Particle counts vs. masses in environmental samples
The derived numbers and masses of the harmonized polymer types detected in individual samples are shown in Fig. 1 and in ESM 2.pdf Table S6. MP were identified in all investigated samples using Py-GC/MS and FTIR. The concentrations determined via Py-GC/MS ranged from 6 to 2525 μg m−3 for treated waste water, 4.2–5.5 μg m−3 for surface water samples, and 8–144 μg kg−1 for marine sediments. For these samples, the FTIR results range from 39 to 37223 N m−3 for treated waste water, 8–20 N m−3 marine water, and 143–1151 N kg−1 for marine sediments. Both methods found the same trends in MP contamination, with highest MP concentrations found at Oldenburg1708VF of the analyzed waste water samples and HE430_23S for sediments, respectively (see Fig. 1). These similar trends in particle and mass concentrations indicate a good overall comparability of the determined results. In the following, the results obtained by both analytical approaches will be discussed for the individual sample types. Additionally, mass calculations based on FTIR imaging data sets will be compared with those masses determined by Py-GC/MS.
Treated waste water samples
The treated waste water (TWW) samples indicate a relatively high level of contamination (Fig. 1 left panel). FTIR imaging or Py-GC/MS analysis resulted in similar trends for five of seven samples. Two samples (Oldenburg1308NF and Oldenburg1708NF) showed differences either in absence or presence of MP. While the overall trend in MP abundances is similar, the polymer composition deviated. Here, FTIR indicated a particularly high presence of the PMMA/PUR group, and Py-GC/MS detected higher shares of PE and PVC. While the PP shares were comparable, the relative PET and PS contents varied between both methods. Conversion of FTIR into mass data resulted in an overall predominance of polyolefins and moreover for PMMA/PUR for Oldenburg1708VF. This mass calculation resulted in masses up to seven times higher (Oldenburg1708VF) compared with those determined via Py-GC/MS. Furthermore, the estimated masses reflect a high share of PP while PVC and PS were underrepresented or even missing on a relative scale. Regarding the procedural blank, only low numbers and small sizes (< 50 μm) of six different MP types were detected with FTIR (see ESM 2.pdf Table S6). Here, the mass of individual polymers was too small in most cases to be even detected by Py-GC/MS, and traces of PVC were quantifiable.
Marine sediment samples
In sediments, intermediate MP contamination levels were found (see Fig. 1 middle panel). The general trend revealed in particle abundances of MP by FTIR imaging was reflected in MP mass concentrations analyzed via Py-GC/MS with the highest quantity at HE430_23S and the lowest at HE430_20S. The determined polymer composition is less variable for Py-GC/MS with a predominance of PE in all samples. PVC was detected in all samples, PMMA in two, while PP and PS are found in HE430_23S only. MP composition detected by FTIR showed the presence of PVC, PE, PMMA/PUR, PP, PES(T), and PA in all samples. In two of them (23S and 5S), PVC and PE are particularly abundant while PE is missing in HE430_20S.
Conversion of FTIR particles into masses led to a dominance of PE and PVC at least for the highest contaminated sediment (He430_23S) and He430_5S, while He430_20S contained PET and PVC. The overall MP mass range of the calculated and the measured data is approximately comparable, even though the calculated masses for HE430_5S are very low.
Marine surface waters
At low MP contamination levels, high variations in relative abundances and polymer compositions are observed between the two analytical methods. Here, various polymer types are detected via FTIR imaging while Py-GC/MS is restricted to three or one cluster (Fig. 1, right column). In contrast, conversion of FTIR particle counts into masses reduces the polymer types to almost one prominent type, PE, and traces of others.
Reasons for different relative abundances
Measured particle abundances and mass trends for distinct polymers often differ, as expected, in the presented sample set. This is underlined by the calculated masses from FTIR imaging data. Here, abundance trends derived from the FTIR particle numbers often do not follow the calculated masses (see Fig. 1). At this point, the particle size and shape used for mass calculation becomes highly relevant. To highlight this general issue, the assigned FTIR polymer types in numbers and the resulting calculated masses are opposed as a function of their respective size class. For the TWW samples, only those containing larger numbers of particles > 200 μm are shown in Fig. 2. All other TWW samples are shown in ESM 2.pdf Fig. S4 and for detailed results of all samples ESM 3.tar.
In all cases, calculated masses were mainly driven by particles > 100 μm, which also led to the observed PE-PP-ratio inversion, e.g., for sample Oldenburg1308VF between particle and mass-related data. The high PP masses were caused mainly by a few large-sized PP particles, while the major part of PE particles is assigned to smaller sizes. The fact that masses complementarily determined by Py-GC/MS were at least one order of magnitude below the calculated ones indicates that the particle volume assumed of these PP particles led to an overestimation, which will be discussed later. Oldenburg1708VF also contains PUR/PMMA particles > 200 μm, which give correspondingly high masses.
For the samples Oldenburg1308VF and Holdorf1708, large particle counts but small particle sizes of PMMA/PUR and PVC were reflected in a low calculated mass equivalent.
A similar trend was observed for the marine sediment samples (see Fig. 3). Again, mass calculations were mainly influenced by the presence of particles > 100 μm, here in particular from the polymer types PE, PET, and PVC. Interestingly, with lower overall particle abundances, the measured (Py-GC/MS) and calculated MP masses fell in the same mass range.
For the marine water samples (see Fig. 4), the influence of particle size on mass estimation was even stronger. Again, the mass calculation changes the polymer composition remarkably compared with the FTIR particle abundance results. The diversity of polymer types for FTIR is mainly driven by particles with sizes < 50 μm, while the calculated masses are dominated by larger PE particles (Fig. 4) in a similar order of magnitude as the PE share determined by Py-GC/MS. In contrast, for HE430_7P, only very few particles of PE, PET, PVC, and PA < 50 μm, and one PA particle < 100 μm were detected. The latter represent the calculated mass exclusively. In contrast, Py-GC/MS measurement detected and quantified PVC only.
Figures 2, 3, and 4 point out an additional aspect regarding the FTIR analysis data and the qualitative polymer compositions resulting from these data, as the compositions can vary considerably in the respective size classes. Accordingly, the lower measurement limit of the instrument should not be underestimated when considering the overall relative polymer compositions based on particle counts.
Limit of quantification
Overall, most of the polymer diversity represented in the FTIR data is related to particles < 75 μm and their respective high abundances, but is almost lost after mass calculation due to their minor mass impact, as already pointed out in recent literature [46, 48]. Direct mass measurements by Py-GC/MS do not show this diversity which is due to targeted measurements further discussed later but also due to the limit of quantification (LOQ) (at the time the measurements were performed). Working with solid standards, the LOQ for Py-GC/MS was set by the available balance and ranges dependent on the polymer type between 0.7 and 1 μg absolute. Reduced to one single particle, this weight is roughly equivalent to a size between 50 and 200 μm. This has a large influence if the polymer compositions are investigated as the smaller particles have a stronger impact on the polymer composition for FTIR as it shows a high variability depending on the investigated size classes compared with Py-GC/MS. In contrast, the limit of detection (LOD equivalent to S/N > 3) is mostly far below 1 μg, again polymer-dependent and equivalent to much lesser particle sizes (cf. ESM 2.pdf Table S5). In ESM 2.pdf Table S7, an overview is given which polymers were quantified and detected via Py-GC/MS for the individual polymer types.
Too little masses are most plausibly the reason why PET, prominent in particle counts, does hardly appear on a mass scale. The exception (HE430_20S), where mass-relevant particles are present, might be traced back to the additional point that FTIR combines a larger number of PEST types in the database while Py-GC/MS just targeted PET in these measurements. The lack of PA detected in none of the samples by Py-GC/MS but frequently present in FTIR measurement and mass calculation (HE430_7P) might have a similar reason. While FTIR detected PA as a group, Py-GC/MS addressed only PA6. As both data sets can be reassessed in the future with extended polymer data/reference sets for better data harmonization, this finding cannot be finally valuated. The calculated PA amount in case of the sample HE430_7P falls below the LOQ for PA6 in the Py-GC/MS analysis, if this PA would be PA6. However, sample volume equivalent to the related polymer particle mass on the Anodisc filter is not sufficient to use the potential of Py-GC/MS for an informative MP composition in a reasonable way, here.
PVC, PS, and PP
Independent of sample origin, some differences in polymer composition were observed that have to be discussed on a more general level.
Disregarding the respective method, PVC was detected in almost all samples, but a systematic link between determined particle size and measured masses was, except for the sediment samples (see Figs. 1 and 3), often missing. PVC represent a consistent mass share in the Py-GC/MS measurement of the TWW samples (see Fig. 1), but even if the particle counts show the presence of PVC, these consistently small particles (<< 50 μm) have no impact on the mass calculation, and even few particles of 150–200 μm at sample Holdorf1708 did not count relative to the polyolefins. An outstanding example was Holdorf1308. Here, high masses of PVC were determined by Py-GC/MS but no PVC particle was detected via FTIR. The FTIR raw data of this particular sample (see ESM 1.xlsx) showed a high abundance of plant fibers over the full particle size range and elevated coal particles albeit < 75 μm were detected. Both might be potential precursors for benzene, the PVC indicator compound that is fairly weak regarding its polymer specificity. This potential interference needs further examination in this particular case, but can be almost excluded for the other TWW samples. This general discrepancy between FTIR and Py-GC/MS measurements needs further investigation in future studies.
Four of the analyzed samples (three WWTP and one sediment) show notably PS shares with Py-GC/MS detection, while they show low and small particle numbers in FTIR. As FTIR should be able to detect the related PS particles anyway, it is much more plausible that the PS detected by Py-GC/MS on the basis of its highly specific styrene trimer indicator product is derived from a PS copolymer, i.e., a styrene acrylate commonly used for paints and consumer products and possibly included in the PUR/PMMA/paint cluster of the FTIR data. An inconsistency pointing in the same direction was observed for sample Oldenburg1708NF where FTIR detected a highly mass-relevant PP particle while Py-GC/MS detected PP at trace levels only. This discrepancy was further investigated (see ESM 2.pdf paragraph S3, Fig. S5 and Fig. S6 for details) and the result indicates that this particle may either be a copolymer of PE and PP or a highly branched polymer with PE backbone. These particles as well as the PS masses stand exemplarily for actual limitations of the applied databases or method. The by now extensive FTIR reference database enables a critical re-investigation of the respective particle spectra. Even though Py-GC/MS data can be reinvestigated easily as well, the pyrograms and respective indicator ion(s) of the further suspected polymer types have to be known previously for a targeted search. This was not the case here as the number of polymers was restricted to nine representatives. Since the data were measured with an internal standard, a retrospective analysis might be performed at given times.
In case of selected samples, a further plausible explanation for the observed differences could be given. The presence of fine red material (Oldenburg1308NF, see ESM 2.pdf Fig. S7a) or a fine opaque material (Holdorf1308, see ESM 2.pdf Fig. S7b), respectively, endured the applied sample treatment and hampered the FTIR measurements. This might have led to additional minor findings by FTIR by covering MP particles.
Target of the measurement
Finally, two key aspects have to be kept in mind when FTIR and Py-GC/MS polymer data are compared:
As a result of highly developed spectral libraries and optimal particle separation out from each other, spectroscopically generated MP data represent often a broad suggestion of highly diverse polymer types that must be critically reviewed either manually or automatically. Accordingly, polymers are clustered to an acceptable extent to achieve an arguable set of polymers that enable further sample comparison. Clustering arguments base on spectral similarities in some cases includes different polymers in one cluster due to almost overlapping spectroscopic signals.
Even though extended pyrogram libraries exist for more than 1000 polymers and over 500 additives (F-Search, FrontierLab), they rely on single (particle) measurements. Py-GC/MS of environmental samples is a bulk measurement of the whole sample. The generated pyrograms sum up all generated indicator pyrolysis products disregarding their original precursor polymer. Ideally, any potential interference of natural organic materials should be excluded by preceded, adequate sample clean up. The resulting signal of a so-called polymer-specific indicator ion condenses all polymers or copolymers related to one respective polymer backbone. For quantification, this is finally expressed as the calibrated pure base polymer disregarding the original polymer type.
While FTIR detects the overall chemical absorption pattern directly related to functional groups inside the polymer after IR excitation, Py-GC/MS detects selected decomposition products of involved polymer chains as a result of pyrolysis. On macromolecular level, this can be of high importance for copolymers. Blends could be masked for FTIR by one compound with increasing mass ratio. Py-GC/MS detect decomposition products of both polymer types. This is consistent to the findings of Hermabessiere et al.  using Raman spectroscopy for one tested particle and Käppler et al.  for several particles and fibers using μATR-FTIR. Due to the presence of varnishes, it is most likely that these are not solely based on, e.g., acrylates, but may also contain crosslinking agents based on styrene or having chlororubber components. Both types are widely used as metal protection paints (styrene based) or for swimming pools and roofs (chlororubber). Similar results were also found by Hendrickson et al.  using ATR-FTIR on isolated particles for PE and PVC due to the chlorination of PE, which could not always be addressed by solely one technique.
LOD in relation to particle size or masses
In spectroscopy, the LOD is depending on the targeted size and instrumentation. The direct comparison of the determined polymer composition is therefore particle size-dependent as particle sizes typically follow a power law distribution. As already discussed, very small polymer particles are detected and quantified by Py-GC/MS measurements once they exceed a critical mass that defines the LOD and LOQ, respectively. Consequently, the contribution to mass increases with particle size. In addition, the mass is dependent on the shape of the particle (e.g., sphere versus fragment). The determination of this critical mass in relation to particle size and shape for the individual polymer types is one of the next challenges in the harmonization of FTIR and Py-GCMS methods with regard to MP analysis.
In consequence, both measurement principles discussed here have a different target and result in either particle numbers or masses. The quality of polymer detection is dependent on the kind of generated signal, its related quality, and the potentials of its interpretation.
Furthermore, our results indicate that the current mass calculation of Simon et al.  is currently limited if larger particles of complex shape are present and, thus, should be considered as an estimation.
Harmonization Py-GC/MS and mass calculation via FTIR imaging
For a discussion of relative overall composition patterns of a particular sample which is based on particle sizes, a clear hypothesis regarding the weighting of individual size fractions is needed.
While the MP trends of FTIR and Py-GC/MS are in good agreement overall, the derived mass calculations from FTIR data do not agree with the results of Py-GC/MS, as the masses determined were overestimated up to 6 times (Oldenburg1708VF) or underestimated by a factor of 10 (Holdorf1308) excluding the OldenburgNF samples due to the different targets and measurement principles mentioned above. Nevertheless, with decreasing contamination level, the accuracy of the estimation improved.
Still, at the current level, the polymer compositions are not comparable due to the different technical backgrounds. Here, the estimated mass concentrations are mainly underestimated; only for HE430_20S, a higher mass was estimated with different polymer composition compared with Py-GC/MS. Especially, a few larger particles caused severe differences as shown in sample Oldenburg1308VF, where large masses of PP were calculated but much lower when measured via Py-GC/MS. Based on Figs. 2, 3, and 4, it can be concluded that the accuracy of the estimate decreases with increasing particle size, as the underlying eclipse approach may overestimate the mass of the different particle shapes present. Here, the mass calculation is overestimating the particle volume and therefore the mass. Therefore, as suggested by Simon et al. , the results should be treated with care, since in particular large particles have a strong influence on the result.
To overcome these limitations, a modified mass calculation was performed that combines various particle shapes and sizes present into a reference particle. For this purpose, the average particle length and diameter of each polymer type (see ESM 2.pdf Paragraph S4) was calculated and used as a reference particle in terms of mass. The area of the individual particles was divided by the reference area calculating the reference particles represented and multiplied by the reference particle mass. Compared with the mass calculation of Simon et al. , this weighted approach reduces the difference compared with the results of Py-GC/MS (see Fig. 5).
For the sample Holdorf1708, the calculated mass is similar to the one derived by Py-GC/MS while the masses for Oldenburg1308VF and Oldenburg1708VF are calculated lower. In all cases, the mass of PP is overestimated while the mass of PE is underestimated. Still, only a factor of 3 in difference is found indicating a better agreement possible using such a weight on the particle data.
Our results indicate that a calibration between Py-GC/MS and mass estimation may be possible, but this must be addressed by a specially designed investigation, which is currently hampered by the lack of suitable reference material and was therefore not part of this work.