Analysis of trade-offs of post-sorting plastic packaging

Waste sampling

Waste samples were collected in April (PMD) and June (PoSo) 2023 at a Dutch facility located in the northern Netherlands. The facility serves municipalities in the region and processes household packaging waste collected under two parallel collection systems.

Source-separated PMD, comprising plastic packaging, metals and drinking cartons collected separately from residual waste at the household level, is received from municipalities with separate collection implemented, encompassing both urban and rural areas. This PMD waste is subsequently sorted in a mechanical sorting plant. This collection-sorting configuration is implemented in many EU member states under different terminologies (for example, PMD, Leichtverpackung (LVP)), including Austria, Estonia, Germany, Poland, Portugal, Slovakia and Slovenia.

In parallel, mixed residual household waste is collected in a single stream from municipalities without separate PMD collection, including densely populated urban areas, and is subsequently post-sorted to recover recyclable materials and organic fractions. This approach is representative of countries and regions that rely on mixed waste collection combined with downstream material recovery.

Separately collected PMD waste (PMD) and post-sorted PMD fractions (PoSo) recovered from residual household waste were processed at different times but on the same sorting line that includes a drum screen, magnet, eddy current, wind shifter, near-infrared sorters and manual quality control.

Although the precise municipal origin of the individual waste streams cannot be fully resolved, the sampled materials are representative of contemporary European collection and sorting systems that rely on either household-level source separation or mixed waste collection, followed mechanical sorting in an MRF into dedicated recyclable streams (for example, ferrous metals, HDPE rigids, LDPE films).

Bale samples were collected randomly in 60-l waste bags in triplicate by bale and collection system directly from the conveyor belts in the manual quality-control cabin, where operators visually inspect the sorted fractions and remove contaminants to ensure product quality, before baling (PE rigids, PP rigids, mixed plastic). LDPE samples were collected at the plant’s own recycling centre (shredded, before pre-wash bunker), and before baling for PMD and PoSo, respectively (Extended Data Table 1). Target items were selected based on ref. ²⁸, which was adjusted to our observations of the material on the conveyor belt in the manual quality-control cabin. For each target sample, for example, a shampoo bottle, a minimum of ten items were collected, that is, ten shampoo bottles of the same brand, sampled both from PMD as well as from PoSo. Target items were later separated into individual components, for example, an HDPE bottle into bottle, closure and label, and mixed to a composite sample per item component (n = 53; Extended Data Table 2). For PoSo, target items also included ‘non-packaging plastic items’ that potentially appear more frequently or exclusively in this collection scheme, for example, toys, textiles and household plastic items, with the aim to assess whether such non-packaging plastic items would cause additional contamination in recycling. Non-packaging plastic items were observed on both the PP rigid and the mixed plastic belts during post-sorting sampling. At the mixed plastic belt, these were not specifically targeted but included within the general grab samples. In contrast, at the PP rigid belt, non-packaging items were deliberately collected to ensure their representation in the analysis, as their occurrence was less frequent. Non-packaging plastic items were grouped into different categories, for example, ‘PP others’, toys, blisters, and combined to composite samples (n = 5, for example, toys, PP others) or analysed as individual samples (n = 9, for example, textiles) (Extended Data Table 2).

All samples were placed in plastic bags, labelled and stored at 5 °C until further analysis. Items were processed and stabilized within the next 7–10 days.

Waste sample preparation

The sample size of bale samples had to be reduced from 3 × 60 l to approximately 20 l using the quartering method. Target samples were divided into equal fractions. One fraction remained unwashed and the other one fraction was prepared for the measurement of residue content and washing. Non-packaging plastic items were checked for any harmful contents, for example, metal pieces or batteries, and, if present, these were removed. All samples were reduced to flakes by shredding (Shini SH Granulator with sieve size 8 mm). After each cycle, the shredder was cleaned to minimize contamination across samples. Target samples were shredded after washing and drying, as described in the following steps.

Quantification of LAMD

LAMD was determined by quantifying the weight of each sample before and after washing and subsequent drying²⁸. Washing was performed with an industrial-style washing machine at 60 °C with tap water. Depending on the grade of contamination, washing cycles ranged from one to three cycles. Items that are hard to clean owing to their design, for example, bottles, were cut open before washing. For heavily contaminated items, a hand wash with cold water and a scrub was used to remove visible dirt and food residue before the washing cycle. After washing, all samples were dried at atmospheric pressure at 60 °C until constant weight was achieved (typically around 1 day), and weighing was repeated for dried samples. For target items, separable packaging subcomponents such as closures and labels were removed, and weighing was repeated for each washed and dried items’ main and subcomponents. For each target item, mean values are reported (n = 5). Calculations to obtain the average residue content per waste stream are detailed in Extended Data Table 3 and Supplementary Information section 1.2.

Quantification of polymer composition

The polymer composition was determined using attenuated total reflection-Fourier transform infrared (ATR-FTIR) spectroscopy (Nicolet iS20 FTIR Spectrometer). ATR-FTIR settings were programmed to 32 scans with a resolution of 4 cm⁻¹ per analysis. Wavenumber region ranked from 400 cm⁻¹ to 4,000 cm⁻¹.

From washed and shredded bale samples (flakes), 100 flakes were randomly selected. The sample was first homogenized by hand to minimize segregation, then divided into four piles following the quartering method, from which flakes were randomly picked for analysis. Each flake was analysed on both sides, sorted into component-specific containers and weighed. The overall composition was calculated from the average weight of the identified polymers and polymer combinations relative to the total flake weight. Later, the composition was corrected based on the measured LAMD of the sorted bale. Samples were analysed in triplicate (3 × 100 flakes), and the average is reported. On the basis of these values, bale quality, that is, percentage of target polymer in sorted bale, was calculated (Extended Data Table 3). Values are reported as sum of mono and multi-materials, for example, PE mono and PE multi-material.

For the mixed PoSo bale, quantification of polymer composition included an initial visual inspection after sample collection and before further sample preparation. The visual inspection included manual sorting of mixed PoSo sample into polymer fractions based on plastic resin identification codes (PET, HDPE, PVC, LDPE, PP, polystyrene, others), if the codes were visible. For films, this was largely not possible, and films were sorted into LDPE (shopping bags, waste bags) and mixed polyolefins (mixed flexible foils, for example, vegetable wrapping, cheese packaging, shrink wrap). For heavily degraded plastics, a fraction named ‘other plastics’ was added. Films, MPO and other plastics could later be identified based on above described ATR-FTIR method. In addition, non-polymer packaging and other material fractions were included, namely, textiles, paper, WEEE, metals, rubber, toys and residue. Residue included organics and sanitary products. Each manual sorted fraction was weighted on a digital scale with accuracy of ±0.01 kg. After manual inspection and weighing, non-packaging plastic items and other materials were separated from the polymer fractions. Polymer fractions were combined and sample preparation followed the described methods for sorted bales. Non-packaging plastic items were kept separate and analysed as described. For quantification of the polymer composition of the mixed PoSo bale, the results from ATR-FTIR and visual inspection were later combined. The weight ratio of polymer to non-polymer fractions was determined during visual inspection. The ratio was used to rescale the ATR-FTIR polymer results to account for non-packaging plastic items in the overall sorted bale composition. No dedicated hand-sorting was performed on PMD-derived mixed plastic samples; however, no identifiable non-packaging plastic items (for example, textiles, WEEE, toys) were observed during visual inspection of the conveyor belt and the bale sample, although minor contamination (for example, organics or small non-target fragments) cannot be excluded.

Polymer composition of target items including individual item components was analysed (Supplementary Information section 1.3) but excluded from bale composition.

Quantification of elemental composition

For the elemental composition, sample preparation required further size reduction from flakes to powder. Of each shredded sample, approximately 10 g of flakes were randomly selected by the quartering method. The selected flakes were further homogenized by a Fritsch Pulverisette 19 using a sieve cassette with 4 mm square perforation. Cryogenic grinding was applied for further size reduction and homogenization of the samples using a ball mill from Anton Parr (BM500) with liquid nitrogen for cooling. Lastly, the powder was sieved through a 425-μm sieve. The elemental composition of unwashed and washed bale samples, and target items was quantified. If target items consisted of different components, then each component was prepared and analysed separately, and the elemental composition of the main component, for example, PE bottle, and total target item, for example, PE bottle, PP closure, paper label, is reported. The calculation method of total target items is detailed in Supplementary Information section 1.4.

Data normality was assessed with the Shapiro–Wilk test. On the basis of the results, either a two-sided Student’s t-test or Wilcoxon signed-rank test was performed, with P < 0.05 considered significant.

CHN/O analysis

CHN/O analyses were performed using a Flash EA2000 elemental analyser (Interscience) equipped with a thermal conductivity detector using a method described by ref. ⁵⁵. Samples were analysed at least in triplicate. Nitrogen and oxygen levels are included in the set of quality indicators outlined in Extended Data Table 3.

Metal concentration

Metals (aluminium (Al), arsenic (As), calcium (Ca), cadmium (Cd), cobalt (Co), copper (Cu), iron (Fe), kalium (K), lithium (Li), natrium (Na), nickel (Ni), lead (Pb), antimony (Sb), selenium (Se), titanium (Ti), vanadium (V)) were analysed using induced coupled plasma optical emission spectroscopy based on the method described in ref. ³⁹. We used 0.2 ± 0.01 g of sample (powder) for digestion, and complemented the CERTIPUR multi-element standard solution IV (100 mg l⁻¹ in 10% HNO₃) with single-element standard solutions for As, Sb, Se, Ti and V (100 mg l⁻¹ in 10% HNO₃). A comprehensive list of analysed metals, limit of detection, limit of quantification, viewing mode and emissions lines can be found in Supplementary Table 1. Each sample was analysed in triplicate. TMC was adapted from ref. ¹⁴, and is calculated based on the sum of concentrations of measured metals (Extended Data Table 3).

Halogen concentration

Halogen content (chlorine, bromide, fluoride) was determined by combustion ion chromatography using a Metrohm IC Eco equipped with a 863 Compact IC Autosampler and an ICS-3000 conductivity detector operated at 250-μl sample loop volume, using 10.0 mM carbonate/5 mM bicarbonate eluent at 1.0 ml min⁻¹. Before analysis, samples (powder) were pelletized into pellets of 0.02 ± 0.01 g, placed on the carrier glass including an ashless filter paper with 50 μl of 50 m% ammonium nitrate to initiate combustion, and placed into combustion vessels including 8 ml of aqueous solution of 1 mM NaHCO₃ and 8 mM Na₂CO₃ to collect the vapours. Before the oxygen combustion, combustion vessels were put under pressure (20 bar) that builds to a maximum of 65 bar and 225 °C within 10 min during the combustion process of the Anton Paar Multiwave 5000. The resulting solution was diluted with double-distilled water to 25 ml, filtered using a syringe filter (CHROMAFIL, regenerated cellulose membrane of 0.45-μm pore size), and stored at 4 °C until combustion ion chromatography analysis. All samples were analysed in triplicate. Chlorine content is included in the set of quality indicators outlined in Extended Data Table 3.

VOC analysis of unwashed samples

For VOC analysis, sample preparation included sample selection and drying of wet samples for 48 h with a desiccator using CaCO₃ as a drying medium. For analysis, 1.0 ± 0.01 g of unwashed and shredded sample was transferred into a glass vial, sealed and stored in a fridge (5 °C) until analysis. All samples were analysed in triplicate. Aqueous washing was not evaluated as a separate VOC removal step, as previous studies^{56,57,58,59,60,61} and industrial observations (Supplementary Fig. 2) show that washing alone does not consistently reduce VOC levels in polyolefin packaging and may not explain collection-scheme-dependent differences (Supplementary Information section 3.3.3). The VOC analysis was performed using headspace-solid-phase microextraction-gas chromatography–mass spectrometry with a method described in ref. ⁶². Only compounds tentatively identified as VOCs, with a sufficiently high match rating from the National Institute of Standards and Technology mass spectral library, are reported.

We report VOC diversity, that is, total count of identified VOCs, and VOC area count. The VOC area count represents a semi-quantitative measure of the relative abundance of VOCs, derived from the integrated gas chromatography–mass spectrometry signal area across detected VOC peaks (Extended Data Table 3).

Quality indicators

In this study, feedstock quality refers to the measurable characteristics of sorted plastic waste streams that influence their suitability for downstream recycling operations. Feedstock quality is described using eight indicators: purity, carbon content, chlorine, nitrogen, oxygen, TMC, LAMD and VOCs (Extended Data Table 3). Throughout the paper, we distinguish between feedstock quality, which is analytically measured, and recyclate quality, which refers to the anticipated performance of recycled materials based on these feedstock characteristics.

For purity and carbon content, high values are preferred. These indicators are also used to compare and discuss the suitability of sorted bales for the downstream recycling process. For mechanical recycling, a standard mechanical recycling operation is considered, for example, size reduction, purification, washing and extrusion. Feedstocks with high purity, low LAMD and low VOCs are preferred^11,23. For chemical recycling, the discussion is limited to pyrolysis as pyrolysis is a relatively mature technique and, in light of the recently introduced EU regulation 2025/40 targets⁴³, is regarded by the industry as the most likely recycling method to deliver food-grade application recyclates²³. Pyrolysis feedstocks have strict requirements on purity, carbon, nitrogen, oxygen and chlorine content^47,48, while LAMD can be of importance owing to additional and unpredictable contaminants, and VOCs are of less importance. Given the extensive chemical diversity of plastics waste⁶³, the indicators were selected to serve as pragmatic yet robust proxies, enabling assessment of feedstock quality in recycling practice.

Recycling feedstock modelling

We developed a cluster-based material flow analysis^{4,5,54,64,65,66,67} to quantify the availability of recycling feedstock for the EU27+3 and the USA under increasing levels of post-sorting residual waste. The baseline year was 2022 for the EU27+3 and 2018–2022 for the USA, reflecting the data availability of key parameters. Owing to the short lifespan of plastic packaging, annual placed-on-the-market volumes were assumed to equal waste generation in the same year. Countries and states were grouped into four clusters per region using k-means clustering based on reported plastic packaging recycling rates (ref. ¹⁹ and EUROSTAT⁶⁸), access to collection systems (USA, ref. ¹⁹), and proximity to policy targets (EU, European Environment Agency reports), representing system performance archetypes rather than individual jurisdictions (Fig. 4a,b, Supplementary Information section 1.5 and Supplementary Table 6). For each cluster, plastic packaging flows were modelled from generation through separate collection, residual waste and sorting, with cluster-specific collection rates and sorting yields (Supplementary Table 7). EU27+3 clusters were further differentiated into household, and commercial and industrial streams.

Scenario analysis examined the effect of increasing the routing of residual waste shares (10–75%) to post-sorting facilities, without assuming technological improvements in sorting performance. The primary outcome was the quantity of sorted plastic packaging exiting MRFs, used as a proxy for the availability of recycling feedstock. Recycling rates were calculated only ex post for model validation (Supplementary Information section 1.5.6). Parameter uncertainty was addressed by Monte Carlo simulation (10,000 runs) using triangular input distributions. The model was designed to assess relative scenario-driven changes rather than predict exact national recycling rates.

A separate order-of-magnitude CAPEX assessment was conducted to estimate additional sorting infrastructure requirements (Supplementary Information section 1.5.7). Results for the 75% scenario are reported using two complementary indicators: (1) total CAPEX of required post-sorting infrastructure (MRFs) and (2) CAPEX per percentage point increase in recycling feedstock availability. A global sensitivity analysis for (1) and the calculation framework for (2) are provided in Supplementary Information sections 1.5.7–1.5.9 and 3.5.5, and Supplementary Table 10.

For the USA, CAPEX figures were converted using an exchange rate of US$1.13 per €1 based on historical average exchange rates. The cost assessment is intentionally limited to CAPEX for post-sorting infrastructure. It does not aim to evaluate economic feasibility, profitability or market dynamics, which would require additional assumptions beyond the scope of this study.

All simulations were implemented in Analytica 6.4 (Lumina Decision Systems). Key parameters and assumptions are summarized in Supplementary Tables 7–9. Median model values for the baseline and calculation steps are provided in Extended Data Table 6. Detailed method descriptions can be found in Supplementary Information section 1.5; model limitations are discussed in Supplementary Information section 3.5.6.