1 Introduction

Pollution with microplastics is an immense environmental problem. Microplastics are commonly defined as plastic particles with diameters below 5 mm. These small particles enter the environment directly by accidental losses of virgin pellets [1], by washing plastic fibers from clothes [2] and even using cleaning [3] and cosmetic [4] products. Furthermore, bigger pieces of plastic litter gradually degrade into smaller and smaller pieces [5]. Considering the increasing plastic production since the 1950s [6], as well as the versatility, low costs of plastic products and the slow improvement of waste management, an increase of plastic litter is to be expected, which tends to find its way into the marine environment. It was estimated that in 2010 between 4.8 and 12.7 million tons of plastic entered the oceans [7]. Already in the 1980s it has been reported that 80% of marine litter is plastics and that the litter density is exponentially increasing [8]. Similar to the plastic litter, the abundance of microplastic debris in the environment is increasing and will most likely continue to increase in the future [9].

The increasing abundance of microplastics also increases the chances for living beings to interact with them. Due to the various specific weights of microplastic particles, they are found floating on the water surface, in the water column and deposited in the sediment [10], which make them available to different marine species. An additional threat is the persistent organic pollutants (POPs), which are adsorbed by microplastics in high concentrations [11,12,13,14] and can be desorbed after ingestion [15]. Floating microplastics present a transport vector for POPs [11, 12, 16] and species settling on them [17]. The ingestion of microplastics was reported [18, 19], as well as evidence of microplastics entering the circulatory system of animals and cells [20, 21]. There are also some indications that humans could be exposed to microplastics through the food chain [18]. However, the investigation of the impact of microplastics on human health did not confirm any direct negative effects [22,23,24]. To better understand the effects of microplastics on living beings, its distribution and composition in the environment, improved programs of monitoring including improved quantification and detection methods are needed [25, 26].

In the past, several different approaches for sample collection have been used, mainly depending on the type of location of interest (e.g. sediment, water surface) [10]. After the collection, the samples are typically processed to reduce their volume and to remove organic materials to be able to detect single pieces of microplastics more easily. A commercially available device for the separation of microplastics from the sediment with a high efficiency was recently developed [27]. Finally, the processed samples need to be analyzed to determine the content and type of microplastics. Spectroscopic methods are best suited to perform this analysis [28,29,30,31].

The most commonly employed spectroscopic techniques for microplastic detection and identification are Raman and Fourier-transform infrared (FTIR) spectroscopy. The advantage of Raman spectroscopy is its better spatial resolution due to the shorter wavelengths of the excitation laser. However, relatively long integration times (up to 2 s) with several co-scans are needed for sample identification [29]. This would make an automated mapping of a sample time consuming. FTIR spectroscopy was employed in transmission, reflection and attenuated total reflection mode for sample identification. Furthermore, a focal plane array (FPA)-based micro-FTIR imaging was employed in reflection [32] and transmission mode [33], which provided a fast automated imaging of a sample. However, this approach still suffers from some drawbacks: the sample needs to be completely dry, the absorption of thicker samples could be too high when working in transmission [34] and scattering due to the irregularly shaped particles in reflection mode will reduce method’s efficiency [35].

Here, we explore the potential of photoluminescence (PL) spectroscopy for microplastic detection. PL is a process in which light is emitted from a material after it was optically excited. The wavelength of the PL emission is typically longer than the excitation wavelength as the excited carriers lose energy by internal conversion processes. As PL spectra give valuable information about the observed material, it is used in many practical applications, e.g. material science, agriculture, biology, medicine [36].

The first question is of course, whether the photoluminescence spectra of polymers differ from those of natural materials, i.e. whether plastic pieces can be distinguished from sand or a piece of a shell. If this is not the case, PL spectroscopy would not be a promising technique for the detection of microplastics. This is the question which we address here. We investigate undyed plastic samples and non-plastic materials commonly found in the environment, such as sea shells, wood and sand. We note, that the study presented here, represents only the very first step towards the goal of establishing photoluminescence spectroscopy as one of the techniques for microplastic detection. A comprehensive comparative study benchmarking photoluminescence against established techniques like FTIR or Raman spectroscopy is beyond the scope of this first paper on the luminescence from polymers and materials from the marine environment.

Most plastic materials, which contribute to microplastic pollution, are not known to be good PL sources. The same is true for sand or a sea shell. Yet, as we show here, they show some luminescence when illuminated with intense, short-wavelength laser radiation. The PL of plastics such as polyethylene [37,38,39], polypropylene [40], polystyrene [41] and others [42,43,44] has been investigated before. Yet, a systematic comparison of many different plastics, as presented here, has not been reported so far. In our study, we acquired PL spectra from nine different plastic materials and nine other materials typically found in the marine environment. By comparing the recorded PL spectra, we assess the potential of PL spectroscopy for microplastic identification.

2 Experimental setup, materials and methods

In our PL setup (Fig. 1), we used a laser diode emitting light at 405 nm with a maximum output power of 200 mW as an excitation source. The laser light first passes through a 405 nm bandpass filter which suppresses unwanted amplified spontaneous emission of the laser at other wavelengths. A dichroic long-pass mirror with a cut-off wavelength of 425 nm directs the filtered laser light towards the sample. The laser light then passes through an achromatic lens (f = 40.0 mm, NA = 0.30), which focuses the light on a spot with a diameter of approximately 10 µm. The sample is positioned in the focal plane. The same lens collimates the PL light emitted from the excited spot on the sample. After the light passed through the dichroic long-pass mirror and an additional long-pass filter a second achromatic lens (f = 40.0 mm, NA = 0.30) couples it into a multimode optical fiber which guides the light to a compact spectrometer (LR1 from ASEQ instruments). This spectrometer covers a wavelength range from 200 to 1200 nm. Its resolution is approximately 2 nm with a dynamic range of 40 dB.

Fig. 1
figure 1

Scheme of the PL setup. The blue and red arrows indicate the propagation direction of laser and PL light, respectively

Full size image

The 18 investigated samples are listed in Table 1. Half of the samples are plates made out of plastic materials obtained from various industrial suppliers. The other half of the samples are non-plastic materials, which were—except for the sand sample—collected from the marine environment. The samples include common cuttlefish bone (Sepia officinalis), coralline red algae (Lithophyllum racemus), sea snail shells (Monodonta turbinata and Neverita josephina), sea urchin skeleton (Echinocardium cordatum), sea grass leaf (Posidonia oceanica), Mediterranean black mussel shell (Mytilus galloprovincialis), and a piece of wood. For the sand sample, a commercial fine-grained (0.1–0.5 mm) quartz sand (Knauf Bauprodukte GmbH) was used to avoid contamination with plastics or organic materials. The sand was put into a glass test tube, which did not exhibit any detectable PL.

Table 1 List of examined samples with short names (abbreviations), names, minimum, average and maximum peak wavelength of the observed PL spectra
Full size table

For each sample, eight to ten PL spectra were acquired following the same procedure: a sample was first mounted on a three-axis translation stage and then positioned into the focal plane of the sample lens. The excitation power and the integration time of the detector were first set to the minimum value and then both increased until PL signal was detected. The spectra were acquired from different parts of the sample by moving the sample in the focal plane. Then the laser light was blocked and the background spectrum was recorded, which was later subtracted from the PL signal to avoid artifacts in the signal. For many samples, the signal was noisy, because of the low signal intensity. Therefore, after subtracting the background spectrum, the spectra were smoothed using a Savitzky–Golay filter function. This filter function takes two arguments: frame size and polynomial order. A frame size of approximately 10 nm and the polynomial of the second order were chosen since they resulted in the best compromise between noise filtering and preserving distinct spectral features. For an easy comparison, the smoothed spectra were normalized to unity. The smoothed and normalized spectra corresponding to the same sample were averaged and the standard deviation for every wavelength was calculated. The presented results are the smoothed and normalized average spectra with a shaded area representing the standard deviation.

3 Results and discussion

Figure 2 shows the PL spectra of the 18 examined samples with part (a) showing the spectra of the plastic materials, and part (b) the spectra of non-plastic materials.

Fig. 2
figure 2

Smoothed and normalized spectra of the detected PL light from a different plastic samples and from b different samples of non-plastic origin. The shaded area around each line indicates the standard deviation. The vertical lines indicate the peak wavelength for individual samples

Full size image

The spectra show a short wavelength cut-off at 425 nm due to the inserted long-pass filters. For all plastic materials, a rather broad emission around 450 nm can be observed. In the PL spectra of PP, PE, PC, PMMA, PET, and PVDF one observes a distinct peak at 462 ± 2 nm. In the spectra of PC and PMMA, an additional peak at approximately 434 ± 2 nm can be noticed. Such features are known from Raman spectroscopy. Therefore, Raman shifts with respect to the measured excitation laser wavelength 405 nm were calculated. The 434 ± 2 nm and 462 ± 2 nm peaks correspond to a Raman shift of 1650 ± 100 cm−1 and 3050 ± 100 cm−1, respectively. The first Raman peak is assigned to the C=O stretching mode, which appears in PC, PET, and PMMA [45, 46], while the latter is attributed to C–H or CH2 modes [47] which cannot be distinguished from one another due to the resolution of our setup. Therefore, we conclude that the origin of the two observed narrow peaks is Raman scattering. In the PC spectrum, another peak can be noticed around 660 nm. The origin of this peak is not known yet. However, it could correspond to phosphorescence exhibited by the sample. Further analysis to reveal the origin of this peak should be carried out but is out of scope of this paper.

A straightforward approach to distinguish between plastic and non-plastic materials from their PL spectra is comparing the wavelength of the maximum intensity peak. These peak wavelengths are listed in Table 1. The comparison shows that the spectra of plastic samples typically peak at shorter wavelengths than the spectra of non-plastic samples. Yet, the peak wavelengths for some plastic materials are not too different from the peak wavelengths of non-plastic materials. In the case of PC and S. officinalis, they even coincide. Consequently, a distinction between plastic and non-plastic materials cannot be made based on the peak wavelength alone. To overcome this problem and to facilitate classification of the plastic type, we represent the data with a simple approach using more spectral information than just the peak wavelength.

The idea of this approach is to define two variables, which can be plotted one against another, in such a way that the data form well separated clusters belonging to different material types. This way the information of the broad spectra is represented in a single two-dimensional plot. We defined the two parameters (\(P_{1}\) and \(P_{2}\)) as follows:

$$P_{1} = \frac{{I_{{\lambda_{1} }} }}{{I_{{\lambda_{0} }} }},$$
$$P_{2} = \frac{{I_{{\lambda_{2} }} }}{{I_{{\lambda_{0} }} }},$$

where \(I_{{\lambda_{0} }}\), \(I_{{\lambda_{1} }}\) and \(I_{{\lambda_{2} }}\) are the PL intensities at wavelengths \(\lambda_{0}\), \(\lambda_{1}\) and \(\lambda_{2}\), respectively. The wavelengths \(\lambda_{0}\), \(\lambda_{1}\) and \(\lambda_{2}\) were selected as 427 nm, 462 nm and 552 nm, respectively, so that a clear separation between plastic and non-plastic samples was observed as shown in Fig. 3.

Fig. 3
figure 3

\(P_{2}\) as a function of \(P_{1}\). The squares correspond to plastic materials and the stars to the non-plastic materials. Each symbol represents one individual measurement

Full size image

Using this approach, a clear separation between plastic and non-plastic samples was possible. Accordingly, the problem of not being able to distinguish S. officinalis form PC based on the peak wavelength could be solved using this approach. The majority of plastic samples correspond to small values of parameter P2 (smaller than 2.5). Only PP exceeds this value; however, its P1 value is greater than 10.5. No non-plastic materials correspond to values of P1 greater than 10.5 and at the same time P2 smaller than 4. Furthermore, clear clusters of PP (10.5 < P1 < 12.5; 2 < P2 < 3.5), PMMA (10.5 < P1 < 14; 0.5 < P2 < 2), PE (7.5 < P1 < 10.5; 1 < P2 < 2), PET (5.8 < P1 < 6.5; 1.3 < P2 < 1.6), and PS (5 < P1 < 6; 1.8 < P2 < 2.2) were formed. However, points corresponding to PA6 and PA66 (6.4 < P1 < 7.5; 1.7 < P2 < 2.2) resulted in a single cluster similar to PC and PVDF (4.0 < P1 < 5.8; 0.9 < P2 < 1.5), which would make classification of these plastic types using this approach difficult. On the one hand, PA6 and PA66 have a similar chemical structure; therefore, formation of a single cluster might be expected. On the other hand, the chemical structure of PC and PVDF is quite different; therefore, formation of a single cluster was unexpected. This indicates that using this approach, an exact identification of the plastic type was not possible. Some overlapping of clusters can also be observed for non-plastic materials, for which, however, exact sample classification is not of priority. Nevertheless, clear differentiation between plastics and non-plastics was possible.

4 Conclusion and outlook

In the present study it was shown that the PL of different plastic and non-plastic macroscopic samples can be observed. Not only was it possible to distinguish between plastic and non-plastic materials using a simple clustering approach, but we could also distinguish between different kinds of plastics up to certain extent using suitable parameters. However, overlapping of some clusters hampers clear discrimination between some plastic types. Nevertheless, considering that the excitation laser light could be focused onto a small spot with radius of few micrometers, microscopic particles could also be examined. Hence, the study presented here is a promising starting point for the analysis of microplastics using PL spectroscopy and for the development of a low-cost monitoring device in combination with a reliable separation method like the one described by Imhof et al. [27].

For future research, different sources of excitation and their combination should be investigated for providing more spectral information and, therefore, facilitate material discrimination. For example, using a laser with shorter wavelength will most likely lead to broader and more intense luminescence spectra. Another option is performing time-resolved PL, which would provide the information about temporal PL profile. However, this approach would require more expensive equipment. Additionally, a more complex representation of the data could improve the distinction between different material types. This could be done by introducing further parameters, i.e. one or more degrees of freedom to our clustering approach. Future studies will focus on this and on further more sophisticated spectral analysis and classification methods (e.g. principal component analysis, machine learning). Additionally, microplastic particles of different sizes and shapes should be analyzed to determine the resolution limit of the presented method. Furthermore, we intend to study plastic particles that are dyed, aged or incrusted by organogenic materials or POPs. These modifications of the pure plastics will likely influence the PL spectra and make identification of microplastics troublesome. Yet, other spectroscopic methods are affected by these modifications as well. Hence, our findings motivate further development of PL spectroscopy technique for microplastic detection and a large systematic and comparative study on the impact of these effects on PL, Raman and FTIR spectroscopy. Also the measurement speed of these methods should be compared.

Our long-term vision is the development of an automated photoluminescence setup for the detection and analysis of microplastics.