Polymer Type Identiﬁcation of Marine Plastic Litter Using a Miniature Near-Infrared Spectrometer (MicroNIR)

: Plastic pollution in the marine environment has turned into an important research topic in recent decades. Until recently, studies were often based on visual assessment only, which is not enough to draw any conclusion about the chemical nature of found plastic items and could lead to incorrect results. Standardized, fast, and e ﬃ cient low-cost methods for marine plastic litter identiﬁcation are urgently needed to monitor the occurrence and distribution worldwide. In this paper, we demonstrate that a miniaturized handheld near-infrared spectrometer—MicroNIR—can be used for on-site identiﬁcation of di ﬀ erent plastic polymers. A database containing polymer spectra of the most produced and reported polymer types in the marine environment was created including polyethylene (PE), polypropylene (PP), polyethylene terephthalate (PET), polystyrene (PS), polyvinyl chloride (PVC), polyamide (PA), polycarbonate (PC), polyurethane (PUR), and Silicone. Using spectral match value (SMV, included in the instrument software) for spectra analysis resulted in an accurate classiﬁcation of all nine polymer types. The method was used for the identiﬁcation of marine macro-, meso-, and microplastic litter collected on beaches in sediments and seawater and enabled the correct identiﬁcation of marine plastic litter for macro-, meso- (96%), and microplastics (73%) with exception of totally black items and items less than 1 mm in size. The method and instrumentation presented here are very well suited to support “Citizen Science” marine litter monitoring projects during beach cleaning and similar activities.


Introduction
Contamination of the World Oceans by synthetic non-biodegradable litter has become a high-profile environmental concern. Of this debris, plastics make up the largest quantity, which is related to increased production of anthropogenic materials and growing dependence on plastic products [1]. Global production rates for 2003 were estimated at 299 million tons [2], while in 2018, this increased to more than 350 million tons [3]. According to recent forecasts, only 23% of plastic produced in 2020 will be recycled. Taken together, these numbers indicate the risk of increased plastic pollution.
Plastics in the environment can be classified by size into macro-(over 25 mm), meso- (25-5 mm), and microplastics (<5 mm). Microplastics may enter the ocean directly as primary products such as granules, pellets, fibers, and powders or through the fragmentation of macroplastics under the influence of biological, chemical, and physical processes producing microplastics. This type of microplastic is often referred to as secondary microplastics [4]. It is important to establish the origin, trajectory, and fate

Spectrometers
Two different spectrometers were used in the study: • A bench-top Fourier Transform Infrared Spectrometer (Spotlight 400 FTIR, Frontier ATR; PerkinElmer, Waltham, MA, USA), working in the range of 650-4000 nm, equipped with an attenuated total reflection. • An ultra-compact device MicroNIR 1700ES (Figure 1), working in the range of 908-1676 nm, distributed by Viavi Solutions-Milpitas, CA, USA.
The MicroNIR dimensions are 45 mm in diameter and 42 mm in height, weighing about 60 g, and it is equipped with a 128-pixel detector array, which records data with a nominal spectral resolution of 6.25 nm. The system is composed by two small tungsten light bulbs as the radiation source and a linear-variable filter (LVF) directly connected to a linear indium gallium arsenide (InGaAs) array detector. The MicroNIR can be directly connected to a USB port of any laptop.
FTIR was used to ground-truth the applicability of a pocket NIR spectrometer for plastic type identification, as it is the most commonly used method for chemical characterization of potential plastics [11,18].
Appl. Sci. 2020, 10, x FOR PEER REVIEW 4 of 15 identification system of plastic solid waste based on NIR spectroscopy in combination with support vector machine [15], with machine learning models [16] or hyper spectral imaging [17].
In this paper, we describe our experience in the application of a simple method to detect polymer type down to the micro range (1 mm) using a portable NIR spectrometer MicroNIR that is expected to be a major step towards large-scale monitoring of marine plastic litter.

Spectrometers
Two different spectrometers were used in the study:  The MicroNIR dimensions are 45 mm in diameter and 42 mm in height, weighing about 60 g, and it is equipped with a 128-pixel detector array, which records data with a nominal spectral resolution of 6.25 nm. The system is composed by two small tungsten light bulbs as the radiation source and a linear-variable filter (LVF) directly connected to a linear indium gallium arsenide (InGaAs) array detector. The MicroNIR can be directly connected to a USB port of any laptop.
FTIR was used to ground-truth the applicability of a pocket NIR spectrometer for plastic type identification, as it is the most commonly used method for chemical characterization of potential plastics [11,18].

MicroNIR Library and Method of Identification
All spectra were collected using the MicroNIR Pro software version 3.0 (Viavi Solutions Inc., Santa Rosa, CA, United States) in the spectral range of 908-1676 nm. A 99% diffuse reflectance panel was used for the full range reference value, and a zero reference was taken by moving the MicroNIR into the open air. All spectra of small items (<2 cm) and transparent items were taken against the 99% diffusive reflectance panel.

MicroNIR Library and Method of Identification
All spectra were collected using the MicroNIR Pro software version 3.0 (Viavi Solutions Inc., Santa Rosa, CA, USA) in the spectral range of 908-1676 nm. A 99% diffuse reflectance panel was used for the full range reference value, and a zero reference was taken by moving the MicroNIR into the open air. All spectra of small items (<2 cm) and transparent items were taken against the 99% diffusive reflectance panel.
To create a library, a minimum of ten spectra were collected for each polymer type. To study the spectra and methods transferability, three different MicroNIR spectrometers connected and powered by different computers and laptops were used to collect data from a set of 10 polymer types (Table 1).
Spectra processing and data analysis were performed using the MicroNIR Pro software. Standard normal variate (SNV) and second derivatives were applied to all spectra. Spectral match value (SMV) was used for further data analysis. SMV is a simple method to perform classification of different materials for identification [19]. Several spectra of different polymers were acquired and added to the different polymer libraries, and the mean spectra for each polymer type was calculated. The individual polymer libraries were compiled into an overall method, correlation coefficients were calculated for each polymer type library, and a cross-correlation table was created.
Measurement of new unknown samples takes several seconds; the method is easy to use also with little pre-knowledge of NIR spectrometry. The result of the measurement is presented with threshold coefficients, where the highest value indicates the best match.

Industrial Reference Polymers
To create a library of reference material (RM) spectra, the most common 10 polymer materials were selected as samples: polyethylene terephthalate (PET), high density polyethylene (HDPE), polyvinyl chloride (PVC), low density polyethylene (LDPE), polypropylene (PP), polystyrene (PS), polyamide (PA), polycarbonate (PC), polyurethane (PUR), and Silicone (Table 2). Information on the prevalence of a different type of plastic in the marine environment was also taken into account. Each polymer type was represented by material from 2-3 different manufacturers and by items with different sizes, ranging from 0.25 mm powder to 1.5 cm fragments (Table 2, Figure 2). All used reference materials were transparent or white with exception of light-gray PET pellets. In case of spherical particles and pellets, spectra of one to ten items together were collected. The spectra of wood and cotton often found in marine samples were also added to the library. Two methods of identification were created based on the reference material library, RM_1 with 6 polymers with RIC 1-6 and RM_2 with 9 polymers with RIC 1-7. To create a library, a minimum of ten spectra were collected for each polymer type. To study the spectra and methods transferability, three different MicroNIR spectrometers connected and powered by different computers and laptops were used to collect data from a set of 10 polymer types (Table 1).
Spectra processing and data analysis were performed using the MicroNIR Pro software. Standard normal variate (SNV) and second derivatives were applied to all spectra. Spectral match value (SMV) was used for further data analysis. SMV is a simple method to perform classification of different materials for identification [19]. Several spectra of different polymers were acquired and added to the different polymer libraries, and the mean spectra for each polymer type was calculated. The individual polymer libraries were compiled into an overall method, correlation coefficients were calculated for each polymer type library, and a cross-correlation table was created.
Measurement of new unknown samples takes several seconds; the method is easy to use also with little pre-knowledge of NIR spectrometry. The result of the measurement is presented with threshold coefficients, where the highest value indicates the best match.

Industrial Reference Polymers
To create a library of reference material (RM) spectra, the most common 10 polymer materials were selected as samples: polyethylene terephthalate (PET), high density polyethylene (HDPE), polyvinyl chloride (PVC), low density polyethylene (LDPE), polypropylene (PP), polystyrene (PS), polyamide (PA), polycarbonate (PC), polyurethane (PUR), and Silicone (Table 2). Information on the prevalence of a different type of plastic in the marine environment was also taken into account. Each polymer type was represented by material from 2-3 different manufacturers and by items with different sizes, ranging from 0.25 mm powder to 1.5 cm fragments (Table 2, Figure 2). All used reference materials were transparent or white with exception of light-gray PET pellets. In case of spherical particles and pellets, spectra of one to ten items together were collected. The spectra of wood and cotton often found in marine samples were also added to the library. Two methods of identification were created based on the reference material library, RM_1 with 6 polymers with RIC 1-6 and RM_2 with 9 polymers with RIC 1-7.

Household Plastics
The methods of identification were tested on standard household plastic (HP) with known RIC (Table 3). Both transparent samples and color-added samples were identified. Household plastics with copolymer composition (copolymers of PS and PA) were also tested using the MicroNIR. Obtained spectra of household plastic were added to the database, and a new method (HP_1) was created to control if color additives affect the classification of different polymer types.

Environmental Samples
The method was applied to determine composition of marine plastic litter collected at different occasions: 1. On

Results
The reproducibility and data transferability were evaluated by collecting spectra of several reference material polymers from different manufacturers, with different particles size, and distance to the instrument's lamps and detector by three different MicroNIR instruments connected to three different laptops. As shown in Figure 4, the different conditions of spectra acquisition do not affect the results and data obtained with different instruments, and computers can easily be merged in one database. The accuracy of the identification by MicroNIR on site was verified using a Perkin Elmer ATR-FTIR bench top spectrometer in the laboratory.

Results
The reproducibility and data transferability were evaluated by collecting spectra of several reference material polymers from different manufacturers, with different particles size, and distance to the instrument's lamps and detector by three different MicroNIR instruments connected to three different laptops. As shown in Figure 4, the different conditions of spectra acquisition do not affect the results and data obtained with different instruments, and computers can easily be merged in one database.

Results
The reproducibility and data transferability were evaluated by collecting spectra of several reference material polymers from different manufacturers, with different particles size, and distance to the instrument's lamps and detector by three different MicroNIR instruments connected to three different laptops. As shown in Figure 4, the different conditions of spectra acquisition do not affect the results and data obtained with different instruments, and computers can easily be merged in one database.

Reference Materials Library
The initial library consisted of six main polymer types (RIC 1-6, method RM_1). Application of SMV to raw spectra did not obtain adequate polymers separation, as some spectra were very similar in this wavelength range ( Figure 5). For example, the threshold coefficient for PE and PP needed to be set to 0.95. Processing the spectra first by SNV followed by applying a second derivative function, largely improved separation capacity.
It was also shown that HDPE and LDPE could only be differentiated using a very high threshold coefficient (>0.995, Table 4). Therefore, we decided to join HDPE and LDPE in one type, PE. The highest correlation coefficient between other polymers was found for PP and HDPE, LDPE-0.53 and 0.60, respectively. Appl. Sci. 2020, 10, x FOR PEER REVIEW 9 of 15 (a) (b) Figure 5. Raw spectra of polymers with RIC 1-6 (a) and polymers with RIC 7 together with wood and cotton spectra (b).

Minimum Size for Detection with MicroNIR
Minimum size of particle for polymer type identification with MicroNIR was found to be around 0.45 mm, wherein the particle should be placed exactly in the center of the diffuse reflectance panel and between two MicroNIR lamps ( Figure 6).  As the second step, the other four polymers (from category RIC 7) but representing the second most produced and most frequently found in marine environments were added to the library. These were PUR, PA, PC, and Silicone, resulting in a nine-polymer library, method RM_2. For this method, the highest correlation was found between PET and PC (0.61) and for PE and PP (0.56) ( Table 5). The method applied resulted in complete classification of all nine polymer types included in the library. Correlation coefficients between polymer types with RIC 1-6 were not changed significantly after the addition of the other four polymers to the library (Tables 4 and 5). Spectra of wood and cotton were also added to the library as they are often found and are sometimes difficult to distinguish from plastic items. Spectra of these materials differ significantly from spectra of studied polymers ( Figure 5); the addition of items in the library did not affect identification of other polymers. The presence of wood and cotton in the database is useful in case of studies of plastic in seawater, since they present very often in samples, and this will reduce any uncertainty in identification.

Minimum Size for Detection with MicroNIR
Minimum size of particle for polymer type identification with MicroNIR was found to be around 0.45 mm, wherein the particle should be placed exactly in the center of the diffuse reflectance panel and between two MicroNIR lamps ( Figure 6).
(a) (b) Figure 5. Raw spectra of polymers with RIC 1-6 (a) and polymers with RIC 7 together with wood and cotton spectra (b).

Minimum Size for Detection with MicroNIR
Minimum size of particle for polymer type identification with MicroNIR was found to be around 0.45 mm, wherein the particle should be placed exactly in the center of the diffuse reflectance panel and between two MicroNIR lamps ( Figure 6).

Household Colored Plastic Library
To examine the effect of plastic color and/or any additivities on plastic type identification, method PM_2 was used to identify the composition of normal household plastic. Many different colored plastics and a variety of polymers were tested in the household (Table 3). Using the MicroNIR and method PM_2, we were able to identify all samples except for totally black items. Dark blue, dark green, dark silver, and dark brown items, although resulting in low signals from the MicroNIR, were correctly classified after spectra processing, and the characteristic peaks are clearly visible (Figure 6).
A new method for polymer type identification, HP_1, was created based only on household samples. Correlation coefficients between different polymers in the method HP_1 were very similar to the method RM_2 (Tables 5 and 6). For both methods, the highest coefficient was found between PET and PC, and the lowest between PET and PS. The main difference between the two methods was found for PVC. This can be explained by a lower number of PVC household items included in the

Household Colored Plastic Library
To examine the effect of plastic color and/or any additivities on plastic type identification, method PM_2 was used to identify the composition of normal household plastic. Many different colored plastics and a variety of polymers were tested in the household (Table 3). Using the MicroNIR and method PM_2, we were able to identify all samples except for totally black items. Dark blue, dark green, dark silver, and dark brown items, although resulting in low signals from the MicroNIR, were correctly classified after spectra processing, and the characteristic peaks are clearly visible (Figure 6).
A new method for polymer type identification, HP_1, was created based only on household samples. Correlation coefficients between different polymers in the method HP_1 were very similar to the method RM_2 (Tables 5 and 6). For both methods, the highest coefficient was found between PET and PC, and the lowest between PET and PS. The main difference between the two methods was found for PVC. This can be explained by a lower number of PVC household items included in the library and by the fact that polymer composition of PVC, which often includes copolymers.

Copolymer Type Identification
Several copolymers with known composition were tested. Two copolymers of PS were included: acrylonitrile butadiene styrene (ABS) a wide used copolymer made from 15-35% acrylonitrile, 5-30% butadiene, and 40-60% styrene, and styrene-acrylonitrile (SAN) consisted of 70-80% of styrene and 20-30% of acrylonitrile. Identifications using the MicroNIR resulted in PS as the identification result with 0.92 and 0.93 threshold coefficients, correspondingly. Another example of copolymer composition identification was tested using synthetic material with 80% of nylon and 20% of elastane, which was identified as PA. These results showed that the main component of the copolymer was identified by using the MicroNIR.

Marine Plastic Litter Identification
Eighteen plastic fragments consisting of different colors and sizes were isolated from beach and coastal water during a beach cleaning (Figure 2a, Table 7). Through visual assessment, it was not possible to define what original product was or what type of polymer the fragments consisted of except for one fragment of a PET bottle and several fishing lines. Using the MicroNIR, we were able to quickly establish seven different polymer types on site. We were able to correctly identify the polymer type of 17 of the 18 fragments with exception of one black fragment of PVC (Table 7), which was confirmed by ATR-FTIR. Although some of the fragments were dirty and/or old (weathered), this did not affect the ability of the instrument to identify them.
The developed method was further applied on 14 plastic fragments found in sediment in the Eurasian Arctic Ocean. These samples were measured on board the R/V "Akademik Mstislav Keldysh" during the 2019 Arctic expedition cruise. The size of fragments varied from 0.4 to 7 cm. All fragments polymer composition was correctly identified using the MicroNIR and consisted of eight PE fragments, four PP, one PET, and one PVC confirmed by FTIR in the laboratory following the cruise.
In addition, 41 plastic fragments in sand samples from Norwegian beached around the Oslo fjord were analyzed using MicroNIR. Average size of found items was 3.5 mm, with a size range of 1-5 mm; all fragments were relatively clean. MicroNIR analysis correctly identified the composition of 40 fragments, where PE had 25 items, PP 13, and PS two. Comparison with ATR-FTIR showed that MicroNIR analysis gave correct results for all fragments with the exception of one undefined black sample of PE.
In Table 8, the results of analysis using both MicroNIR and ATR_FTIR of 47 plastic floating items collected by surface trawling in the Eurasian Arctic are given. Three meso plastic fragments with size 10-30 mm and another 44 items in the microplastics range with sizes from 0.2-5 mm and average 2.5 mm were examined. Although many particles were covered with organic matter, 28 particles were defined correct on board the R/V "Akademik Mstislav Keldysh" using the MicroNIR, 17 items were not classified including to black fragments, nine thin fibers, and eight fragments with either length < 1 mm or extremely dirty fragments. Finally, 66 floating meso and micro plastic particles (average size 4.6 mm, range 1.2-12 mm) were collected by trawling in the Atlantic Ocean. Identification was only done with MicroNIR on board, and no confirmation analysis was performed. Here, 47 particles were identified of which 34 were PE, 11 PP, and two PVC. In total, 19 particles were undefined including eight black fragments, six fibers, and five small items with lengths less than 1.5 mm and a width < 1 mm.

Discussion
Using the handhold MicroNIR spectrometer, two libraries of polymer material spectra and two methods of polymer type identification were created. One spectral library based on industrial reference materials and the other on household plastic with known polymer composition according to RIC codes. Correlation coefficients between studied polymer types are very similar for both libraries. The spectral database and method created from household plastic samples (with known polymer composition based on RIC) worked well as the method created from virgin reference polymer materials and pre-production pellets for polymer type identification of plastic litter. Color and other additives did not affect the efficiency of the polymer identification, the method also found that most dark colored plastics could be correctly identified using MicroNIR with the exception of completely black materials. Furthermore, the identification of samples with a copolymer composition resulted in correct identification of the main component of the copolymer although with weaker correlation coefficients with the spectra of the polymer present in the spectral library.
In this study, we have defined results of polymer identification with correlation coefficient > 0.7 as true and results with correlation coefficient < 0.4 as a lower limit, indicating that there is no match with the items and polymers included in the library. Results with coefficients between 0.4-0.7 may indicate that the sample might be a copolymer with no easily distinguishable primary polymer type (if these single polymers composition is less than 50%). For example, the evaluation of a sample from Crete that looked like a kitchen oilcloth and was made of two different layers (an unknown fabric and PVC) resulted in the low correlation coefficient, 0.48 due to a "mixed" spectra of the combination of materials.
This work shows that MicroNIR can be excellently used on site, down to the microplastic range. The lower particle size that can be detected was found to be 0.45 mm for reference materials. For real environmental samples the lower size limit was somewhat higher, and it was only possible to identify fragments larger than 1 mm. The reason for this could be related to the marine litter items being covered by organic matter, and especially for smaller particles, this results in distorted NIR spectra, which is critical for polymer identification. Although no further investigations in this study were done, procedures for organic matter removal and decomposition before measurements with MicroNIR are available [11]. One drawback of the use of MicroNIR on site is that it was not possible to identify thin fibrous materials. However, if fibers are long enough, the material can be concentrated by rolling the fiber into a ball and placing it in the middle of the MicroNIR camera window. This will allow the sampler to acquire spectra with enough quality to correctly identify the polymer composition.
In this work, a total of 186 items of marine plastic litter, with different colors and sizes, were studied. The MicroNIR along with the created databases and methods were able to correctly identify 96% of the larger marine macro-and mesoplastic litter (42 of 44 fragments) with exception of two black items. Seventy-three percent (104 of 142 items) of the smaller microplastic litter items were correctly identified with the MicroNIR method. Undefined items were mostly black (8%), fibers (10.5%), or fragments, which were either <1 mm or very dirty particles (8.5%). The results, however, indicate that MicroNIR can be excellently used to identify marine plastic litter with size >1 mm even on site or on board of research vessels. In addition, the use of the MicroNIR is easy and efficient when the right methods and databases are established and could be an excellent tool to be used in Citizen Science projects, adding extra value to marine registration protocols by filling the knowledge gap on the origin and sources of meso-and larger microplastics.
The focus of the study was to create a method of identification of the nine most used polymer types-PET, PE, PP, PVC, PS, PA, PC, PUR, and Silicone. Combining results from five different randomly performed studies revealed that PE fragments amounted to 51% of all found plastic items: PP-24%, PVC-9%, PET-5%, PS, PA, PUR-3-4% each, and PC-1%, showing the potential of using MicroNIR methods for marine litter identification. To improve results of copolymer identification, more polymer types could be added to the library, for example, Acryl, Teflon, and ABS. Several other options to achieve even better correlation coefficient for larger datasets containing more polymers or copolymers including the use of more advanced statistics (including principal component analysis) have only been partly examined. Iterative development of this instrument allows the stored spectra of the samples to be re-analyzed with new methods or larger databases.
Comparison of the described application of MicroNIR with another common technique such as FTIR spectroscopy [11,18] shows both advantages and disadvantages of the MircoNIR method. The use of MicroNIR in the field-at beaches and on-board vessels-showed good results for marine plastic litter identification, correctly identifying 96% of macro-and mesoplastics and 73% of microplastics. However, MicroNIR is not able to identify totally black items and items smaller than 1 mm. Conventional IR spectroscopy cannot capture spectra of completely black objects that absorb the signal [17]. For black items and items smaller than 1 mm, using more advanced methods, i.e., FTIR, is needed. FTIR is also needed for identification of copolymer composition where high resolution allows a detection of differences between spectra.
On the other hand, fast and inexpensive plastic identification methods are needed to support the growing number of marine litter studies. The inclusion of plastic type identification within these studies is greatly needed to support knowledge and direct mitigation strategies towards combating identified materials [11,21]. Recently, usage of miniature devices for plastic identification is actively developing for different purposes, mainly for plastic waste identification [14][15][16]. Previous studies with MicroNIR have focused on pharmaceutical analysis [22,23], narcotic material identification [24], and authentication of seafood [25].
The method presented here for the identification of marine plastic litter with MicroNIR is a fast, easy, and efficient low-cost method that can be used in the field, and it does not need qualified personnel to operate it. In this way, plastic identification with MicroNIR is very well suited to support "Citizen Science" projects, as shown here for the marine environment. The development of a handheld, easy-to-use NIR camera for the identification of polymers in the meso-and micro range (>1 mm) could also play a major role to gather more reliable data for source identification and subsequent mitigation of plastic and litter pollution to the environment.