Satellite-based maximum entropy modelling for identifying potential soil microplastics hotspots
The pervasive presence of microplastics in terrestrial ecosystems has emerged as a pressing environmental concern. Recent studies have identified soil as a major sink for microplastics contamination, potentially surpassing oceanic levels by factors ranging from 4 to 23-fold. The small size of microplastics and the complexity of soil matrices as sink substrates pose challenges for quantifying soil pollution. As a result, current analytical methods are limited in efficiency, making large-scale environmental assessments unfeasible. The vertical incorporation of microplastics into soil, along with the challenges of recognizing microscopic objects in satellite images, restricts the practicality of using remote sensing for direct large-scale environmental assessments. Hence, a more comprehensive approach is necessary to tackle these challenges. One potential solution involves utilizing satellite imagery combined with a maximum entropy model. By integrating locations where microplastic presence has been confirmed and extracted from soil samples, the maximum entropy model can establish a connection between satellite-derived environmental predictors and the presence of microplastics in soil. The aim of this research was to assess the practicality and viability of employing this approach in a real-world setting.
To test our approach, we designed a case study covering wider administrative area of the City of Osijek, Croatia. For training data, we utilized 31 sampled locations where soil microplastics have been confirmed through previous research, along with environmental variables primarily derived through signal enhancement of Sentinel-based imagery. After literature review, a preliminary list of 31 environmental predictor variables was generated, covering various facets of microplastics input to the soil and their dispersion in the environment. These were tested for variance inflation factor (VIF) and spatial autocorrelation to identify statistically significant variables for model calibration. To relate environmental variables to microplastics presence, we leveraged maximum entropy model. The best-performing model underwent additional testing using various permutation tests to evaluate its robustness. We identified 4491 different sets of three environmental variables eligible for further examination. We employed each combination to train maximum entropy models using 5-fold cross-validation to identify the most robust model. Additional testing included jackknife cross-validation to identify and remove outlier samples.
The best performing model, with an AUC under the ROC of 0.863, was the one trained using combination of environmental predictors including land cover (CLC+ Backbone raster product), soil moisture derived from Sentinel-1 imagery, and catchment areas determined through hydrological analysis of the digital elevation model. The output prediction map clearly delineates areas that highly likely represent pollution hotspots. This research demonstrates the feasibility of utilizing satellite imagery, in conjunction with topological analysis and maximum entropy models, to conduct large-scale environmental assessment and accurately pinpoint hotspots of soil microplastics contamination. This approach could significantly aid future stakeholders since the EU has taken proactive steps as of 2018 to tackle soil microplastics pollution, by implementing regulations, action plans, and initiatives to prevent plastic pellet loss. Furthermore, the European Commission has incorporated impact assessments into its decision-making process regarding microplastics. Advanced environmental monitoring techniques offer potential in tracking progress and quantifying effectiveness of forthcoming measures.
