Microplastic detection in arable soil using a 3D Laser Scanning Confocal Microscope coupled with a Machine-Learning Algorithm

Abstract

In Europe, about 0.71 million tonnes of agricultural plastic were intentionally used in 2019. Most widely used were plastic films (about 75%), which are dominated by light density polyethylene (LDPE). Especially LDPE plastic films for mulching covers in direct contact arable soil to increase temperature and reduce evaporation. Thereby, microplastic is detached from the mulch film via mechanical and environmental weathering. Another microplastic pathway in arable soil is the application of sewage sludge. Depending on land use, a 4 to 23 times higher microplastic contamination in soils than in the sea is estimated. Obviously, microplastic input to soils is critically high, but an accurate quantification is still lacking. This is partly caused by challenges in detection and analysis of microplastic in soils. First, it is challenging to extract microplastic from a matrix of organic and inorganic particles of similar size. Second, the well-established spectroscopic methods (e.g., Raman and FTIR) for detecting microplastics in water samples are sensitive to soil organic matter, and they are very time-consuming. Eliminating very stable organic particles (e.g., lignin) from soil samples without affecting the microplastic to be measured is another challenge. Hence, a robust analytical approach to detect microplastic in soils is needed. In this context, we developed a methodological approach that is based on a high-throughput (25 g soil sample) density separation scheme for measurements in a 3D Laser Scanning Confocal Microscope (Keyence VK-X1000, Japan) and subsequently using a Machine-Learning algorithm to classify and analyze microplastic in soil samples. Our aim is to develop a method for a fast screening of microplastic particle numbers in soils while avoiding the use of harmful substances (e.g., ZnCl2) or prolonged organic carbon destruction. For method development, we contaminate a standard soil (LUFA type 2.1 - sand: 86.6% sand, 9.7% silt, 3.7% clay, 0.58% organic carbon; and LUFA type 2.2 - loamy sand: 72.6% sand, 16.8% silt, 10.7% clay, 1.72% organic carbon) with different concentrations of transparent LDPE microplastic (< 700 µm), LDPE microplastic originating from black mulch film (< 400 µm) and microplastic originating from Bio-degraded black mulch film (< 250 µm). For density separation, three non-toxic, easy to handle mediums were compared for the best microplastic output: distilled water (ρ = 1.0 g/cm3), 26% NaCl solution (ρ = 1.2 g/cm3), and 41% CaCl2 solution (ρ = 1.4 g/cm3). The separated microplastic plus organic particles and some small mineral particles were scanned using a 3D Laser Scanning Confocal Microscope. For each sample, the 3D Laser Scanning Confocal Microscope generates three different main outputs: color, laser intensity, and surface characteristics. Based on these data outputs, a Machine-Learning algorithm distinguishes between the mineral, organic, and microplastic particles. It was found that color changes of microplastics due to soil contact challenge the classification but can be compensated by surface characteristics that become an essential input parameter for the detection. The presented methodological approach provides an accurate and high-throughput microplastic assessment in soil systems, which is critically needed to understand the boundaries of sustainable plastic application in agriculture.