchilage et al. (2025) used red and infrared (IR) bands from S-2 imagery to develop plastic-detection indices. The former focused on naturally occurring marine plastic debris, while the latter tested the indices using experimentally deployed plastic targets. Similarly, Acharki and Kozhikkodan Veettil (2023) proposed spectral indices based on blue, green, red, NIR, and shortwave infrared (SWIR) bands from multiple satellites to identify plastic-covered greenhouse farms. Several studies also validated satellite-derived plastic detection indices by integrating multispectral imagery from other platforms. For example, spectral indices from Sentinel-2 data were confirmed using true-color composite images from UAV (Themistocleous et al. 2020; Sakti et al. 2023) and ARS platforms (Cillis et al. 2022). Other studies focused exclusively on multispectral imagery from UAV (Jeong et al. 2024) or SS platforms (De Giglio et al. 2021), demonstrating the capability of multispectral sensors across various platforms to detect plastic waste using spectral bands comparable to those of S-2 or WorldView-2.

Studies using hyperspectral imagery employed specialized sensors that capture narrow, contiguous bands, particularly within the 1.2–1.8 µm range, to enhance sensitivity to plastic signatures while reducing atmospheric interference. Hyperspectral imagery in the NIR and SWIR ranges was used to classify plastic polymers based on their distinct absorbance spectra. These studies were conducted on ARS platforms (Freitas et al. 2022) and UAV platforms (Balsi et al. 2021; Pérez-García et al. 2024). However, satellite-based hyperspectral data, which typically have lower spatial resolution, presented limitations. For example, PRISMA imagery was able to classify plastic into general categories but not by polymer type (Kremezi et al. 2021; Taggio et al. 2022). Consequently, some studies combined multispectral and hyperspectral data to enhance classification performance. Zhou et al. (2021) validated plastic detection classifiers by converting four airborne hyperspectral datasets to WorldView-3 spectral resolution, while Schmidt et al. (2023) combined S-2 and PRISMA data to improve spectral discrimination of plastic materials.

Twenty-three studies employed VIS-range (red, green, blue) data to create true-color composite images captured by cameras mounted on various platforms. Of these, 15 used UAV-based imagery to classify plastic based on characteristics such as shape, color, and material type. True-color images from Complementary Metal–Oxide–Semiconductor (CMOS) cameras provided detailed visual features that facilitated the identification of plastic litter in beach environments (Escobar-Sánchez et al. 2021; Antara et al. 2024) and riverine areas (Geraeds et al. 2019). Furthermore, the combination of true-color and VIR imagery allowed for more accurate classification of beach litter by material and size, aligned with the Oslo-Paris convention (OSPAR) protocol (Andriolo et al. 2020; Pinto et al. 2021) and UNEP guidelines (Andriolo et al. 2021). Studies using ARS platforms also employed Digital Single-Lens Reflex (DSLR) cameras to visually characterize floating marine litter (Garcia-Garin et al. 2020a) and stranded debris along the shoreline (Winans et al. 2023).

Two studies focused specifically on IR-only imagery. These studies utilized SWIR hyperspectral cameras to identify polymer types in marine plastics, such as those found in the North Pacific Ocean (Garaba et al. 2018) and along the Scottish coastline (Cocking et al. 2022).

In contrast, active sensors emit their own energy and detect its interaction with objects. Four studies employed active sensors, with three using microwave radar and one using sonar. For instance, Serafino and Bianco (2024) compared radar backscatter signals from plastic debris and other materials using coastal radar stations, while Simpson et al. (2022) monitored temporal variations in backscatter signals and surface roughness associated with large plastic accumulations via satellite radar. Additionally, Flores et al. (2022) identified plastic waste by analyzing sonar signal reflections within the water column and flowing rivers.

Classification methods

The reviewed literature reveals four primary classification methods in RS for plastic waste identification: PI, UL, SL, DL, and CM.

SL emerged as the most widely applied method, employed in 41 studies (50%). These studies used labeled datasets to train algorithms for predictive classification. Specifically, 18 of these studies adopted SL techniques based on spectral signal classification, including thresholding and spectral unmixing, to identify plastics among various surface materials. For instance, threshold-based detection of spectral index values was utilized to identify marine plastic debris from satellite imagery (Ciappa 2022; Berto et al. 2025) or UAV imagery (Antara et al. 2024). Pixel values from satellite (Ciappa 2022) and UAV (Antara et al. 2024) imagery were also used for plastic waste thresholding. Spectral unmixing techniques, which decompose pixel spectra to estimate the contribution of pure material endmembers, were applied to identify ocean plastics using multispectral imagery from satellite imagery (Topouzelis et al. 2020b; Papageorgiou et al. 2022). Similarly, hyperspectral unmixing algorithms that exploited polymer-specific absorption features improved sub-pixel plastic detection in ARS data (Garaba et al. 2018).

Additionally, 9 studies compared the performance of various SL classification algorithms for plastic waste detection. For example, Freitas et al. (2021) and Sannigrahi et al. (2022) compared support vector machines (SVM) and random forest (RF) classifiers, concluding that SVM outperformed RF in marine plastic detection. Similarly, Acuña-Ruz et al. (2018) and Baburaj et al. (2023) found SVM to be more accurate than other models for classifying plastics in UAV and satellite imagery, respectively. A further group of 14 studies tested specific SL classifiers, for example, RF was employed as a multi-class classifier to detect plastic waste in agricultural (Iordache et al. 2022), urban (Page et al. 2020), and riverine (G.N. and Muthu 2024) contexts. In addition, matched filter was applied to identify and isolate plastic signals within image datasets across urban (Aguilar et al. 2025), agricultural, and marine environments (Estrela et al. 2025).

CM—involving at least two classification techniques—were used in 21 studies (26%). Eight of these studies compared digital classifications with visual interpretations, using PI results as references to evaluate platform performance. For example, Topouzelis et al. (2020a) and Cillis et al. (2022) used visual classification to estimate plastic coverage per pixel, comparing results with spectral unmixing and spectral index-based classifications of plastic targets. Martin et al. (2018) assessed the efficiency, reliability, and accuracy of VC, PI, and SL methods for classifying beach plastic waste.

Nine additional studies in this group focused on comparing classification methods or developing hybrid models to enhance detection accuracy. For instance, Zhou et al. (2022) compared SL and DL performance, reporting that RF exhibited the most robust and reliable results, even under noise and uncertainty from unknown plastics or background surfaces. Vadivel et al. (2025) developed an integrated model combining SL with DL training to classify plastic leakage in riverine environments. The remaining four studies evaluated the utility of visual surveys and manual image screening (MIS), often involving CSOs. For example, Merlino et al. (2021) compared VC and MIS based on type, color, material, size, and source of plastic items, with MIS conducted by CSOs in secondary school classrooms.

DL was used in 11 studies (13%), primarily through CNNs that automatically learn image features or apply object detection frameworks such as YOLO (you only look once). For example, Garcia-Garin et al. (2021) applied CNN-based models via a web-oriented application to detect and quantify floating marine macro-litter (FMML) from aerial imagery. Armitage et al. (2022) implemented the YOLOv5 framework to detect floating plastics using closed-circuit television (CCTV) footage from ocean ports.

PI or MIS appeared in eight studies (10%). This approach relies heavily on user experience and standardized protocols. For instance, Andriolo et al. (2021) conducted visual interpretation of UAV imagery in accordance with UNEP/Intergovernmental Oceanographic Commission (IOC) guidelines, involving both experts and trained non-expert operators familiar with common litter types in the study area. Similarly, Manzolli and Portz (2024a) proposed best-practice protocols for monitoring marine macro-litter and evaluating waste-reduction efforts in coastal areas.

UL was the least common approach, applied in only one study to detect artificial plastic target using hyperspectral UAV and S-2 images (Fronkova et al. 2024).

Polymer type classification

Of the total corpus of reviewed studies, 45% (37 studies) analyzed plastic polymer classification using RS data. These studies classified plastics into five major polymer types: PE, PET, PS, PP, and OTH.

A total of 35 studies focused on the classification of polyethylene, including both LDPE and HDPE. Among LDPE types, plastic bags were the most frequently studied items (15 studies). These studies utilized RS techniques to classify plastic bags commonly discarded along beaches and in marine environments. For instance, UAV images of beach litter were analyzed through visual interpretation (Manzolli and Portz 2024a) and digital classification methods (Antara et al. 2024) to estimate litter coverage, while plastic bag fragments observed in oceanic ARS imagery were classified to assess their density and distribution (Garcia-Garin et al. 2020a). Plastic-covered greenhouses were the subject of seven studies, typically in large-scale agricultural zones. Acharki and Kozhikkodan Veettil (2023) mapped greenhouse plastic waste in vegetable and fruit farming areas using satellite imagery to quantify post-harvest plastic pollution. LDPE plastic sheets were also recognized in RS imagery (Garcia-Garin et al. 2020b; Prodanov and Bekova 2023). Regarding HDPE, eleven studies identified plastic containers, and seven studies detected buoys. For example, plastic containers were classified using UAV data in coastal areas of Cambodia (Wolf et al. 2020), while fisheries-related buoys were detected in Hawaii and the Pacific Ocean using ARS imagery (Garaba et al. 2018; Winans et al. 2023). Additional HDPE waste types such as drink drums (Martin et al. 2018) and octopus pots (Pinto et al. 2021) were also successfully identified.

PET was the second most studied polymer group, with 19 studies using various RS platforms. Plastic bottles were the primary PET waste analyzed in 17 of these studies, either as individual items or experimental samples. For instance, Andriolo et al. (2020, 2021) used UAV data and PI techniques to classify discarded beach bottles and assess their density and type based on inter-operator consistency. Other studies, such as those by Themistocleous et al. (2020) and Topouzelis et al. (2020a), deployed large numbers of plastic bottles in marine environments to test RS classification performance using both satellite and UAV data. PET plastic sheets were also used in classification trials (Zhou et al. 2022).

Three additional polymer groups had similar representation in the literature: PP with 12 studies, PS with 11 studies, and OTH with 9 studies. Eight of the PS studies focused on styrofoam, primarily found on beaches, with UAV data commonly used for detection. For instance, styrofoam fragments were digitally classified from UAV imagery collected at Sardinia’s beaches, Italy (Balsi et al. 2021) and Batu Belig Beach, Indonesia (Antara et al. 2024) to estimate plastic occurrence, while Acuña-Ruz et al. (2018) also employed very high-resolution satellite data to distinguish styrofoam fragments from other plastic debris based on shape and color.

For PP, rope debris was the dominant waste type identified in eight studies. These items, often discarded from fishing activities in coastal and marine areas, require very high-resolution data and are typically analyzed through both visual interpretation and digital classification. Moy et al. (2018) used aerial photo-photointerpretation to detect ropes along the Hawaiian coastline, while Freitas et al. (2021) employed UAV and ARS platforms to identify floating rope debris in the North Atlantic Ocean near Portugal.

The final group, labeled as “OTH,” includes plastic waste types not covered by the four main groups. Notably, seven studies focused on nylon fishing nets, which were classified using various RS techniques. For example, Garaba et al. (2018) used ARS data to detect and monitor fishing nets for marine plastic pollution management. Martin et al. (2018) analyzed UAV data to identify beach waste, including fishing nets, in support of seasonal waste management planning.

Remote sensing component analysis for plastic monitoring

The relationships among key parameters in plastic waste monitoring were extracted from a structured dataset and compiled into a column-based spreadsheet. This information was then visualized using sunburst diagrams (Fig. 2) created in Microsoft Excel. The left diagram illustrates area-based classifications linking plastic location, land cover types, and RS platform, while the right diagram presents classifications relating to RS platforms, classification methods, and polymer types.

Fig. 2.

Sunburst diagrams illustrating hierarchical parameters used in plastic waste monitoring: (left) spatial classification by location, land cover type, and RS platform; (right) plastic identification by RS platform, classification method, and polymer type

Plastic location in relation to land cover types and RS platforms (Fig. 2, left)

The majority of studies investigating plastic waste in natural water bodies focused on ocean environments, with SRS being the predominant platform, used in 19 studies. The second most common approach involved CP in five studies, integrating SRS with UAVs in three cases and with ARS in two cases. The number of studies employing ARS, UAV, or SS as standalone platforms was identical, with each platform used in three studies. In river environments, UAVs were the most frequently used platform (3 studies), whereas SRS and SS each appeared in two study. Within lake environments, both UAV and CP (SRS + UAVs) platforms were employed in only one study each. One study examined a mixed water environment that included both river and ocean land covers.

In terrestrial environments, beaches were the most studied land cover type. UAV platforms were dominant (10 studies), followed by ARS and SS (2 studies each), while SRS and CP (SRS + UAVs) appeared in only one study. In agricultural areas, SRS was the primary platform (7 studies), with one additional study employing CP (SRS + ARS). Similarly, urban areas were primarily investigated using SRS (5 studies), while one urban-focused study used CP (ARS + UAVs). In floodplain environments, one study adopted a CP approach (SRS + UAVs).

For studies addressing both land and water environments, the majority involved urban and floodplain areas in combination with river, lake, or agriculture zones. Among the CP studies, two applied dual-platform integrations (SRS + ARS and SRS + UAVs), whereas three adopted a tri-platform approach combining SRS, ARS, and UAVs. Two studies used SRS—one covering ocean and beach environments and another spanning ocean, agricultural, and urban areas. Two additional studies used UAVs to monitor beaches alongside oceans or lakes. Furthermore, one study employed SS to monitor both floodplain and river environments.

These results highlight a clear dominance of SRS in large-scale monitoring across oceanic, agricultural, and urban environments, while UAV platforms remain prevalent for high-resolution monitoring in beach and riverine contexts. The limited use of ARS, SS, and CP suggests potential areas for future research, particularly in integrating complementary platforms to improve spatial and temporal coverage in plastic waste detection.

RS platforms in relation to classification methods and polymer types (Fig. 2, right)

A total of 37 studies were identified that classified plastic waste by polymer type. PE was the most frequently detected polymer, reported in 35 studies, followed by polyethylene terephthalate PET in 19 studies. All RS platforms were capable of detecting PE, and every classification method applied across the dataset successfully identified it. SL was the most widely used technique for PE identification (15 studies), followed by CM in ten studies.

UAV platforms were the most common for polymer classification (14 studies), followed by SRS (9 studies). Within the UAV group, CM and PI techniques were applied in four studies each, consistently identifying PE and PET across all cases. DL was employed in three of these studies, particularly for detecting PS, while PI was used to classify all polymer types. SL was applied in three UAV studies to identify both PE and PS.

Studies using SRS completely also relied predominantly on SL (8 of 9 studies), with detection limited mainly to a single polymer type—typically PE (6 studies). PS and OTH were each classified in only one study. In addition, one SRS study applied DL to classify PE.

The ARS and CP groups comprised five and six studies, respectively. ARS platforms were used to classify multiple polymer types, including PE (five studies) and PP (three studies), with deep learning methods applied in two of these studies. Notably, PP was classified using various methods from ARS data including PI, PL, and DL. The CP group used CM in four studies (SRS + ARS; SRS + UAVs)—one of which classified all polymer types, while the remaining studies focused specifically on PET and/or PE. In addition, two CP studies that combined UAVs with SRS or ARS utilized SL for polymer classification.

Finally, SS platforms (3 studies) applied either SL, DL, or CM, primarily for the classification of PET and PE.

Overall, the results indicate a dominance of UAV and SRS platforms for polymer-specific classification, with SL emerging as the most widely applied technique across platforms. While PE and PET were consistently identified, the relatively limited detection of other polymer types—particularly PS, PP, and OTH—suggests that further methodological integration, such as combining DL with multi-platform datasets, could improve classification diversity and accuracy.

Correspondence analysis of remote sensing technologies

The dataset of parametric components related to RS technologies for plastic waste monitoring was simplified by counting the number of research studies corresponding to each element. These frequency counts were organized into a contingency table, which formed the basis for the CA. The resulting biplot is shown in Fig. 3, where RS platforms are represented in red, and their corresponding associations with monitoring components—such as plastic location, land cover, sensor type, spectral range, classification method, and polymer type—are shown in blue.

Fig. 3.

Correspondence analysis (CA) results for the different RS platforms (red) and their relationship with plastic location, landcover, sensor, spectral range, classification method, and polymer type (blue) used to monitor plastic waste in entire environment

Prior to conducting the CA, the statistical significance of the relationship between rows and columns was tested using the Chi-square test, which yielded a Chi-square value of 325.15 with a p-value < 2.2 × 10⁻¹⁶, indicating a strong association among variables. The CA effectively captured the variation in the dataset, with dimension 1 accounting for 69.1% of the total variance and dimension 2 for 23.4%. Together, these two dimensions explained 92.5% of the total variance, surpassing the commonly accepted threshold of 80% for meaningful two-dimensional interpretation.

As shown in Fig. 3, the CA biplot reveals four distinct research groupings based on RS platform usage (treated as the dependent variable):

Group 1 (upper right quadrant) comprises studies that use UAVs to monitor plastic waste, particularly in beaches and other local coastal environments. These studies are associated with VIS and IR wavelength ranges and commonly employ PI or sight-based observation methods. They also emphasize polymer-level identification, including materials such as HDPE, PET, PP, and PS.

Group 2 (upper left quadrant) comprises studies that use SS to monitor plastic waste in riverine environments, including both river and floodplain areas, typically employing OTH (other) sensor types. These studies are further associated with active sensing modes, MW spectral ranges, and UL methods, and they generally do not include polymer-type identification.

Group 3 (lower left quadrant) comprises studies that use SRS to monitor plastic waste across both terrestrial and aquatic environments, including agricultural fields, urban areas, landfills, and parts of the ocean. These studies are typically associated with the VIR spectral range and passive sensor systems and frequently apply SL methods. However, they generally do not focus on polymer-type identification.

Group 4 (lower right quadrant) comprises studies that employ ARS platforms to monitor plastic waste, primarily in grassland and lake environments. These studies commonly apply DL methods and demonstrate the capability to classify specific polymer types, including PE, PS, and PP.

Overall, the correspondence analysis highlights the diversity of RS platform applications and their unique strengths in detecting plastic waste and associated parameters. It also reinforces the importance of tailoring RS strategies to specific environmental contexts and detection objectives.

Discussion

Geographic concentration and environmental dependencies

The detailed analysis of this paper reveals significant geographic concentration in current research, with European countries serving as study areas in over 50% of studies, followed by Asian countries in over 20%. Greece and Italy emerge as the most extensively studied countries, with Greece’s Mediterranean coastal waters, particularly Mytilene on Lesvos Island and Tsamakia beach, representing the most intensively studied marine environments (Topouzelis et al. 2019, 2020b; Kremezi et al. 2021). Although this concentration has facilitated algorithm development through controlled experimental conditions, it severely limits the generalizability of findings to diverse global environments.

The success of RS applications appears highly dependent on environmental conditions. Mediterranean studies demonstrate that controlled conditions—such as deep, dark-water backgrounds and artificial plastic targets—yield optimal detection results, while complex natural environments with mixed surface materials present significant challenges (Themistocleous et al. 2020; Basu et al. 2021). Environmental factors and background surface complexity substantially impact classification accuracy across all platforms. Studies conducted in The Gulf of Gera exemplify this by illustrating how natural phenomena can interfere with plastic classification algorithms (Papageorgiou et al. 2022; Papageorgiou and Topouzelis 2024), highlighting the gap between experimental validation and operational deployment.

Platform performance and technological limitations

SRS platforms demonstrate superior capabilities for large-scale monitoring with consistent temporal coverage, making them particularly suitable for tracking plastic accumulation trends in marine environments. The extensive use of S-2 data across Mediterranean studies has helped to established benchmarks for plastic detection (Sannigrahi et al. 2022; Pathira Arachchilage et al. 2025), though their performance remains highly dependent on controlled conditions. In contrast, UAV platforms excel in high-resolution detection processes and detailed polymer classification in coastal and riverine environments, with successful study examples ranging from Portuguese beaches (Andriolo et al. 2021; Pinto et al. 2021) to Hawaiian coastlines (Moy et al. 2018; Winans et al. 2023). However, their limited spatial coverage constrains regional-scale monitoring applications. Recent satellite missions such as NASA’s EMIT (2022) and PACE (2024) now provide expanded spectral coverage from the visible to the SWIR range, offering new opportunities for detecting both floating and land-based plastics.

Recent studies have demonstrated that hyperspectral sensors, particularly when mounted on UAV and airborne platforms, offer enhanced spectral discrimination of plastic polymers. These sensors can capture narrow, contiguous bands in the visible to SWIR regions, allowing differentiation of polymers such as PE, PP, and PS based on their unique absorption features (Garaba et al. 2018; Balsi et al. 2021; Freitas et al. 2022; Pérez-García et al. 2024). Such hyperspectral systems bridge the gap between laboratory spectroscopy and large-scale satellite applications, providing material-level precision in plastic waste identification.

LiDAR applications for plastic-debris monitoring are still in the early exploratory stage. Recent developments include active-optical LiDAR systems that use pulsed laser light to generate high-resolution 3D surface information, enabling improved detection and characterization of floating debris. Initial studies, such as (Raimondi et al. 2024), demonstrate the potential of LiDAR to distinguish plastics from other marine litter.

Active sensors remain underutilized (5% of studies), yet microwave-based systems are increasingly recognized for their potential to detect floating debris regardless of cloud cover or illumination conditions. Both spaceborne Synthetic Aperture Radar (SAR) systems (e.g., S-1 and TerraSAR-X) and ground-based X-band radar have shown sensitivity to surface-roughness and backscatter changes that enable differentiation of plastic debris from natural materials (Simpson et al. 2022; Serafino and Bianco 2024). These findings suggest that radar data could be used to complement optical monitoring by providing continuous, weather-independent coverage of plastic debris movement and accumulation.

The predominant use of SL methods (50% of studies) indicates reliance on labeled training data, while the emergence of combination platform approaches (16% of studies) reflects a growing recognition of the need for more integrated monitoring strategies (Topouzelis et al. 2020a; Sakti et al. 2023). Despite technological advances, several persistent challenges remain: spectral confusion between plastic materials and natural objects, lack of standardized detection protocols across platforms, and limited capabilities for detecting smaller plastic debris (micro- and mesoplastics).

CSOs represent an emerging component in plastic waste monitoring systems, contributing to data collection enhancements and environmental education initiatives. Their integration with RS workflows has demonstrated effectiveness in UAV-based classification tasks, offering potential for costs reduction and improved classification accuracy through collaborative validation approaches (Merlino et al. 2021; Almeida et al. 2023). The implementation of standardized protocols such as UNEP and OSPAR guidelines could facilitate the integration of scientific methodologies with community-based monitoring efforts (Andriolo et al. 2021; Pinto et al. 2021). However, maintaining data consistency across different geographical regions and ensuring adequate training for varied operator skills levels remain ongoing challenges that require systematic approaches to quality assurance and standardization.

Research evolution and future directions

The progression from controlled experimental setups using artificial plastic targets to monitoring naturally occurring debris represents an important evolution in the field. However, the continued reliance on controlled conditions in many studies suggests that operational applications in complex, real-world environments remain challenging. Limited representation of research in urban environments (4% of studies) and riverine systems (10% of studies) represents missed opportunities for addressing plastic pollution monitoring where waste management challenges are most acute.

Innovative sensor applications demonstrate expanding detection capabilities beyond traditional optical methods. The use of Adaptive Resolution Imaging Sonars along Dutch rivers (Flores et al. 2022), coastal radar systems in Italian ports (Serafino and Bianco 2021), and bridge-mounted cameras for river monitoring (De Giglio et al. 2021; Iordache et al. 2022) show potential for specialized sensor deployments in challenging environments where optical detection may be limited.

Future research should prioritize developing robust algorithms capable of reliable performance across diverse environmental conditions rather than optimized performance in controlled settings. The integration of multi-platform approaches (Zhou et al. 2021, 2022), standardization of detection protocols, and expansion to underrepresented geographic regions. The Integrated Marine Debris Observing System (IMDOS) provides an example of a multi-scale, collaborative framework that links in situ measurements, remote sensing, and modeling to track plastic movement from source to sink. Incorporating such global initiatives can help address data gaps and promote standardized approaches to plastic-pollution monitoring and assessment.

Conclusion

This systematic review of 82 studies published between 2018 and 2025 revealed significant technological advances in RS for plastic waste monitoring alongside critical gaps that must be addressed for operational implementation. SRS emerged as the most widely used platform (45% of studies), demonstrating effectiveness for large-scale marine monitoring under controlled conditions, while UAV platforms (23% of studies) excel in high-resolution coastal applications where detailed characterization is required.

The CA identified four distinct research clusters optimized for specific environmental conditions but also exposed significant geographic bias with European countries dominating research and Mediterranean environments receiving disproportionate attention. Although this concentration has facilitated robust algorithm development, it limits applicability to diverse global conditions where plastic pollution challenges are most severe.

Environmental context emerged as critical for detection success, with performance varying significantly between controlled experimental conditions and complex natural environments. The predominant focus on macro- and megaplastics, currently dictated by technological limitations, represents a significant constraint for comprehensive environmental assessments. Current approaches cannot address the full spectrum of plastic pollution impacts due to detection gaps for smaller debris categories in particular.

The emergence of combination platform approaches indicates the growing recognition of the need for integrated monitoring strategies, though standardized protocols across platforms remain absent. While promising technological directions include specialized sensors for challenging environments and citizen science integration, the transition from experimental validation to operational deployment requires addressing fundamental challenges related to environmental complexity and geographic representation.

Integrating citizen science with RS technologies represents a promising pathway for scaling monitoring efforts while fostering community engagement in environmental stewardship. Future research is necessary to develop standardized co-designed protocols enhanced training frameworks and mobile platform technologies to systematically integrate community-based observations with RS data streams, ensuring that participatory monitoring is a sustainable component of comprehensive plastic pollution surveillance systems.

Future research priorities should focus on geographic diversification to underrepresented regions, development of robust algorithms for complex real-world environmental conditions, establishment of standardized multi-platform integration protocols, and transition from controlled experimental validation to systematic validation using naturally occurring debris. The success of RS-based plastic monitoring ultimately depends on developing comprehensive frameworks that combine technological innovation with global applicability, enabling effective plastic pollution management strategies at local, regional, and global scales.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

Yootthapoom Potiracha conceptualized the study, conducted the systematic review, performed the analysis, and drafted the manuscript. Roger C. Baars contributed to the conceptual framework, provided critical revisions, and supervised the research. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data availability

The data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval

This is not applicable.

Consent to participate

This is not applicable.

Consent for publication

This is not applicable.

Competing interests

The authors declare no competing interests.